Datasets:

TomTBT

/

pmc_open_access_xml

like 4

[ "<title>1. PHOTODYNAMIC THERAPY: BACKGROUND</title>", "<p>The use of light in the treatment of disease has been known for many centuries and can be traced\nback over 4000 years to the ancient Egyptians [##REF##3064295##1##]. The Egyptian people used a \ncombination of the orally ingested Amni Majus plant and sunlight to\nsuccessfully manage vitiligo: a skin disorder of unknown cause. The\nactive ingredient of this plant (psoralen, ##FIG##0##Figure 1##) is now successfully employed\nin the worldwide treatment of psoriasis [##REF##3064295##1##–##REF##25048431##4##].</p>", "<p>\nPhotodynamic therapy (PDT) is a treatment involving light and\na chemical substance (a photosensitiser), used in conjunction with molecular\noxygen to elicit cell death. More explicitly, photodynamic\ntherapy is a selective treatment modality for the local destruction of diseased\ncells and tissue. The selectivity is based on the ability of the\nphotosensitiser to preferentially accumulate in the diseased tissue and\nefficiently generate singlet oxygen or other highly reactive species such as\nradicals, which induce target cell death.</p>", "<p>The principle of photodynamic therapy is based on a multi-stage process (##FIG##1##Figure 2##). The first of\nthese stages (##FIG##1##Figure 2(a)##) sees the administration of a photosensitiser with negligible\ndark toxicity, either systemically or topically, in the absence of light. When the optimum ratio of photosensitiser in diseased <italic>versus</italic> healthy tissue is achieved, the photosensitiser is \n(##FIG##1##Figure 2(c)##) activated by exposure to a carefully regulated dose of light which is shone directly onto the diseased tissue for a specified\nlength of time. The light dose is regulated in order to allow a sufficient amount\nof energy to be delivered to activate the photosensitiser, but at the same time\nthe dose should be small enough to minimise damage inflicted on neighbouring\nhealthy tissue. It is the activated form\nof the photosensitiser which evokes a toxic response in the tissue, resulting\nin cell death. The success of photodynamic therapy lies in the prolonged\naccumulation of photosensitiser in diseased tissue, relative to the more rapid\nclearance from normal tissue cells.</p>", "<p>Photodynamic therapy is commonly practiced in the treatment of a number of cancers, including those present in the head and neck, the lungs, bladder, and\nparticular skin cancers [##REF##10529768##5##–##UREF##4##16##]. \nIt has also been successfully used in the\ntreatment of non-cancerous conditions such as age-related macular degeneration\n(AMD), psoriasis, atherosclerosis, and has shown some efficacy in anti-viral treatments\nincluding herpes [##UREF##0##2##, ##UREF##1##3##, ##REF##10529768##5##–##REF##9637138##7##, ##REF##15267226##9##, ##REF##25048432##17##–##UREF##6##20##].</p>", "<p>Photodynamic therapy carries advantages for both the patient and the physician: \nthe need for delicate surgery and lengthy recuperation periods is minimised, \nalong with minimal formation of scar tissue and disfigurement. However, photodynamic \ntherapy is not without its drawbacks: a major limitation is the associated general\nphotosensitisation of skin tissue.</p>" ]

[]

[]

[]

[]

[ "<p>Recommended by Michael J. Cook</p>", "<p>Photodynamic therapy (PDT) is a treatment modality that has been used in the successful treatment of a number of diseases and disorders, including age-related macular degeneration (AMD), psoriasis, and certain cancers. PDT uses a combination of a selectively localised light-sensitive drug (known as a photosensitiser) and light of an appropriate wavelength. The light-activated form of the drug reacts with molecular oxygen to produce reactive oxygen species (ROS) and radicals; in a biological environment these toxic species can interact with cellular constituents causing biochemical disruption to the cell. If the homeostasis of the cell is altered significantly then the cell enters the process of cell death. The first photosensitiser to gain regulatory approval for clinical PDT was Photofrin. Unfortunately, Photofrin has a number of associated disadvantages, particularly pro-longed patient photosensitivity. To try and overcome these disadvantages second and third generation photosensitisers have been developed and investigated. This Review highlights the key photosensitisers investigated, with particular attention paid to the metallated and non-metallated cyclic tetrapyrrolic derivatives that have been studied <italic>in vitro</italic> and <italic>in vivo</italic>; those which have entered clinical trials; and those that are currently in use in the clinic for PDT.</p>" ]

[ "<title>2. HISTORY OF PHOTODYNAMIC THERAPY</title>", "<p>Reports of contemporary photodynamic therapy came first in the investigations led by\nFinsen in the late nineteenth century [##UREF##3##8##]. \nFinsen successfully demonstrated phototherapy by employing heat-filtered light from a carbon-arc lamp (the “Finsen lamp”) in the treatment of a tubercular condition of the skin known as \n<italic>lupus vulgaris</italic>, for which he won the Nobel Prize in Physiology or\nMedicine in 1903 [##UREF##3##8##]. But it was not until the early twentieth \ncentury that reports of photodynamic therapy for the treatment of cancer patients (with\nsolid tumours) were made by von Tappeiner's group in Munich \n[##UREF##0##2##, ##REF##25048431##4##, ##UREF##2##6##, ##REF##15267226##9##]. \nIn 1913 another German scientist, Meyer-Betz, described the major stumbling block of \nphotodynamic therapy. After injecting himself with haematoporphyrin (Hp, a photosensitiser),\nhe swiftly experienced a general skin sensitivity upon exposure to sunlight—a\nproblem still persistent with many of todays' photosensitisers [##UREF##0##2##, ##UREF##1##3##, ##REF##9637138##7##, ##UREF##5##18##].</p>", "<p>Further studies, investigating the accumulation of haematoporphyrin and the purified\nhaematoporphyrin derivative (HpD) in tumours, culminated in the late 1980s with\nthe photosensitiser Photofrin (##FIG##2##Figure 3##). A photosensitiser which,\nafter further purification, was first given approval in 1993 by the Canadian\nhealth agency for use against bladder cancer and later in Japan, USA and parts\nof Europe for use against certain cancers of the oesophagus and non-small cell\nlung cancer [##UREF##1##3##–##REF##15267226##9##, ##REF##25048432##17##, ##UREF##5##18##, ##UREF##7##21##, ##UREF##8##22##].</p>", "<p>Photofrin was\nfar from ideal and carried with it the disadvantages of prolonged patient\nphotosensitivity and a weak long-wavelength absorption (630 nm) [##UREF##2##6##, ##REF##9637138##7##, ##UREF##7##21##]. This\nled to the development of improved (second generation) photosensitisers,\nincluding Verteporfin (a benzoporphyrin derivative, also known as Visudyne) and\nmore recently, third generation photosensitisers based around targeting\nstrategies, such as antibody-directed photosensitisers [##REF##25048431##4##, ##REF##10529768##5##, ##REF##9637138##7##, ##UREF##5##18##, ##REF##9498864##19##, ##REF##15812551##23##–##REF##17343613##25##].</p>", "<title>3. CYCLIC TETRAPYRROLIC CHROMOPHORES\nAND PHOTOSENSITISERS</title>", "<p>Cyclic tetrapyrrolic molecules are good examples of fluorophores (see Glossary) and photosensitisers. \nPhotosensitisers are molecules, which, when excited by light energy, can utilise the energy \nto induce photochemical reactions to produce lethal toxic agents. In a cellular environment, \nthese agents (reactive oxygen species (ROS) and radicals) ultimately result in cell \ndeath and tissue destruction (##FIG##3##Figure 4##) [##REF##10529768##5##–##REF##15267226##9##].\nPhotosensitisers are absorbed into cells all over the body and alone are harmless, that is, in the absence of light, and usually oxygen they have no effect on healthy or abnormal tissue. Ideally, they should be retained by diseased tissue, \nparticularly tumours, for longer periods of time in comparison to healthy tissue; thus it is\nimportant to carefully time light exposure and ensure that activation only occurs when \nthe ratio of photosensitiser is greater in diseased tissue than in healthy tissue, thereby \nminimising unwanted damage to surrounding non-cancerous cells \n[##UREF##1##3##, ##REF##9498864##19##].</p>", "<p>Photosensitisers also have alternative applications. They have been employed in the \nsterilisation of blood plasma and water in order to remove blood-borne viruses and microbes \nand have been considered for agricultural uses, including herbicides and insecticides \n[##REF##10529768##5##, ##REF##15267226##9##, ##UREF##9##26##–##UREF##10##28##].</p>", "<title>4. PHOTOCHEMISTRY:\nPHOTOCHEMICAL PROCESSES</title>", "<p>Only when a photosensitiser is in its excited state (³Psen*) \ncan it interact with molecular oxygen (³O₂) and produce radicals\nand activated oxygen species (ROS), crucial to the Type II mechanism which is\nthought to predominate in PDT (see below). These species include singlet oxygen\n(¹O₂), hydroxyl radicals \n(^•OH), and superoxide (O₂\n⁻) ions and can interact with\ncellular components including unsaturated lipids; amino acid residues; and\nnucleic acids. If sufficient oxidative damage ensues, this will result in\ntarget-cell death (only within the immediate area of light illumination).</p>", "<title>5. PHOTOCHEMICAL MECHANISMS</title>", "<p>When a chromophore, such as a cyclic tetrapyrrolic\nmolecule, absorbs a photon of electromagnetic radiation in the form of light\nenergy, an electron is promoted into a higher-energy molecular orbital, elevating\nthe chromophore from the ground state (S₀) into a short-lived,\nelectronically excited state (S_{<italic>n</italic>}) composed of a number of\nvibrational sub-levels (S_{<italic>n</italic>}′). The excited chromophore can lose\nenergy by rapidly decaying through these sub-levels <italic>via</italic> internal conversion (IC) to populate the first excited\nsinglet state (S₁), before quickly relaxing back to the ground state (##FIG##5##Figure 5##).</p>", "<p>The decay from the excited singlet state (S₁)\nto the ground state (S₀) is <italic> via</italic>\n<bold>fluorescence (S₁ → S₀</bold>).\nSinglet state lifetimes of excited fluorophores are very short (<italic>τ</italic>\n_fl. = 10⁻⁹–10⁻⁶ seconds) since transitions\nbetween the same spin states (S → S or T → T)\nconserve the spin multiplicity of the electron and, according to the Spin Selection\nRules, are therefore considered “allowed” transitions [##UREF##1##3##, ##UREF##2##6##, ##UREF##3##8##]. Alternatively, an excited singlet state\nelectron (S₁) can undergo spin inversion and populate the\nlower-energy first excited triplet state (T₁) <italic>via </italic> intersystem crossing (ISC); a\nspin-forbidden process, since the spin of the electron is no longer conserved\n[##UREF##11##29##–##UREF##15##34##]. The excited electron can then undergo a second spin-forbidden\ninversion and depopulate the excited triplet state (T₁) by decaying\nback to the ground state (S₀) <italic>via </italic>\n<bold>phosphorescence (T₁ → S₀)</bold> [##UREF##11##29##–##UREF##15##34##]. Owing\nto the spin-forbidden triplet to singlet transition, the lifetime of\nphosphorescence (<italic>τ</italic>\n_{<italic>P</italic>} = 10⁻³ − 1 second) is considerably longer than that of\nfluorescence.</p>", "<title>6. PHOTOSENSITISERS AND PHOTOCHEMISTRY</title>", "<p>Tetrapyrrolic photosensitisers in the excited singlet state (¹Psen*, S_&gt;0)\nare relatively efficient at undergoing intersystem crossing and can\nconsequently have a high triplet-state quantum yield, ##FIG##6##Box 2## (Φ_T 0.62 (tetraphenylporphyrin (TPP), methanol)), 0.83 (etiopurpurin, benzene), 0.71\n(tetrasulphonated TPP, D₂O), and 0.47 (tetrasulphonated zinc\nphthalocyanine, methanol)) [##UREF##3##8##, ##REF##10907495##35##, ##UREF##16##36##]. The longer lifetime of this species is\nsufficient to allow the excited triplet state photosensitiser to interact with\nsurrounding bio-molecules, including cell membrane constituents [##REF##10529768##5##, ##REF##25048432##17##].</p>", "<title>7. PHOTOCHEMICAL REACTIONS</title>", "<p>Excited triplet-state\nphotosensitisers can react in two ways defined as Type-I and Type-II processes.\nType-I processes can involve the excited singlet or triplet photosensitiser (¹Psen*,\nS₁; ³Psen*, T₁), however due to the\nshort lifetime of the excited singlet state, the photosensitiser can only react\nif it is intimately associated with a substrate, in both cases the interaction\nis with readily oxidisable or reducable substrates. Type-II processes involve\nthe direct interaction of the excited triplet photosensitiser (³Psen*,\nT₁) with molecular oxygen (³O₂, ³Σ_g)\n[##REF##10529768##5##–##UREF##3##8##, ##REF##25048432##17##, ##UREF##5##18##, ##UREF##17##37##].</p>", "<p>Type-I processes can be divided into two further mechanisms;\nType I(i) and Type I(ii). The first of these mechanisms (i) involves\nthe transfer of an electron (oxidation) from a substrate molecule to the\nexcited state photosensitiser (Psen*), generating a photosensitiser\nradical anion (Psen^•−) and a substrate radical cation\n(Subs^•+). The majority of the radicals produced\nfrom Type-I(i) reactions react instantaneously with oxygen, generating a\ncomplex mixture of oxygen intermediates. For example, the photosensitiser\nradical anion can react instantaneously with molecular oxygen (³O₂)\nto generate a superoxide radical anion (O₂\n^•−), which can go on to produce the\nhighly reactive hydroxyl radical (OH^•, ##FIG##7##Figure 6##), initiating a cascade of cytotoxic free radicals; this process is common in\nthe oxidative damage of fatty acids and other lipids [##REF##25048432##17##, ##UREF##5##18##]. Some of the more\ncommon Type-I(i) reactions are shown in ##FIG##7##Figure 6##.</p>", "<p>The second Type-I process (ii) involves the transfer \nof a hydrogen atom (reduction) to the excited state photosensitiser (Psen*).\nThis generates free radicals capable of rapidly reacting with molecular oxygen\nand creating a complex mixture of reactive oxygen intermediates, including\nreactive peroxides (##FIG##8##Figure 7##). Once again, this can trigger a torrent of cytotoxic\nevents, culminating in cell damage and death.</p>", "<p>On the other hand, Type-II processes involve the\ndirect interaction of the excited triplet state photosensitiser (³Psen*)\nwith ground state molecular oxygen (³O₂, ³Σ_g,\n##FIG##9##Figure 8##); a spin allowed transition—the excited state\nphotosensitiser and ground state molecular oxygen are of the same spin state\n(T, ##FIG##5##Figure 5##).</p>", "<p>When the excited photosensitiser collides with a\nmolecule of molecular oxygen, a process of triplet-triplet annihilation takes \nplace (³Psen*<italic> →</italic>\n¹Psen and ³O₂\n<italic> →</italic>\n¹O₂).\nThis inverts the spin of one of molecular oxygens (³O₂) outermost\nantibonding electrons, generating two forms of singlet oxygen (¹Δ_g and ¹Σ_g, ##FIG##11##Figure 9##), while\nsimultaneously depopulating the photosensitiser's excited triplet state (T₁ → S₀, ##FIG##5##Figure 5##). The higher-energy\nsinglet oxygen state (¹Σ_g, 157kJ mol⁻¹ &gt; ³Σ_g) is very short-lived (¹Σ_g ≤ 0.33 milliseconds (methanol),\nundetectable in H₂O/D₂O) and rapidly relaxes to the lower-energy\nexcited state (¹Δ_g, 94kJ mol⁻¹ &gt; ³Σ_g) [##UREF##16##36##]. It is, therefore, this lower-energy form of\nsinglet oxygen (¹Δ_g) which is implicated in\ncell injury and cell death [##UREF##18##38##].</p>", "<p>The highly-reactive oxygen species (¹O₂)\nproduced <italic>via</italic> the Type-II\nprocess act near to their site of generation and within a radius of action of\napproximately 20 nm, with a typical lifetime of approximately 40 nanoseconds in\nbiological systems [##UREF##0##2##, ##REF##9637138##7##, ##REF##25048432##17##]. However, it has recently been suggested that (over\na 6 microsecond period) singlet oxygen can diffuse up to approximately 300 nm <italic>in vivo </italic> [##REF##17323398##39##–##UREF##19##41##]. Singlet\noxygen can theoretically only interact with proximal molecules and structures\nwithin this radius [##REF##25048432##17##]. ROS are known to initiate a large number of\nreactions with biomolecules, including amino acid residues in proteins, such as\ntryptophan; unsaturated lipids like cholesterol and nucleic acid bases,\nparticularly guanosine and guanine derivatives, ##FIG##10##Box 3##, with the latter base more\nsusceptible to ROS [##UREF##0##2##, ##REF##10529768##5##, ##UREF##3##8##, ##REF##25048432##17##, ##UREF##16##36##, ##REF##25048553##42##–##UREF##22##45##]. These\ninteractions cause damage and potential destruction to cellular membranes and\nenzyme deactivation, culminating in cell death [##UREF##3##8##].</p>", "<p>It is highly probable that in the presence of\nmolecular oxygen, and as a direct result of the photoirradiation of the\nphotosensitiser molecule, both Type-I and II pathways play a pivotal role in\ndisrupting cellular mechanisms and cellular structure. Nevertheless, there is\nconsiderable evidence to suggest that\nthe Type-II photo-oxygenation process predominates in the induction of cell\ndamage, a consequence of the interaction between the irradiated photosensitiser\nand molecular oxygen [##UREF##0##2##, ##REF##10529768##5##, ##UREF##3##8##, ##UREF##5##18##, ##UREF##20##43##, ##REF##6234604##46##]. It has been suggested,\nhowever, that cells <italic>in vivo </italic> are\npartially protected against the effects of photodynamic therapy by the presence\nof singlet oxygen scavengers (such as histidine) and that certain skin cells\nare somewhat resistant to photodynamic therapy in the absence of molecular\noxygen; further supporting the proposal that the Type-II process is at the\nheart of photoinitiated cell death [##REF##25048432##17##, ##UREF##20##43##, ##REF##8206681##47##, ##REF##3088562##48##].</p>", "<p>The efficiency of Type-II processes is dependent upon the triplet state lifetime <italic>τ</italic>\n_T (see Glossary under luminescence\nlife time)\nand the triplet quantum yield (Φ_T) of the photosensitiser.\nBoth of these parameters have been implicated in the effectiveness of a\nphotosensitiser in phototherapeutic medicine; further supporting the distinction\nbetween Type-I and Type-II mechanisms. However, it is worthy to note that the\nsuccess of a photosensitiser is not exclusively dependent upon a Type-II\nprocess taking place. There are a number of photosensitisers whose excited\ntriplet lifetimes are too short to permit a Type-II process to occur. For\nexample, the copper metallated octaethylbenzochlorin photosensitiser (##FIG##12##Figure 10##)\nhas a triplet state lifetime of less than 20 nanoseconds and is still deemed to\nbe an efficient photodynamic agent [##REF##8506398##13##, ##UREF##20##43##].</p>", "<title>8. PHOTOSENSITISERS—IDEAL PHOTOSENSITISERS</title>", "<p>Although a number of different photosensitising compounds,\nsuch as methylene blue (see Glossary), rose bengal,\nand acridine (##FIG##13##Figure 11##), are known to be efficient singlet oxygen generators (and\ntherefore potential photodynamic therapy agents), a large number of\nphotosensitisers are cyclic tetrapyrroles or structural derivatives of this \nchromophore; in particular porphyrin, chlorin, bacteriochlorin, expanded porphyrin, and\nphthalocyanine (PCs)\nderivatives (##FIG##14##Figure 12##). This is possibly because cyclic tetrapyrrolic\nderivatives have an inherent similarity to the naturally occurring porphyrins\npresent in living matter—consequently they have little or no toxicity\nin the absence of light [##UREF##0##2##, ##REF##10529768##5##, ##REF##25048432##17##, ##UREF##5##18##, ##UREF##16##36##, ##UREF##21##44##, ##REF##11377804##49##].</p>", "<p>Porphyrins are a group of naturally occurring and \nintensely coloured compounds, whose name is drawn from the Greek word <italic>porphura</italic>, the Greek word\nfor purple [##UREF##23##50##–##UREF##25##52##]. These molecules are known to be involved in a number\nof biologically important roles, including oxygen transport and photosynthesis,\nand have applications in a number of fields, ranging from fluorescence imaging\nto medicine [##UREF##0##2##, ##REF##10529768##5##, ##REF##25048432##17##, ##REF##25048553##42##]. Porphyrins are classified as tetrapyrrolic\nmolecules, with the heart of the skeleton a heterocyclic macrocycle, known as a\nporphine. The fundamental porphine frame consists of four pyrrolic sub-units\nlinked on opposing sides (<italic>α</italic>-positions, numbered 1, 4, 6, 9, 11, 14,\n16, and 19, ##FIG##15##Figure 13##) through four methine (CH) bridges (5, 10, 15, and 20),\nknown as the <italic>meso</italic>-carbon\natoms/positions (##FIG##15##Figure 13##). The resulting conjugated planar macrocycle may be\nsubstituted at the <italic>meso</italic>- and/or <italic>β</italic>-positions\n(2, 3, 7, 8, 12, 13, 17, and 18): if the <italic>meso-\n</italic> and <italic>β</italic>-hydrogens are substituted with non-hydrogen\natoms or groups, the resulting compounds are known as porphyrins.</p>", "<p>The inner two\nprotons of a free-base porphyrin can be removed by strong bases such as\nalkoxides, forming a dianionic molecule; conversely, the inner two pyrrolenine nitrogens can\nbe protonated with acids such as trifluoroacetic acid affording a dicationic intermediate\n(##FIG##16##Figure 14##). The tetradentate anionic species can readily form\ncomplexes with most metals.</p>", "<title>9. PORPHYRIN ABSORPTION SPECTROSCOPY</title>", "<p>On account of their highly conjugated skeleton,\nporphyrins have a characteristic ultra-violet visible (UV-VIS) spectrum (##FIG##17##Figure 15##). The spectrum typically consists of an intense, narrow absorption band (<italic>ε</italic> &gt; 200000 l mol⁻¹cm⁻¹) at around 400 nm, known as the\nSoret or B band, followed by four longer wavelength (450–700 nm), weaker absorptions (<italic>ε</italic> &gt; 20000 l mol⁻¹cm⁻¹ (free-base porphyrins)) referred to as the Q bands [##UREF##2##6##, ##REF##25048432##17##, ##UREF##23##50##, ##UREF##26##53##, ##UREF##27##54##].</p>", "<p>The Soret band arises from a <bold>strong</bold> electronic transition from the (porphyrin) ground state to the second excited singlet state (S₀ → S₂, ##FIG##18##Figure 16##); whereas the Q band is a result of a <bold>weak</bold> transition to the first excited singlet state (S₀ → S₁).\nThe dissipation of energy <italic>via</italic>\ninternal conversion (IC) is so rapid that fluorescence is only observed from depopulation of the first excited singlet state to the lower-energy ground state (S₁ → S₀).</p>", "<title>10. SECOND-GENERATION PHOTOSENSITISERS</title>", "<title>10.1. Ideal photosensitiser properties</title>", "<p>The key characteristic of any photodynamic sensitiser is its ability to preferentially accumulate in diseased tissue and, <italic>via</italic> the generation of cytotoxic species, induce a desired biological effect. In\nparticular, a good photodynamic sensitiser should adhere to the following criteria: </p>", "<p>have strong absorption with a high extinction\ncoefficient in the red/near infrared region of the electromagnetic spectrum (600–850 nm)—allows deeper tissue penetration [##REF##10529768##5##–##REF##9637138##7##, ##REF##25048432##17##, ##UREF##16##36##, ##UREF##20##43##],</p>", "<p>be effective generators of singlet\noxygen and other ROS,</p>", "<p>have suitable photophysical characteristics: a high-quantum yield of triplet formation (Φ_T ≥ 0.5); a high singlet oxygen quantum yield (Φ_Δ ≥ 0.5); a relatively long triplet state lifetime (<italic>τ</italic>\n_T, microsecond range); and a high triplet-state energy (≥ 94 KJ mol⁻¹) [##UREF##1##3##, ##UREF##3##8##, ##UREF##5##18##, ##UREF##16##36##, ##UREF##29##56##]. To date the parameters Φ_T = 0.83 and Φ_Δ = 0.65 (haematoporphyrin); Φ_T = 0.83 and Φ_Δ = 0.72 (etiopurpurin); and Φ_T = 0.96 and Φ_Δ = 0.82 (tin etiopurpurin) have been achieved\n[##UREF##0##2##, ##UREF##16##36##], </p>", "<p>have minimum dark toxicity and negligible cytotoxicity in the absence of light,</p>", "<p>exhibit greater retention in diseased/target tissue over healthy tissue, </p>", "<p>present rapid clearance from the body, </p>", "<p>be single, well-characterised\ncompounds, with a known and constant composition, </p>", "<p>have a short and high yielding synthetic route (with easy translation into multi-gram scales/reactions), </p>", "<p>have a simple and stable drug formulation, </p>", "<p>be soluble in biological media,\nallowing direct intravenous administration and transport to the intended target. Failing this, a hydrophilic delivery system should be sought enabling efficient and effective transportation of the photosensitiser to the target site <italic>via</italic> the bloodstream.</p>", "<p>While the major disadvantages associated with the first generation photosensitisers HpD and Photofrin (skin\nsensitivity and weak absorption at 630 nm) have not prevented the treatment of some cancers and other diseases, they have markedly reduced the successful application of these photosensitisers to a wider field of disease. The development of second generation photosensitisers, designed to minimise the drawbacks of the first generation photosensitisers, was key to the development\nof photodynamic therapy. A number of new photosensitisers were therefore developed to overcome these short comings.</p>", "<p>\n<statement id=\"head50\"><title>5-Aminolaevulinic acid</title><p>The 5-Aminolaevulinic acid (ALA) is a prodrug used in\nthe clinic to treat and image a number of superficial cancers and tumours (see Tables\n##TAB##1##2## and ##TAB##2##3##) [##REF##10529768##5##–##REF##15267226##9##, ##REF##15288239##11##, ##REF##25048432##17##, ##UREF##5##18##]. ALA on its own is not a photosensitiser, but a key\nprecursor in the biosynthesis of the naturally occurring porphyrin, haem (##FIG##20##Scheme\n1##).</p><p>Haem is synthesised in every energy-producing cell in the body\nand is a key structural component of haemoglobin, myoglobin, and other haemproteins. The\nimmediate precursor to haem is protoporphyrin IX (PPIX), an effective\nphotosensitiser. Haem itself is not a photosensitiser, due to the coordination\nof a paramagnetic ion (iron; see Glossary; see also diamagnetic species) in the centre of the macrocycle, causing\nsignificant reduction in excited state lifetimes [##REF##10529768##5##–##REF##15267226##9##, ##REF##15288239##11##].</p><p>The haem molecule is synthesised from glycine and\nsuccinyl coenzyme A (succinyl CoA). The rate-limiting step in the biosynthesis\npathway is controlled by a tight (negative) feedback mechanism in which the\nconcentration of haem regulates the production of ALA. However, this controlled\nfeedback can be by-passed by artificially adding excess exogenous ALA to cells.\nThe cells respond by producing PPIX (photosensitiser) at a faster rate than the\nferrochelatase enzyme can convert it to haem [##REF##10529768##5##–##REF##15267226##9##, ##REF##15288239##11##, ##REF##25048432##17##, ##UREF##5##18##].</p><p>ALA, marketed as Levulan (DUSA Pharmaceuticals\nIncorporated, Toronto, Canada), has shown promise in photodynamic\ntherapy (tumours) <italic>via </italic> both\nintravenous and oral administration, as well as through topical administration\nin the treatment of malignant and non-malignant dermatological conditions,\nincluding psoriasis, Bowen's disease, and Hirsutism (Phase II/III\nclinical trials, see Glossary) [##REF##10529768##5##–##REF##15267226##9##, ##REF##15288239##11##, ##UREF##5##18##].</p><p>ALA shows a more rapid accumulation in comparison to other intravenously\nadministered sensitisers [##REF##10529768##5##–##REF##15267226##9##, ##REF##15288239##11##]. Typical peak tumour accumulation levels post-administration for\nPPIX are usually achieved within several hours; compare this with other\n(intravenously administered) photosensitisers which may take up to 96 hours to\nreach peak levels and one of the main advantages of ALA can be clearly seen. ALA\nis also excreted\nmore rapidly from the body (∼24 hours) than other photosensitisers, minimising\npatient photosensitivity [##REF##10529768##5##–##UREF##3##8##, ##REF##15288239##11##].</p><p>In an attempt to overcome the poor bioavailability\nwhen ALA is applied topically, esterified ALA\nderivatives with \nimproved pharmacological properties have been examined [##REF##10529768##5##–##UREF##3##8##, ##REF##15288239##11##]. A methyl ALA ester (Metvix) is now being marketed by Photocure ASA (Oslo, Norway)\nas a potential photosensitiser for basal cell carcinoma and other skin lesions\n[##REF##10529768##5##, ##UREF##2##6##, ##REF##15267226##9##, ##REF##25048432##17##]. Benzyl (Benvix) and hexyl ester (Hexvix) derivatives are also\nregistered by Photocure ASA for the treatment of gastrointestinal cancers and\nfor the diagnosis of bladder cancer [##REF##15267226##9##].</p></statement>\n</p>", "<p>\n<statement id=\"head55\"><title>Verteporfin</title><p>The second generation photosensitiser, benzoporphyrin\nderivative monoacid ring A (BPD-MA, ##FIG##19##Figure 17##) has been developed by QLT Phototherapeutics\n(Vancouver, Canada) under the trade name Visudyne (Verteporfin, for injection)\nand, in collaboration with Ciba Vision Corporation (Duluth, GA, USA), has\nundergone Phase III clinical trials (USA) for the photodynamic\ntreatment of wet age-related macular degeneration (AMD, see Glossary) and cutaneous non-melanoma\nskin cancer [##UREF##1##3##, ##REF##10529768##5##–##REF##9637138##7##, ##REF##15267226##9##, ##REF##3480384##57##–##UREF##31##59##]. Verteporfin is currently marketed by Novartis\nPharmaceuticals Corporation (NJ,\nUSA).</p><p>The chromophore of BPD-MA has a red-shifted and\nintensified long-wavelength absorption maxima at approximately 690 nm. Tissue\npenetration by light at this wavelength is 50% greater than that achieved for\nPhotofrin (<italic>λ</italic>\n_max. = 630 nm) [##REF##10529768##5##, ##UREF##32##60##].</p><p>Verteporfin has further advantages over the first\ngeneration sensitiser Photofrin. It is rapidly absorbed by the tumour (optimal\ntumour-normal tissue ratio 30–150 minutes post-intravenous\ninjection) and is rapidly cleared from the body, minimising patient\nphotosensitivity (1-2 days) [##REF##10529768##5##, ##REF##7585658##61##].</p></statement>\n</p>", "<p>\n<statement id=\"head60\"><title>Purlytin</title><p>Tin etiopurpurin, a chlorin photosensitiser (##FIG##21##Figure 18##), is marketed\nunder the trade name Purlytin by Miravant Medical Technologies (Santa Barbara,\nCalif, USA) [##REF##10529768##5##–##REF##15267226##9##, ##REF##3334995##62##]. Purlytin has also undergone Phase II\nclinical trials (USA) for cutaneous metastatic breast cancer and Kaposi's\nsarcoma in patients with AIDS (acquired immunodeficiency syndrome) [##UREF##1##3##, ##REF##9637138##7##].\nPurlytin has been used successfully to treat the non-malignant conditions\npsoriasis and restenosis [##REF##10529768##5##].</p><p>Chlorins (##FIG##14##Figure 12##) are distinguished from the\nparent porphyrins by a reduced exocyclic double bond. The result of the reduced\nbond is a decrease in the symmetry of the conjugated macrocycle, leading to an\nincreased absorption in the long-wavelength portion of the visible region of\nthe electromagnetic spectrum (650–680 nm). More\ncorrectly, Purlytin is a purpurin; a degradation product of chlorophyll [##UREF##3##8##, ##REF##15267226##9##, ##REF##11749485##12##].</p><p>Purlytin has a tin atom chelated in its central\ncavity which causes a red-shift of approximately 20–30 nm (with respect to\nPhotofrin and non-metallated etiopurpurin, <italic>λ</italic>\n_max.SnEt₂ = 650 nm) [##UREF##2##6##, ##REF##15267226##9##, ##REF##11749485##12##]. Purlytin has been reported to localise in skin and produce a photoreaction\n7–14 days post-administration\n[##UREF##2##6##, ##REF##15267226##9##].</p></statement>\n</p>", "<p>\n<statement id=\"head65\"><title>Foscan</title><p>Tetra(<italic>m</italic>-hydroxyphenyl)chlorin\n(<italic>m</italic>THPC, ##FIG##22##Figure 19##) has been developed\nand entered into clinical trials (USA and Europe) under the trade name Foscan \nby Scotia Pharmaceutics (Guildford, Surrey, UK) and BioLitec Pharma Limited (Dublin, Ireland) [##UREF##1##3##, ##REF##10529768##5##–##REF##15267226##9##, ##REF##15288239##11##, ##UREF##5##18##]. Foscan, also known as Temoporfin,\nhas been evaluated as a phototherapeutic agent against head and neck cancers in\nthese trials [##REF##10529768##5##]. It has also been investigated in clinical trials for\nmalignant and non-malignant diseases, including gastric and pancreatic cancers,\nhyperplasia, field sterilisation after cancer surgery and for the control of\nantibiotic-resistant bacteria, in the USA, Europe, and the Far East [##REF##10529768##5##, ##REF##15267226##9##, ##REF##15288239##11##].</p><p>Foscan has a singlet oxygen quantum yield comparable\nto other chlorin photosensitisers but the low drug and light doses (approximately\n0.1 mg kg⁻¹ and as low as 5 J cm⁻², resp.) required to\nachieve photodynamic responses (equivalent to Photofrin, 2–5 mg kg⁻¹,\n100–200 J cm⁻²; therefore Foscan is approximately 100 times more photoactive than\nPhotofrin), potentially make Foscan one of the most potent second generation\nphotosensitisers currently under investigation [##REF##10529768##5##, ##REF##9637138##7##, ##REF##15267226##9##].</p><p>Unfortunately, Foscan can render patients photosensitive\nfor up to 20 days after initial illumination [##UREF##2##6##, ##REF##10421887##63##, ##UREF##33##64##]. One solution to this\nproblem would be to use lower drug doses.</p></statement>\n</p>", "<p>\n<statement id=\"head70\"><title>Lutex</title><p>Lutetium texaphyrin, marketed under the trade name\nLutex and Lutrin (Pharmacyclics, Calif, USA), is a “texas-sized” porphyrin [##REF##10529768##5##–##REF##15267226##9##, ##UREF##5##18##, ##UREF##34##65##, ##UREF##35##66##].\nTexaphyrins (first synthesised in 1987 by Sessler and his group) are expanded\nporphyrins that have a penta-aza core (##FIG##23##Figure 20##). The result of this macrocyclic\nmodification is a strong absorption in the 730–770 nm region of\nthe electromagnetic spectrum [##REF##15267226##9##, ##REF##11749485##12##]. This region is particularly important\nsince tissue transparency is optimal in this range. As a result, Lutex-based\nPDT can (potentially) be carried out more effectively at greater depths and on\nlarger tumours [##REF##10529768##5##, ##UREF##2##6##].</p><p>Lutex has entered Phase II\nclinical trials (USA) for evaluation against breast cancer and malignant\nmelanomas [##UREF##2##6##, ##REF##9579539##67##].</p><p>A Lutex derivative, Antrin, has also undergone Phase I\nclinical trials (USA) for the prevention of restenosis (see Glossary) of vessels after cardiac\nangioplasty by photoinactivating foam cells that accumulate within arteriolar\nplaques [##UREF##2##6##, ##UREF##36##68##]. A second Lutex derivative, Optrin, is in Phase I trials for AMD\n[##REF##10529768##5##].</p><p>Texaphyrins are being developed further by Pharmacyclics as\nradiosensitisers (Xcytrin, see Glossary) and chemosensitisers (see Glossary) [##REF##10529768##5##]. Xcytrin, a gadolinium texaphyrin (motexafin gadolinium), has been evaluated in Phase III clinical trials against brain metastases and Phase I clinical trials (USA) for primary brain tumours [##REF##10529768##5##].</p></statement>\n</p>", "<p>\n<statement id=\"head75\"><title>ATMPn</title><p> 9-Acetoxy-2,7,12,17-tetrakis-(<italic>β</italic>-methoxyethyl)-porphycene\n(##FIG##24##Figure 21##) has been evaluated by Glaxo Dermatology (GlaxoWellcome, NC, USA)\nand Cytopharm (Calif, USA) as a photodynamic therapy agent for dermatological\napplications against psoriasis vulgaris and superficial non-melanoma skin cancer\n[##REF##10529768##5##, ##UREF##37##69##–##REF##9128760##72##].</p></statement>\n</p>", "<p>\n<statement id=\"head80\"><title>Zinc phthalocyanine CGP55847</title><p>A liposomal formulation of zinc phthalocyanine\n(CGP55847, ##FIG##25##Figure 22##), developed by QLT Phototherapeutics (Vancouver, Canada)\nand sponsored by Ciba Geigy (Novartis, Basel, Switzerland), has undergone clinical\ntrials (Phase I/II, Switzerland) against squamous cell carcinomas of the upper\naerodigestive tract [##REF##10529768##5##, ##UREF##5##18##, ##UREF##40##73##, ##UREF##41##74##]. Phthalocyanines (PCs) (##FIG##14##Figure 12##) are related\nto tetra-aza porphyrins. Instead of four bridging carbon atoms at the <italic>meso-</italic>positions, as for the porphyrins,\nPCs have four nitrogen atoms linking the pyrrolic sub-units together. PCs\nfurther differ from porphyrins through the presence of an extended conjugate\npathway: a benzene ring is fused to the <italic>β</italic>-positions\nof each of the four-pyrrolic sub-units. These benzene rings act to strengthen\nthe absorption of the chromophore at longer wavelengths (with respect to\nporphyrins). The absorption band of PCs is almost two orders of magnitude\nstronger than the highest Q band of haematoporphyrin [##REF##11749485##12##]. These favourable\ncharacteristics, along with the ability to selectively functionalise their\nperipheral structure, make PCs favourable photosensitiser candidates [##REF##12859150##10##, ##REF##12203905##75##–##REF##16906790##78##].</p><p>A sulphonated aluminium PC derivative (Photosense, ##FIG##26##Figure 23##) has also entered clinical trials (Russian Academy of Medical Sciences, and\nthe surgical clinic of the Moscow Medical Academy, Moscow, Russia) against\nskin, breast, and lung malignancies and cancer of the gastrointestinal tract [##REF##10529768##5##, ##UREF##5##18##, ##UREF##43##79##–##UREF##45##81##].\nSulphonation significantly increases PC solubility in polar solvents including\nwater, circumventing the need for alternative delivery vehicles [##REF##15267226##9##, ##REF##11749485##12##, ##UREF##5##18##, ##UREF##46##82##].</p><p>A third PC under investigation is a silicon complex, PC4.\nThis photosensitiser is being examined for the sterilisation of blood\ncomponents at the New York Blood Centre (VI Technologies Incorporated (Vitex),\nMelville, NY, USA), against human colon, breast, and ovarian cancers and\nagainst glioma [##REF##10529768##5##, ##REF##8451285##83##–##REF##15965990##89##].</p><p>A shortcoming of many of the metallo-PCs is their tendency\nto aggregate in aqueous buffer (pH 7.4), resulting in a decrease, or total loss,\nof their photochemical activity. This behaviour can be minimised in the presence\nof detergents [##REF##11749485##12##].</p><p>Metallated cationic porphyrazines (PZ), including\nPdPZ⁺, CuPZ⁺, CdPZ⁺, MgPZ⁺, AlPZ⁺,\nand GaPZ⁺, have been developed and also tested <italic>in vitro </italic> on V-79 (Chinese hamster\nlung fibroblast) cells. Results have suggested these photosensitisers are\ncapable of inducing substantial dark toxicity [##REF##11749485##12##].</p></statement>\n</p>", "<p>\n<statement id=\"head85\"><title>Naphthalocyanines</title><p>Naphthalocyanines (NCs, ##FIG##27##Figure 24##) are an extended PC\nderivative. They have an additional benzene ring attached to each isoindole sub-unit\non the periphery of the PC structure. Subsequently, NCs absorb strongly at even\nlonger wavelengths (approximately 740–780 nm) than PCs\n(670–780 nm), further\nincreasing the depth NC photosensitisers can be effectively used at. This absorption\nin the near infrared region makes NCs good candidates for photodynamic\ntreatment of highly pigmented tumours, including melanomas, which present\nsignificant problems with respect to transmission of visible light.</p><p>However, a number of problems are associated with NC\nphotosensitisers. NCs are generally less stable than their PC relatives: they\nreadily decompose in the presence of light and oxygen; and metallo-NCs, which\nlack axial ligands, have a tendency to form H-aggregates in solution [##REF##11749485##12##, ##UREF##48##90##].\nThese aggregates are photoinactive, thus compromising the photodynamic efficacy\nof NCs [##REF##11749485##12##]. The main investigations into NCs as photodynamic therapy agents\nhave been carried out by Kenney and co-workers, van Lier's group and the Bulgarian\nAcademy of Sciences (Sofia, Bulgaria) (see below).</p></statement>\n</p>", "<p>\n<statement id=\"head90\"><title>Functional groups</title><p>Altering the peripheral functionality of\nporphyrin-type chromophores can also have an effect on photodynamic activity.</p><p>Diamino platinum porphyrins show high anti-tumour activity,\ndemonstrating the combined effect of the cytotoxicity of the platinum complex\nand the photodynamic activity of the porphyrin species [##REF##11749485##12##, ##UREF##49##91##].</p><p>Positively charged PC derivatives have also been\ninvestigated [##REF##11749485##12##, ##UREF##33##64##, ##UREF##42##76##, ##REF##17517508##77##]. Cationic species are believed to selectively\nlocalise in the vital sub-cellular organelle, the mitochondrion. Mitochondria\nare key to the survival of a cell; being the site of oxidative phosphorylation,\nand hence are potentially important PDT targets.</p><p>Zinc and copper cationic derivatives have been\ninvestigated. Although, the positively charged zinc complexed PC was found to\nbe less photodynamically active than its neutral counterpart <italic>in vitro</italic> against V-79 cells [##REF##11749485##12##].</p><p>Water-soluble cationic porphyrins bearing\nnitrophenyl, aminophenyl, hydroxyphenyl, and/or pyridiniumyl functional groups\nexhibit varying cytotoxicity to cancer cells <italic>in vitro </italic>, depending on the nature of the metal ion (Mn, Fe, Zn,\nNi), and on the number and type of functional groups [##REF##11749485##12##, ##REF##17517508##77##, ##UREF##50##92##]. The manganese\npyridiniumyl derivative has shown the highest photodynamic activity, while the\nnickel analogue is photoinactive (##FIG##28##Figure 25##) [##REF##11749485##12##, ##UREF##50##92##].</p><p>Another metallo-porphyrin complex, the iron chelate,\nwas found to be more photoactive (towards HIV and simian immunodeficiency virus\nin MT-4 cells) than the manganese complexes; the zinc derivative was found to\nbe photoinactive [##REF##11749485##12##, ##REF##1384504##93##].</p><p>The hydrophilic sulphonated porphyrins and PCs\n(AlPorphyrin and AlPC) compounds were tested for photodynamic activity [##UREF##51##94##].\nThe disulphonated analogues (with adjacent substituted sulphonated groups, ##FIG##29##Figure 26##) exhibited greater photodynamic activity than their di-(symmetrical), mono-,\ntri- and tetra-sulphonated counterparts; tumour activity increased with\nincreasing degree of sulphonation [##UREF##3##8##, ##REF##16906790##78##].</p></statement>\n</p>", "<title>11. THIRD-GENERATION PHOTOSENSITISERS</title>", "<p>The poor\nsolubility of many photosensitisers in aqueous media, particularly at\nphysiological pH, prevents their intravenous delivery directly into the\nbloodstream. It would be advantageous therefore, if a delivery model could be\nconceived which would allow the transportation of these (otherwise potentially\nuseful) photosensitisers to the site of diseased tissue.</p>", "<p>Work has\nrecently focused on designing systems to effect greater selectivity and\nspecificity on the photosensitiser in order to enhance cellular uptake [##REF##9637138##7##, ##UREF##18##38##]. A number of possible delivery strategies have been suggested, ranging\nfrom the use of oil-in-water (o/w) emulsions to liposomes and nanoparticles as\npotential carrier vehicles [##UREF##1##3##, ##REF##9637138##7##, ##UREF##5##18##, ##UREF##16##36##, ##REF##25048437##95##, ##UREF##52##96##]. There is concern however, that although the use of these systems may increase the therapeutic\neffect observed as a result of photodynamic therapy, the carrier system may\ninadvertently decrease the “observed” singlet oxygen quantum yield (Φ_Δ) of the encapsulated\nphotosensitiser: the singlet oxygen generated by the photosensitiser would have\nto diffuse out of the carrier system; and since it (singlet oxygen) is believed\nto have a narrow radius of action, singlet oxygen may not reach the target and\nelicit its desired effect [##UREF##5##18##]. It may also be possible that, if the size of the\ncarrier is not sufficiently small or that the carrier system does not fully\ndissolve in physiological media, the incidence/exciting light may not be\nappropriately absorbed and light scattering may be significant, thus\ninadvertently reducing the singlet oxygen yield. An alternative delivery method\nwhich would remove this problem is the use of targeting moieties. Typical\ntargeting strategies have included the investigation of photosensitisers\ndirectly attached to biologically active molecules such as antibodies [##REF##15812551##23##–##REF##17343613##25##].\nThese third generation photosensitisers are currently showing promise <italic>(in vitro)\n</italic> against colorectal tumour cells [##REF##16685457##24##].</p>", "<p>\n<statement id=\"head95\"><title>Metallation</title><p>A wide range of metals have been used to form\ncomplexes with photosensitiser macrocycles, with variable photodynamic results. \nA number of the second generation photosensitisers described earlier contain a\nchelated central metal ion. The main metals which have been used are transition\nmetals, although a number of photosensitisers co-ordinated to group 13 (Al,\nAlPcS₄) and group 14 (Si, SiNC, and Sn, SnEt₂) metals\nhave also been synthesised.</p><p>There seems to be no consistent observation as to the\npotential success of metallated photosensitisers. Indeed, a wide range of\nphotosensitisers are metallated, but the metal ion does not confer definite\nphotoactivity on the photosensitiser. Copper (II), cobalt (II), iron (II), and\nzinc (II) complexes of Hp are all photoinactive in contrast to metal-free\nporphyrins [##REF##11749485##12##]. Yet the reverse has been observed for texaphyrin and PC\nphotosensitisers; only the metallo-complexes have demonstrated efficient\nphotosensitisation [##REF##11749485##12##].</p><p>The presence and nature of the central metal ion,\nbound by a number of photosensitisers, strongly influences the photophysical\nproperties of the photosensitiser [##REF##11749485##12##, ##UREF##33##64##, ##REF##17517508##77##]. Chelation of paramagnetic metals\nto a PC chromophore appears to shorten triplet lifetimes (down to nanosecond\nrange), generating variations in the triplet quantum yield and triplet lifetime\nof the photoexcited triplet state of the metallated PC (mPC) [##REF##11749485##12##, ##UREF##33##64##, ##REF##17517508##77##, ##UREF##53##97##].</p><p>Intersystem crossing (ISC) is an important parameter\nof photosensitisers. The triplet quantum yield and lifetime of a\nphotosensitiser are directly related to the efficiency of singlet oxygen\ngeneration; a key component in the success of a photosensitiser [##UREF##53##97##].</p><p>Certain heavy metals are known to enhance ISC.\nGenerally, diamagnetic metals promote ISC and have a long triplet lifetime [##UREF##33##64##, ##REF##17517508##77##, ##UREF##53##97##].\nIn contrast, paramagnetic species deactivate excited states, reducing the\nexcited-state lifetime and preventing photochemical reactions from taking place\n[##UREF##53##97##]. However, there are well-known exceptions to this generalisation,\nincluding copper octaethylbenzochlorin [##REF##8506398##13##].</p><p>For many of the metallated paramagnetic texaphyrin\nspecies, triplet-state lifetimes are down in the nanosecond range [##UREF##53##97##]. These\nresults are also mirrored by metallated PCs. PCs metallated with diamagnetic\nions, such as Zn²⁺, Al³⁺, and Ga³⁺, generally\nyield photosensitisers with desirable quantum yields and lifetimes (Φ_T 0.56, 0.50 and 0.34 and <italic>τ</italic>\n_T 187, 126 and 35 <italic>μ</italic>s,\nresp.) [##REF##11749485##12##, ##UREF##53##97##]. The ZnPC photosensitiser (ZnPcS₄) has a singlet\noxygen quantum yield of 0.70; nearly twice that of most other mPCs (Φ_Δ at least 0.40) [##REF##11749485##12##, ##UREF##5##18##]. Hence,\nthe latter diamagnetic complexes should be strong candidates for PDT.</p><p>Since the heavy metal effect (see Glossary) is known to promote ISC,\ntheoretically, it should be possible to enhance the photophysical properties (Φ_T,\nΦ_Δ, and <italic>τ</italic>\n_T)\nof any photosensitiser <italic>via </italic>\nmetallation. In practice, this is not the case. Only one metallo-porphyrin\nphotosensitiser (copper octaethylbenzochlorin) has shown photodynamic promise,\nthe remaining efficient porphyrin photosensitisers are metal-free [##REF##8506398##13##]. The\nreverse of this behaviour is observed for PCs and texaphyrins; only the\n(diamagnetic) metallated complexes have exhibited potential as photosensitisers\n[##REF##12859150##10##, ##REF##11749485##12##]. The metal-free analogues have shown no promise as photosensitisers [##REF##11749485##12##].</p></statement>\n</p>", "<p>\n<statement id=\"head100\"><title>Expanded metallo-porphyrins</title><p>Expanded porphyrins have a larger central binding\ncavity, increasing the number of potential metals it can accommodate.</p><p>Diamagnetic metallo-texaphyrins have shown encouraging\nphotophysical properties; high triplet quantum yields and efficient generation\nof singlet oxygen [##REF##11749485##12##, ##UREF##33##64##, ##REF##17517508##77##]. In particular, the zinc and cadmium derivatives\nhave shown triplet quantum yields close to unity [##REF##11749485##12##]. In contrast, the\nparamagnetic metallo-texaphyrins, Mn-Tex, Sm-Tex, and Eu-Tex, have undetectable\ntriplet quantum yields. This behaviour is parallel with that observed for the\ncorresponding metallo-porphyrins [##REF##11749485##12##].</p><p>The cadmium-texaphyrin derivative has shown <italic>in vitro </italic> photodynamic activity against\nhuman leukemia cells and Gram positive (<italic>Staphylococcus</italic>)\nand Gram negative (<italic>Escherichia coli</italic>)\nbacteria [##UREF##54##98##–##UREF##57##101##]. Although follow-up studies have been limited with this\nphotosensitiser due to the toxicity of the complexed cadmium ion.</p><p>A zinc-metallated <italic>seco</italic>-porphyrazine (##FIG##30##Figure 27##) has been developed with a high\nquantum singlet oxygen yield (Φ_Δ 0.74) [##UREF##58##102##]. This expanded porphyrin-like photosensitiser has shown the best\nsinglet oxygen photosensitising ability of any of the reported <italic>seco-</italic>porphyrazines. Platinum and\npalladium derivatives have also been synthesised with singlet oxygen quantum\nyields of 0.59 and 0.54, respectively, (##FIG##30##Figure 27##) [##UREF##58##102##].</p></statement>\n</p>", "<p>\n<statement id=\"head105\"><title>Metallochlorins/bacteriochlorins</title><p>The tin (IV) purpurins were found to be more active\nwhen compared with analogous zinc (II) purpurins, when evaluated against human\ncancers [##REF##10529768##5##–##REF##9637138##7##, ##REF##15267226##9##, ##UREF##5##18##, ##UREF##59##103##, ##UREF##60##104##].</p><p>Sulphonated benzochlorin derivatives have\ndemonstrated a reduced phototherapeutic response against murine leukemia L1210\ncells <italic>in vitro</italic> and transplanted\nurothelial cell carcinoma in rats, whereas the tin (IV) metallated\nbenzochlorins exhibited an increased photodynamic effect in the same tumour\nmodel (##FIG##31##Figure 28##) [##REF##8248325##105##].</p><p>The previously mentioned copper octaethylbenzochlorin\n(##FIG##12##Figure 10##) demonstrated an unexpected result. Despite an undetectable triplet\nstate, it appears to be more <italic>photoactive</italic> towards leukemia cells <italic>in vitro </italic>\nand a rat bladder tumour model [##REF##8506398##106##–##REF##8165237##108##]. Suggestions for this unusual effect\nhave pointed to interactions between the cationic iminium group and\nbiomolecules [##REF##9796439##109##]. Such interactions may allow electron-transfer reactions to\ntake place <italic>via</italic> the short-lived\nexcited singlet state and lead to the formation of radicals and radical ions.\nThe copper-free derivative exhibited a tumour response with short intervals\nbetween drug administration and photodynamic therapy. Increased <italic>in vivo</italic> activity was observed with\nthe zinc benzochlorin analogue [##REF##9796439##109##].</p></statement>\n</p>", "<p>\n<statement id=\"head110\"><title>Metallo-phthalocyanines</title><p>The photophysical properties of PCs are strongly\ninfluenced by the presence and nature of the central metal ion [##REF##11749485##12##, ##UREF##5##18##, ##UREF##33##64##, ##REF##17517508##77##]. Co-ordination\nof transition metal ions gives metallo-complexes with short triplet lifetimes (nanosecond\nrange), resulting in different triplet quantum yields and lifetimes (with\nrespect to the non-metallated analogues) [##REF##11749485##12##]. The diamagnetic metals, such as\nzinc, aluminium, and gallium, generate metallo-phthalocyanines (MPC) with high\ntriplet quantum yields (Φ_T ≥ 0.4)\nand short lifetimes (ZnPCS₄\n<italic>τ</italic>\n_T = 490 Fs and AlPcS₄\n<italic>τ</italic>\n_T = 400 Fs) and high singlet oxygen quantum yields (Φ_Δ ≥ 0.7)\n[##REF##11749485##12##, ##UREF##5##18##, ##UREF##33##64##, ##REF##17517508##77##, ##REF##3426922##110##]. As a result, ZnPc and AlPc have been evaluated as second\ngeneration photosensitisers active against certain tumours [##REF##11749485##12##].</p></statement>\n</p>", "<p>\n<statement id=\"head10000\"><title>Metallo-naphthocyaninesulfobenzo-porphyrazines (M-NSBP)</title><p>Aluminium has been successfully coordinated to M-NSBP \n(##FIG##32##Figure 29##). The resulting complex has shown photodynamic activity against\nEMT-6 tumour-bearing Balb/c mice (disulphonated analogue demonstrated greater\nphotoactivity than the mono-derivative) [##UREF##61##111##].</p></statement>\n</p>", "<p>\n<statement id=\"head20000\"><title>Metallo-naphthalocyanines</title><p>Wöhrle and co-workers (Bulgaria) have concentrated their investigations on a zinc NC with various amido substituents.\nThey observed the best phototherapeutic response (Lewis lung carcinoma in mice)\nwith a tetrabenzamido analogue [##UREF##62##112##–##UREF##64##114##]. Kenney's group in the USA have studied\ncomplexes of silicon (IV) NCs (##FIG##33##Figure 30##) with two axial ligands in\nanticipation the ligands would minimise aggregation [##UREF##65##115##]. In particular, they\ninvestigated the disubstituted analogues as potential photodynamic agents [##REF##8127942##116##, ##REF##8577866##117##].\nKenney's results suggested that a siloxane NC substituted with two\nmethoxyethyleneglycol ligands is an efficient photosensitiser against Lewis\nlung carcinoma in mice and that SiNC[OSi(i-Bu)₂-n-C₁₈H₃₇]₂ is effective against Balb/c mice MS-2 fibrosarcoma cells [##REF##2257228##118##, ##UREF##66##119##]. van Lier\nand his group in Canada\nhave also extensively investigated siloxane NCs as agents for photodynamic\ntherapy [##REF##10529768##5##, ##UREF##42##76##]. van Lier's research on these compounds suggests that they are\nefficacious photosensitisers against EMT-6 tumours in Balb/c mice also [##REF##8308868##120##, ##REF##8570740##121##].\nThe ability of certain metallo-NC derivatives (AlNc) to generate singlet oxygen\nis weaker than the analogous (sulphonated) metallo-PCs (AlPC); reportedly 1.6–3 orders of\nmagnitude less [##REF##11749485##12##].</p><p>It can be seen from the above examples that\ngeneralisation(s) between the nature of the parent chromophore; the\npresence/absence of a central metal ion; and the desirable photophysical\nproperties required for a successful photosensitiser are difficult to make. In\nthe porphyrin systems, the zinc ion appears to hinder the photodynamic activity\nof the compound; whereas, in the higher/expanded <italic>π</italic>-systems, dyes chelated with the same\nmetal ion are observed to form complexes with good to high\nphotophysical/photodynamic properties.</p><p>In order to try and address these observations,\nSessler and his group undertook an extensive study into the metallated\ntexaphyrins, investigating the “influence of large metal cations on the\nphotophysical properties of texaphyrins.” They particularly studied “the effect\nof metal cations on the photophysical properties of coordinating ligands.” The\ngroup concentrated on the lanthanide (III) metal ions, Y, In, Lu, Cd, Nd, Sm,\nEu, Gd, Tb, Dy, Ho, Er, Tm, and Yb [##UREF##53##97##].</p><p>Sessler and co-workers observed that when diamagnetic\nLu (III) was complexed to texaphyrin, an effective photosensitiser (Lutex) was\ngenerated. When they substituted the paramagnetic Gd (III) ion for the Lu\nmetal, photodynamic activity was lost. As a result, the group investigated a\nrange of diamagnetic and paramagnetic ions [##UREF##53##97##].</p><p>Sessler further reported a correlation between the\nexcited-singlet and triplet state lifetimes and the rate of ISC of the\ndiamagnetic texaphyrin complexes, Y(III), In (III), and Lu (III), and the\natomic number of the cation [##UREF##53##97##].</p><p>Paramagnetic metallo-texaphyrins were observed to display\nrapid ISC. Greater effects on the rates of triplet decay were also observed,\nand the triplet lifetimes were strongly affected by the choice of metal centre\n[##UREF##53##97##]. The diamagnetic ions (Y, In, and Lu) were recorded as having triplet\nlifetimes ranging from 187, 126, and 35 <italic>μ</italic>s, respectively. Comparable lifetimes\nfor the paramagnetic species (Eu-Tex 6.98 <italic>μ</italic>s, Gd-Tex 1.11, Tb-Tex &lt; 0.2, Dy-Tex \n0.44 × 10⁻³, Ho-Tex 0.85 × 10⁻³, Er-Tex 0.76 × 10⁻³,\nTm-Tex 0.12 × 10⁻³, and Yb-Tex 0.46) were obtained [##UREF##53##97##].</p><p>Sessler and his group were only able to measure the\ntriplet quantum yields for three of the paramagnetic complexes (see ##TAB##0##Table 1##).\nThe results were significantly lower than the diamagnetic metallo-texaphyrins [##UREF##53##97##].</p><p>The data collected from Sessler and co-workers experiments\nsuggests that, in general, singlet oxygen quantum yields closely follow the\ntriplet quantum yields.</p><p> Their experimental data leads to the conclusion that various\ndiamagnetic and paramagnetic texaphyrins investigated have independent\nphotophysical behaviour with respect to a complex's magnetism. The diamagnetic\ncomplexes were characterised by relatively high fluorescence quantum yields,\nexcited-singlet and triplet lifetimes, and singlet oxygen quantum yields; in\ndistinct contrast to the paramagnetic species investigated [##UREF##53##97##].</p><p> Results suggested that the +2 charged diamagnetic species exhibit\na direct relationship between their fluorescence quantum yields, excited state\nlifetimes, rate of ISC, and the atomic number of the metal ion. The greatest\ndiamagnetic ISC rate was observed for Lu-Tex; a result ascribed to the heavy\natom effect. The heavy atom effect also held for the Y-Tex, In-Tex, and Lu-Tex\ntriplet quantum yields and lifetimes. The triplet quantum yields and lifetimes\nboth decreased with increasing atomic number. The singlet oxygen quantum yield\ncorrelated with this observation [##UREF##53##97##].</p><p> The photophysical properties displayed by the paramagnetic\nspecies were more complex. A simple correlation between the observed\ndata/behaviour and the number of unpaired electrons located on the metal ion\ncould not be made. For example, the ISC rates and the fluorescence lifetimes\ngradually decreased with increasing atomic number, the Gd-Tex, and Tb-Tex\nchromophores showed (despite having a larger number of unpaired electrons)\nslower rates of ISC and longer lifetimes than Ho-Tex or Dy-Tex. Sessler\nsuggested that charge transfer or intermolecular energy transfer is taking\nplace from higher excited states (such as S₂) [##UREF##53##97##].\n</p></statement>\n</p>", "<title>12. SUMMARY</title>", "<p>A variety of second generation photosensitisers have\nbeen developed and evaluated against a range of clinical applications (see Tables\n##TAB##1##2## and ##TAB##2##3##). The metallation of a number of these chromophores has generated a variety\nof photosensitisers with improved photophysical properties. The effectiveness\nof these metallo-photosensitisers depends largely (but not definitively) on the\nnature of the co-ordinated central metal ion. Chromophores chelated to\ndiamagnetic transition metals and lanthanide ions have shown the greatest\npotential as photodynamic agents, a consequence of the heavy metal effect\nenhancing the rate of ISC. As a result, a number of these metallated\ntetrapyrrole-based macrocycles are currently photosensitisers of choice,\nparticularly the zinc (II), aluminium (III), and tin (IV) complexes.</p>" ]

[ "<title>GLOSSARY</title>", "<p>\nAMD is the leading cause of blindness in humans over the age of 50. AMD is characterised by a rapid growth of abnormal blood vessels under the central retina causing scarring, and an accelerated loss of visual acuity [##UREF##30##58##].</p>", "<p>Chemosensitisers are drugs or chemicals which are used to enhance the therapeutic effects of anti-cancer (chemotherapy) drugs. They make the tumour cells more sensitive to the effects of chemotherapy.</p>", "<p>Diamagnetic species is a species with no unpaired electrons, that is, all electrons are spin-paired.</p>", "<p>Fluorophore is generally a molecule capable of absorbing light energy when irradiated at a specific wavelength and emitting energy at longer wavelengths.</p>", "<p>Heavy atom effect enhances coupling between the excited-singlet (S₁) and excited-triplet (T₁) states. It is the enhancement of a spin-forbidden process by the presence of an atom of high atomic molecular weight. Mechanistically, it responds to a spin-orbit coupling enhancement produced by a heavy atom. Spin-forbidden and spin-allowed processes are highlighted in <xref ref-type=\"sec\" rid=\"sec7\">Section 7</xref>.</p>", "<p>Luminescence lifetime is the average time a molecule spends in an excited state (S_{<italic>n</italic>} &gt; 0 or T_{<italic>n</italic>} &gt; 0).</p>", "<p>Methylene blue is used to sterilize/decontaminate freshly frozen plasma units by inactivating extracellularly enveloped viruses (such as HIV), as well as methaemoglobinaemia [##REF##10529768##5##, ##REF##15267226##9##].</p>", "<p>Paramagnetic species is a species with one or more unpaired electrons.</p>", "<p>Phase I clinical trials are used to determine pharmacokinetic properties (metabolism, elimination, and preferred method of administration) and a safe dosage range, and identify any side effects of a new drug. They are performed on a small number of people (20–80).</p>", "<p>Phase II clinical trials are performed on a larger group of people (100–300). The drug is further evaluated to test its effectiveness and any side effects.</p>", "<p>Phase III clinical trials: the drugs effectiveness is confirmed and comparisons are made to more commonly used treatment modalities in a range of 1000–3000 people. Potential side effects are monitored.</p>", "<p>Phase IV clinical trials are post-marketing observations and evaluations.</p>", "<p>Photochemotherapy is a combination of a chemical substance and light to treat disease.</p>", "<p>Phototherapy is a term used to describe the treatment of disease by a series of (photo-) chemical processes initiated by light.</p>", "<p>Radiosensitisers are drugs which boost the effect of radiation therapy (radiotherapy) by making the tumour tissue more vulnerable to the applied radiation. They may be used alone or in conjunction with other drugs.</p>", "<p>Restenosis is the renarrowing of a coronary artery after angioplasty or stenting.</p>" ]

[ "<fig id=\"fig1\" position=\"float\"><label>Figure 1</label><caption><p>Structure of psoralen.</p></caption></fig>", "<fig id=\"fig2\" position=\"float\"><label>Figure 2</label><caption><p>Photosensitiser administration.</p></caption></fig>", "<fig id=\"fig3\" position=\"float\"><label>Figure 3</label><caption><p>Structure of Photofrin, <italic>n</italic> = 1–9.</p></caption></fig>", "<fig id=\"fig4\" position=\"float\"><label>Figure 4</label><caption><p>Photosensitiser initiated cell death.</p></caption></fig>", "<fig id=\"boxx1\" position=\"float\"><label>Box 1</label></fig>", "<fig id=\"fig5\" position=\"float\"><label>Figure 5</label><caption><p>Modified Jablonski energy diagram.</p></caption></fig>", "<fig id=\"boxx2\" position=\"float\"><label>Box 2</label></fig>", "<fig id=\"fig6\" position=\"float\"><label>Figure 6</label><caption><p>Type-I process (i).</p></caption></fig>", "<fig id=\"fig7\" position=\"float\"><label>Figure 7</label><caption><p>Type-I process (ii).</p></caption></fig>", "<fig id=\"fig8\" position=\"float\"><label>Figure 8</label><caption><p>Type-II process.</p></caption></fig>", "<fig id=\"boxx3\" position=\"float\"><label>Box 3</label></fig>", "<fig id=\"fig9\" position=\"float\"><label>Figure 9</label><caption><p>Molecular orbital diagram of oxygen (³Σ_g, ¹Δ_g,\nand ¹Σ_g).</p></caption></fig>", "<fig id=\"fig10\" position=\"float\"><label>Figure 10</label><caption><p>A copper metallated photosensitiser.</p></caption></fig>", "<fig id=\"fig11\" position=\"float\"><label>Figure 11</label><caption><p>Examples of non-porphyrinic photosensitisers.</p></caption></fig>", "<fig id=\"fig12\" position=\"float\"><label>Figure 12</label><caption><p>Porphine, chlorine, bacteriochlorin, expanded porphyrin, and phthalocyanine\nstructures.</p></caption></fig>", "<fig id=\"fig13\" position=\"float\"><label>Figure 13</label><caption><p>Porphine macrocycle.</p></caption></fig>", "<fig id=\"fig14\" position=\"float\"><label>Figure 14</label><caption><p>Porphyrin dianionic and dicationic species.</p></caption></fig>", "<fig id=\"fig15\" position=\"float\"><label>Figure 15</label><caption><p>Typical porphyrin absorption spectrum [##UREF##28##55##, (modified)].</p></caption></fig>", "<fig id=\"fig16\" position=\"float\"><label>Figure 16</label><caption><p>Modified Jablonski energy diagram.</p></caption></fig>", "<fig id=\"fig17\" position=\"float\"><label>Figure 17</label><caption><p>Verteporfin.</p></caption></fig>", "<fig id=\"sch1\" position=\"float\"><label>Scheme 1</label><caption><p>Simplified haem biosynthesis.</p></caption></fig>", "<fig id=\"fig18\" position=\"float\"><label>Figure 18</label><caption><p>Purlytin.</p></caption></fig>", "<fig id=\"fig19\" position=\"float\"><label>Figure 19</label><caption><p>Foscan.</p></caption></fig>", "<fig id=\"fig20\" position=\"float\"><label>Figure 20</label><caption><p>Lutex.</p></caption></fig>", "<fig id=\"fig21\" position=\"float\"><label>Figure 21</label><caption><p>ATMPn.</p></caption></fig>", "<fig id=\"fig22\" position=\"float\"><label>Figure 22</label><caption><p>Zinc phthalocyanine CGP55847.</p></caption></fig>", "<fig id=\"fig23\" position=\"float\"><label>Figure 23</label><caption><p>Photosense.</p></caption></fig>", "<fig id=\"fig24\" position=\"float\"><label>Figure 24</label><caption><p>A naphthalocyanine.</p></caption></fig>", "<fig id=\"fig25\" position=\"float\"><label>Figure 25</label><caption><p>Water-soluble cationic metallated porphyrins.</p></caption></fig>", "<fig id=\"fig26\" position=\"float\"><label>Figure 26</label><caption><p>5,10-Di-(4-sulphonatophenyl)-15,20-diphenylporphyrinato aluminium chloride.</p></caption></fig>", "<fig id=\"fig27\" position=\"float\"><label>Figure 27</label><caption><p>Zinc metallated <italic>seco-</italic>porphyrazine, platinum, and palladium derivatives.</p></caption></fig>", "<fig id=\"fig28\" position=\"float\"><label>Figure 28</label><caption><p>Sulphonated and tin (IV) benzochlorin derivatives.</p></caption></fig>", "<fig id=\"fig29\" position=\"float\"><label>Figure 29</label><caption><p>An aluminium naphthalocyaninesulfobenzoporphyrazine.</p></caption></fig>", "<fig id=\"fig30\" position=\"float\"><label>Figure 30</label><caption><p>A siloxane naphthalocyanine.</p></caption></fig>", "<fig id=\"boxx4\" position=\"float\"><label>Box 4</label></fig>" ]

[ "<table-wrap id=\"tab1\" position=\"float\"><label>Table 1</label><caption><p>Quantum triplet yields\nof texaphyrin metallated with paramagnetic and diamagnetic lanthanides (data reproduced from [##UREF##53##97##]).</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\"> Paramagnetic MTex </th><th align=\"center\" colspan=\"2\" rowspan=\"1\">Φ_T\n</th><th align=\"center\" rowspan=\"1\" colspan=\"1\"> Diamagnetic MTex </th></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Eu-Tex</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">0.090</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">0.563</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">Y-Tex</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Gd-Tex</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">0.156</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">0.500</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">In-Tex</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Yb-Tex</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">0.126</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">0.340</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">Lu-Tex</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"tab2\" position=\"float\"><label>Table 2</label><caption><p>Summary of a collection of different photosensitiser\ntypes and their absorption data.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\"> CLASS OF PHOTOSENSITISER </th><th align=\"left\" rowspan=\"1\" colspan=\"1\"> LONGEST WAVELENGTH ABSORPTION/nm</th><th align=\"left\" rowspan=\"1\" colspan=\"1\"> EXTINCTION COEFFICIENT/ M⁻¹cm⁻¹\n</th><th align=\"left\" rowspan=\"1\" colspan=\"1\"> DRUGS IN CLINICAL TRIAL (Phase I–III)</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">DRUGS APPROVED FOR PDT (PRECLINICAL AND CLINICAL) </th></tr></thead><tbody><tr><td align=\"left\" rowspan=\"3\" colspan=\"1\">Porphyrins</td><td align=\"left\" rowspan=\"3\" colspan=\"1\">620–640</td><td align=\"left\" rowspan=\"3\" colspan=\"1\">3,500</td><td align=\"left\" rowspan=\"3\" colspan=\"1\">—</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Photofrin</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> Levulan</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> Metvix</td></tr><tr><td align=\"center\" colspan=\"5\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">(Expanded Porphyrins) </td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Porphycenes</td><td align=\"left\" rowspan=\"1\" colspan=\"1\"> 610–650</td><td align=\"left\" rowspan=\"1\" colspan=\"1\"> 50,000</td><td rowspan=\"1\" colspan=\"1\"/><td align=\"left\" rowspan=\"1\" colspan=\"1\">ATMPn</td></tr><tr><td rowspan=\"6\" colspan=\"1\">Texaphyrins</td><td rowspan=\"6\" colspan=\"1\">730–770</td><td rowspan=\"6\" colspan=\"1\"> 40,000</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">Lu-Tex</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Optrin</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> Antrin</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Xcytrin</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Benzvix</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Hexvix</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"center\" colspan=\"5\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"2\" colspan=\"1\">Chlorins</td><td align=\"left\" rowspan=\"2\" colspan=\"1\">650–690</td><td align=\"left\" rowspan=\"2\" colspan=\"1\">40,000</td><td align=\"left\" rowspan=\"1\" colspan=\"1\"> Foscan\n</td><td align=\"left\" rowspan=\"2\" colspan=\"1\">Visudyne</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\nPuryltin</td></tr><tr><td align=\"center\" colspan=\"5\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Bacteriochlorins</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">730–800</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">150,000</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">—</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">—</td></tr><tr><td align=\"center\" colspan=\"5\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"3\" colspan=\"1\">Phthalocyanines</td><td align=\"left\" rowspan=\"3\" colspan=\"1\">680–780</td><td align=\"left\" rowspan=\"3\" colspan=\"1\">200,000</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\nCGP55847</td><td align=\"left\" rowspan=\"3\" colspan=\"1\">—</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\nPC4</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\nPhotosense</td></tr><tr><td align=\"center\" colspan=\"5\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Naphthalocyanines</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">740–780</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">250,000</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">—</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">—</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"tab3\" position=\"float\"><label>Table 3</label><caption><p>Summary of a range of photosensitisers and their clinical applications.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\"> TRADE NAME </th><th align=\"center\" rowspan=\"1\" colspan=\"1\"> MARKETING\nCOMPANY </th><th align=\"center\" rowspan=\"1\" colspan=\"1\"> PRE-/CLINICAL APPLICATION</th><th align=\"left\" rowspan=\"1\" colspan=\"1\"> COUNTRIES APPROVED IN </th></tr></thead><tbody><tr><td align=\"left\" rowspan=\"12\" colspan=\"1\">Photofrin</td><td align=\"center\" rowspan=\"12\" colspan=\"1\">QLT Phototherapeutics</td><td align=\"center\" rowspan=\"12\" colspan=\"1\">Oesophageal, lung, bladder and cervical dysplasia</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Canada (1993),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">The Netherlands (1994),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Japan (1994),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">USA (1995),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">France (1996),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Germany (1997),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Finland (1999), </td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">UK (1999),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Sweden (2000), </td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Italy (2000), </td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Ireland (2000),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Poland (2000)</td></tr><tr><td align=\"center\" colspan=\"4\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"3\" colspan=\"1\">Levulan</td><td align=\"center\" rowspan=\"3\" colspan=\"1\">DUSA Pharmaceuticals</td><td align=\"center\" rowspan=\"1\" colspan=\"1\"> Actinic keratosis\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">USA (1999), </td></tr><tr><td align=\"center\" rowspan=\"1\" colspan=\"1\">Actinic\nkeratosis and basal-cell carcinoma</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Sweden (2001),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\"> Europe (2001)</td></tr><tr><td align=\"center\" colspan=\"4\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"2\" colspan=\"1\">Metvix</td><td align=\"center\" rowspan=\"2\" colspan=\"1\">Photocure ASA</td><td align=\"center\" rowspan=\"2\" colspan=\"1\">Actinic keratosis and basal-cellcarcinoma\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Sweden (2001), </td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Europe (2001)</td></tr><tr><td align=\"center\" colspan=\"4\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"6\" colspan=\"1\">Visudyne</td><td align=\"center\" rowspan=\"6\" colspan=\"1\">QLT Phototherapeutics</td><td align=\"center\" rowspan=\"3\" colspan=\"1\">Wet-AMD</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Europe (2001),\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">USA (2000),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Canada\n(2000), </td></tr><tr><td align=\"center\" rowspan=\"3\" colspan=\"1\"> Subfoveal choroidal neovascularisation</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Europe (2001),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">USA(2001),</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Canada (2001)</td></tr><tr><td align=\"center\" colspan=\"4\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">ATMPn</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">GlaxoWellcome and Cytopharm</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">\nPsorasis and non-melanoma\nskin cancer</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Germany (1997)</td></tr><tr><td align=\"center\" colspan=\"4\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Purlytin</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">Miravant Medical Technologies</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">\nPsorasis and restenosis</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">USA (1998)</td></tr><tr><td align=\"center\" colspan=\"4\" rowspan=\"1\">\n<hr/>\n</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Foscan</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">BioLitec Pharmaceuticals</td><td align=\"center\" rowspan=\"1\" colspan=\"1\">\nHead and neck cancers</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Europe (2001)</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"MBD2008-276109.001\"/>", "<graphic xlink:href=\"MBD2008-276109.002\"/>", "<graphic xlink:href=\"MBD2008-276109.003\"/>", "<graphic xlink:href=\"MBD2008-276109.004\"/>", "<graphic xlink:href=\"MBD2008-276109.boxx1\"/>", "<graphic xlink:href=\"MBD2008-276109.005\"/>", "<graphic xlink:href=\"MBD2008-276109.boxx2\"/>", "<graphic xlink:href=\"MBD2008-276109.006\"/>", "<graphic xlink:href=\"MBD2008-276109.007\"/>", "<graphic xlink:href=\"MBD2008-276109.008\"/>", "<graphic xlink:href=\"MBD2008-276109.boxx3\"/>", "<graphic xlink:href=\"MBD2008-276109.009\"/>", "<graphic xlink:href=\"MBD2008-276109.010\"/>", "<graphic xlink:href=\"MBD2008-276109.011\"/>", "<graphic xlink:href=\"MBD2008-276109.012\"/>", "<graphic xlink:href=\"MBD2008-276109.013\"/>", "<graphic xlink:href=\"MBD2008-276109.014\"/>", "<graphic xlink:href=\"MBD2008-276109.015\"/>", "<graphic xlink:href=\"MBD2008-276109.016\"/>", "<graphic xlink:href=\"MBD2008-276109.017\"/>", "<graphic xlink:href=\"MBD2008-276109.sch001\"/>", "<graphic xlink:href=\"MBD2008-276109.018\"/>", "<graphic xlink:href=\"MBD2008-276109.019\"/>", "<graphic xlink:href=\"MBD2008-276109.020\"/>", "<graphic xlink:href=\"MBD2008-276109.021\"/>", "<graphic xlink:href=\"MBD2008-276109.022\"/>", "<graphic xlink:href=\"MBD2008-276109.023\"/>", "<graphic xlink:href=\"MBD2008-276109.024\"/>", "<graphic xlink:href=\"MBD2008-276109.025\"/>", "<graphic xlink:href=\"MBD2008-276109.026\"/>", "<graphic xlink:href=\"MBD2008-276109.027\"/>", "<graphic xlink:href=\"MBD2008-276109.028\"/>", "<graphic xlink:href=\"MBD2008-276109.029\"/>", "<graphic xlink:href=\"MBD2008-276109.030\"/>", "<graphic xlink:href=\"MBD2008-276109.boxx4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[]

[]

[]

[]

[]

[]

[ "<p>Treatment of diseases with natural and synthetic\nmaterials has been an aspiration of mankind since the dawn of human\ndevelopment. From the use of willow-bark to the marketing\nof aspirin, a steady move from folk remedies to the use of chemistry and biology\nto develop new therapies has been observed.\nIn terms of metal-containing drugs, the platinum-containing drug\ncisplatin has long been the most effective metal-containing anticancer drug on the market.</p>", "<p>However, severe side effects of conventional drugs\nare associated with the inability to distinguish between healthy and cancer\ncells. Hence, a concerted world-wide effort is in progress to discover and characterise new drugs that may distinguish between healthy and cancer or other diseased cells. New techniques of drug delivery are sought and the use of natural products, proteins, antibodies, and synthetic polymers as drug delivery devices capable of targeting a diseased site is being\ninvestigated.</p>", "<p>These issues are nicely illustrated by macrocycles such as porphyrins,\nphthalocyanines, and related systems. Some of these compounds exhibit selective absorption by cancer cells and have the ability to photosensitize formation of singlet oxygen.\nThese attributes have led to the development of alternative cancer\ntreatments known as photodynamic therapy. Sadly, many potentially good new therapeutic agents often never leave the designers' laboratory due to some pharmacological problems associated with\nits in vivo use. The use of drug delivering devices, including\nwater-soluble synthetic polymeric drug delivery systems, may help overcome many\npharmacological drug-related problems, including those of solubility, specificity,\nand biocompatibility, factors that currently prevent many potentially good\ntherapeutic agents from reaching clinics.</p>", "<p>The focus of this special issue is the synthesis, characterisation,\nphysical studies, and application of synthetic metal-containing complexes and\nnatural occurring proteins in serious human diseases such as cancer, diabetes,\narthritis, viral disease, malaria, and tuberculosis with special focus on the\nfollowing:</p>", "<p>\n<list list-type=\"alpha-lower\"><list-item><p>porphyrins, phthalocyanines, and related complexes in photodynamic cancer therapy;</p></list-item><list-item><p>proteins, enzymes, and synthetic polymeric drug delivery systems in the treatment of cancer and other diseases;</p></list-item><list-item><p>coordination and organometallic compounds in cancer, arthritis, malaria, and viral disease.</p></list-item></list>\n</p>", "<p>Towards these goals, L. Josefsen and R. Boyle describe in their review article the development and application of metal-based photosensitisers,\nincluding porphyrins and phthalocyanines, in photodynamic therapy. Four other publications highlight different aspects of porphyrin-based macrocyclic photosensitisers. S. Lee et al. focus on the cellular\nuptake and toxicity of thiotetra (ethylene glycol) monomethyl ether-functionalized\nporphyrazines. J.-Y. Liu et al. focus on in vitro photodynamic activity of novel amphiphilic zinc(II) phthalocyanines\nbearing oxyethylene-rich substituents. E. Antunes and T.\nNyokong highlight the syntheses and photophysical properties of\ntetraazatetrabenzcorrole photosensitizers. Sakamoto et al. present a fundamental study of zinc\nbis(1,4-didecylbenzo)-bis(2,3-pyrido)porphyrazine for application in\nphotodynamic therapy of cancer.</p>", "<p>Considering polymeric drug delivery systems, South African\nE. Neuse's excellent review describes the use of synthetic polymers as\nmetal-containing drug delivery vehicles in medicine. M. David Maree et al.\nprovided a fine treatise on why biocompatible synthetic polymeric drug delivery\nsystems are becoming increasing popular as drug delivering devices. They also demonstrate the principles behind these systems in a practical study utilising ferrocene and phthalocyanine derivatives anchored on a water-soluble polymeric drug carrier derived from lysine and aspartic\nacid. Italians Longo and Vasapollo demonstrated the use of phthalocyanine-based molecularly imprinted polymers as nucleoside receptors. X. Sun et al. report on the identification of proteins related to nickel homeostasis in <italic>Helicobater pylori</italic> by immobilized metal affinity chromatography and two-dimensional gel electrophoresis. P. Nagababu reported DNA-binding and photocleavage studies\nof cobalt (III) ethylenediamine pyridine complexes.</p>", "<p>M. Blackie and K. Chibale's excellent minireview focuses on metallocene antimalarials. The development of new metal-containing chemotherapeutic drugs is highlighted by contributions from M. Hogan et al. in their synthetic and cytotoxic study of new titanocene analogous. \nS. Mahepal et al. look at the in vitro antitumor activity of novel, mitochondrial-interactive, gold-based lipophilic cations. Sathisha et al. describe bis-isatin thiocarbohydrazone metal complexes and their antitumor activity against Ehrlich Ascites Carcinoma in Swiss albino mice. In conclusion, J. M. Botha and A. Roodt report mechanistic studies on trans-aquatetracyanooxo complexes of Re(V) and Tc(V) and discuss the implications of their results for nuclear\nmedicine.</p>", "<p>We hope that this special issue will stimulate new research in all areas\nof metal-containing drug research and that it will help to focus research\nefforts of new and experienced researchers on key problems in the exciting field\nof metal-containing drug research.</p>", "<p content-type=\"signature-group\">\n<named-content content-type=\"signature\"><italic>Jannie C. Swarts</italic>\n<italic>Jannie C. Swarts</italic>\n</named-content>\n</p>", "<p content-type=\"signature-group\">\n<named-content content-type=\"signature\"><italic>Michael J. Cook</italic>\n<italic>Michael J. Cook</italic>\n</named-content>\n</p>", "<p content-type=\"signature-group\">\n<named-content content-type=\"signature\"><italic>Edward N. Baker</italic>\n<italic>Edward N. Baker</italic>\n</named-content>\n</p>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

{ "acronym": [], "definition": [] }

[]

[]

[]

[ "<title>Discussion</title>", "<p>The protocol presented here provides researchers with a concise, easy-to-follow outline of how to obtain thin cryostat sections of small, difficult-to-manage, tissue pieces, such as biopsies and brain slices for further studies to be performed, such as various staining methods, in situ hybridization, or immunohistochemistry.</p>" ]

[]

[ "<p>Correspondence to: Miles G. Cunningham at <email>mcunningham@mclean.harvard.edu</email></p>", "<p>Many investigations in neuroscience, as well as other disciplines, involve studying small, yet macroscopic pieces or sections of tissue that have been preserved, freshly removed, or excised but kept viable, as in slice preparations of brain tissue. Subsequent microscopic studies of this material can be challenging, as the tissue samples may be difficult to handle. Demonstrated here is a method for obtaining thin cryostat sections of tissue with a thickness that may range from 0.2-5.0 mm. We routinely cut 400 micron thick Vibratome brain slices serially into 5-10 micron coronal cryostat sections. The slices are typically first used for electrophysiology experiments and then require microscopic analysis of the cytoarchitecture of the region from which the recordings were observed. We have constructed a simple device that allows controlled and reproducible preparation and positioning of the tissue slice. This device consists of a cylinder 5 cm in length with a diameter of 1.2 cm, which serves as a freezing stage for the slice. A ring snugly slides over the cylinder providing walls around the slice allowing the tissue to be immersed in freezing compound (e.g., OCT). This is then quickly frozen with crushed dry ice and the resulting wafer can be position easily for cryostat sectioning. Thin sections can be thaw-mounted onto coated slides to allow further studies to be performed, such as various staining methods, in situ hybridization, or immunohistochemistry, as demonstrated here.</p>" ]

[ "<title>Protocol</title>", "<p>Prepare mold from tape for OCT platform.</p>", "<p>Fill mold with OCT. Freeze within cryostat or by using crushed dry ice.</p>", "<p>Remove tape from around frozen OCT platform.</p>", "<p>Align marks on freezing chuck and cryostat mounting stage and lock in chuck.</p>", "<p>Section through OCT platform until surface is flat.</p>", "<p>Remove resurfaced OCT platform and place on cryostat freezing stage.</p>", "<p>Place tissue sample (previously cryopreserved with 30% glycerol or sucrose in PBS) in OCT.</p>", "<p>Prepare freezing column with outer ring projecting about 5 mm above top of column forming well for OCT.</p>", "<p>Carefully position tissue sample onto center of freezing column surface and slowly add OCT until well is filled.</p>", "<p>Surround freezing column with crushed dry ice. Tissue and OCT should completely freeze within 20-60 seconds.</p>", "<p>As preparation increases in temperature, the outer ring can be removed while the sample remains frozen.</p>", "<p>Slide sample off freezing column sideways and place in cryostat.</p>", "<p>Place drop of OCT on surface of OCT platform and position specimen (tissue down) applying firm pressure. Specimen will quickly freeze onto OCT platform.</p>", "<p>Secure chuck onto cryostat mounting stage with marks aligned.</p>", "<p>Section through OCT superficial to the tissue specimen.</p>", "<p>Thaw mount thin sections onto glass slides and store frozen or at room temperature.</p>", "<p>Immunoreactions can be performed for tissue mounted on glass slides.</p>", "<p>Reagent is pooled onto slide, can be gently agitated, and may be covered if light-sensitive.</p>", "<p>Subsequent stages of the reaction are easily performed by inverting slide into waste receptacle, wicking the slide, and then applying the next reagent.</p>" ]

[]

[ "<fig id=\"Fig_194\" position=\"anchor\"><alternatives></alternatives></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"jove-3-194-thumb\"/>" ]

[ "<media id=\"Video_194\" xlink:href=\"jove-3-194.flv\" xlink:type=\"simple\" position=\"anchor\" mimetype=\"text\" mime-subtype=\"plain\"/>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Mutations causing gene fusions, loss of function or misregulation of transcriptional corepressor proteins have been implicated both in genetically inherited diseases and cancer ##REF##17609497##[1]##, ##REF##16390317##[2]##. BCOR (BCL6 corepressor) is a transcriptional corepressor that was originally identified by its ability to interact with the site specific transcriptional repressor BCL6 ##REF##10898795##[3]##. BCOR is found in an 800 kDa complex in which at least two proteins have chromatin modifying activity: the PcG transcriptional repressor protein, RNF2, is a histone H2A E3 ubiquitin ligase and FBXL10 (JHDM1B, KDM2B) is a Jmjc histone demethylase in addition to its presumed ubiquitin E3 ligase activity ##REF##16943429##[4]##, ##REF##17296600##[5]##. BCL6 plays critical roles in specific immunological processes involving B and T cells, including germinal center formation and the generation and maintenance of memory T cells ##REF##9110977##[6]##–##REF##9171827##[10]##. In addition, testicular germ cell apoptosis and defects in erythropoiesis have been reported in BCL6-deficient mice ##REF##15661395##[11]##, ##REF##11092811##[12]##. Deregulated expression of BCL6, due to chromosomal translocations or point mutations, is associated with the formation of approximately one sixth of non-Hodgkin's lymphomas ##REF##15202519##[13]##, ##REF##10626250##[14]##. In cell lines derived from such patients, BCOR is detected at a number of BCL6 target genes suggesting that BCOR is likely to play a role in mediating BCL6 driven lymphomagenesis. Since its original discovery, BCOR has been shown to directly interact with the transcriptional regulator AF9 (MLLT3), and to be in a complex with ENL (MLLT1). AF9 and ENL are known MLL (trithorax) fusion partners in acute leukemias ##REF##17957188##[15]##. AF9 itself is a regulator of Hox gene expression and skeletal development ##REF##12776190##[16]##, ##REF##12242306##[17]##.</p>", "<p>In addition, BCOR has been shown to play multiple roles in the complex process of human development. Females who are heterozygous for X-linked <italic>Bcor</italic> mutations have the rare Oculofaciocardiodental (OFCD) syndrome, the primary subtype of OMIM #300166 microphthalmia, syndromic 2 (MCOPS2) ##REF##15004558##[18]##. Congenital disorders in patients suffering from OFCD include cataracts, microphthalmia, and cardiac, dental and digital anomalies ##REF##15004558##[18]##, ##UREF##0##[19]##. In the hematopoietic lineage of OFCD patients, <italic>Bcor</italic> clearly is critical as 90–100% of surviving white blood cells show inactivation of the X- chromosome carrying the mutant allele of <italic>Bcor</italic>\n##REF##15004558##[18]##. Presumably, the selective disadvantage, caused by an active X-chromosome harboring a <italic>Bcor</italic> mutation, is less severe in some tissues leading to variable phenotypic effects in a mosaic fashion. Male OFCD patients do not exist and are presumed to die in early development. Almost all <italic>Bcor</italic> mutations in OFCD patients result in premature stop codons that are thought to cause nonsense-mediated decay of the mRNA ##REF##15004558##[18]##. The second form of MCOPS2, Lenz microphthalmia, results from a single missense mutation (p. P85L) in the fourth coding exon of <italic>Bcor</italic> and is inherited in an X-linked recessive pattern ##REF##15004558##[18]##. Patients with Lenz microphthalmia suffer from microphthalmia/anophthalmia, mental retardation, and skeletal and other anomalies ##REF##15004558##[18]##, ##REF##12116202##[20]##.</p>", "<p>The severity and breadth of these two MCOPS2 syndromes illustrate the important roles <italic>Bcor</italic> plays during development. Currently two animal models have successfully recapitulated developmental abnormalities similar to those found in patients with OFCD and Lenz microphthalmia. RNAi knock down of <italic>Bcor</italic> in zebrafish (<italic>Danio rerio</italic>) results in colobomatous eye defects and perturbations in somite, skeletal and neural tube development ##REF##15004558##[18]##. Colobomas, microphthalmia and cardiac abnormalities were also seen in morpholino knock down of <italic>Bcor</italic> in frogs (<italic>Xenopus tropicalis</italic>). Additionally, laterality defects were observed in <italic>Bcor</italic> morpholino injected frogs, similar to those present in a subset of OFCD patients ##REF##17517692##[21]##.</p>", "<p>The expression of <italic>Bcor</italic> during mouse development correlates well with tissues and organs in adversely affected patients with OFCD or Lenz microphthalmia ##REF##17344103##[22]##. Three separate promoters differentially control the expression of <italic>Bcor</italic> through embryonic development and into adulthood. Section <italic>in situ</italic> analysis shows that <italic>Bcor</italic> mRNA is expressed strongly in extraembryonic tissues during gastrulation and embryonic turning, suggesting a significant role in placental formation. <italic>Bcor</italic> expression is upregulated in the embryo proper at embryonic day 9 in mice starting in the tail, limb buds and branchial arches. During the later fetal stages of mouse development, <italic>Bcor</italic> is expressed in multiple tissues including strong expression in lens and retina of the eye and the neural tube ##REF##17344103##[22]##.</p>", "<p>The abnormalities present in <italic>Bcor</italic> knockdown fish and frogs, combined with the severe clinical presentation of MCOPS2 syndromes in humans, clearly illustrate the importance of <italic>Bcor</italic> in development. The developmental requirement for <italic>Bcor</italic>, together with its potential role in BCL6-related lymphomagenesis, underscores the importance of generating and analyzing mutant alleles of <italic>Bcor</italic>. To this end, we have analyzed the effect of two independent <italic>Bcor</italic> loss of function alleles on early development in mice and in ES cell differentiation. We found that <italic>Bcor</italic> exhibits a parent-of-origin effect, in that only paternal transmission results in viable offspring. Since the paternal X chromosome is exclusively inactivated in the extraembryonic tissue, this strongly suggests <italic>Bcor</italic> is required for extraembryonic development. In addition, loss of <italic>Bcor</italic> results in compromised development during <italic>in vitro</italic> differentiation of embryonic stem cells.</p>" ]

[ "<title>Materials and Methods</title>", "<title>Animals</title>", "<p>All experimental protocols involving mice described in this publication have been approved by the University of Minnesota Institutional Animal Care and Use Committee. The mice were not kept in specific pathogen free (SPF) rodent housing, thus they were exposed to multiple pathogens that are present at sub clinical levels in conventional rodent housing. Due to continual pathogen exposure, PNA+ CD19+ germinal center positive cell populations were already present in our mice thus, we did not specifically induce germinal center reactions through immunization.</p>", "<title>Generation of <italic>Bcor</italic> mutant mice</title>", "<p>A mouse <italic>Bcor</italic> cDNA fragment containing sequences from within exon 4 to within exon 7 was used to screen the RPCI-22 129S6/Sv EvTac Bac library (Stratagene) ##REF##10645956##[63]##, and positive BAC clones were used to clone <italic>Bcor</italic> genomic sequence. The left and right homology arms were cloned into the backbone vector pDZ157 ##REF##11040213##[64]##. The final targeting construct is diagrammed in ##FIG##0##Figure 1A##. The <italic>Bcor</italic> targeting vector was linearized with <italic>PmeI</italic> and electroporated into CJ7 ES cells (originally derived from the 129S1 strain). One homologous recombinant was identified from 768 G418-resistant colonies by Southern hybridization using a DNA probe from sequences upstream of the targeting vector to screen genomic DNA digested with <italic>BamH1</italic>. Homologous recombination was confirmed on both ends of the targeted region by Southern hybridization. Probes for Southern hybridization were generated from a BAC clone containing <italic>Bcor</italic> genomic sequence by PCR using primers VBp525/VBp526 (5′ probe, left arm) and VBp582/ VBp583 (3′ probe, right arm). The targeted ES cell clone containing the <italic>Bcor^Neo</italic> allele was injected into C57Bl/6 blastocysts to generate chimeras. Chimeric males were bred with C57Bl/6 females to generate heterozygotes carrying the <italic>Bcor^Neo</italic> allele. <italic>Bcor^Neo</italic>\n^/+ females were bred with male β-<italic>actin-Flpe</italic> transgenic mice to delete the frted sequence and generate heterozygotes carrying the <italic>Bcor^Fl</italic> allele. <italic>Bcor^Fl/+</italic> females were bred with male β-<italic>actin-Cre</italic> transgenic mice to delete the floxed sequence and generate heterozygous <italic>Bcor^Δ3/+</italic> and hemizygous <italic>Bcor^Δ3/Y</italic> deletion mutants.</p>", "<p>VBp525 <named-content content-type=\"gene\">5′-CTCTACTTGCTCAGTCTGCCTGCAATG-3′</named-content>\n</p>", "<p>VBp526 <named-content content-type=\"gene\">5′-AAGTCGACACATTTCCTTTGTTAGCAG-3′</named-content>\n</p>", "<p>VBp582 <named-content content-type=\"gene\">5′-GAGTTGTATCTCATAAATTGTGGTTG-3′</named-content>\n</p>", "<p>VBp583 <named-content content-type=\"gene\">5′-CTGTCATTCACTTTGAGCCTGGTGT-3′</named-content>\n</p>", "<p>The XE541 ES cell line was created by the BayGenomics genetrap consortium through insertional genetrap mutagenesis of the wild type E14 ES cell line (originally derived from the 129P2 strain). The genetrap mutation resides in the sixth intron of the <italic>Bcor</italic> gene and results in a loss of function allele we named <italic>Bcor^Gt</italic>. XE541 ES cells harboring the <italic>Bcor^Gt</italic> allele were injected into C57Bl/6 blastocysts by the University of Minnesota Mouse Genetics Laboratory to generate chimeric mice.</p>", "<title>Genotyping</title>", "<p>The <italic>Bcor⁺</italic>, <italic>Bcor^Fl</italic> and <italic>Bcor</italic>\n^{Δ<italic>3</italic>} alleles were detected by polymerase chain reaction (PCR) from tail clip genomic DNA using primer set VBp573 /VBp981. The described PCR generates an 1122 base pair (bp) amplicon for the <italic>Bcor^Fl</italic>, a 993 bp amplicon for the <italic>Bcor⁺</italic> allele and a 332 bp amplicon for the <italic>Bcor</italic>\n^{Δ<italic>3</italic>}. The Bcor^Neo allele was detected by separate PCR reaction using primer set, VBp1023/VBp981 and generates a 693 bp amplicon.</p>", "<p>VBp573 <named-content content-type=\"gene\">5′-GCCTGAAGTAGCTGACATGTCTCTGAT-3′</named-content>\n</p>", "<p>VBp981 <named-content content-type=\"gene\">5′-AAAGCCCTAGGAACTACTTGGAGGC-3′</named-content>\n</p>", "<p>VBp1023 <named-content content-type=\"gene\">5′-CTATCGCCTTCTTGACGAGTTCTTC-3′</named-content>\n</p>", "<title>Western Blot Analyses</title>", "<p>In both western blot analyses performed, mouse eyes or ES cells were incubated in lysis buffer (1× phosphate buffered saline (PBS), 10% glycerol, 0.5% Nonidet-P40, 2 mM dithiothreitol (DTT), 0.2 mM phenylmethanesulphonylfluoride (PMSF) and 1× Complete Protease Inhibitor (Roche)), sonicated, normalized by Bradford Assay and resolved on a NuPAGE 3–8% Tris-Acetate gel (Invitrogen). Proteins were transferred overnight at 4°C to a nitrocellulose membrane, blocked with non-fat dry milk and incubated with polyclonal anti-BCOR ##REF##16943429##[4]## and monoclonal anti-B-Actin antibodies (Abcam 8226).</p>", "<title>Generation of “Rescued” <italic>Bcor^Fl/+</italic> Embryonic Stem Cells</title>", "<p>To remove the <italic>Neo</italic> selection cassette and restore BCOR expression, <italic>Bcor^Neo</italic>\n^/Y ES cells were transfected with a mouse codon-optimized FLPo site specific recombinase as described by Raymond and Soriano ##REF##17225864##[47]##. Five clonally selected rescued lines in which the <italic>Neo</italic> cassette had been deleted were identified by PCR screening of gDNA with primer sets VBp573 /VBp981 and VBp1023/VBp981.</p>", "<title>Flow Cytometry and Hemoglobin Analyses</title>", "<p>White blood cells were harvested into Hank's buffered saline solution (HBSS) at 4°C from the spleens of sacrificed <italic>Bcor^Neo</italic>, <italic>Bcor^Gt</italic> and appropriate control mice. Cells were pelleted at 300×g at 4°C for this step and all subsequent centrifugation steps. Red blood cells were lysed and white blood cells were then pelleted and resuspended in a 1∶3 dilution of clone 2.4G2 hybridoma (ATCC HB-197) supernatant for 10 minutes at room temperature to block Fc receptors. Cells were then counted and 2×10⁶ cells were respuspended in 200 uL of FACs buffer (PBS, 1% FCS, and 0.02% azide, pH 7.2) in a 96 well round bottom plate for antibody staining. Cells were stained with a 1∶200 dilution of appropriate primary antibody or lectin (Anti-Ly9.1-FITC (BD Pharmingen), Anti-CD19-PE (eBioscience), Anti-CD4-PerCP (BD Pharmingen), Anti-CD8-APC (eBioscience) and PNA-biotin (BD Pharmingen)) in FACs buffer for 20 minutes at room temperature and washed 1 time with 200 uL of FACs buffer. Cells incubated with PNA-biotin were then incubated with streptavidin-APC for 20 minutes at room temperature. All samples were then washed another 2 times with FACs buffer, fixed in 1% paraformaldehyde and stored at 4°C until flow cytometry was performed on the following day. Cell events were recorded on a flow cytometer (FACSCalibur; BD Biosciences) and analyzed with FlowJo software (Tree Star Inc.). Clarified hemolysates prepared from the erythrocytes of the same mice were resolved by Triton-acid-urea gel electrophoresis and visualized by Coomassie blue staining ##REF##646104##[32]##, ##REF##9787139##[65]##\n</p>", "<title>ES Cell Culture and Differentiation</title>", "<p>ES cells were maintained in 37°C 5% CO2 incubator on irradiated mouse embryonic fibroblasts (MEFs) in Knock Out Dulbecco's Modified Eagles Medium (KO-DMEM, Invitrogen) supplemented with (15% ES cell certified fetal bovine serum (ES-FBS, Hyclone), 0.1 mM non-essential amino acids (NEAA), 2 mM L-glutamine, 0.1 mM 2-mercaptoethanol (BME), 100 U/ml penicillin, 100 µg/ml streptomycin (Invitrogen) and 1000 U/mL ESGRO (Chemicon)). Protocols for differentiation of ES cells were essentially as described by Zhang et. al. ##REF##14696336##[41]##, ##REF##14696346##[42]## with minor modifications. Briefly, ES cells were passaged one time onto 0.1% gelatinized tissue culture plates to remove MEFs. After 2 days, cells were trypsinized for 5 minutes to a single cell suspension, stopped with fetal bovine serum, counted and plated at low density (5–10×10⁴ cells/mL depending on the experiment) in embryoid body (EB) differentiation media (Iscove's Modified Dulbecco's Medium (IMDM) supplemented with 15% fetal bovine serum (Atlas Biologicals), 50 ug/mL Ascorbic Acid (Sigma), 2 mM L-glutamine, 4.5×10⁻⁴ M monothioglycerol (MTG) in biological triplicate for each sample on ultra low attachment tissue culture plates (Corning)). EBs were maintained at 37°C 5% CO2 for duration of indicated experiments. Primitive erythrocyte colony forming assays utilized EBs, as described above, incubated until day 4.5. EBs were then trypsinized to a single cells suspension, stopped with FBS, counted and replated at 5×10³ cells/mL in triplicate in primitive erythrocyte differentiation media containing (IMDM, 1% methylcellulose (Sigma), 10% plasma derived serum (Animal Technologies Inc.), 12.5 ug/mL Ascorbic Acid (Sigma), 2 mM L-glutamine (Invitrogen), 200 ug/mL transferrin (Roche), 4.5×10⁻⁴ M MTG (Sigma), 2 U/mL Erythropoietin (Epogen, Roche) and 5% protein free hybridoma media –II (Invitrogen). On day 4 secondary differentiation plates were counted for the presence of primitive erythrocyte colonies based on morphology and color.</p>", "<title>Gene Expression Analysis</title>", "<p>Total RNA was extracted with TRIzol (Invitrogen), normalized by spectroscopy and reverse transcribed using either M-MLV or SuperScript III (Invitrogen) according to the manufacturer's protocol. Real-time quantitative PCR was performed using FastStart SYBR Green Master mix (Roche) on a Stratagene M×3000P (Stratagene). Gene expression was determined using the relative quantitation method ##REF##11846609##[66]## and Hprt expression was used to normalize all sample template concentrations. Denaturing curves were performed on all reactions to verify homogeneity of the amplified product. The following gene specific primers were used.</p>", "<p>Hprt For <named-content content-type=\"gene\">5′-AGCTACTGTAATGATCAGTCAACG-3′</named-content>\n</p>", "<p>Hprt Rev <named-content content-type=\"gene\">5′-AGAGGTCCTTTTCACCAGCA-3′</named-content>\n</p>", "<p>Oct4 For <named-content content-type=\"gene\">5′-GAAGCAGAAGAGGATCACCTTG-3′</named-content>\n</p>", "<p>Oct4 Rev <named-content content-type=\"gene\">5′-TTCTTAAGGCTGAGCTGCAAG-3′</named-content>\n</p>", "<p>Nanog For <named-content content-type=\"gene\">5′-CCTCAGCCTCCAGCAGATGC-3′</named-content>\n</p>", "<p>Nanog Rev <named-content content-type=\"gene\">5′-CCGCTTGCACTTCATCCTTTG-3′</named-content>\n</p>", "<p>Brachyury For <named-content content-type=\"gene\">5′-CTCACCAACAAGCTCAATGG-3′</named-content>\n</p>", "<p>Brachyury Rev <named-content content-type=\"gene\">5′-GGTCTCGGGAAAGCAGTGGC-3′</named-content>\n</p>", "<p>Flk1 For <named-content content-type=\"gene\">5′-CACCTGGCACTCTCCACCTTC-3′</named-content>\n</p>", "<p>Flk1 Rev <named-content content-type=\"gene\">5′-GATTTCATCCCACTACCGAAAG-3′</named-content>\n</p>", "<p>Fgf5 For <named-content content-type=\"gene\">5′-CAAAGTCAATGGCTCCCACGAAG-3′</named-content>\n</p>", "<p>Fgf5 Rev <named-content content-type=\"gene\">5′-CTACAATCCCCTGAGACACAGCAAATA-3′</named-content>\n</p>", "<p>Bmp4 For <named-content content-type=\"gene\">5′-CACTGTGAGGAGTTTCCATCACGAAG-3′</named-content>\n</p>", "<p>Bmp4 Rev <named-content content-type=\"gene\">5′-GGATGCTGCTGAGGTTGAAGAGGA-3′</named-content>\n</p>", "<title>Statistical analysis</title>", "<p>P values in ##FIG##2##Figure 3## were determined using a Student's t-test, using the natural log of the percent contribution values, assuming unequal variances.</p>" ]

[ "<title>Results</title>", "<title>Generation and analysis of a conditional <italic>Bcor</italic> allele</title>", "<p>To investigate the role of the X-linked <italic>Bcor</italic> gene in development, we generated a conditional <italic>Bcor</italic> allele, <italic>Bcor^Fl</italic>, in which exon 3 is flanked by loxP sites to allow its removal via expression of Cre recombinase (##FIG##0##Figure 1A## and targeting verification ##FIG##0##Figure 1B##). We hypothesized that excision of exon 3 (<italic>Bcor</italic>\n^{Δ<italic>3</italic>}) would result in a frame shift and a premature stop codon that should cause severe carboxy-terminal deletion of the BCOR protein and/or elimination of the mRNA by nonsense-mediated decay (##FIG##0##Figure 1C##). Based on human OFCD patients we expected that heterozygous female mice carrying this deletion in all tissues would recapitulate the phenotype of human OFCD patients and that hemizygous male mice would not be viable. Unexpectedly, breeding <italic>Bcor^Fl/+</italic> mice to ß-actin-Cre mice generated apparently normal female (<italic>Bcor</italic>\n^{Δ<italic>3/+</italic>}) and male (<italic>Bcor</italic>\n^{Δ<italic>3/Y</italic>}) mice. Western blot analysis of BCOR protein from the eyes of these mice and wild type controls revealed that an alternative start codon, 3′ to the engineered frame shift, was used for translation, generating BCOR protein with a predicted 71 amino acid amino-terminal deletion (##FIG##0##Figure 1C##). Thus, although this amino-terminal region is conserved, with 69% identity to the <italic>Xenopus</italic> protein and 90% identity to that of human, it is not essential.</p>", "<title>Generation and analysis of <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> alleles in mice</title>", "<p>Although the <italic>Bcor^Fl</italic> mutation did not provide the expected null allele, during the initial targeting of the <italic>Bcor</italic> locus we fortuitously created a loss of function <italic>Bcor</italic> allele (<italic>Bcor^Neo</italic>) that provided insights into <italic>Bcor</italic> function both in mice and in ES cell differentiation. We included a Neomycin (<italic>Neo</italic>) selection cassette in the <italic>Bcor</italic> targeting vector. Often the <italic>Neo</italic> cassette is excised during mRNA splicing to generate a normal transcript. However, in some cases a proportion of mRNAs are spliced into and then out of <italic>Neo</italic>, and the insertion of <italic>Neo</italic> sequences can result in a hypomorphic or null allele ##REF##11584291##[23]##, ##REF##9635194##[24]##. Phenotypic effects of such aberrant splicing in chimeric animals would be most readily observed for X-linked genes like <italic>Bcor</italic> since XY ES cells are routinely used for gene targeting. Upon generation of chimeric mice with our <italic>Bcor^Neo/Y</italic> ES cells (##FIG##1##Figure 2A##) we observed two prominent phenotypes suggesting that the <italic>Neo</italic> cassette was indeed interfering with <italic>Bcor</italic> expression and function (##FIG##1##Figure 2B##). First, 3 out of 10 <italic>Bcor^Neo</italic> chimeras had kinked and sometimes shortened tails, suggesting a possible neural tube defect ##REF##16126904##[25]##. Second, 10 out of 10 chimeras exhibited a consistent bias in coat color contribution of the 129 (Agouti) ES cells to the sides of the bodies and legs of the mice. Since Agouti locus expressing cells in the skin are derived from the ectodermal lineage, this indicates a selective bias against <italic>Bcor^Neo/Y</italic> cells to contribute to certain ectodermal lineages. The same tail and coat color phenotype was also observed in 3 out of 3 chimeric mice that we generated from the Bay Genomics ES cell line, XE541 (<italic>Bcor^Gt</italic>) (##FIG##1##Figure 2B##). The XE541 ES cell line harbors an insertional gene trap mutation (ß-Geo) in the 6^th intron of mouse <italic>Bcor</italic>, which is predicted to result in an N-terminal BCOR-ß-Geo fusion protein (##FIG##1##Figure 2A##).</p>", "<title>Molecular analysis of <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> alleles</title>", "<p>To establish the molecular nature of the expression defects of the <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> alleles in ES cells, we examined <italic>Bcor</italic> mRNA splicing patterns by RT-PCR and proteins levels by western blot analysis. The <italic>Bcor^Neo</italic> allele ES cells had a mixture of transcripts spliced to exclude <italic>Neo</italic> (##FIG##1##Figure 2A##, left upper splicing pattern and ##SUPPL##0##Figure S1A##), which encode intact BCOR, together with transcripts spliced to incorporate the <italic>Neo</italic> cassette (##FIG##1##Figure 2A##, left lower splicing pattern and ##SUPPL##0##Figure S1A##). Sequencing of RT-PCR products demonstrated that the two regions of the Phosphoglycerate Kinase-Neomycin (Pgk-<italic>Neo</italic>) cassette were recognized as exons by the splicing machinery (##SUPPL##0##Figure S1B##). The first exon was 86 nucleotides from the Pgk promoter and the second utilized a splice acceptor site within the 5′ UTR of the <italic>Neo</italic> gene cassette and a donor site within the 3′UTR. This latter exon was identical to that used by other previously reported targeted genes ##REF##9635194##[24]##. The result of this splicing is the introduction of a premature in frame stop codon derived from the Pgk promoter sequence. Splicing into <italic>Neo</italic> predominates, as full length BCOR protein levels are not detectable by western blot analysis in the <italic>Bcor^Neo</italic> allele ES cells as compared to the parental ES cell line CJ7 (##FIG##1##Figure 2C##). We conclude that the <italic>Bcor^Neo</italic> allele is a severe loss of function allele but because we did detect some correctly spliced transcript it may not be a null allele.</p>", "<p>The transcripts from the <italic>Bcor^G</italic>\n^t allele are a mixture of those that splice over the gene trap cassette (##FIG##1##Figure 2A##, right upper splicing panel and ##SUPPL##0##Figure S1B##) which encode intact BCOR and those that splice into the gene trap ß-Geo cassette, truncating the gene (##FIG##1##Figure 2A##, right, lower splicing pattern and ##SUPPL##0##Figure S1B##). By western blot analysis the large BCOR-ß-Geo fusion protein and multiple breakdown products are observed (##FIG##1##Figure 2C##). Although a faint band corresponding to full length BCOR is present this may represent a co-migrating BCOR-ß-Geo degradation product. The BCOR-ß-Geo fusion protein lacks the region required for interaction with the PcG proteins ##REF##16943429##[4]## and therefore is predicted to lack transcriptional repression activity. Although the fusion protein appears to be unstable, the total amount of protein is significantly higher than the level of intact BCOR found in the parental ES cell line E14 (##FIG##1##Figure 2C##). Consistent with this we found, using quantitative-RT-PCR, that the <italic>Bcor-ß-Geo</italic> fusion transcript level is 10-fold higher than the level of intact <italic>Bcor</italic> transcript found in the parental ES cell line. This suggests that intact BCOR negatively autoregulates its own transcription while the BCOR-ß-Geo fusion is unable to function in this capacity. In summary, both the <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> alleles represent severe loss of function mutations and result in kinked and shortened tails and a coat color bias in chimeric animals.</p>", "<title>Parent-of-origin effect of <italic>Bcor^Neo</italic> allele</title>", "<p>To further characterize the effect of these two loss of function mutations in mice we bred the male chimeras to wild type females to obtain germline transmission of the alleles. This was only successful with the <italic>Bcor^Neo</italic> allele and we were able to produce <italic>Bcor^Neo/+</italic> females. (Because <italic>Bcor</italic> is X-linked, <italic>Bcor^Neo/Y</italic> males cannot be generated from male chimeras.) Germline transmission of the <italic>Bcor^Gt</italic> allele did not occur and there are at least two possible explanations for this. First, the <italic>Bcor^Gt</italic> allele may not be compatible with germline development. Second, since only three <italic>Bcor^Gt</italic> chimeric mice were obtained and bred, it is possible that upon generation of more chimeras germline transmission would occur. We bred the female <italic>Bcor^Neo/+</italic> mice extensively but, in contrast to the chimeric males, they never produced any <italic>Bcor^Neo/+</italic> offspring (##FIG##1##Figure 2A and D##). (<italic>Bcor^Neo/Y</italic> males, based on the lack of male OFCD patients, are predicted to die during embryonic development.) However, we were able to breed the female <italic>Bcor^Neo/+</italic> mice to FLPe recombinase expressing mice to remove <italic>Neo</italic>, and got viable <italic>Bcor^Fl/+</italic> and <italic>Bcor^Fl/Y</italic> progeny (##FIG##1##Figure 2A##). In mice, X chromosome inactivation differs between the embryo and the extraembryonic tissue. The embryo undergoes random X-inactivation, in contrast to the extraembryonic tissue where the paternally-derived X chromosome is inactivated ##REF##1152998##[26]##. The female lethality observed, arising when the mutant X-linked allele comes from the mother, likely results from paternally-imprinted X-inactivation of the wild type <italic>Bcor</italic> allele in extra-embryonic tissue of the female offspring (##FIG##1##Figure 2D## and see discussion). Thus, our results are consistent with a requirement for <italic>Bcor</italic> in extraembryonic tissue where <italic>Bcor</italic> is normally highly expressed ##REF##17344103##[22]##. The practical consequence of this parent-of-origin effect was that we could not generate extensive numbers of <italic>Bcor^Neo/+</italic> mice and thus phenotypic analysis was limited to the <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> allele-containing chimeras and the remaining six aging female <italic>Bcor^Neo/+</italic> mice that were generated from the two germline-transmitting male chimeras we obtained.</p>", "<title>Effect of <italic>Bcor^Neo</italic> allele on eye development</title>", "<p>In OFCD congenital cataracts have been described in all patients. We examined the eyes of the remaining <italic>Bcor^Neo/+</italic> mice when they ranged in age from 7 to 22 months. Using a slit lamp, the mice were examined for lens opacification, which is indicative of catararacts. Eight out of twelve eyes (67%) in the six <italic>Bcor^Neo/+</italic> mice displayed lens opacification. Although we do not know the age of onset, two of <italic>Bcor^Neo/+</italic> mice that were 7 months of age had bilateral lens opacification. We did not have aged matched C57Bl/6∶129 mixed background control animals. However, the frequency of cataracts in the <italic>Bcor^Neo/+</italic> mice was substantially higher than that reported for either C57Bl/6 or 129 inbred stains (C57Bl/6 is 25% at 14 months ##REF##10870527##[27]## and 129 0% at 13–17 months, as detected by slit lamp analysis ##REF##17943656##[28]##). Over seventy percent of OFCD patients have microphthalmia. By visual inspection we only observed one <italic>Bcor^Neo</italic> allele-containing chimera with obvious microphthalmia. However, we did not perform postmortem globe measurements so we cannot rule out the possibility that additional mice had more subtle microphthalmia.</p>", "<title>Analysis of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> cell contribution to hematopoietic development</title>", "<p>In OFCD patients 90–100% of surviving white blood cells show inactivation of the X chromosome carrying the mutant allele of <italic>Bcor</italic>, indicating that cells expressing the mutant allele of <italic>Bcor</italic> are at a selective disadvantage ##REF##15004558##[18]##. Because the majority of white blood cells are neutrophils and T cells (and to a lesser extent B cells) this strongly suggests that <italic>Bcor</italic> plays an important role in the development of these hematopoietic lineages.</p>", "<p>We hypothesized that the loss of function mutations in the <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cell would result in reduced or blocked contribution to hematopoietic lineages in our chimeric mice. The contributions of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> 129 strain ES cells to splenic B and T cells of 129/B6 chimeric animals were analyzed by fluorescence activated cell sorting (FACS). We used the 129 strain-specific B and T cell surface antigen Ly9.1 ##UREF##1##[29]##, ##REF##10970093##[30]##, the commonly used T-cell markers CD4 and CD8, and the B-cell marker CD19. We found that Ly9.1 positive cells can contribute to CD4, CD8 and C19 positive hematopoietic populations in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> chimeric mice (##FIG##2##Figure 3A and B##). The 129 contribution to each hematopoietic cell type tested was reduced relative to two control chimeras in three <italic>Bcor^Neo/Y</italic> and three <italic>Bcor^Gt/Y</italic> chimeric mice. CD4+, CD8+ and CD19+ were reduced compared to control chimeras on average by 87%, 86% and 85%, respectively. Thus, the <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cells can contribute to the B and T lineages but they appear to be at a selective disadvantage, as in OFCD patients, suggesting that <italic>Bcor</italic> is important for murine B and T cell development.</p>", "<p>Because BCOR is a known corepressor of BCL6, an important regulator of germinal center formation in B-cell maturation, we hypothesized that <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> cell contribution to germinal center formation might be blocked or reduced in chimeras. To test this, we looked for the presence of PNA, a germinal center marker, in the CD19+ and Ly9.1+ cell population in cells harvested from the spleens of both mutant chimeric lines. Similar to the B and T cell analysis, we observed an 87% reduction compared to control chimeras in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> contribution to CD19+ PNA+ cells suggesting a compromised germinal center reaction (##FIG##2##Figure 3B##). To assess the ability of <italic>Bcor^Gt/Y</italic> and <italic>Bcor^Neo/Y</italic> ES cells to contribute to adult erythrocyte lineages in chimeras, a hemoglobin analysis was performed, relying on genotype differences between host and ES-derived cells. Erythrocytes derived from the 129 genetic background express ß-major and ß-minor globin, while B6 erythrocytes express ß-single globin ##REF##284410##[31]##, ##REF##646104##[32]##. We found that <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cells can contribute to mature red blood cells but the contribution is severely reduced compared to the controls (##FIG##2##Figure 3C##).</p>", "<title>Analysis of <italic>in vitro</italic> differentiation potential of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cells</title>", "<p>The absence of OFCD males indicates an early and essential role for <italic>Bcor</italic> in human development. The parent-of-origin effect strongly suggests that <italic>Bcor</italic> plays an essential role in extraembryonic tissue but <italic>Bcor</italic> may play additional roles earlier in development. We used our two loss of function ES cell lines in <italic>in vitro</italic> differentiation assays to test whether <italic>Bcor</italic> plays a role in ES cell differentiation. ES cells cultured in suspension in the absence of leukemia inhibitory factor (LIF) undergo primary differentiation after 2 to 4 days to form simple embryoid bodies (EBs) ##REF##3897439##[33]##, ##REF##8608017##[34]##. These contain an outer endoderm layer surrounding an inner cell mass. Around day 4, differentiation of columnar epithelium with a basal lamina and formation of a central cavity occurs. Such cystic EBs bear similarities to the egg cylinder-stage embryos ##REF##7585945##[35]##–##REF##1055416##[37]##. By 6 days of differentiation, EBs are comparable to early organogenesis-stage embryos (E7.5) ##REF##10368935##[38]##. EB cells can be dispersed and induced with appropriate growth conditions to undergo secondary differentiation into particular cell lineages ##REF##8608017##[34]##, ##REF##15905405##[39]##.</p>", "<p>Both the severe skewing of X-inactivation seen in OFCD patients and the limited contribution of <italic>Bcor</italic> mutant ES cells to hematopoietic lineages in our chimeric animals indicate that <italic>Bcor</italic> plays a role in hematopoiesis. Primitive erythrocytes are the earliest hematopoietic cell types to form <italic>in vivo</italic>\n##REF##8608017##[34]##, ##REF##8417345##[40]##. Therefore, to determine whether <italic>Bcor</italic> plays an early role in hematopoiesis, we first examined the ability of dispersed embryoid bodies from our <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cell lines to differentiate into primitive erythrocyte (EryP) colonies in the presence of appropriate factors ##REF##14696336##[41]##, ##REF##14696346##[42]##. We found EryP colony formation with CJ7 derived <italic>Bcor^Neo/Y</italic> and E14 derived <italic>Bcor^Gt/Y</italic> cell lines is reduced relative to the matched parental ES cell lines (98% and 97% respectively, ##FIG##3##Figure 4A##).</p>", "<p>We next used quantitative real-time polymerase chain reaction (qRT-PCR) to examine expression of Flk1 (KDR), an early marker of hematopoietic differentiation potential that is required for primitive and definitive hematopoiesis ##REF##16140143##[43]##, ##REF##9200616##[44]##. RNA samples were collected from ES cells that were allowed to differentiate into embryoid bodies for 0 to 5.5 days. We found that the induction of Flk1 expression in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> cell lines is delayed by one day relative to the parental cell lines and in the time course of the experiment Flk1 expression only reaches ∼40% of the peak expression seen in the parental cell lines (##FIG##3##Figure 4B##). Moving back in developmental time, we examined expression of Brachyury (T), a marker for the entire mesodermal lineage which in combination with Flk1 marks the hemangioblast cell population ##REF##16140143##[43]##. We saw a delay in expression of about 1–2 days but expression does eventually peak at levels comparable to the parental ES cell lines (##FIG##3##Figure 4C##). Finally, knowing that <italic>Bcor</italic> is expressed in ES cells, we examined the expression of Oct3/4 (Pou5f1), a marker of ES cell pluripotency ##REF##7958450##[45]##, ##REF##9814708##[46]##, and found a one to two day delay in repression of Oct3/4 (##FIG##3##Figure 4D##).</p>", "<p>To confirm that these effects on ES cell differentiation are the result of the <italic>Neo</italic> insertion in <italic>Bcor</italic>, we removed the <italic>Neo</italic> cassette from the <italic>Bcor^Neo</italic> ES cell line. We generated three independent clonal <italic>Bcor^Fl/Y</italic> ES cell lines via expression of Flpo recombinase ##REF##17225864##[47]## in <italic>Bcor^Neo</italic> ES cells. We found that these three <italic>Bcor^Fl/Y</italic> “rescued” lines restore the timing of expression of various ES cell and differentiation markers to that of the parental wild type ES cells (##FIG##4##Figure 5##) demonstrating that the effects on differentiation were due to the interruption of <italic>Bcor</italic> by the <italic>Neo</italic> cassette. Included in the analysis were a second ES cell marker, <italic>Nanog</italic>, the primitive ectoderm marker, <italic>Fgf5</italic>, ##REF##10049358##[48]##, ##REF##7657163##[49]## and the extraembryonic ectoderm marker, <italic>Bmp4</italic>\n##REF##7585945##[35]##, ##REF##1794310##[50]##, ##REF##1794311##[51]##. In the <italic>Bcor^Neo</italic> cells we observed a delay in the downregulation of the ES cell markers <italic>Oct3/4</italic> and <italic>Nanog</italic>, whereas <italic>Fgf5</italic> was still up regulated appropriately but its down regulation was delayed by two days. Subsequently, activation of <italic>Bmp4</italic>, <italic>Brachyury</italic> and <italic>Flk1</italic> were all delayed by one to two days. For the genes that are normally expressed earlier in differentiation, the rescued lines restore both the timing and the level of gene expression to that of the parental cell line. However with <italic>Brachyury</italic> and <italic>Flk1</italic>, even though the normal timing of gene expression was restored, inexplicably the level was increased beyond the parental line. We also compared the proliferation rates and cell cycle profile of the parental CJ7 ES cell line with the <italic>Bcor^Neo/Y</italic> and the three rescued <italic>Bcor^Fl/Y</italic> ES cell lines and detected no significant differences (data not shown).</p>", "<p>Together these ES cell differentiation and expression studies indicate that <italic>Bcor</italic> plays a role in the regulation of gene expression very early in ES cell differentiation and in the mesodermal and ectodermal lineages.</p>" ]

[ "<title>Discussion</title>", "<p>In this study we have used a combination of mouse molecular genetics and ES cell differentiation to identify roles of <italic>Bcor</italic> in early development. These studies provide insights into the possible causes of male embryonic lethality in OFCD, the hematopoietic phenotype of female OFCD patients and the parent-of-origin effect in mice. We have two major findings, as discussed below.</p>", "<p>First, we found that in mice the mutant <italic>Bcor^Neo</italic> allele exhibits a parent-of-origin effect, which together with the high level of <italic>Bcor</italic> expression in extraembryonic tissue ##REF##17344103##[22]## strongly suggest an essential requirement for <italic>Bcor</italic> in extraembryonic tissue. The X chromosome has been implicated in causing several malformations of the placenta ##REF##12900566##[52]## and our work suggests that <italic>Bcor</italic> is likely to be one of the genes responsible.</p>", "<p>There are two alternative but less likely explanations for the parent-of-origin effect that we have considered. First, it is possible that <italic>Bcor</italic> is required very early in development (beginning as early as the 4 to 8 cell stage) when the whole embryo is subject paternal X chromosome inactivation. Once the inner cell mass has formed the cells of the embryo proper are subject to random X inactivation. Paternal X-inactivation is maintained and becomes more complete in the trophectoderm and primitive endoderm, which go on to form to the extraembryonic tissue. However, given the incomplete nature of the paternal imprinting during the 4 cell to blastocyst stage it is most likely that sufficient levels of wild type <italic>Bcor</italic> are transcribed to support development.</p>", "<p>A second, albeit unlikely, alternative explanation for the parent-of-origin effect observed in the mouse is that <italic>Bcor</italic> is imprinted such that only the maternal allele is expressed in the embryo proper. However, if such imprinting ##REF##15908953##[53]##–##REF##15820547##[55]## occurred in humans, the mother to daughter transmission seen in OFCD patients ##UREF##2##[56]##, ##REF##16829040##[57]## could not take place since the daughter would only express the mutant <italic>Bcor</italic> allele and would be equivalent to the inviable hemizygous males.</p>", "<p>Our second major finding was that in ES cell differentiation studies <italic>Bcor</italic> is required for formation of primitive erythrocytes and proper expression of genes that regulate ES cell pluripotency and genes that drive ectodermal and mesodermal development. These defects together with the apparent limited contribution of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> cells to B and T cells and adult erythrocytes may provide an explanation for the severe skewing of X-inactivation seen in peripheral white blood cells of OFCD patients and indicate a critical early role for <italic>Bcor</italic> in helping to establish hematopoietic development. In data not included, we do not see significant changes in expression of the few endoderm specific genes we tested. Since the number of genes tested was not exhaustive, Bcor may be affecting the expression of other genes important for endoderm development that are yet unidentified.</p>", "<p>One of the genes that we found to be misregulated in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cell differentiation assays was <italic>Brachyury</italic>. Mice heterozygous for the <italic>Brachyury</italic> T mutation exhibit a variable short-tailed phenotype ##REF##8436292##[58]## while other mutations such as <italic>Brachyury</italic> T¹³⁷ have a kinked or bent tail ##REF##10723731##[59]##. Thus mis-expression of <italic>Brachyury</italic> is a possible explanation for the kinked and shortened tails we observed in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> chimeras. The extraembryonic ectoderm marker, <italic>Bmp4</italic>, was also misregulated in <italic>Bcor^Neo/Y</italic> ES cell differentiation assays. <italic>Bmp4</italic> knock out mice do not express <italic>Brachyury</italic> and display impaired mesoderm differentiation, including reduced extraembryonic mesoderm blood island formation and disorganized posterior structures, further implicating <italic>Bcor</italic> in mesoderm development ##REF##7657163##[49]##.</p>", "<p>Polycomb group proteins (PcG) have been shown to play a central role in stem cell maintenance and lineage specification ##REF##17481880##[60]##. The BCOR complex contains several PcG proteins, including RNF2, an H2A ubiquitin ligase, and NSPC1, a BMI1 homolog ##REF##16943429##[4]##. RNF2 has been shown to be present at over 1,200 targets in ES cells ##REF##16625203##[61]##. At many genes, the PcG complex, PRC2, serves to recruit RNF2 containing complexes. However, about one quarter of ES cell RNF2 targets are not co-occupied with PRC2 components, suggesting RNF2 can be recruited via other means. We hypothesize that BCOR, via interaction with site specific transcription factors, can recruit RNF2 and other BCOR complex components to regulate expression of genes in ES cells and their differentiating progeny. Recently, NSPC1 was identified as a protein that when over-expressed can partially rescue the phenotypic properties of undifferentiated mouse ES cells under differentiation-inducing conditions ##REF##16621925##[62]##. Over-expression of NSPC1 may be sequestering or diluting the activity of the Bcor complex thereby preventing full transcriptional repression of target genes. This, together with our data, suggests that inappropriate expression of BCOR complex components affects ES cell differentiation.</p>", "<p>Our studies suggest two possible causes of male embryonic lethality in OFCD. First, inappropriate regulation of key developmental genes in <italic>Bcor^−/Y</italic> embryos may cause lethality before implantation. Second, male embryonic lethality may be due to improper development of extraembryonic derived tissue, potentially resulting in placental failure or incomplete chorioallantoic fusion. Conditional inactivation of Bcor in a spatial and temporal manner will be critical in future experiments in order to bypass the strict requirement for Bcor function in early development.</p>" ]

[]

[ "<p>Conceived and designed the experiments: JAW CMC AMK VJB. Performed the experiments: JAW CMC AMK. Analyzed the data: JAW CMC VJB. Wrote the paper: JAW AMK VJB.</p>", "<p>\n<italic>Bcor</italic> (BCL6 corepressor) is a widely expressed gene that is mutated in patients with X-linked Oculofaciocardiodental (OFCD) syndrome. BCOR regulates gene expression in association with a complex of proteins capable of epigenetic modification of chromatin. These include Polycomb group (PcG) proteins, Skp-Cullin-F-box (SCF) ubiquitin ligase components and a Jumonji C (Jmjc) domain containing histone demethylase. To model OFCD in mice and dissect the role of Bcor in development we have characterized two loss of function <italic>Bcor</italic> alleles. We find that Bcor loss of function results in a strong parent-of-origin effect, most likely indicating a requirement for <italic>Bcor</italic> in extraembryonic development. Using <italic>Bcor</italic> loss of function embryonic stem (ES) cells and in vitro differentiation assays, we demonstrate that <italic>Bcor</italic> plays a role in the regulation of gene expression very early in the differentiation of ES cells into ectoderm, mesoderm and downstream hematopoietic lineages. Normal expression of affected genes (<italic>Oct3/4</italic>, <italic>Nanog</italic>, <italic>Fgf5</italic>, <italic>Bmp4</italic>, <italic>Brachyury</italic> and <italic>Flk1</italic>) is restored upon re-expression of <italic>Bcor</italic>. Consistent with these ES cell results, chimeric animals generated with the same loss of function <italic>Bcor</italic> alleles show a low contribution to B and T cells and erythrocytes and have kinked and shortened tails, consistent with reduced <italic>Brachyury</italic> expression. Together these results suggest that <italic>Bcor</italic> plays a role in differentiation of multiple tissue lineages during early embryonic development.</p>" ]

[ "<title>Supporting Information</title>" ]

[ "<p>We thank members of the Zarkower and Bardwell labs for many helpful discussions; Anna Petryk and Lisa Schimmenti for help in initially characterizing our chimeric mice; Erick Mendenhall and Meri Firpo for insights into differentiating mouse embryonic stem cells; Laura Bursch, Troy Baldwin and Keli Hippen for help with immunological analyses; and Eric Russel and Zhennig He for help with hemoglobin analyses. We thank the University of Minnesota Mouse Genetics Laboratory for ES cell injection.</p>" ]

[ "<fig id=\"pone-0002814-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0002814.g001</object-id><label>Figure 1</label><caption><title>Generating <italic>Bcor</italic>\n^Δ\n^{<italic>3</italic>/+} and <italic>Bcor</italic>\n^Δ\n^{<italic>3/Y</italic>} mice.</title><p>(A) Diagram of targeting construct and recombination strategy. The <italic>Bcor⁺</italic> allele was targeted by homologous recombination to generate <italic>Bcor^Neo</italic> embryonic stem cells. This allele contains loxP sites flanking the third exon, and a <italic>Neo</italic> resistance cassette flanked by frt sites, internal to the loxP sites immediately downstream of exon 3. Expression of Flp and/or Cre site specific recombinases <italic>in vivo</italic> or <italic>in vitro</italic> can be used to excise the intervening sequence between their respective recognition sites, frt and loxP, generating <italic>Bcor^Fl</italic> and <italic>Bcor</italic>\n^{Δ<italic>3</italic>} alleles. PGK-Neo (phosphoglycerate kinase promoter–driven neomycin resistance gene cassette) (B) Southern blot and PCR analyses of targeted and wild type ES cell clones. Genomic DNA from individual ES cell clones was digested with the restriction enzymes, <italic>BamH1</italic> (left arm) and <italic>BglII</italic> (right arm). The 5′ external probe (Left Arm) hybridizes to 10.7-kb (<italic>Bcor^+/Y</italic>) and 9.5-kb (<italic>Bcor^Neo/Y</italic>) <italic>BamH1</italic> fragments. Given the weak left arm signal, confirmatory PCR was performed on genomic DNA from wild type and targeted ES cells using a forward primer 5′ to the left arm and a reverse primer in the neomycin coding sequence. The 3′ external probe (Right Arm) hybridizes to 12.3-kb (<italic>Bcor^+/Y</italic>) and 10.5-kb (<italic>Bcor^Neo/Y</italic>) <italic>BglII</italic> fragments. (C) The <italic>Bcor</italic>\n^{Δ<italic>3</italic>} allele encodes an N-terminally truncated version of BCOR. The anticipated and actual translation open reading frames for the <italic>Bcor</italic>\n^{Δ<italic>3</italic>} allele are diagrammed (top). Western blot analysis on the eyes of <italic>Bcor^+/+</italic>, <italic>Bcor</italic>\n^{Δ<italic>3</italic>}/+ and <italic>Bcor</italic>\n^{Δ<italic>3/Y</italic>} mice (bottom) reveals the presence of an N-terminally truncated version of BCOR (red arrows) in both the <italic>Bcor</italic>\n^{Δ<italic>3</italic>}/+ and <italic>Bcor</italic>\n^{Δ<italic>3/Y</italic>} lanes, migrating at a slightly lower molecular weight in comparison to wild type BCOR (black arrows).</p></caption></fig>", "<fig id=\"pone-0002814-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0002814.g002</object-id><label>Figure 2</label><caption><title>Transmission of the <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> alleles in mice.</title><p>(A) Diagram of <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> allele transmission. Location of each genomic alteration is shown, respective to the <italic>Bcor</italic> locus (top). The splicing pattern of each allele is indicated with the more predominant splicing pattern drawn below each allele in a darker line. See ##SUPPL##0##Supplementary Figure 1## for detailed analysis. The mating schemes for <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> mouse lines are depicted below the genomic alterations. (B) <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> chimeric mice and <italic>Bcor^Neo/+</italic> mice display similar phenotypes. Coat color contribution bias (orange dashed outline) and malformed kinked tail phenotype (black arrows) were present in both <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> chimeric mouse lines. <italic>Bcor^Neo/+</italic> offspring of <italic>Bcor^Neo</italic> chimeric fathers also displayed the malformed kinked tail phenotype. (C) Western blot analysis of CJ7 (<italic>Bcor^Neo/Y</italic> parent), <italic>Bcor^Neo/Y</italic>, <italic>Bcor^Fl/Y</italic> (rescued line), E14 (<italic>Bcor^Gt/Y</italic> parent) and <italic>Bcor^Gt/Y</italic> ES cell lines. The top panel was probed with an anti-BCOR antibody and anti-ACTIN antibody was used as a loading control on the bottom panel. BCOR protein is undetectable in <italic>Bcor^Neo/Y</italic> ES cells. <italic>Bcor^Fl/Y</italic> rescued ES cells show wild type levels of BCOR. <italic>Bcor^Gt/Y</italic> ES cells show a large BCOR-ß-Geo fusion protein with multiple breakdown products. (D) Parent-of-origin effect of <italic>Bcor^Neo</italic> allele. Matings between <italic>Bcor^Neo/+</italic> females (offspring of <italic>Bcor^Neo</italic> male chimeras) and wild type males do not produce any offspring carrying the <italic>Bcor^Neo</italic> allele. In mice, the paternal X chromosome is selectively inactivated in extraembryonic tissue whereas in the embryo proper X inactivation occurs randomly (grey text indicates paternally imprinted X inactivation). The female lethality observed likely results from paternally imprinted X-inactivation of the wild type <italic>Bcor</italic> allele in extraembryonic tissue.</p></caption></fig>", "<fig id=\"pone-0002814-g003\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0002814.g003</object-id><label>Figure 3</label><caption><title>Analyses of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> cell contribution to hematopoietic development in chimeric mice.</title><p>(A) The <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> 129 strain derived ES cells can contribute to the B and T lineages. White blood cells from the spleens of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> and appropriate controls were stained with antibodies to the 129 strain specific surface antigen Ly9.1 and the T cell markers CD4 and CD8 and B cell marker CD19. Cells were analyzed by flow cytometry and results are displayed as density plots with black boxes gating T or B cell populations of 129 origin. B6 and 129 strain wild type mice control for Ly9.1 specificity. CJ7 and E14 129 mouse strain derived ES cells were used to generate control chimera 1 (Con. Ch. 1) and control chimera 2 (Con. Ch. 2). Flow cytometry analyses were performed on chimeric control mice and on three mice for each mutant chimeric line, of which an example scatter plot is displayed for each. (B) Compared to chimeric controls, <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> 129 strain derived ES cell contribution to B and T cells is reduced. Graphs of scatter plot data show 129 strain contributions to each hematopoietic lineage. Black bars indicate the average contribution of control chimeric mice 1 and 2 (Con, black +) and the combined average of three <italic>Bcor^Neo</italic> (Neo, red +) and three <italic>Bcor^Gt</italic> (Gt, blue +) mice. Contribution to germinal center B cells is also reduced compared to controls as indicated by the final plot showing the PNA+ (peanut agglutinin; germinal center marker) percentage of CD19+ 129 derived B cells. P values were ≤0.05 for all hematopoietic cell populations tested (CD4; 0.004, CD8; 0.034, CD19; 0.023 and PNA; 0.001). (C) <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cells can contribute to adult erythrocytes in chimeric mice but contribution is reduced in comparison to controls. Coomassie stained gel of hemolysates from erythrocytes of mice described above in panels A and B. <italic>Bcor^Neo</italic> and <italic>Bcor^Gt</italic> chimeric mice (Neo 1, 2 and 3 and Gt 1, 2 and 3) primarily express B6 derived ß -single globin where as chimeric control mice 1 and 2 primarily express 129 derived ß -major and ß -minor globin. The hemoglobin control lane (Hbb Con) was prepared from samples that contain ß -single, ß -major and ß -minor globin.</p></caption></fig>", "<fig id=\"pone-0002814-g004\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0002814.g004</object-id><label>Figure 4</label><caption><title>\n<italic>In vitro</italic> differentiation of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cell lines.</title><p>(A) <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cell lines display a reduced ability to form primitive erythrocyte colonies in comparison to control CJ7 and E14 parental lines. ES cell were differentiated into embryoid bodies for 4.5 days, dispersed and replated in methylcellulose primitive erythrocyte colony forming assays. Error bars represent the standard error of the mean for each ES cell line differentiated (CJ7 and <italic>Bcor^Neo/Y</italic>; n = 3, E14 and <italic>Bcor^Gt/Y</italic>; n = 5). (B–D) qRT-PCR analyses of gene expression were performed on embryoid body differentiations (0–5.5 days) of <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> ES cell lines and respective parental control lines CJ7 and E14 (<italic>Bcor^+/Y</italic>). Differentiations were performed in triplicate excluding the day 0 time point. Error bars represent the standard deviation within each time point of the samples analyzed. (B) The expression of hematopoietic lineage marker, <italic>Flk1</italic>, is delayed approximately 1 day in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> differentiating embryoid bodies. (C) The expression of mesoderm and early hematopoietic lineage marker <italic>Brachyury</italic> is similarly delayed 1–2 days in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> differentiating embryoid bodies. (D) The repression of pluripotency marker, <italic>Oct3/4</italic>, is delayed 1–2 days in <italic>Bcor^Neo/Y</italic> and <italic>Bcor^Gt/Y</italic> differentiating embryoid bodies.</p></caption></fig>", "<fig id=\"pone-0002814-g005\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0002814.g005</object-id><label>Figure 5</label><caption><title>\n<italic>In vitro</italic> differentiation of <italic>Bcor^Fl/Y</italic> “rescued” ES cell lines.</title><p>(A–C) <italic>Bcor^Fl/Y</italic> “rescued” ES cell lines were created through removal of the <italic>Neo</italic> selection cassette via Flpo recombinase <italic>in vitro</italic>. <italic>Bcor^Fl/Y</italic> ES cells were clonally selected, screened for <italic>Neo</italic> removal and assessed for rescue of BCOR protein expression by western blot (##FIG##1##Figure 2C##). qRT-PCR analyses of gene expression during embryoid body differentiation (0–6.5 days) of <italic>Bcor^+/Y</italic> (CJ7, wild type parent line), <italic>Bcor^Neo/Y</italic> and three independent <italic>Bcor^Fl/Y</italic> “rescued” ES cell lines. Differentiations were performed in triplicate excluding the day 0 time point. <italic>Bcor^Fl/Y</italic> results are the average of 3 independent <italic>Bcor^Fl/Y</italic> “rescued” lines each differentiated in triplicate. (A) The repression of pluripotency markers, <italic>Oct3/4</italic> and <italic>Nanog</italic>, is restored in <italic>Bcor^Fl/Y</italic> differentiating embryoid bodies in comparison to <italic>Bcor^Neo/Y</italic>. (B) Repression of the primitive ectoderm marker, <italic>Fgf5</italic>, and expression of extraembryonic ectoderm marker, <italic>Bmp4</italic>, is restored in <italic>Bcor^Fl/Y</italic> differentiating embryoid bodies. (C) Timing of expression of the early mesoderm and hematopoietic lineage markers, <italic>Brachyury</italic> and <italic>Flk1</italic>, is restored in <italic>Bcor^Fl/Y</italic> differentiating embryoid bodies; however expression levels are restored to a higher magnitude with respect to <italic>Bcor^+/Y</italic> controls (CJ7, wild type parent line).</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"pone.0002814.s001\"><label>Figure S1</label><caption><p>Analysis of splicing pattern in <italic>BcorNeo/Y</italic> and <italic>BcorGt/Y</italic> ES cells. (A) Clone 2B1 ES cells, containing the <italic>BcorNeo/Y</italic> allele, aberrantly splice two portions of the Pgk-Neo coding sequence into the <italic>Bcor</italic> transcript. S1–S4 shows the sequence surrounding the splice junctions as determined by sequencing of reverse transcription PCR products generated by primers A–D. Amplicon AB and CD were sequenced to determine splice junctions S1–S3 and S4, respectively. The predominant splice pattern found is indicated by bolded black splice connector lines. VBp1114 (A) <named-content content-type=\"gene\">ATGCTTTCTGCAACCCCTCTGTAT</named-content>, VBpNeoRev (B) <named-content content-type=\"gene\">TCGGCAGGAGCAAGGTGAGAT</named-content>, VBpNeoFor (C) <named-content content-type=\"gene\">CCGGTTCTTTTTGTCAAGACCG</named-content>, VBp1116 (D) <named-content content-type=\"gene\">TTGTATCCCAGGCGGTGTTTTG</named-content>. (B) Clone XE541 ES cells, containing the <italic>BcorGt/Y</italic> allele, predominantly splice from exon 6 of <italic>Bcor</italic> into the splice acceptor of the genetrap cassette. Reverse transcription PCR of XE541 ES cell total RNA using primers F and G, generates the expected amplicon of 529 base pairs. Wild type splicing can also be detected in XE541 ES cells as shown by reverse transcription PCR using primers E and H to generate the expected amplicon of 445 base pairs. The predominant splice pattern found is indicated by bolded black splice connector lines. VBp1616 (E) <named-content content-type=\"gene\">CGTGCAATGATGCGCTTCTC</named-content>, VBp1074 (F) <named-content content-type=\"gene\">AGATTCCAGTCAGCTCAGCCGAGA</named-content>, VBp1071 (G) <named-content content-type=\"gene\">ATTCAGGCTGCGCAACTGTTGGG</named-content>, VBp1617 (H) <named-content content-type=\"gene\">CTTTGGAGATCCGTCTTCGCTT</named-content>.</p><p>(0.97 MB DOC)</p></caption></supplementary-material>" ]

[ "<fn-group><fn fn-type=\"COI-statement\"><p><bold>Competing Interests: </bold>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p><bold>Funding: </bold>Funding for this work was provided by NCI (R01 CA071540), NIH (T32DE07288) and the University of Minnesota Cancer Center. The funding agencies had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"pone.0002814.g001\"/>", "<graphic xlink:href=\"pone.0002814.g002\"/>", "<graphic xlink:href=\"pone.0002814.g003\"/>", "<graphic xlink:href=\"pone.0002814.g004\"/>", "<graphic xlink:href=\"pone.0002814.g005\"/>" ]

[ "<media xlink:href=\"pone.0002814.s001.tif\"><caption><p>Click here for additional data file.</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p><italic>Myxococcus xanthus</italic> is a Gram-negative bacterium that has a complex life cycle that involves vegetative swarming, predation and fruiting body formation (##REF##11207714##Reichenbach, 1999##; ##REF##10547700##Shimkets, 1999##; ##REF##15040179##Kaiser, 2003##; ##REF##15487930##2004##; ##REF##16885457##Berleman et al., 2006##). These behaviours require motility on solid surfaces: S-motility powered by Type IV pili (##REF##8748037##Wu and Kaiser, 1995##) moves cells in groups and A-motility powered by unidentified motor proteins and putative adhesion complexes moves single cells (##REF##17653507##Mignot, 2007##). To achieve directed movements, <italic>M. xanthus</italic> cells periodically reverse so that the leading pole becomes the lagging pole. The frequency of cell reversals is controlled by the <italic>frz</italic> chemosensory pathway and is important for directed cell movements (##REF##3936045##Blackhart and Zusman, 1985##). <italic>frz</italic> mutants (Δ<italic>frzA</italic>, Δ<italic>frzB,</italic> Δ<italic>frzCD,</italic> Δ<italic>frzE,</italic> Δ<italic>frzF</italic>) rarely reverse and are therefore defective in swarming and fruiting body formation, forming frizzy aggregates instead of fruiting bodies on starvation media (##REF##6281244##Zusman, 1982##). In contrast, some constitutively signalling <italic>frz</italic> mutants hyper-reverse, forming very compact colonies with little cell spreading (##REF##3936045##Blackhart and Zusman, 1985##; ##REF##15387825##Bustamante et al., 2004##).</p>", "<p>Frz proteins are homologous to bacterial chemotaxis proteins (##REF##2492105##McBride et al., 1989##; ##REF##17922045##Zusman et al., 2007##). FrzCD, a cytoplasmic methyl-accepting chemotaxis protein (MCP) homologue, lacks the transmembrane and periplasmic domains common to most MCPs and has in its place a unique N-terminal domain. In contrast, the C-terminal domain of FrzCD is similar to other MCPs and contains potential methylation sites (##REF##16359317##Astling et al., 2006##). <italic>In vitro</italic> analysis has shown that FrzCD interacts with FrzE, a histidine kinase (CheA)-response regulator fusion protein by means of two CheW-like proteins, FrzA and FrzB (##UREF##0##Astling, 2003##). When stimulated, FrzE autophosphorylates and transfers a phosphoryl group to the dual response regulator FrzZ, triggering cell reversals for both the A- and S-motility systems (##REF##17581122##Inclán et al., 2007##).</p>", "<p>It is not known how FrzCD receives signals as it lacks the usual signal-binding domain common to most MCPs. Furthermore, <italic>frzCD</italic> N-terminal domain deletion mutants show only minor defects in behaviour (##REF##15387825##Bustamante et al., 2004##). Previously, it was hypothesized that a signal input to the Frz pathway may involve differential methylation of particular sites on the receptor (##REF##15496464##Igoshin et al., 2004##; ##REF##16359317##Astling et al., 2006##). ##REF##16359317##Astling et al. (2006)## identified several putative methylation sites based on sequence in comparison with known <italic>Escherichia coli</italic> methylation sites. They systematically mutated these sites, substituting each EE, QQ, QE and EQ pair with an alanine pair (AA). This work suggested that FrzCD receptor activity could be turned on or off depending on the site that was methylated.</p>", "<p>Based on this work, we hypothesized that differential methylation of FrzCD may be mediated by FrzF, a methyltransferase (CheR) homologue (##REF##2168368##McCleary et al., 1990##) that contains an additional domain with three tetra trico-peptide repeats (TPRs) (##REF##12101179##Shiomi et al., 2002##). In other organisms, TPRs have been shown to mediate protein–protein interactions (##REF##10517866##Blatch and Lassle, 1999##). A Basic Local Alignment Search Tool (<sc>blast</sc>) (##REF##2231712##Altschul et al., 1990##) analysis of all sequenced bacterial genomes revealed that dozens of bacterial species possess putative methyltransferases with one or more TPRs, including several within the α-proteobacteria, β-proteobacteria, δ-proteobacteria and high GC-rich Gram-positive bacteria. In <italic>M. xanthus</italic>, CheR4 and CheR6 are each predicted to contain one TPR (##UREF##1##Scott, 2008##). However, to our knowledge, no function has yet been attributed to TPRs in methyltransferases.</p>", "<p>To investigate the role of the TPR containing domain in FrzCD methylation, we used full-length FrzF and FrzF lacking the TPRs (FrzF^CheR) to methylate FrzCD <italic>in vitro</italic>. We found that indeed the TPRs of FrzF negatively regulate FrzCD methylation. We used mass spectrometry to identify the methylated sites and site-directed mutagenesis to determine the function of these sites <italic>in vivo</italic>. This work showed that each FrzCD methylation site played specific roles in cell motility and behaviour.</p>" ]

[]

[ "<title>Results</title>", "<title>Methylation of FrzCD <italic>in vitro</italic> using purified FrzF and FrzF^CheR</title>", "<p>Previous work showed that methylation of FrzCD is required for swarming and fruiting body formation in <italic>M. xanthus</italic> and that methylation is mediated by FrzF (##REF##2492105##McBride et al., 1989##; ##REF##2168368##McCleary et al., 1990##). FrzF is a complex methyltransferase that contains an N-terminal domain with 31% sequence identity (83 of 271 amino acids) to the methyltransferase (CheR) of <italic>E. coli</italic> and a C-terminal domain with three TPRs (##FIG##0##Fig. 1A##). Because of its high homology to CheR, the N-terminal domain of FrzF was assumed to have methyltransferase activity, but the function of the C-terminal domain was unknown. To determine the activities of full-length FrzF and FrzF^CheR (FrzF lacking its TPR domains), we cloned His-tagged <italic>frzCD, frzF</italic> and <italic>frzF</italic>^{<italic>CheR</italic>} in expression vectors and purified the respective proteins from <italic>E. coli</italic>. FrzCD purified from <italic>E. coli</italic> was unmethylated (##FIG##0##Fig. 1B##, lane 3), indicating that the <italic>E. coli</italic> methyltransferase does not methylate FrzCD. To methylate FrzCD <italic>in vitro</italic>, we incubated FrzCD and the methyl donor S-adenosyl methionine (SAM) with either FrzF or FrzF^CheR. The reactions were monitored by Western immunoblot analysis using purified anti-FrzCD antibodies as methylated FrzCD migrates faster than unmethylated FrzCD in this SDS-PAGE system (##REF##2168368##McCleary et al., 1990##). We found that that both FrzF and FrzF^CheR were able to methylate FrzCD in the presence of SAM (##FIG##0##Fig. 1B##), but that they produced different FrzCD methylation patterns. FrzF^CheR produced a faster migrating band of FrzCD (##FIG##0##Fig. 1## lane 5) than FrzF (##FIG##0##Fig. 1## lane 4). Neither FrzF nor FrzF^CheR were able to methylate FrzCD without SAM (data not shown). As methylated receptors migrate faster than unmethylated receptors in this gel system, we hypothesized that the faster migrating band observed with the FrzF^CheR sample represented a more methylated species of FrzCD.</p>", "<title>Identifying the methylated residues of FrzCD</title>", "<p>As the mobility of FrzCD on polyacrylamide gels does not give us quantitative data on methylation patterns, we analysed the <italic>in vitro</italic> methylated FrzCD by mass spectrometry. We prepared methylated FrzCD samples as described above by incubating purified FrzCD, SAM and either FrzF or FrzF^CheR; as a control, we also prepared non-methylated FrzCD in the same way except that we omitted FrzF. To generate peptides of the optimal size for mass spectrometry, we digested the methylated and unmethylated FrzCD samples with trypsin, chymotrypsin or GluC proteases. The combined tandem mass spectrometry (MS/MS) results accounted for 92.3% total sequence coverage including 100% sequence coverage of the C-terminus of FrzCD (amino acids 136–437), predicted to contain all putative methylation sites (##SUPPL##0##Fig. S1##).</p>", "<p>A comparison between spectra of tryptic FrzCD peptides methylated by FrzF and by FrzF^CheR revealed that FrzF methylated FrzCD on a single glutamate residue and that FrzF^CheR methylated FrzCD on three residues (based on accurate precursor mass measurement and MS/MS data) within the peptide L148-K184. This peptide sequence includes five glutamate residues, all representing possible sites of methylation. However, while fragmentation was sufficient for the identification of the peptide, as well as the presence of methylation, it was not adequate to obtain the sequence coverage necessary to ascertain the specific sites of methylation (data not shown). To address this, FrzCD was digested with the chymotrypsin and GluC proteases. Chymotrypsin allowed us to determine that FrzCD was methylated by FrzF on E182 (##TAB##0##Table 1##, ##SUPPL##0##Fig. S2##). Specifically, the b-ion with the value of 1372.3 indicates that there is a methylated glutamate on the N-terminal side of residue L183 and the b-ion with the value of 1301.1 indicates that no residue is methylated to the N-terminal side of E182. Thus, the only residue that can be methylated on this peptide is E182. However, we found that the peptides were not always methylated. For instance, two of the nine ₁₅₇AASTQHETSSTEQAAA IHETTATMEEL₁₈₃ peptides analysed showed methylation on site E182 (##TAB##0##Table 1##, ##SUPPL##0##Fig. S2##, and data not shown). No methylation was seen on sites E168 or E175 on any of the nine peptides. MASCOT scores for these nine peptides were all greater than 24 indicating that these results are highly significant.</p>", "<p>We found that FrzF^CheR methylated FrzCD on residues E175 and E182 by using a chymotrypsin digest (##TAB##1##Table 2##, ##SUPPL##0##Fig. S3A##). Specifically, b-ion 849.7 showed that the methylated glutamate was on the C-terminal side of residue I173 and b-ion 1040.6 showed that the methylated glutamate was on the N-terminal side of T176. Thus, E175 must be methylated as it is the only glutamate between residues I173 and T176. b-ion 1315.2 showed that another methylated glutamate was located on the C-terminal side of residue E181. Therefore, E182 was a site of methylation (##TAB##1##Table 2##, ##SUPPL##0##Fig. S3A##). Similar to our FrzF data, we found that the FrzF^CheR methylated FrzCD peptides were not always fully methylated. For instance, one of the three ₁₅₇AASTQHETSSTEQA AAIHETTATMEEL₁₈₃ peptides analysed showed methylation on sites E175 and E182 (##TAB##1##Table 2##, ##SUPPL##0##Fig. S3A##). The remaining two peptides showed methylation only on site E182 (Data not shown). MASCOT scores for these three peptides were all greater than 24 indicating that these results are highly significant.</p>", "<p>To find the third site methylated by FrzF^CheR, we used a GluC digest (##TAB##2##Table 3##, ##SUPPL##0##Fig. S3B##). The b-ion 648.2 showed that a glutamate was methylated on the N-terminal site of Q169 and the y-ion 1546.4 showed that a glutamate was methylated on the C-terminal side of T167. In all seven ₁₆₄TSSTEQAAAIHE175 peptides analysed, site E168 was shown to be methylated (data not shown). In one out of four the ₁₆₄TSSTEQAAAIHETTATME181 peptides analysed, both E168 and E175 were found to be methylated (##SUPPL##0##Fig. S3B##). In all eight of the ₁₆₉QAAAIHETTATME181 peptides site E175 was methylated (data not shown). MASCOT scores for these peptides were all greater than 24 indicating that these results are highly significant.</p>", "<p>The sample without FrzF was not methylated at any site (##SUPPL##0##Table S1, Fig. S4##, and data not shown). None of the 14 copies of the ₁₅₇AASTQHETSSTEQA AAIHETTATMEEL₁₈₃ peptide showed methylation (##SUPPL##0##Table S1 and Fig. S2##). MASCOT scores for these 14 peptides were all greater than 24 indicating that these results are highly significant.</p>", "<p>In summary, mass spectrometry showed that <italic>in vitro</italic> FrzF methylates FrzCD on one site (E182) and that FrzF^CheR methylates FrzCD on three sites (E168, E175 and E182). This would suggest that the TPR domain of FrzCD is a regulatory domain that inhibits the methyltransferase activity of FrzF at two specific sites, E168 and E175.</p>", "<title>Isolation and characterization of methylation site point mutants</title>", "<p>To learn the function of these three FrzCD methylation sites, we constructed FrzCD methylation point mutants where we replaced a methylatable glutamate residue with either an aspartate or an alanine residue. These substitutions have been used in other bacteria to mimic unmethylated and methylated glutamates, respectively, as aspartate residues, which cannot be methylated, maintain the negative charge of a glutamate residue and alanine residues are neutrally charged, similar to methylated glutamates (##REF##3041407##Nowlin et al., 1988##; ##REF##2254280##Park et al., 1990##; ##REF##8157631##Shapiro and Koshland, 1994##).</p>", "<p>FrzCD methylation site point mutants were generated by PCR, cloned into the plasmid pCT2, and integrated into the non-essential <italic>crtB</italic> locus of the <italic>M. xanthus</italic> Δ<italic>frzCD</italic> strain. We expressed <italic>frzCD</italic> under control of the <italic>tet</italic> promoter by growing cells in the presence of anhydrotetracycline (##REF##17590236##Mignot et al., 2007##). As a positive control we made strain DZ4717, which contains the non-mutated <italic>frzCD</italic> gene in the <italic>crtB</italic> locus under the control of the <italic>tet</italic> promoter. We found that DZ4717 displayed phenotypes that were comparable to that of the wild-type strain, DZ2 (##SUPPL##0##Fig. S5##). To ensure that all strains expressed the FrzCD protein, we examined each strain by immunoblot analysis using the anti-FrzCD antibody. ##FIG##1##Figure 2## shows that FrzCD was expressed at the same level in all strains and that any changes in phenotype were not due to altered expression.</p>", "<p>We then determined the effects of mutating the methylation sites on <italic>M. xanthus</italic> fruiting body formation (##FIG##2##Fig. 3##) and swarming (##FIG##3##Fig. 4##), phenotypes that require a functioning Frz pathway. For reference, we also examined the phenotypes of the hypo-reversing Δ<italic>frzCD</italic> mutant and the hyper-reversing <italic>frzCD</italic>^{Δ<italic>6</italic>–<italic>183</italic>} mutant. As shown previously (##REF##15387825##Bustamante et al., 2004##), the Δ<italic>frzCD</italic> control strain exhibited reduced swarming on rich media, formed tangled aggregates instead of fruiting bodies on starvation media, and reversed infrequently. In contrast, the <italic>frzCD</italic>^{Δ<italic>6</italic>–<italic>183</italic>} control strain did not swarm on rich media, was unable to form aggregates on starvation media, and reversed much more frequently as single cells than wild type (##FIG##2##Figs 3## and ##FIG##3##4##; ##TAB##3##Table 4##). The methylation point mutants displayed varied phenotypes: some behaved like wild type, whereas others displayed Δ<italic>frzCD</italic>, <italic>frzCD</italic>^{Δ<italic>6</italic>–<italic>183</italic>} or intermediate phenotypes.</p>", "<p>The most severe single mutant phenotypes were observed when we eliminated methylation at E182. The <italic>frzCD</italic>^{<italic>E182D</italic>} mutant, similar to Δ<italic>frzCD</italic>, showed reduced swarming (##FIG##3##Fig. 4##), formed tangled aggregates instead of fruiting bodies (##FIG##2##Fig. 3I##), and had a severely reduced single cell reversal frequency when compared with DZ4717 (##TAB##3##Table 4##). The <italic>frzCD</italic>^{<italic>E182A</italic>} mutant displayed an intermediate swarming phenotype between DZ4717 and the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant (##FIG##3##Fig. 4##), was able to form fruiting bodies (##FIG##2##Fig. 3H##), but still showed a reduced single cell reversal frequency (##TAB##3##Table 4##). Thus, FrzCD methylation site 182 is important for both social behaviours and single cell reversals.</p>", "<p>Single mutations in E175 or E168 resulted in less severe defects than single mutations in site E182. E to A or E to D mutations at site 175 allowed cells to swarm like DZ4717 (##FIG##3##Fig. 4##) and to form fruiting bodies (##FIG##2##Figs 3F,G##). Yet both of these mutants showed a decrease in reversal frequency when compared with DZ4717 (##TAB##3##Table 4##). The <italic>frzCD</italic>^{<italic>E168D</italic>} mutant displayed no obvious defects. While the <italic>frzCD</italic>^{<italic>E168A</italic>} mutant displayed a 30% decrease in swarming (##FIG##3##Fig. 4##), it showed normal single cell and fruiting body behaviour. Thus, site E175 is required for single cell reversal frequency, but neither site E168 nor E175 must be methylated for proper fruiting body formation.</p>", "<p>In <italic>E. coli,</italic> methylation site mutants display additive phenotypes; defects increase in severity as additional methylation sites are mutated (##REF##7862632##Shapiro et al., 1995##). To test if methylation site mutations resulted in additive defects in <italic>M. xanthus</italic>, we constructed double and triple methylation site point mutants. We expected the double and triple mutants to display more severe defects than the single mutants. However, we found that the <italic>frzCD</italic>^{<italic>E168D,E175D</italic>} and <italic>frzCD</italic>^{<italic>E168D,E175D,E182D</italic>} mutants displayed DZ4717 phenotypes under all conditions tested (##FIG##2##Fig. 3K and M##, ##FIG##3##Fig. 4##, ##TAB##3##Table 4##). This was surprising as the <italic>frzCD</italic>^{<italic>E175D</italic>} and the <italic>frzCD</italic>^{<italic>E182D</italic>} mutants displayed reduced single cell reversal frequencies compared with DZ4717. These results suggest that the overall number of methylatable FrzCD residues is not crucial, but rather the pattern of FrzCD methylation is the determining factor for receptor function.</p>", "<p>We were also surprised to find that the <italic>frzCD</italic>^{<italic>E168A,E175A</italic>} and <italic>frzCD</italic>^{<italic>E168A,E175A,E182A</italic>} mutants displayed severe defects under all conditions tested, as the single site mutants, E168A and E175A displayed no or subtle defects. These mutants swarmed less than Δ<italic>frzCD</italic> (##FIG##3##Fig. 4##), hyper-reversed compared with DZ4717 (##TAB##3##Table 4##), and formed smaller and less dispersed fruiting bodies than DZ4717 during development (##FIG##2##Fig. 3J and L##). The <italic>frzCD</italic>^{<italic>E168A,E175A</italic>} and <italic>frzCD</italic>^{<italic>E168A,E175A,E182A</italic>} mutants displayed phenotypes that were intermediate between DZ4717 and the hyper-reversing <italic>frzCD</italic>^{Δ<italic>6</italic>–<italic>183</italic>} mutant.</p>" ]

[ "<title>Discussion</title>", "<p>FrzCD, the principal MCP for the Frz chemosensory pathway, plays a central role in regulating cellular motility and behaviour in <italic>M. xanthus</italic>. Previous studies showed that methylation of FrzCD by FrzF is critical to function of FrzCD, as cells lacking FrzF, the unusual TPR domain containing methyltransferase, rarely reverse and are unable to swarm, form fruiting bodies, or move towards nutrients (##REF##3936045##Blackhart and Zusman, 1985##; ##REF##15387825##Bustamante et al., 2004##). We were curious to learn the role of the TPRs in FrzF and how methylation regulates FrzCD receptor activity.</p>", "<p>By analysing <italic>in vitro</italic> methylated FrzCD by mass spectrometry, we found that FrzF methylates FrzCD on one residue, E182, and that FrzF^CheR methylates FrzCD on three residues, E168, E175 and E182. When we compared the methylated residues of FrzCD with methylated residues of the <italic>E. coli</italic> MCP, Tar, we found both similarities and differences (##FIG##4##Fig. 5##). Both FrzCD and Tar have methylation sites spaced seven residues apart. It has been shown that methylation sites spaced at seven residue intervals are located on the same face of the α-helix and are accessible to the methyltransferase (##REF##2875460##Terwilliger et al., 1986##).</p>", "<p>Several differences exist between the methylation sequences of Tar and FrzCD. Tar is methylated on glutamine (Q) or glutamate (E) residues that are located within QQ, EE, QE or EQ methylation pairs and methylation occurs on the second residue of the methylation pair (##REF##6213619##Kehry and Dahlquist, 1982##; ##REF##6309776##Terwilliger et al., 1983##; ##REF##6330075##Terwilliger and Koshland, 1984##). In FrzCD, only site E182 conforms precisely to the <italic>E. coli</italic> paradigm; it is methylated on the second residue of an EE methylation pair. Site E175 is the most unusual of the FrzCD methylation sites; it is not located in a typical methylation pair. Site E175 is flanked instead by histidine (H) and threonine (T) residues (##FIG##4##Fig. 5##). This is only the second instance of a receptor being methylated outside of a typical methylation pair. In <italic>Thermotoga maritima</italic>, the receptor TM0429c is methylated on the glutamate in a TE pair (##REF##16707700##Perez et al., 2006##). FrzCD site E168 is also somewhat atypical because methylation is occurring on the first residue of a methylation pair (##FIG##4##Fig. 5##). Although unusual, other receptors have been shown to be methylated on the first residue of a methylation pair (##REF##16707700##Perez et al., 2006##). We found it interesting that FrzCD site E182, which conforms precisely to the <italic>E. coli</italic> consensus sequence, was methylated by both FrzF and FrzF^CheR<italic>in vitro</italic>, but that sites E168 and E175, which differ from the <italic>E. coli</italic> consensus sequence, were only methylated by FrzF^CheR. It is possible that methylation preferentially occurs on sites that adhere more closely to the methylation consensus sequence found in <italic>E. coli</italic>.</p>", "<p>To learn how methylation affects the activity of FrzCD we constructed mutations in each of the identified methylation sites. A previous study had mutated each QQ, EE, EQ and QE pair in FrzCD to a double AA (##REF##16359317##Astling et al., 2006##). Because FrzCD sites E168 and E182 were located within EQ and EE pairs, respectively, the FrzCD^E168AQ169A and FrzCD^E181AE182A mutants were constructed, previously. ##REF##16359317##Astling et al. (2006)## found by analysing single cell reversals that the FrzCD^E168AQ169A mutant appeared to hyperactivate FrzCD, whereas the FrzCD^E181AE182A mutant resulted in a hypoactive FrzCD signally state. Although these results were promising, we were concerned that the double AA mutation may result in a more severe phenotype than a single mutant. Additionally, we identified a FrzCD methylation site that was located outside of a QQ, EE, QE, EQ methylation pair and were interested to learn its function. We were also interested in learning the phenotype of both the gain and loss of methylation at each site; therefore, we constructed mutants where we replaced each methylatable glutamate with an alanine to mimic methylation or an aspartate to mimic demethylation (##REF##3032955##Nowlin et al., 1987##; ##REF##8157631##Shapiro and Koshland, 1994##).</p>", "<p>We found that site 182 (the site methylated by both FrzF and FrzF^CheR) displayed the most profound phenotypes of the single mutants. In fact, the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant behaved similarly to the Δ<italic>frzCD</italic> mutant suggesting that methylation at this residue is critical for activating the Frz pathway. Furthermore, the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant phenotype also resembled the Δ<italic>frzE</italic> and the <italic>frzE</italic>^{<italic>H49A</italic>} (kinase dead) mutant phenotypes suggesting that FrzCD^{<italic>E182D</italic>} may inhibit the histidine kinase activity of FrzE. The severe mutant phenotype seen in the <italic>frzCD</italic>^{<italic>E182D</italic>} strain may be due to loss of methylation at other nearby methylation sites. In <italic>E. coli</italic>, replacing a methylatable E or Q residue with a D residue has been shown to prevent methylation at nearby sites (##REF##7862632##Shapiro et al., 1995##). This appears to hold true for the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant. We do not see any FrzCD methylation in <italic>M. xanthus</italic> in the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant (##FIG##1##Fig. 2##). As the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant displayed a more severe phenotype than either the <italic>frzCD</italic>^{<italic>E168D,E175D,E182D</italic>} or <italic>frzCD</italic>^{<italic>E168D,E175D</italic>} mutants, we concluded that the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant phenotype was not due to loss of methylation at sites E168 or E175. It is likely that the <italic>frzCD</italic>^{<italic>E182D</italic>} mutant lacks methylation at unidentified sites. We believe that additional FrzCD methylation sites exist because the <italic>frzCD</italic>^{<italic>E168D,E175D,E182D</italic>} mutant still displays a methylated band of FrzCD (##FIG##1##Fig. 2##). We anticipate that several additional methylation sites may be uncovered when glutamine residues are converted to glutamates by a methylesterase <italic>in vivo</italic> (##REF##358191##Stock and Koshland, 1978##). Additionally, some methylation sites may only be modified <italic>in vivo</italic> under certain physiological conditions.</p>", "<p>Mutations in FrzCD site E175 resulted in less severe mutant phenotypes compared with mutations in FrzCD site E182. Strains containing an aspartate or an alanine at site 175 displayed a decrease in single cell reversal frequency, but showed no obvious defect in social swarming or fruiting body development. These results suggest that site E175 is critical for coordinating single cell behaviour, but is dispensable for groups of cells. This is interesting because it suggests that FrzCD is able to regulate single cell behaviour and social behaviour independently.</p>", "<p>FrzCD site 168 appears to play a role in regulating social swarming. The <italic>frzCD</italic>^{<italic>E168A</italic>} mutant displayed a reduction of social swarming, but did not have an obvious defect in fruiting body formation and single cells reversed normally. The <italic>frzCD</italic>^{<italic>E168D</italic>} mutant did not display any mutant phenotypes. Thus, site E168 appears to be a minor regulator of FrzCD when methylated alone. However, when both FrzCD sites E168 and E175 were replaced with alanines, <italic>M. xanthus</italic> cells displayed profound affects; cells were unable to swarm, they were defective in fruiting body formation, and single cells hyper-reversed. Thus, simultaneous methylation at these two sites seems to activate FrzCD. It was interesting that the <italic>frzCD</italic>^{<italic>E168D,E175D</italic>} mutant did not display any mutant phenotypes. Thus, the mutation at site 168 can rescue the hypo-reversing phenotype of the <italic>frzCD</italic>^{<italic>E175D</italic>} mutant.</p>", "<p>In sum, we found that the TPRs of FrzF inhibit its ability to methylate FrzCD <italic>in vitro</italic>. Additionally, each site of FrzCD methylation plays a unique role in regulating FrzCD activity. Third, FrzCD appears to be able to control single cell and social behaviours independently. Last, our results confirm previous work by ##REF##16359317##Astling et al. (2006)##, which suggested that FrzCD methylation could both turn on and off receptor activity and that the pattern of methylation determines the activity of FrzCD, not the quantity of methylation.</p>", "<p>Based on our current knowledge of the Frz pathway, we have proposed a model that addresses a putative input into FrzCD, the regulation of FrzCD activity, and how FrzCD activity regulates downstream signalling and cell behaviour. We believe that one input into FrzCD can be the methylation of FrzCD by FrzF. FrzF may be controlled by its TPRs. As FrzF^CheR methylates FrzCD more than FrzF <italic>in vitro</italic>, it is possible that the TPRs prevent the methyltransferase domain from methylating specific residues of FrzCD (such as E168 and E175) by forming a physical barrier between the methyltransferase domain and certain binding sites on FrzCD. To relieve this inhibition, the TPRs may bind to another protein. This binding may place the TPRs in a conformation in which they no longer impede the methyltransferase domain, thus allowing FrzF to methylate additional FrzCD residues. We hypothesize that a FrzF TPR-binding partner is up-regulated or activated as cells proceed through development because FrzCD methylation has been shown to increase as <italic>M. xanthus</italic> cells develop (##REF##8335650##McBride and Zusman, 1993##). Additionally, we propose that the TPR-binding partner is upregulated by AsgA, CsgA, FruA and DevT and downregulated by RodK because FrzCD methylation is reduced in <italic>asgA</italic>, <italic>csgA</italic>, <italic>fruA</italic> and <italic>devT</italic> mutants, and increased in <italic>rodK</italic> mutants (##REF##8610100##Sogaard-Andersen and Kaiser, 1996##; ##REF##9791131##Geng et al., 1998##; ##REF##11872704##Boysen et al., 2002##; ##REF##15882426##Rasmussen et al., 2005##).</p>", "<p>The second part of our model explains how the activity of FrzCD can be controlled through methylation. We propose that methylation can both turn on and turn off FrzCD activity. For instance, if FrzCD is methylated on one site (E182) its activity is inhibited, whereas if it is methylated on additional residues (E168 and E175) its activity is activated. These hypotheses are based on our FrzCD methylation point mutant data where mutations in site E182 were similar to a deletion of FrzCD and mutations that mimicked methylation at both 168 and 175 resembled a ‘hyperactive’ <italic>frzCD</italic> mutant (<italic>frzCD</italic>^{Δ<italic>6−183</italic>}).</p>", "<p>The final part of our model involves how FrzCD activity regulates downstream proteins. It has been proposed that the CheA homologue, FrzE, stimulates cellular reversals by autophosphorylating and transferring a phosphoryl group to the dual receiver domain containing protein, FrzZ (##REF##2123853##McCleary and Zusman, 1990##; ##REF##17581122##Inclán et al., 2007##). We propose that certain methylation sites lead to an active form of FrzCD and this stimulates FrzE kinase activity, which leads to an increase in cellular reversals. Conversely, methylation states of FrzCD that inhibit activity result in a reduction of FrzE kinase activity and a reduction in cellular reversals.</p>" ]

[]

[ "<p>Re-use of this article is permitted in accordance with the Creative Commons Deed, Attribution 2.5, which does not permit commercial exploitation.</p>", "<p><italic>Myxococcus xanthus</italic> is a gliding bacterium with a complex life cycle that includes swarming, predation and fruiting body formation. Directed movements in <italic>M. xanthus</italic> are regulated by the Frz chemosensory system, which controls cell reversals. The Frz pathway requires the activity of FrzCD, a cytoplasmic methyl-accepting chemotaxis protein, and FrzF, a methyltransferase (CheR) containing an additional domain with three tetra trico-peptide repeats (TPRs). To investigate the role of the TPRs in FrzCD methylation, we used full-length FrzF and FrzF lacking its TPRs (FrzF^CheR) to methylate FrzCD <italic>in vitro</italic>. FrzF methylated FrzCD on a single residue, E182, while FrzF^CheR methylated FrzCD on three residues, E168, E175 and E182, indicating that the TPRs regulate site-specific methylation. E168 and E182 were predicted consensus methylation sites, but E175 is methylated on an HE pair. To determine the roles of these sites <italic>in vivo</italic>, we substituted each methylatable glutamate with either an aspartate or an alanine residue and determined the impact of the point mutants on single cell reversals, swarming and fruiting body formation. Single, double and triple methylation site mutants revealed that each site played a unique role in <italic>M. xanthus</italic> behaviour and that the pattern of receptor methylation determined receptor activity. This work also shows that methylation can both activate and inactivate the receptor.</p>" ]

[ "<title>Experimental procedures</title>", "<title>Bacterial strains and culture conditions</title>", "<p>The strains used in this study are listed in ##TAB##4##Table 5##. <italic>M. xanthus</italic> was grown in CYE medium, which contains 10 mM morpholinepropanesulphonic acid (MOPs, pH 7.6), 1% (w/v) Bacto Casitone (BD Biosciences), 0.5% Bacto yeast extract and 4 mM MgSO₄ (##REF##416222##Campos et al., 1978##). For <italic>M. xanthus</italic> swarming phenotypes 5 μl of cells at 4 × 10⁹ cells ml⁻¹ were spotted on CYE plates containing 0.5% Bacto agar (BD Biosciences). For developmental phenotypes 5 μl of cells at 2 × 10⁹ cells ml⁻¹ were spotted on CF agar which contains 1.5% Bacto agar, 10 mM MOPs (pH 7.6), 0.015% Bacto Casitone, 8 mM MgSO₄, 1 mM KH₂PO₄, 0.02% NH₄SO₄, 0.2% sodium citrate and 0.1% pyruvate (##REF##98366##Hagen et al., 1978##).</p>", "<title>Reversal frequency analysis</title>", "<p>Cells were grown to mid-log phase and 10 μl of culture were spotted on 1/2 CTT (##REF##16592422##Hodgkin and Kaiser, 1977##) with 1.5% agar. The cells were covered immediately with an oxygen permeable membrane (Yellow Springs Instrument) and allowed to settle for 1 h. Cells were filmed using a Labophot 2 microscope and a micropublisher 3.3 RTV digital camera (Q imaging) and reversal frequencies were analysed by eye. Only moving cells that did not touch another cell during filming were analysed. A students <italic>t</italic>-test (two-tailed, type 3) was used to determine if the reversal frequency of each strain differed from wild type. Strains with a <italic>P</italic>-value less than 0.005 were considered to have different reversal frequencies from wild type. For each strain, a minimum of five movies were taken on two separate days.</p>", "<title>Protein purification from <italic>E. coli</italic></title>", "<p>To obtain soluble protein, <italic>frzF</italic> and <italic>frzF</italic>^{<italic>cheR</italic>} amplified by PCR from genomic DNA and inserted into pET28a (Novagen) at the EcoRI and HindIII sites. The plasmid was sequenced and analysed using the Sequencher program (Gene Codes) and transformed into the <italic>E. coli</italic> Tuner protein expression strain (Novagen). Cells were grown to mid-log phase at 37°C and expression was induced by adding 1 mM of IPTG and transferring cells to 18°C for 16 h. Cells were isolated by centrifugation at 9000 <italic>g</italic>. Cells were lysed by sonication (Branson sonifier 450) (3 × 1 min output 4) in 20 mM Tris pH 7.2, 20 mM imidizole, 10% glycerol, 0.1% CHAPS buffer, 500 mM NaCl lysis buffer supplemented with 2% mammalian protease inhibitor cocktail (Sigma). Insoluble material was removed by centrifugation (20 min, 17 000 <italic>g</italic>) and filtration through a 0.22 micron filter (Corning 115 ml filter system). Cell lysate was injected onto a 5 ml HisTrap HP Nickel column (GE Healthcare) via an AKTA FPLC (GE Healthcare). Protein was washed with 50 ml of lysis buffer, then 50 ml of lysis buffer with 60 mM, 100 mM and 150 mM imidizole. FrzF^CheR was eluted using lysis buffer with 250 mM imidizole. The protein was concentrated using 30 000 molecular weight cutoff Centriprep columns (Amicon).</p>", "<title><italic>In vitro</italic> methylation of FrzCD</title>", "<p>Each FrzCD methylation reaction mixture contained 3.5 μM FrzCD, 6 μM SAM, 7 μM FrzF or FrzF^CheR, 10 mM TrisHCl pH 7.0, 1% glycerol and 50 mM KCl. Reactions took place at 32°C for 4 h. FrzCD methylation was observed initially by SDS-PAGE analysis and then by Mass Spectrometry (see below). Each reaction was repeated in triplicate on four separate days.</p>", "<title>Sample preparation for mass spectrometry</title>", "<p>Methylation of receptors was performed as described above. Methylated and unmethylated species were separated by SDS-PAGE. Gels were stained with Coloidal Coomassie (Invitrogen) and bands corresponding to unmethylated and methylated FrzCD were excised and diced into ∼1–2 mm³ pieces. The gel slices were destained twice by incubation with 50 μl of 100 mM ammonium bicarbonate and 50 μl of acetonitrile at 37°C for 10 min The gel slices were then dehydrated by incubation with 50 μl of acetonitrile at 37°C for 5 min. FrzCD contains no cysteine residues, so reduction and alkylation were not performed. Gel slices were incubated overnight at 37°C in 50 μl of 100 mM ammonium bicarbonate containing 150 ng of trypsin (Promega), chymotrypsin (Roche) or GluC (Roche). Peptides were extracted from the gel slices by incubation with 30 μl of extraction solution (2% acetonitrile and 1% formic acid) at 37°C for 30 min. The solution was removed and a second extraction was performed with 12 μl of acetonitrile and 12 μl of extraction solution at 37°C for 30 min. The peptide extract was dried down and reconstituted in 25 μl of 0.1% trifluoroacetic acid (TFA).</p>", "<title>Liquid chromatography/mass spectrometry</title>", "<p>High-performance liquid chromatography grade water and acetonitrile (Optima) and formic acid (Acros Organics) were purchased from Fisher Scientific. TFA was purchased from Sigma-Aldrich. Peptides were desalted and concentrated on a reversed-phase cartridge (Zorbax C₁₈; 5 mm by 0.3 mm i.d.; 5 μm; Agilent) then loaded onto a reverse-phase column (ProteoPep^TM C18, 5 cm by 50 μm i.d.; 300 Å; 5 μm; New Objective) using an Ultimate^TM nanoliquid chromatography system (Dionex/LC Packings) coupled to a ThermoFinnigan Orbitrap tandem mass spectrometer equipped with a nanospray source (Michrom Bioresources). The column was equilibrated for 5 min in 94% solvent A (0.1% formic acid) and 6% solvent B (90% acetonitrile, 0.1% formic acid). Solvent B was increased linearly to 40% at 40 min, 60% at 45 min and 100% B at 45.1 min, where it was held for 3 min. It was then set back to its initial solvent composition (6% B), where it was held for the duration of the run (60 min). MS survey scans were performed in the orbitrap followed by subsequent MS/MS scans of the three most abundant ions fragmented in the linear ion trap, with a dynamic exclusion of 30 s. Data files (.dta) for MS/MS spectra were generated by Bioworks Browser 3.2 <sc>efi</sc> software (Thermo Fisher Scientific), converted to MASCOT generic format (.mgf), and searched against a proteomic database for <italic>M. xanthus</italic>. Peptides identified by MASCOT, with or without methyl modifications, at a significance score above 95% confidence, were validated manually. No fixed modifications were used in the searches while methionine oxidation, asparagine and glutamine deamidation, and aspartate and glutamate methylation were used as variable modifications. Three replicates of each sample were analysed.</p>", "<title>Construction of mutants</title>", "<p>All strains and plasmids are listed in ##TAB##4##Table 5##. Point mutations were made using PCR. <italic>M. xanthus</italic> genomic DNA from strain DZ2 was used as a template for PCR. Oligonucleotides were prepared by operon. Complementary forward and reverse primers were used to amplify the gene with platinum HiFi Taq (Invitrogen). The primers for the coding strand are listed below with the lower case indicating the altered codon:</p>", "<p>For <italic>frzCD</italic>^{<italic>E168D</italic>} catgagacgtcctccacggaccaggcggcggccatccacg, For <italic>frzCD</italic>^{<italic>E168A</italic>} catgagacgtcctccacggcgcaggcggcggccatccacg, For <italic>frzCD</italic>^{<italic>E175D</italic>} caggcggcggccatccacgacacgaccgccaccatggaggag, For <italic>frzCD</italic>^{<italic>E175A</italic>} caggcggcggccatccacgcgacgaccgccaccatggaggag, For <italic>frzCD</italic>^{<italic>E182D</italic>} Cacgagacgaccgccaccatggaggacctgaagcacgcgtcggcgc, For <italic>frzCD</italic>^{<italic>E182A</italic>} CACCATGGAGgcgctgaagcacgc. For <italic>frzCD</italic>^{<italic>E168D E175D</italic>} we used pAS215 as template and primer catgagacgtcctccacggaccaggcggcggccatccacg, For <italic>frzCD</italic>^{<italic>E168A E175A</italic>} we used pAS212 and primer catgagacgtcctccacggcgcaggcggcggccatccacg, For <italic>frzCD</italic>^{<italic>E168D E175D E182D</italic>} we used pAS218 as template and primer cacgagacgaccgccaccatggaggacctgaagcacgcgtcggcgc, For <italic>frzCD</italic>^{<italic>E168A E175A E182A</italic>} we used pAS217 (E168A E174A) and primer CACCATGGAGgcgCTGAAGCACGC.</p>", "<p>Once each half of the insert was constructed the two PCR products per mutant were used as template and the primers. GATATCCAGCTGCCCGAGGAGGACGATG and GGCCAGTGCCAAGCTTCATTACTAGTCG were used to construct the complete insert. The insert was then placed into the digested (SmaI and HindIII) pCT2 (##REF##17590236##Mignot et al., 2007##) plasmid via the In-Fusion reaction (Clonetech). The resulting plasmids were confirmed by DNA sequencing.</p>", "<p>Plasmids containing the wild-type <italic>frzCD</italic> and the point mutations were inserted into the <italic>crtB</italic> locus of <italic>M. xanthus</italic> strain Δ<italic>frzCD</italic> and confirmed by PCR and sequencing. In all figures ‘wild type’ corresponds to <italic>6His::frzCD</italic> in the <italic>crtB</italic> locus. To induce expression of <italic>frzCD</italic> and the point mutants<italic>,</italic> 50 μg ml⁻¹ of anhydrotetracycline HCl (Reidel-de Haen) was used in all media.</p>", "<title>Immunoblot analysis of FrzCD</title>", "<p><italic>Myxococcus xanthus</italic> strains were grown to mid-exponential phase, concentrated by centrifugation, and resuspended in 1 × SDS loading buffer lacking coloured dye. Cells were lysed by 10 s of sonication (Branson sonifier 450) on ice. Protein concentration was determined using the BCA (bicinchoninic acid) method (reagents from Pierce). Cells were resuspended to the same concentration in 2 × SDS loading buffer. Thirty micrograms of total protein were loaded per lane on 10% Tris HCl ready gels (Bio-Rad). After electrophoresis, the gel was transferred to a nitrocellulose membrane. Blots were probed with anti-FrzCD antibody as described (##REF##2168368##McCleary et al., 1990##) and with the antirabbit alexa fluor 680 (Molecular Probes) secondary antibody. Blots were visualized using an infrared imaging system (LiCor) and results were analysed using Odyssey software. The data were confirmed by three independent experiments done in duplicate.</p>" ]

[ "<p>We would like to acknowledge the contributions of Vivian Trang who isolated several point mutant strains. We would like to thank Lori Kohlstaedt (UC Berkeley Mass Spectrometry Facility) for mass spectrometry help and advice. We would like to thank Zusman lab and the Andrews lab past and present specifically Yuki Inclán, John Merlie, David Astling, Emilia Mauriello and Tam Mignot for many helpful suggestions and discussions throughout this work. We would like to thank Emilia Mauriello and Sophie Laurent for critically reading this manuscript. This work was supported by grants from the National Institutes of Health (GM20509) to DRZ and from the NIH/NCRR National Resource for Proteomics and Pathways (P41-18627) to PCA. AES was supported by a graduate fellowship from the National Science Foundation.</p>", "<title>Supplementary material</title>", "<p>This material is available as part of the online article from: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.blackwell-synergy.com/doi/abs/10.1111/\">http://www.blackwell-synergy.com/doi/abs/10.1111/</ext-link></p>", "<p>j.1365-2958.2008.06323.x</p>", "<p>(This link will take you to the article abstract).</p>", "<p>Please note: Blackwell Publishing is not responsible for the content or functionality of any supplementary materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.</p>" ]

[ "<fig id=\"fig01\" position=\"float\"><label>Fig. 1</label><caption><p>Methylation of FrzCD by FrzF and FrzF^CheR<italic>in vitro</italic>.</p><p>A. Cartoons show the domain organization of the wild-type FrzF protein and the FrzF^CheR protein, which lacks FrzF amino acids 450–580. TPR indicates a tetra-trico peptide repeat. Numbers refer to the amino acid position in full-length FrzF.</p><p>B. His-tagged FrzCD was expressed and purified from <italic>E. coli</italic> and incubated <italic>in vitro</italic> with purified FrzF or FrzF^CheR and S-adenosyl methionine (SAM) for 4 h at 32°C. Following the reaction, FrzCD was analysed by SDS-PAGE and Western immunoblotting using purified α-FrzCD antibodies. The white arrowhead shows the mobility of unmethylated FrzCD; the black arrowheads indicate methylated FrzCD. The position of the molecular weight markers is indicated on the left.</p></caption></fig>", "<fig id=\"fig02\" position=\"float\"><label>Fig. 2</label><caption><p>FrzCD methylation point mutants are stably expressed. An immunuoblot of vegetative cell extracts from the methylation point mutants probed with the α-FrzCD antibody. Thirty micrograms of total protein from whole cell extracts were loaded per lane. The white arrowhead shows the mobility of unmethylated FrzCD; the black arrowhead indicates methylated FrzCD. DZ4717 (<italic>6His::frzCD</italic>) is the positive control strain used in this study. All point mutations were made in the DZ4717 background.</p></caption></fig>", "<fig id=\"fig03\" position=\"float\"><label>Fig. 3</label><caption><p>Effect of FrzCD E to A and E to D methylation site mutations on aggregation and fruiting body formation. FrzCD methylation site glutamates (E) 168, 175 and 182 were changed to alanine (A) or aspartate (D) residues by site directed mutagenesis as described in <italic>Experimental procedures</italic>. Cells were spotted at 4 × 10⁹ cells ml⁻¹ on CF fruiting agar and incubated for 4 days at 32°C. Reference strains (DZ4717 (<italic>6His::frzCD</italic>), Δ<italic>frzCD</italic>, and <italic>frzCD</italic>Δ<italic>6</italic>^{−<italic>183</italic>}) are shown in A–C. FrzCD E to A mutants are shown in the middle column (D, F, H, J and L) and FrzCD E to D mutants are shown in the right column (E, G, I, K and M). The small corner inset in each picture is a 7 × magnification of a portion of the original.</p></caption></fig>", "<fig id=\"fig04\" position=\"float\"><label>Fig. 4</label><caption><p>Effect of FrzCD E to A and E to D methylation site mutations on vegetative swarming. Reference strains [DZ4717 (<italic>6His::frzCD</italic>), Δ<italic>frzCD</italic> and <italic>frzCD</italic>^{Δ<italic>6</italic>–<italic>183</italic>}] are shown at the top (striped bars), FrzCD E to A mutants are shown in the left panel (dark grey), and FrzCD E to D mutants are shown in the right panel (light grey). The horizontal axis represents the relative distance swarmed by each strain compared with wild type. Error bars represent the standard deviation of the mean. Five microlitres of 4 × 10⁹ cells ml⁻¹ were spotted on CYE nutrient plates containing 0.4% agar and swarm expansion was measured after 3 days incubation at 32°C. Data shown are from two independent experiments with a total of seven measurements per strain.</p></caption></fig>", "<fig id=\"fig05\" position=\"float\"><label>Fig. 5</label><caption><p>Similarities between the methylation site sequences of <italic>M. xanthus</italic> FrzCD and <italic>E. coli</italic> Tar. The FrzCD peptide containing identified methylated residues is shown aligned with a methylated peptide from the <italic>E. coli</italic> receptor, Tar. Residues that are identical or that share similar charges are indicated on the second row by a letter (amino acid) or ‘+’ respectively. Sites of methylation are indicated by a box and an arrow. Numbers indicate amno acid position in FrzCD.</p></caption></fig>" ]

[ "<table-wrap id=\"tbl1\" position=\"float\"><label>Table 1</label><caption><p>Ions from MS/MS spectrum of a chymotryptic fragment<xref ref-type=\"table-fn\" rid=\"tf1-1\">a</xref> of FrzCD show FrzF methylates site E182.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\">AA<xref ref-type=\"table-fn\" rid=\"tf1-2\">b</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Theoretical<xref ref-type=\"table-fn\" rid=\"tf1-3\">c</xref> b-ions<xref ref-type=\"table-fn\" rid=\"tf1-4\">d</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Observed<xref ref-type=\"table-fn\" rid=\"tf1-5\">e</xref> b-ions</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Theoretical y-ions<xref ref-type=\"table-fn\" rid=\"tf1-6\">f</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Observed y-ions</th></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">I173</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">842.9 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">843.0 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" colspan=\"5\" rowspan=\"1\">H174</td></tr><tr><td align=\"left\" colspan=\"5\" rowspan=\"1\">E175</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T176</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1026.5 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1026.3 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T177</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">824.4</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">824.0</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">A178</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1112.5 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1112.5 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">723.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">723.1</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T179</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1163.0 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1163.0 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">652.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">652.1</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">M180</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1236.6 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1236.9 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">551.2</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">551.1</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">E181</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1301.1 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1301.0 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">E182#</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1372.6 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1372.3 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" colspan=\"5\" rowspan=\"1\">L183</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"tbl2\" position=\"float\"><label>Table 2</label><caption><p>Ions from a MS/MS spectrum of a chymotryptic fragment<xref ref-type=\"table-fn\" rid=\"tf2-1\">a</xref> of FrzCD show FrzF^CheR methylates sites E175 and E182.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\">AA<xref ref-type=\"table-fn\" rid=\"tf2-2\">b</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Theoretical<xref ref-type=\"table-fn\" rid=\"tf2-3\">c</xref> b-ions<xref ref-type=\"table-fn\" rid=\"tf2-4\">d</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Observed<xref ref-type=\"table-fn\" rid=\"tf2-5\">e</xref> b-ions</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Theoretical y-ions<xref ref-type=\"table-fn\" rid=\"tf2-6\">f</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Observed y-ions</th></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">I173</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">849.9 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">849.7 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" colspan=\"5\" rowspan=\"1\">H174</td></tr><tr><td align=\"left\" colspan=\"5\" rowspan=\"1\">E175#</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T176</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1040.5 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1040.6 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T177</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1091.0</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1091.4</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">824.4</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">824.2</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">A178</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1126.5 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1126.6 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">723.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">722.9</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T179</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1177.0 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1177.1 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">652.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">652.0</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">M180</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1250.6 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1251.0 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">551.2</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">551.0</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">E181</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1315.1 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1315.2 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">E182#</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1372.6 (2+)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1372.3 (2+)</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr></tbody></table></table-wrap>", "<table-wrap id=\"tbl3\" position=\"float\"><label>Table 3</label><caption><p>Ions from an MS/MS spectrum of a GluC digested fragment<xref ref-type=\"table-fn\" rid=\"tf3-1\">a</xref> show FrzF^CheR methylates FrzCD on sites E168 and E175.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\">AA<xref ref-type=\"table-fn\" rid=\"tf3-2\">b</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Theoretical<xref ref-type=\"table-fn\" rid=\"tf3-3\">c</xref> b-ions<xref ref-type=\"table-fn\" rid=\"tf3-4\">d</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Observed<xref ref-type=\"table-fn\" rid=\"tf3-5\">e</xref> b-ions</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Theoretical y-ions<xref ref-type=\"table-fn\" rid=\"tf3-6\">f</xref></th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Observed y-ions</th></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">E168</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1546.7</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1546.4</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Q169</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">648.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">648.2</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1403.6</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1403.4</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">A170</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">719.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">719.2</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1275.6</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1275.3</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">A171</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1204.6</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1204.4</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">A172</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">861.4</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">861.2</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1133.5</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1133.3</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">I173</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">974.5</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">974.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1062.5</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1062.3</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">H174</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1111.5</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1111.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">949.4</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">949.3</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">E175#</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1254.6</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1255.3</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T176</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1355.6</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1355.4</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">669.3</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">669.1</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T177</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1456.7</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1456.4</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">568.2</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">568.1</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">A178</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1527.7</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1527.5</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">467.2</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">467.1</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">T179</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1628.8</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1628.5</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">396.1</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">396.1</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">M180</td><td rowspan=\"1\" colspan=\"1\"/><td rowspan=\"1\" colspan=\"1\"/><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">295.1</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">295.1</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"tbl4\" position=\"float\"><label>Table 4</label><caption><p>Effect of methylation site mutations on single cell reversals.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\">Strain</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Average reversals in 30 min (# cells)</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Average reversals in 30 min E to D (# cells)</th></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">DZ4717<xref ref-type=\"table-fn\" rid=\"tf4-1\">a</xref></td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1.58 (84)</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD</italic></td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">0.20 (59)</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"><italic>frzCD</italic>Δ^{<italic>6</italic>−<italic>183</italic>}</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">12.14 <italic>(</italic>24)</td><td rowspan=\"1\" colspan=\"1\"/></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Site 168</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1.67* (49)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1.57* (74)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Site 175</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">0.82 (44)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">0.25 (58)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Site 182</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">0.58 (29)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">0.22 (34)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Sites 168 + 175</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">6.25 (70)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1.34* (59)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Sites 168 + 175 + 182</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">4.63 (56)</td><td align=\"char\" char=\".\" rowspan=\"1\" colspan=\"1\">1.18* (42)</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"tbl5\" position=\"float\"><label>Table 5</label><caption><p>Strains and plasmids used in this study.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><th align=\"left\" rowspan=\"1\" colspan=\"1\">Strain or plasmid</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Relevant feature</th><th align=\"left\" rowspan=\"1\" colspan=\"1\">Source</th></tr></thead><tbody><tr><td align=\"left\" colspan=\"3\" rowspan=\"1\"><italic>M. xanthus</italic></td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ2</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Wild type</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Laboratory collection</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> VB197b</td><td align=\"left\" rowspan=\"1\" colspan=\"1\"><italic>frzCD</italic>^{Δ<italic>6−183</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">##REF##15387825##Bustamante et al. (2004)##</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4480</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">##REF##15387825##Bustamante et al. (2004)##</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4717</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD</italic>, <italic>6His::frzCD</italic> integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4707</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E168A</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4708</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E175A</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4709</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E182A</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4710</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E168D</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4711</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E175D</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4712</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E182D</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4713</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E168A E175A</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4714</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E168D E175D</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4718</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E168A E175A E182A</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> DZ4719</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Δ<italic>frzCD 6His::frzCD</italic>^{<italic>E168D E175D 182D</italic>} integrated in <italic>crtB</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" colspan=\"3\" rowspan=\"1\"><italic>E. coli</italic></td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> Top10</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">General cloning strain</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Invitrogen</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> Tuner</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Protein expression strain</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Novagen</td></tr><tr><td align=\"left\" colspan=\"3\" rowspan=\"1\">Plasmids</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pET28a</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Expression plasmid</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Novagen</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS201</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pET28a with <italic>6His::frzF</italic>^{<italic>cheR</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pCT2</td><td align=\"left\" rowspan=\"1\" colspan=\"1\"><italic>M. xanthus</italic> genomic integration plasmid</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">##REF##17590236##Mignot et al. (2007b)##</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS210</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::FrzCD</italic></td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS211</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E168A</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS212</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E175A</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS213</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E182A</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS214</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E168D</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS215</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E175D</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS216</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E182D</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS217</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E168A E175A</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS218</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E168D E175D</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS221</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E168A E175A E182A</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"> pAS222</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">pCT2 with <italic>6His::frzCD</italic>^{<italic>E168D E175D E182D</italic>}</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">This study</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"SD1\"></supplementary-material>" ]

[ "<table-wrap-foot><fn id=\"tf1-1\"><label>a</label><p>No ions were observed for residues ₁₅₇AASTQHETSSTEQAAA172 of the chymotryptic peptide ₁₅₇AASTQHETSSTEQAAAIHETTATMEEL₁₈₃ (##SUPPL##0##Fig. S2##), so these residues were removed from the left column for simplicity.</p></fn><fn id=\"tf1-2\"><label>b</label><p>Amino acids are indicated by their one letter code; numbers represent their position in FrzCD. The # symbol represents an amino acid that is methylated.</p></fn><fn id=\"tf1-3\"><label>c</label><p>Theoretical ions are calculated by dividing the predicted mass of an ion by the ion's charge, m/z = (M + nH⁺)/n.</p></fn><fn id=\"tf1-4\"><label>d</label><p>A b-ion is an N-terminal charged fragment generated after ion activation causes a peptide bond to break.</p></fn><fn id=\"tf1-5\"><label>e</label><p>Observed ions were found by digesting FrzCD with chymotrypsin and using tandem MS/MS.</p></fn><fn id=\"tf1-6\"><label>f</label><p>A y-ion is a C-terminal charged fragment generated after ion activation causes a peptide bond to break.</p></fn><fn><p>All ions are 1+ charged unless otherwise indicated in parentheses.</p></fn></table-wrap-foot>", "<table-wrap-foot><fn id=\"tf2-1\"><label>a</label><p>No ions were observed for residues ₁₅₇AASTQHETSSTEQAAA172 of the chymotryptic peptide ₁₅₇AASTQHETSSTEQAAAIHETTATMEEL₁₈₃ (##SUPPL##0##Fig. S3A##), so these residues were removed from the left column for simplicity.</p></fn><fn id=\"tf2-2\"><label>b</label><p>Amino acids are indicated by their one letter code; numbers represent their position in FrzCD. The # symbol represents an amino acid that is methylated.</p></fn><fn id=\"tf2-3\"><label>c</label><p>Theoretical ions are calculated by dividing the predicted mass of an ion by the ion's charge, m/z = (M + nH⁺)/n.</p></fn><fn id=\"tf2-4\"><label>d</label><p>A b-ion is an N-terminal charged fragment generated after ion activation causes a peptide bond to break.</p></fn><fn id=\"tf2-5\"><label>e</label><p>Observed ions were found by digesting FrzCD with chymotrypsin and using tandem MS/MS.</p></fn><fn id=\"tf2-6\"><label>f</label><p>A y-ion is a C-terminal charged fragment generated after ion activation causes a peptide bond to break.</p></fn><fn><p>All ions are 1+ charged unless otherwise indicated in parentheses.</p></fn></table-wrap-foot>", "<table-wrap-foot><fn id=\"tf3-1\"><label>a</label><p>No ions were observed for residues ₁₆₄TSST167 of the GluC-generated fragment ₁₆₄TSSTEQAAAIHETTATME181 (##SUPPL##0##Fig. S3B##), so these residues were removed from the left column for simplicity.</p></fn><fn id=\"tf3-2\"><label>b</label><p>Amino acids are indicated by their one-letter code; numbers represent their position in FrzCD. The # symbol represents an amino acid that is methylated.</p></fn><fn id=\"tf3-3\"><label>c</label><p>Theoretical ions are calculated by dividing the predicted mass of an ion by the ion's charge, m/z = (M + nH⁺)/n.</p></fn><fn id=\"tf3-4\"><label>d</label><p>A b-ion is an N-terminal charged fragment generated after ion activation causes a peptide bond to break.</p></fn><fn id=\"tf3-5\"><label>e</label><p>Observed ions were found by digesting FrzCD with chymotrypsin and using tandem MS/MS.</p></fn><fn id=\"tf3-6\"><label>f</label><p>A y-ion is a C-terminal charged fragment generated after ion activation causes a peptide bond to break.</p></fn><fn><p>All ions are 1+ charged.</p></fn></table-wrap-foot>", "<table-wrap-foot><fn id=\"tf4-1\"><label>a</label><p>DZ4717 has the same reversal frequency as wild type (see ##SUPPL##0##Fig. S5##).</p></fn><fn><p>Asterisks indicate that reversals are statistically the same as DZ4717 (Student's <italic>t</italic>-test). Values that were statistically different from DZ4717 had <italic>P</italic>-values less than 0.005 (Student's <italic>t</italic>-test).</p></fn></table-wrap-foot>" ]

[ "<graphic xlink:href=\"mmi0069-0724-f1\"/>", "<graphic xlink:href=\"mmi0069-0724-f2\"/>", "<graphic xlink:href=\"mmi0069-0724-f3\"/>", "<graphic xlink:href=\"mmi0069-0724-f4\"/>", "<graphic xlink:href=\"mmi0069-0724-f5\"/>" ]

[ "<media xlink:href=\"mmi0069-0724-SD1.pdf\" xlink:type=\"simple\" id=\"N0x1c937e0N0x3579e70\" position=\"anchor\" mimetype=\"application\" mime-subtype=\"pdf\"/>" ]

{ "acronym": [], "definition": [] }

[ "<title>1. INTRODUCTION</title>", "<p>Organ\ntransplantation in the clinic became a reality in 1954 when Merrill, Murray,\nand Harrison performed the first successful human vascular organ graft, a\nkidney transplant [##REF##13278189##1##, ##REF##6371266##2##]. However, the donor\nand recipient were monozygotic twins, obviating the need for immunosuppression\nfor organ graft survival. With the development of immunosuppressive regimens,\nthe same group 5 years later performed the first kidney allograft transplantation\nbetween unrelated individuals; that graft survived for 20 years [##UREF##0##3##]. Although successful\ngraft survival was achieved, it rapidly became clear that all immunosuppressive drugs, even the newer\ngenerations of immunosuppressive regimens, are toxic [##REF##16567244##4##, ##REF##17981639##5##]. Immunosuppressive drugs are also known to increase the risk of\ninfection and neoplasia [##REF##14662323##6##, ##REF##11833147##7##], and their associated side effects often lead to\npatient noncompliance [##REF##11072057##8##]. Since most patients eventually reject transplanted allografts\neither acutely or through a process of chronic rejection [##REF##7619392##9##–##REF##16418687##11##], these deleterious\nside effects make organ transplantation a therapy in which the risk/benefit\nratio must be carefully weighed.</p>", "<p> To overcome issues associated with chronic\nimmunosuppression, investigators have focused on approaches that lead to the\ninduction of tolerance to transplanted organ allografts [##REF##9922369##12##]. Operationally, transplantation tolerance is defined as the\nsurvival of a donor allogeneic graft in the absence of immunosuppression. Most\ntransplantation tolerance induction protocols take advantage of information\nresulting from studies on the natural mechanisms by which the immune system prevents\nself-reactivity and autoimmune disease. Two major forms of natural tolerance\nhave been identified: central tolerance and peripheral tolerance.</p>" ]

[]

[]

[]

[]

[ "<p>Recommended by Eiji Matsuura</p>", "<p>Transplantation of allogeneic organs has proven to be an effective therapeutic for a large variety of disease states, but the chronic immunosuppression that is required for organ allograft survival increases the risk for infection and neoplasia and has direct organ toxicity. The establishment of transplantation tolerance, which obviates the need for chronic immunosuppression, is the ultimate goal in the field of transplantation. Many experimental approaches have been developed in animal models that permit long-term allograft survival in the absence of chronic immunosuppression. These approaches function by inducing peripheral or central tolerance to the allograft. Emerging as some of the most promising approaches for the induction of tolerance are protocols based on costimulation blockade. However, as these protocols move into the clinic, there is recognition that little is known as to their safety and efficacy when confronted with environmental perturbants such as virus infection. In animal models, it has been reported that virus infection can prevent the induction of tolerance by costimulation blockade and, in at least one experimental protocol, can lead to significant morbidity and mortality. In this review, we discuss how viruses modulate the induction and maintenance of transplantation tolerance.</p>" ]

[ "<title>2. CENTRAL TRANSPLANTATION TOLERANCE</title>", "<p>In\n1953, Peter Medawar et al. obtained the first experimental evidence that the\nestablishment of allogeneic hematopoietic chimerism leads to the induction of central\ntolerance and permits permanent acceptance of skin allografts [##REF##13099277##13##]. Inspired by the work\ndone in freemartin cattle by Owen in 1945 [##REF##17755278##14##] and the clonal\nselection theory subsequently proposed by Burnet and Fenner [##UREF##1##15##], Medawar demonstrated in\nmice that the transfer of allogeneic hematopoietic cells in utero could induce\ntolerance to skin transplanted from the original donor later in life [##REF##13099277##13##].</p>", "<p> Medawar's\nobservation led Main and Prehn to experimentally induce hematopoietic chimerism\nby treating mice with whole-body irradiation (WBI) and allogeneic bone marrow\ncells, followed by transplantation with donor-strain-matched skin allografts [##REF##13233946##16##]. This protocol successfully\ninduced tolerance to skin allografts, conclusively linking the establishment of\nhematopoietic chimerism with subsequent allograft survival. However, despite\nthe long-term survival of skin allografts on mice treated with WBI and\nallogeneic bone marrow, animals eventually develop lethal graft-versus-host\ndisease (GVHD), a reaction where passenger leukocytes in the donor bone marrow\nor graft mount an immune response against the host. Therefore, modern conditioning\nprotocols to induce central tolerance have been designed to address the common\nobjectives of (1) establishing hematopoietic chimerism using a relatively\nbenign preconditioning protocol that (2) prevents the development of GVHD.</p>", "<p> Despite\nthese common objectives, modern conditioning regimens can differ quite\nsignificantly in their methodology. In preclinical models of hematopoietic\nchimerism, conditioning regimens span the spectrum from myeloablative\nprotocols, which often entail lethal irradiation and subsequent stem\ncell rescue, to noncytoreductive treatments that do not require irradiation, for\nexample, costimulation blockade [##UREF##2##17##–##REF##10861026##19##]. Between these two\nextremes are protocols that significantly weaken the recipient's immune system\nthrough selective antibody-mediated elimination of specific immune populations (e.g., CD4⁺ and CD8⁺ T cells) coupled with targeted irradiation (e.g., thymic irradiation) [##REF##2562984##20##]. These latter protocols\nare often considered nonmyeloablative. In clinical trials, successful\nnonmyeloablative approaches have recently been described [##REF##18216356##21##, ##REF##18216355##22##]. Stable renal allograft\nfunction in recipients for as long as five years after complete withdrawal of immunosuppressive\ndrugs was observed in recipients in which hematopoietic chimerism was\nestablished [##REF##18216356##21##, ##REF##18216355##22##]. These reports\ndocument that in humans, as in rodents, establishment of hematopoietic\nchimerism is a robust approach for the development of central tolerance and the\npermanent survival of donor-specific allografts.</p>", "<title>3. PERIPHERAL TRANSPLANTATION TOLERANCE</title>", "<p>The second major form of tolerance is peripheral\ntolerance. Different from central tolerance in which hematopoietic chimerism\nleads to the clonal deletion of antigen-specific cells during development,\nperipheral tolerance targets pre-existing cells that have already been\ngenerated. To induce tolerance in this population, fundamental insights into\nhow naive antigen-specific T cells become activated have led to protocols\ndesigned to prevent this process. Naive T cell activation is initiated by the\ninteraction of the antigen-specific T cell receptor (TCR) with a peptide\npresented by the MHC. This interaction conveys specificity leading to the\nactivation of only antigen-specific T cells. This signal is often termed as “signal\n1” (##FIG##0##Figure 1##). Following\nTCR-peptide/MHC ligation, a T cell then receives a number of costimulatory\nsignals [##REF##16611323##23##–##REF##16918529##25##]. A key costimulatory\nsignal in this pathway that permits the activated naive T cells to become\nfunctional effector/memory T cells is provided by CD28-CD80/86 interaction [##UREF##3##26##], which has often been\nreferred to as “signal 2.” In early studies, it was shown in vitro that T cells that receive\nsignals through their TCR in the absence of engagement of the CD28-CD80/86\ncostimulation pathway became nonresponsive, a state of T cell nonresponsiveness\noften referred to as anergy [##REF##9922369##12##, ##UREF##4##27##]. Following induction\nof signal 2, cytokines are produced that impart the final signal for T cell\nactivation, and this is termed as “signal 3” [##REF##16824119##24##, ##REF##17513722##28##, ##REF##12732656##29##]. Although these three\ncritical signals are required for the full activation of T cells, additional\nsignals such as those derived from CD40-CD154 interaction can have potent\neffects on the activation of naive T cells (##FIG##0##Figure 1##).</p>", "<p> The\nexistence of a comparable in vivo\nstate of T cell nonresponsiveness has been debated for years until it was independently\nshown to exist by Ohashi et al. [##REF##1901764##30##] and Oldstone et al. [##REF##1901765##31##] using two very similar\nexperimental systems. These investigators generated double-transgenic mice that\nexpressed (1) lymphocytic choriomeningitis virus (LCMV) glycoprotein (GP) [##REF##1901764##30##] or nucleoprotein (NP) [##REF##1901765##31##] under the control of\nthe rat insulin promoter, and (2) a transgenic TCR that recognizes a peptide\nfrom the transgenic LCMV protein. In unmanipulated mice, the transgenic T cells\nmigrate from the thymus into the peripheral tissues and encounter their cognate\nantigen, but they remain nonresponsive to islets expressing GP or NP. However,\nLCMV infection reverses this state of nonresponsiveness, leading to a diabetic\nphenotype resulting from the destruction of pancreatic islets expressing the\ntransgenic protein [##REF##1901764##30##, ##REF##1901765##31##]. These data support a\nmechanism where the LCMV-reactive T cells in naive mice encounter antigen in\nthe absence of costimulation and become nonresponsive (tolerant), and further\nshow that environmental perturbation can break this nonresponsive state. This model\nserves as the conceptual basis for the induction of peripheral transplantation\ntolerance, where the in vivo disruption\nof the costimulatory process—referred to as\ncostimulation blockade—leads to the\ninduction of tolerance in an antigen-specific manner [##REF##9922369##12##].</p>", "<p>Costimulation\nblockade therapies can target several different steps in the process of T cell\nactivation. However, the CD40-CD154 pathway linking signal 1 to signal 2 has been\nidentified to be a critical step in the activation of naive T cells. Anti-CD154\nmAb blocks the interaction between CD154 on the T cell and CD40 on the APC [##REF##9368772##32##, ##UREF##5##33##], and prevents the differentiation\nbetween naive T cells and effector/memory T cells [##UREF##5##33##] (##FIG##0##Figure 1##).</p>", "<p> In peripheral tolerance induction protocols,\nanti-CD154 monotherapy significantly improves islet [##REF##7568172##34##] and cardiac [##REF##8560571##35##] allograft survival in\nmice and islet allograft survival in nonhuman primates [##REF##11707732##36##–##REF##15816883##39##]. In combination with a\ndonor-specific transfusion (DST), anti-CD154 monoclonal antibody (mAb) induces\npermanent islet [##REF##7568172##34##] and prolonged skin [##REF##9256196##40##] allograft survival in\nmice. DST provides selective activation of the alloantigen-specific T cells,\nand we have shown that the subsequent blockade of costimulation by anti-CD154\nmAb leads to selective depletion of only the specific alloantigen-reactive CD8⁺ T cells [##REF##11714833##41##, ##REF##10605049##42##]. Another reagent,\nCTLA4-Ig, binds to the costimulatory molecules CD80/86 on the APC. This blocks\nits interaction with CD28 on the T cell, preventing signal 2. CTLA4-Ig\nmonotherapy induces the survival of xenogeneic islets [##REF##1323143##43##] and allogeneic cardiac\ngrafts [##REF##8228826##44##]. Not surprisingly, the\ncombination of anti-CD154 mAb and CTLA4-Ig has shown great potential in prolonging\nskin and cardiac allograft survival in mice [##REF##8632801##45##].</p>", "<p> Effective \nas a peripheral tolerance induction protocol, costimulation blockade protocols based\non blockade of the CD40-CD154 pathway have also been used to establish\nhematopoietic chimerism leading to the generation of central tolerance [##UREF##2##17##–##REF##10861026##19##]. By establishing\nmultilineage hematopoietic chimerism, these noncytoreductive protocols have\nproven to promote robust transplantation tolerance to a variety of solid-organ\nallografts across fully allogeneic barriers when transplanted several weeks\nafter bone marrow transplantation (BMT) [##UREF##2##17##, ##REF##12952928##18##] or being concurrent\nwith BMT [##REF##10861026##19##, ##REF##17564635##46##]. Furthermore, because\ndonor-reactivity against the host is dependent on the CD40-CD154 pathway [##REF##7521888##47##], costimulation blockade\neffectively establishes hematopoietic chimerism in the absence of GVHD [##UREF##2##17##, ##REF##12952928##18##].</p>", "<title>4. VIRUS INFECTION AND\nTRANSPLANTATION TOLERANCE</title>", "<p> As costimulation blockade protocols move\ncloser to clinical reality, there is concern that virus infection during\ntolerance induction may (1) induce tolerance to the virus, (2) prevent the\ninduction or maintenance of tolerance to the organ allograft, or (3) increase\nrisk to the host. Viruses are known to stimulate innate immunity by activating various\npattern recognition receptors (PRRs), such as Toll-like receptors (TLRs) and retinoic\nacid inducible gene-I- (RIG-I-) like receptors (RLRs) [##REF##17979849##48##]. Activation of innate\nimmunity by virus infection leads to the modulation of adaptive immunity, and it\nhas been shown to impair transplantation tolerance induction and allograft\nsurvival [##REF##16869800##49##–##REF##10666251##57##].</p>", "<p>For example, infection with LCMV before [##REF##12682237##54##], at the time of [##REF##11673506##51##, ##REF##10945944##56##], or shortly after costimulation\nblockade for the induction of peripheral tolerance [##REF##10666251##57##] impairs allograft\nsurvival. Mice treated with costimulation blockade rapidly reject skin\nallografts if they are infected with LCMV shortly after skin transplantation [##REF##10666251##57##]. Interestingly, this\neffect appeared to be virus-specific, as infection with vaccinia virus (VV) and\nmurine cytomegalovirus (MCMV) did not engender allograft rejection [##REF##10666251##57##]. Furthermore, skin\nallograft survival is significantly shortened in LCMV-immune mice treated with\na peripheral tolerance induction protocol consisting of DST and anti-CD154 mAb\ncombination therapies [##REF##12682237##54##]. Additionally, TLRs\nand their proinflammatory role in responding to infection and ischemia are\nbeing increasingly seen as a serious obstacle to solid-organ transplantation [##REF##16495791##58##–##REF##16785361##60##].</p>", "<p>Barriers to the induction of hematopoietic\nchimerism and establishment of central tolerance in the setting of viral\ninfection have also been reported. Anti-CD154 mAb, CTLA4-Ig, and busulfan\ntreatment fails to induce bone marrow chimerism and tolerance to skin\nallografts in the setting of multiple viral infections [##REF##12813024##53##]. Moreover, using a\nnonmyeloablative protocol where anti-CD154 mAb treatment was coupled with\nsublethal irradiation, Forman et al.\nobserved that infection with LCMV on the day of BM transplantation not only resulted\nin allograft rejection but also proved lethal to the recipient [##REF##12055213##55##]. Interestingly,\nconditioned recipients that were infected and given syngeneic BM grafts did not\ndie. Recipients of allogeneic BM died by a type I interferon- (IFN-) dependent\nmechanism, whereas mice deficient in the type I IFN receptor survived. The\nrecent deaths of a cluster of human transplant recipients of LCMV-infected\norgans make this finding particularly relevant to the safety and efficacy of\ntolerance induction protocols based on costimulation blockade [##REF##16723615##61##, ##REF##18256387##62##].</p>", "<title>5. INNATE IMMUNE ACTIVATION BY VIRUS INFECTION</title>", "<p>It has\nbeen shown that mice infected with LCMV concurrent to costimulation blockade\ntreatment [##REF##10945944##56##, ##REF##11064087##63##] or persistently infected with\nLCMV clone 13 prior to costimulation blockade treatment [##REF##12421910##52##] rapidly reject skin\nallografts. In a transgenic TCR model, LCMV prevents the deletion of\nalloreactive CD8⁺ T cells that is ordinarily induced by costimulation\nblockade [##REF##10945944##56##, ##REF##11064087##63##]. In this same model system,\ninjection of a TLR agonist similarly prevents the deletion of host alloreactive\nCD8⁺ T cells which are required for skin allograft rejection [##REF##16424185##64##].</p>", "<p>Surprisingly,\nthe TLR4 agonist LPS impairs CD8⁺ T cell deletion and shortens skin\nallograft survival by activating host cells [##REF##16424185##64##] rather than donor cells [##REF##16424185##64##, ##REF##17982052##65##], even though the transgenic CD8\nT cells recognize donor antigen via the direct pathway. Furthermore, LPS\nrequired the expression of the adaptor molecule myeloid differentiation primary\nresponse gene-88 (MyD88) on the recipient to shorten allograft survival [##REF##17982052##65##, ##REF##16970798##66##]. These findings are consistent\nwith clinical data suggesting that TLR4 polymorphisms on the host, but not the\ndonor, correlate with allograft survival [##REF##12773319##67##]. Together, these data suggest\nthat TLR activation induces a soluble mediator that augments host T cell\nactivation, perhaps through a process of bystander activation (see below).</p>", "<p>Numerous\ncytokines are reported to be important in the activation of CD8⁺ T\ncells, including IL-12 [##REF##12732656##29##], TNF<italic>α</italic> [##REF##11739497##68##, ##REF##15383581##69##], and IFN-<italic>α</italic>/<italic>β</italic> [##REF##15814665##70##]. While IL-12 and TNF<italic>α</italic>\nare dispensable for shortened allograft survival induced by LPS in\ncostimulation blockade treatment protocols [##REF##16424185##64##], IFN-<italic>α</italic>/<italic>β</italic> has\nbeen reported to be absolutely essential for LPS to prime CTLs and induce\nallograft rejection [##REF##17982052##65##]. Type I IFNs also proved\nindispensable for allograft rejection mediated by the dsRNA mimetic and TLR3\nagonist poly I:C [##REF##17982052##65##]. Emerging data suggest that\nIFN-<italic>α</italic>/<italic>β</italic>\ncan be induced by viruses through a growing number of pathogen recognition\nreceptor systems [##REF##11607032##71##–##REF##12773480##74##].\nThus the induction of IFN-<italic>α</italic>/<italic>β</italic> by\nvirus infection or TLR ligation has emerged as an important obstacle to the establishment\nof peripheral transplantation tolerance as well as to the maintenance of\nself-tolerance [##REF##17562353##75##].</p>", "<title>6. SIGNALING PATHWAYS INVOLVED\nIN INNATE IMMUNE CELL ACTIVATION\nBY VIRUS INFECTION</title>", "<p> How\ndoes virus-mediated activation of innate immunity lead to the production of IFN-<italic>α</italic>/<italic>β</italic>?\nAt present, the two best-characterized IFN-<italic>α</italic>/<italic>β</italic>-inducing\nviral recognition systems are members of the TLR and the retinoic acid\ninducible gene-I- (RIG-I-) like receptor (RLR) families (##FIG##1##Figure 2##). These receptors are activated by sensing viral nucleic\nacids either in the cytosol (RLR) or in endosomes (TLR) of cells [##REF##17892846##76##]. Cytosolic receptors that\ndetect nucleic acids upon viral infection are expressed ubiquitously by\nnucleated cells, while endosomal receptors, which detect viral particles that\nare engulfed from outside rather than from direct infection, are expressed in\nspecialized cells of the innate immune system such as macrophages and dendritic\ncells [##REF##16979569##77##].</p>", "<p>Cytosolic\nRLRs, exemplified by the proteins RIG-I and melanoma differentiation factor-5\n(MDA5), recognize double stranded RNA (dsRNA) located in the cytosol following\nreplication by an RNA virus, or infection by a dsRNA-genome virus, through\ninteraction with their helicase domains [##REF##17979849##48##]. RLRs contain a caspase\nrecruitment domain (CARD) [##REF##15208624##72##] which links detection of\nviral dsRNA to transcription of IFN-<italic>α</italic>/<italic>β</italic> by forming homotypic interactions with\nthe CARD-containing molecule interferon-<italic>β</italic> promoter stimulator (IPS-1, also\nknown as mitochondrial antiviral signaling protein (MAVS), CARD adaptor\ninducing IFN-B (CARDIF), and virus-induced signaling adaptor (VISA)) [##REF##16127453##79##–##REF##16153868##82##]. Activation of IPS-1 triggers\nmembers of the I<italic>κ</italic>B kinase (IKK) family, specifically TANK-binding kinase 1\n(TBK-1) and IKK<italic>ε</italic> (also known as inducible I<italic>κ</italic>B kinase, IKK-i), to phosphorylate\nand activate interferon regulatory factory (IRF)-3 and/or IRF7 [##REF##16713980##83##–##REF##16306937##88##]. Once activated, IRF3 and\nIRF7 translocate to the nucleus and bind to interferon-stimulated response\nelements (ISREs) to induce the expression of IFN-<italic>α</italic> and IFN-<italic>β</italic>, as well as other\nIFN-inducible genes [##REF##17979849##48##, ##REF##15800576##89##, ##UREF##6##90##].</p>", "<p> It has\nrecently been recognized that cytoplasmic sensing of DNA can also trigger IFN-<italic>α</italic>\nand IFN-<italic>β</italic> production [##REF##16286919##91##–##REF##16301743##93##]. This pathway is thought to\nintersect with the RIG-I and MDA5 pathways at the level of TBK-1 and IKK-I [##REF##16286919##91##], and it requires IRF3 for IFN-<italic>α</italic>/<italic>β</italic> induction [##REF##16413926##92##]. A candidate cytosolic recognition receptor that senses and is activated\nby DNA has been described [##REF##17618271##94##]. This receptor, known as\nDNA-dependent activator of IFN-regulatory factors (DAI), was reported to induce\ntype I IFN upon recognition of bacterial and mammalian as well as viral DNAs [##REF##17618271##94##].</p>", "<p>With\nthe exception of TLR4, all known TLRs that induce type I IFN recognize nucleic\nacids, and are found in the endosomal compartment of cells. These include TLR3,\nTLR7, TLR8, and TLR9. Unlike the cytoplasmic nucleic acid receptors, the\ncellular distribution of endosomal TLRs is much more restricted. TLR7 and TLR9,\nwhich recognize ssRNA [##REF##14976262##95##, ##REF##14976261##96##] and unmethylated DNA that\ncontain CpG motifs [##REF##11130078##97##], respectively, are expressed\nhighly on both conventional (cDC) and plasmacytoid (pDC) dendritic cells. However,\nthey can also be expressed on other hematopoietic cells, including B cells [##REF##15454922##98##, ##REF##14751761##99##]. TLR3, which recognizes dsRNA [##REF##11607032##71##], has a broader distribution\nthan TLR7 and TLR9, and can be found on nonhematopoietic cells such as\nastrocytes and epithelial cells of the cervix, airway, uterus, vagina,\nintestine, and cornea [##REF##17892846##76##, ##REF##15454922##98##–##REF##16497588##100##]. Its expression, however, is thought\nto be highest in cDCs [##REF##17892846##76##, ##REF##16497588##100##].</p>", "<p>Similar\nto the other non-IFN-<italic>α</italic>/<italic>β</italic>-inducing TLRs, TLR3, 7, 8, and 9 are capable of\nactivating both NF-<italic>κ</italic>B and MAPK cascades and triggering the transcription of\nscores of proinflammatory cytokines and chemokines [##REF##17892846##76##, ##REF##14751761##99##, ##REF##16497588##100##]. However, the endosomal TLRs are\nalso capable of signaling through additional cascades, which results in the\nexpression of type I IFNs. Recognition of dsRNA by TLR3 results in the\nactivation of the adaptor molecule Toll/interleukin-1 receptor (TIR) domain-containing\nadaptor protein inducing IFN-<italic>β</italic> (TRIF) [##REF##12539043##101##]. TRIF interacts with tumor\nnecrosis factor receptor-associated factor (TRAF)-3 to activate TBK1 [##REF##16306937##88##] and, as described above,\nleads to the activation of IRF3 and IRF7 and induction of type I IFN. In\ncontrast, the coupling of TLR7 and TLR9 to IFN-<italic>α</italic>/<italic>β</italic> production involves the adaptor\nmolecule MyD88 [##REF##11130078##97##, ##REF##11812998##102##]. Following recognition of\neither ssRNA or unmethylated DNA, a large complex consisting of MyD88, TRAF3,\nTRAF6, IL-1 receptor-associated kinase (IRAK)-4, IRAK-1, IKK-<italic>α</italic>, and IRF-7 is\nrecruited to the TLR [##REF##17979849##48##, ##REF##16306936##87##, ##REF##16306937##88##, ##REF##15361868##103##–##REF##16612387##105##]. Following recruitment of the\ncomplex, cytokines downstream of NF-<italic>κ</italic>B are stimulated, and type I IFN expression\nis induced in an osteopontin (OPN) [##REF##16604075##106##] and IRF7-dependent fashion [##REF##17979849##48##, ##REF##15800576##89##]. Interestingly, stimulation\nof TLR7 and TLR9 in cDCs is capable of inducing the expression of cytokines\nthat are downstream of the NF-<italic>κ</italic>B pathway, such as IL-6 and IL-12. However, only\npDCs are capable of producing IFN-<italic>α</italic> in response to ssRNA and CpG-containing DNA [##REF##17892846##76##]. As exemplified by the\nmultitude of signaling pathways by which TLRs can activate innate immunity, it\nis clear that virus or microbial infection has multiple ways to active innate\nimmunity and modulate the adaptive immune system during tolerance induction.</p>", "<title>7. MECHANISMS OF VIRUS-MEDIATED MODULATION\nOF TRANSPLANTATION TOLERANCE</title>", "<p>There\nare multiple mechanisms by which virus infection or TLR agonists may modulate\ntolerance induction and allograft survival. We will focus on three potential\nmechanisms. First, virus infection can mature APCs to prime non-cross-reactive\nT cells, a process called bystander activation [##REF##15621563##107##, ##REF##16418524##108##]. Second, virus infection may\nstimulate innate immune cells to produce cytokines that suppress\ntolerance-promoting regulatory T cells [##REF##16921386##109##]. Third, virus infection may\nlead to the generation of virus-specific T cells that can cross-react with\nalloantigens, a phenomenon known as heterologous immunity [##REF##16824126##110##].</p>", "<title>7.1. Bystander activation</title>", "<p> A\nmechanism by which virus infection may modulate tolerance induction is through\nbystander activation. As described above, virus infection activates innate\nimmunity, and is able to mature APCs to “license” them to activate non-cross-reactive\nT cells. CD4⁺ T cells are known to play a pivotal role in the\nlicensing of antigen-presenting cells (APCs) [##REF##14657225##111##]. The intercourse between\nantigen-specific CD4⁺ T cells and antigen-presenting APCs is thought\nto be crucial for the generation of a full immune response. In the setting of\nviral infection, virus-specific CD4⁺ T cells facilitate the\nmaturation of virus-presenting APCs via CD154-CD40 interactions. Consequently,\nthe APC is stimulated to upregulate costimulatory molecules, as well as to secrete\nproinflammatory cytokines. These molecules then feed back on the T cell,\nstimulate it to become fully activated, and release additional inflammatory\ncytokines and growth factors. Allospecific T cells that have encountered cognate alloantigen can be activated in this inflammatory milieu even if they do not cross-react with viral antigens. This process is traditionally referred to as bystander activation [##REF##14657225##111##].</p>", "<p>Viruses\nhave also been shown to mature APCs independently of the normally required CD154-CD40\ninteraction. LCMV infection stimulates the upregulation of MHC classes I and\nII, CD40, CD80, and CD86 in the presence of CTLA-4-Ig and anti-CD154 mAb [##REF##11673506##51##]. The molecular mechanisms\nthat govern this process have not been fully elucidated; however, the induction\nof type I IFNs by virus-infected APCs is a likely candidate. \nIFN-<italic>α</italic>/<italic>β</italic> is\nknown to directly induce the maturation of immature DCs, and it results in the\nupregulation of MHC and costimulatory molecules [##REF##9712065##112##, ##UREF##7##113##]. Given that pDCs can produce\nup to a thousand-fold more type I IFN than other cells [##UREF##7##113##, ##REF##14991609##114##], we propose that viral\ndetection by pDCs triggers the release of IFN-<italic>α</italic>/<italic>β</italic> that can in turn act in a paracrine\nor autocrine fashion to mature alloantigen-presenting APCs (##FIG##2##Figure 3##). Thus,\nthese “IFN-<italic>α</italic>/<italic>β</italic>-licensed” alloantigen-presenting APCs could directly stimulate\nalloreactive T cells, even in the presence of costimulation blockade.</p>", "<title>7.2. Regulatory cell suppression</title>", "<p>The induction and maintenance of CD4⁺ regulatory T cells (Tregs) are essential\nto allograft survival [##REF##9616216##115##–##REF##15371665##117##]. Therefore, a second\nmechanism by which viruses could impair tolerance induction is through\nmodulation of the generation or activity of this important T cell subset. In addition\nto priming cells through an IFN-<italic>α</italic>/<italic>β</italic>-dependent\nmechanism, TLR activation also prevents the intragraft recruitment of regulatory\nT cells in an MyD88-dependent manner [##REF##16970798##66##]. This observation extended\nearlier work showing that the MyD88 pathway plays an important role in the\nrejection of minor antigens [##REF##12750407##118##] and cardiac allografts [##REF##17015716##119##].</p>", "<p>IL-6 is\nan MyD88-dependent cytokine that has emerged as a candidate mediator for\nimpairing regulatory T cell generation and function; its production is\ndiminished in untreated [##REF##17015716##119##]—as well as\nLCMV-infected [##REF##15724245##120##]—mice deficient in\nMyD88. CD4⁺ T cells develop a FoxP3⁺ regulatory T cell phenotype\nwhen they are activated in the presence of TGF-<italic>β</italic>. However, when CD4⁺ T cells are\nactivated in the presence of TGF-<italic>β</italic>\nand IL-6, this regulatory phenotype is suppressed and the cells develop a\nproinflammatory TH17 cell phenotype [##REF##16648838##121##] (##FIG##3##Figure 4##). Therefore, virus infection may precipitate allograft\nrejection by preventing the generation of Tregs following costimulation\nblockade and instead favor development of proinflammatory effector T cells.</p>", "<p>IL-6\nhas also been shown to be important in regulating antigen-specific adaptive\nimmune responses via additional mechanisms. Pasare et al. demonstrated that microbial induction of the TLR pathway\non DCs enabled effector T cells to overcome suppression by CD4⁺CD25⁺ regulatory cells [##REF##12532024##122##] (##FIG##3##Figure 4##). They reported\nthat secretion of soluble mediators (principally IL-6) by TLR-activated DCs\ncould render effector T cells refractory to Treg-mediated regulation, permitting\nactivation of antigen-specific T cells in the presence of regulatory T cells. Hence,\nvirus infection may trigger allograft rejection by compromising key regulatory\nmechanisms such as preventing the generation of regulatory T cells by\ncostimulation blockade as well as by enabling alloreactive T cells to escape\nTreg-mediated suppression.</p>", "<title>7.3. Heterologous immunity</title>", "<p>The classic view of clonal T cell\nactivation is that one TCR interacts with one cognate antigen. However, we now\nunderstand that TCR binding is degenerate, and can recognize multiple related\nand unrelated antigens. The ability of an antigen-specific T cell to\ncross-react with multiple antigens, known as heterologous immunity [##REF##16824126##110##], can influence\nimmunodominance, protective immunity, and immunopathology during subsequent\nviral infections [##REF##16824126##110##, ##UREF##8##123##, ##REF##15528078##124##].</p>", "<p>In\nstudies of peripheral tolerance induction, of particular interest to transplant\nscientists is the observation that virus-specific T cells cross-react with\nalloantigens (##FIG##4##Figure 5##) [##REF##2418107##125##, ##REF##8093891##126##]. Yang et al. have reported that acute infection with VV, MCMV, or\narena viruses LCMV and pichinde virus (PV) resulted in the spontaneous\ngeneration of cytotoxic lymphocytes (CTLs) with cytolytic activity towards\nallogeneic cells [##REF##3487705##127##, ##REF##2537363##128##]. These results were further\nsupported by Nahill and Welsh [##REF##8093891##126##], who used limiting dilution\nanalyses to demonstrate that T cell clones specific for virus-infected\nsyngeneic cells also kill uninfected allogeneic targets. Our report using\nvirus-specific tetramers and an intracellular cytokine assay confirmed the\nfindings that LCMV infection led to the generation of virus-specific CD8 T\ncells that cross-react with alloantigens, and further showed that virus-immune\nmice were refractory to the induction of tolerance by costimulation blockade [##REF##10666251##57##]. Others have also reported\nthat virus-immune mice are refractory to tolerance induction by costimulation\nblockade [##REF##12813024##53##]. Because memory T cells are\nresistant to the induction of tolerance by costimulation blockade [##REF##15621563##107##, ##REF##16418524##108##], our data suggest that the\nallo-cross-reactive virus-specific memory T cells may precipitate the rejection\nof allografts even in the presence of costimulation blockade.</p>", "<title>8. VIRUS INFECTION AND ESTABLISHED\nALLOGRAFT SURVIVAL</title>", "<p>Surprisingly, mice infected with\nLCMV one day after transplantation also exhibit shortened allograft survival [##REF##10666251##57##]. Interestingly, the longer time\nafter transplantation is, the less impact LCMV infection has on allograft\nsurvival. The deletion of alloreactive CD8⁺ T cells is thought to be\ncomplete at this time [##REF##11714833##41##, ##REF##10605049##42##], making it improbable that\nLCMV is interfering with deletion. However, because costimulation blockade\nprotocols are only implemented during the peritransplant period, it is possible\nthat LCMV infection shortly after transplantation prevents the generation of\nregulatory T cells, which have been shown to require up to 30 days after\ncostimulation blockade to develop [##REF##11801641##129##]. Further research is\nnecessary to elucidate the mechanisms by which LCMV shortens allograft survival\nduring the post-transplantation timeframe.</p>", "<title>9. SUMMARY</title>", "<p>Viral infection presents a potent barrier to the induction of transplantation\ntolerance. In this review, we have discussed potential mechanisms by which\nviral infection modulates organ allograft survival in the setting of\ntransplantation tolerance. We have briefly summarized data on three mechanisms\nby which viral infection may mediate these effects: bystander activation,\nmodulation of Tregs, or heterologous immunity. Recognition of viruses by\npattern recognition receptors on innate immune cells can also directly\nstimulate the maturation of APCs, and thus may lead to bystander activation and\nlicensing of alloreactive T cells. Activation of APCs by viruses may trigger\nthe release of cytokines such as IL-6 that can prevent the generation and/or\nfunction of regulatory T cells that are essential for transplantation\ntolerance. Finally, heterologous immunity may be responsible for the\ndiscrepancy that has been encountered when tolerance strategies that work in specific\npathogen-free rodent models fail when translated to nonhuman primates and to humans [##REF##16861933##130##], which have been exposed to a\nvariety of pathogens and thus have large memory T cell pools. Understanding the \ncellular and molecular mechanisms by which viruses and other microbial\norganisms modulate transplantation tolerance may lead to novel approaches that\nimprove the efficacy of allogeneic organ transplantation.</p>" ]

[ "<title>ACKNOWLEDGMENTS</title>", "<p>This \nwork is supported in part by the National Institutes of Health Research Grant\nno. AI42669, the American Diabetes Association Grant no. 7-05-PST-02, the\nJuvenile Diabetes Research Foundation, and a Diabetes Endocrinology Center\nResearch Grant\nDK32520 from the National Institutes of Health. The contents of this\npublication are solely the responsibility of the authors and do not necessarily\nrepresent the official views of the National Institutes of Health. D. M. Miller\nand T. B. Thornley contributed equally to this work.</p>" ]

[ "<fig id=\"fig1\" position=\"float\"><label>Figure 1</label><caption><p>\n<italic>Costimulation blockade</italic>. Activation of a T cell involves a series of interactive steps with an APC. The first signal imparts antigen specificity and commences when the TCR engages the antigen/MHC complex presented by the APC. This signal is commonly referred to as “signal 1.” In subsequent steps, the T cell receives a number of costimulatory signals, including those following interaction of CD154 on T cells with CD40 on APCs, which matures the APC to upregulate expression of CD80/86. The interaction of CD28 with CD80/86 is termed as “signal 2” and activates APCs to secrete cytokines, which provide the final activation signals to the T cell; this step is commonly referred to as “signal 3.” Protocols based on costimulation blockade can prevent T cell activation by targeting steps in the T cell activation cascade. Anti-CD154 mAb blocks the interaction between CD154 and CD40, and prevents the APC from upregulating CD80/86, blocking full APC activation. This prevents the secretion of proinflammatory cytokines, thus depriving the T cell of signal 3. As a result of costimulation blockade, the T cell does not develop an activated phenotype, and consequently becomes nonresponsive (tolerant) to allogeneic antigens.</p></caption></fig>", "<fig id=\"fig2\" position=\"float\"><label>Figure 2</label><caption><p>\n<italic>Pathogen recognition systems</italic>. The innate immune system senses viral pathogens by recognizing distinct pathogen-associated molecular patterns (PAMPs) using various pattern recognition receptors (PRRs). Two of the best-characterized virus-sensing PRRs include member of the Toll-like receptors (TLRs) and retinoic acid inducible gene-I- (RIG-I-) like receptors (RLRs) families. These PRRs couple the recognition of viral PAMPs to the induction of proinflammatory cytokines through various signaling cascades. The cytosolic RNA helicase receptors MDA5 and RIG-I initiate the cascade by recruiting the Cardif/TBK1 complex after sensing viral RNA. This activates the kinase TBK1 to phosphorylate interferon regulatory factor (IRF)-3 and IRF7, resulting in their nuclear translocation and the transcription of IFN<italic>α</italic>/<italic>β</italic>. The cell surface receptor TLR4, in partnership with CD14, couples the recognition of respiratory syncytial virus fusion protein [##REF##11062499##78##] to cytokine induction by signaling through the MyD88-dependent as well as the MyD88-independent pathways. The endosomal TLRs, TLR7, TLR8, and TLR9 also signal through MyD88 to activate inflammatory cytokines such as TNF, IL-6, and IFN-<italic>α</italic>/<italic>β</italic>. The other endosomal TLR (TLR3) signals through the MyD88-independent pathway via the TIR domain-containing adaptor molecule TRIF. Via TRIF, TLR3 signaling can activate NF-kB using TRAF6, and in addition, can induce type I IFN expression probably via TRAF3, TBK1, and IRF3.</p></caption></fig>", "<fig id=\"fig3\" position=\"float\"><label>Figure 3</label><caption><p>\n<italic>Bystander activation of alloreactive T cells by “virus-licensed” APCs</italic>. Upon viral infection, detection of pathogen-associated molecular patterns by PRRs can stimulate the production of inflammatory cytokines such as IFN-<italic>α</italic>/<italic>β</italic>, TNF-<italic>α</italic>, and IL-6. These cytokines activate alloantigen-processing APCs in a paracrine or autocrine fashion to upregulate MHC classes I and II, as well as costimulatory molecules, such as CD80 and CD86. The heightened expression of costimulatory molecules elicits the proliferation and differentiation of alloreactive T cells.</p></caption></fig>", "<fig id=\"fig4\" position=\"float\"><label>Figure 4</label><caption><p>\n<italic>Modulation of regulatory mechanisms by virus infection</italic>. Regulatory T cells play a crucial role in transplantation tolerance to allogeneic organs. Regulatory mechanisms that prevent immune attack on allogeneic tissues may be compromised in the setting of viral infection by at least two mechanisms. Release of inflammatory cytokines by virus-infected cells can prevent the differentiation of uncommitted naive CD4⁺ T cells into Tregs. Naive CD4⁺ T cells can differentiate into regulatory T cells in the presence of TGF-<italic>β</italic>. However, in the presence of TGF-<italic>β</italic> and proinflammatory cytokines such as IL-6, and perhaps IL-21, naive T cells can be skewed to turn into effector T cells such as the IL-17-producing TH17 cells. In a separate mechanism, release of cytokines such as IL-6 by infected APCs can render alloreactive effector cells refractory to suppression by regulatory T cells.</p></caption></fig>", "<fig id=\"fig5\" position=\"float\"><label>Figure 5</label><caption><p>\n<italic>Heterologous immunity; cross-reactivity between viral and allogeneic antigens.</italic> Unlike the very small proportion of naive T cells that can respond to any given pathogen (reported to be ~1:200 000), the frequency of T cells that directly recognize allogeneic antigens, such as MHC, is thought to be within 1:100–1:10. A proportion of those TCRs that recognize alloantigens, therefore, may have arisen as a result of virus infection that induces virus-specific T cells that cross-react with allo-MHC. Activation of these T cells may result in the recognition of MHC molecules found on donor tissues, such as the endothelium of transplanted organs, precipitating allograft rejection.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"CDI2008-742810.001\"/>", "<graphic xlink:href=\"CDI2008-742810.002\"/>", "<graphic xlink:href=\"CDI2008-742810.003\"/>", "<graphic xlink:href=\"CDI2008-742810.004\"/>", "<graphic xlink:href=\"CDI2008-742810.005\"/>" ]

[]

{ "acronym": [ "APC:", "BMT:", "CARD:", "CARDIF:", "cDC:", "CTL:", "DC:", "DAI:", "dsRNA:", "DST:", "GP:", "GVHD:", "IFN:", "IKK:", "IKK-I:", "IPS-1:", "IRAK:", "IRF:", "ISRE:", "LCMV:", "mAb:", "MAVS:", "MDA5:", "MCMV:", "MyD88:", "NP:", "OPN:", "PRR:", "PV:", "RIG-I:", "RLR:", "pDC:", "TBK-1:", "TCR:", "TIR:", "TLR:", "TRAF:", "Treg:", "TRIF:", "VISA:", "VV:", "WBI:" ], "definition": [ "Antigen presenting cell", "Bone marrow transplantation", "Caspase recruitment domain", "CARD adaptor inducing IFN-B", "Conventional dendritic cell", "Cytotoxic T lymphocytes", "Dendritic cell", "DNA-dependent activator of\nIFN-regulatory factors", "Double stranded RNA", "Donor-specific transfusion", "Glycoprotein", "Graft-versus-host disease", "Interferon", "IκB kinase", "Inducible IκB kinase", "Interferon-β promoter stimulator", "IL-1 receptor-associated kinase", "Interferon regulatory factory", "Interferon-stimulated response element", "Lymphocytic choriomeningitis virus", "Monoclonal antibody", "Mitochondrial antiviral signaling protein", "Melanoma differentiation factor-5", "Murine cytomegalovirus", "Myeloid differentiation primary response gene-88", "Nucleoprotein", "Osteopontin", "Pattern recognition receptor", "Pichinde virus", "Retinoic acid inducible gene I", "RIG-I-like receptor", "Plasmacytoid dendritic cell", "TANK-binding kinase 1", "T cell receptor", "Toll/interleukin-1 receptor", "Toll-line receptor", "Tumor necrosis factor\nreceptor-associated factor", "Regulatory T cell", "TIR-domain-containing adaptor\nprotein inducing IFN-β\n", "Virus-induced signaling adaptor", "Vaccinia virus", "Whole-body irradiation" ] }

[]

[]

[]

[]

[]

[]

[ "<p>Each year, the International Society for Computational Biology (ISCB; <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.iscb.org\">http://www.iscb.org</ext-link>) gives two major awards to leaders and innovators in the field of bioinformatics. The awards committee, composed of current and past directors of the Society and previous award winners, has announced that the 2008 Accomplishment by a Senior Scientist Award will be given to David Haussler of the University of California Santa Cruz, Santa Cruz, California, United States, and the 2008 Overton Prize for outstanding achievement in early to midde career will be awarded to Aviv Regev of MIT and the Broad Institute of MIT and Harvard, Cambridge, Massachusetts, United States. Both award winners are Associate Editors for <italic>PLoS Computational Biology</italic>, an official journal of ISCB.</p>", "<p>Søren Brunak, director of the Center for Biological Sequence Analysis at Technical University of Denmark and ISCB Awards Committee chair, admitted that the choice of award winners had not been easy. “These awards are a sign of recognition of achievement not just from ISCB, but from the whole bioinformatics community,” he said. “It is a significant honor to receive one.” He highlighted the fact that the 2008 award winners had made major contributions to both the development of algorithms and the application of those algorithms to “real life” biological problems.</p>", "<p>Both awards will be presented at the Society's flagship international conference, ISMB (Intelligent Systems for Molecular Biology) 2008, where the winners will give <italic>Keynote Presentations</italic>. The Sixteenth Annual ISMB Conference will take place July 19–23 in Toronto, Canada.</p>", "<title>ISCB 2008 Accomplishment by a Senior Scientist Award: David Haussler</title>", "<p>Bioinformaticians worldwide owe a debt to David Haussler (see ##FIG##0##Image 1##) for his significant contributions to the design of algorithms they use every day. In this, he is like his two immediate predecessors as Senior Scientist Accomplishment Award winners, Temple D. Smith and Michael S. Waterman, the eponymous joint authors of the Smith–Waterman algorithm for local alignment of DNA or protein sequence fragments. “Haussler's group was one of the pioneers of machine learning in bioinformatics, introducing Hidden Markov Models for the statistical analysis of patterns in biological data,” says Brunak. However, Haussler's recent achievements have been more in the application of bioinformatics methods than in their development. Since 1999, he has been one of the principal figures in sequencing, and later analysing, the human genome and those of other mammals, and in mining this genomic information for insight into vertebrate evolutionary history.</p>", "<p>Haussler originally trained as a mathematician, graduating magna cum laude from Connecticut College and obtaining prizes for mathematics at both Bachelor's and Master's levels. His first encounter with computational biology came in graduate school, at the University of Boulder in Colorado, where he had the good fortune to study for his Ph.D. under Andrzej Ehrenfeucht. “Andrzej is an extraordinary man,” says Haussler. “In our weekly research seminars, we would discuss topics ranging from dinosaur flight to abstract graph theory. He taught me that I should never be constrained by disciplinary boundaries, and never be frightened to tackle big problems. The word “bioinformatics” didn't exist when I was a graduate student, but we were doing it.” Two of his fellow students, Gary Stormo and Gene Myers, have also gone on to have distinguished careers in the field. Stormo, now professor of genetics at the University of Washington in St. Louis, and Deputy Editor-in-Chief of <italic>PLoS Computational Biology</italic>, has made significant contributions to the study of DNA–protein interactions and the prediction of nucleic acid structure and function; Myers was one of the inventors of the BLAST program, a key innovator in shotgun sequencing, and a principal architect of Celera's draft sequence of the human genome.</p>", "<p>Haussler's first years as an independent investigator were devoted to rather abstruse studies in pattern recognition and machine learning, focusing on modelling the way the brain learns. He only shifted from computational neuroscience back to bioinformatics when Anders Krogh joined him at Santa Cruz as a post-doc. Characteristically, Haussler underestimates his own role in their joint achievements. “Anders was an exceptional post-doc, who has gone on to have an exceptional career as an independent scientist. He came to my lab to work on machine learning, but soon discovered that these methods could be applied to biological sequence analysis, to classifying proteins into families and recognising genes in fragments of DNA.” Krogh is co-author of acclaimed and popular textbook <italic>Biological Sequence Analysis: Probabilistic Models of Proteins and Nucleic Acids</italic>. Other members of Haussler's group applied machine learning techniques to the classification of microarray data, including the development of one of the first expression-based methods for distinguishing tumor from normal cells.</p>", "<p>Late in 1999, a phone call changed the direction of Haussler's research. “I was called by Eric Lander, one of the leaders of the public human genome sequencing project, and asked to apply my HMM methodology to identifying the genes in the then newly sequenced human DNA,” he explains. At that time, the public project was in a “full-on race” with Celera to publish an initial working draft of the sequence. Haussler joined the international public effort, rapidly recruiting a team of talented young bioinformaticians that included Jim Kent, winner of ISCB's 2003 Overton Prize.</p>", "<p>Barely six months after Haussler joined the project, both teams—the publicly funded one and Celera's—were ready to release their first genome drafts into the public domain. Haussler well recalls July 7, 2000, when the complete draft genome sequence was posted on the University of Santa Cruz' Web server. “Seeing the waterfall of As, Gs, Cs, and Ts pouring off our server was an emotional moment,” he says. “We were witnessing the product of more than three billion years of evolution, sequences passed down from the beginning of life to present-day humans.” This excitement was shared by the worldwide scientific community; Internet traffic on the Santa Cruz server reached 0.5 terabytes per day then: a record that still stands.</p>", "<p>Raw DNA sequence, however, is not much use on its own, and Haussler has dedicated the first years of the new millennium to mapping and analysing that sequence. The first release of Santa Cruz' genome browser went online shortly after the human sequence was released, and it now includes twenty complete vertebrate genome sequences, plus those of a few representative invertebrates. “The publication of the second vertebrate genome—that of the mouse—gave us the first real sequence-based insights into the mechanisms of vertebrate evolution,” he says. “And we could also use evolutionary theory and sequence analysis to answer a central question: how much of the mammalian genome is ‘junk’?” Assuming that fewer inter-species substitutions are found in functional DNA than in non-functional DNA, Haussler's team in the mouse genomics consortium were able to estimate that at least 5% of a mammalian genome is functionally important. This value has been confirmed as more complete sequences have emerged. “We may think that 5% is a small value, but it is particularly interesting in that less than 1.5% of the genome codes for proteins. There is still a question over the function of much of the 3.5% that is conserved but does not form protein-coding genes.” Other questions that have attracted Haussler's attention include the analysis of hyper-conserved DNA sequences that remain virtually unchanged in divergent species, and the genetic changes that distinguish humans from apes. While most researchers in this field have concentrated on gene gain during evolution, Haussler and his team recently identified twenty-six genes that are well-established in the vertebrate lineage but that were lost in the latter stages of human evolution.</p>", "<title>ISCB 2008 Overton Prize: Aviv Regev</title>", "<p>Brunak describes the 2008 Overton Prize winner Aviv Regev (see ##FIG##1##Image 2##) as “a role model for how theoretical computer science can be applied to understanding biological organisms as systems.” Like Haussler, she has worked on both the development of computational methods and their applications, and she has already made significant contributions to both fields. Trained initially at Tel Aviv University, Tel Aviv, Israel, she entered an interdisciplinary undergraduate program but knew that her interests lay in bioinformatics “from Day One.” And she made her first contribution to the field while still an undergraduate, developing mathematical models for the evolution of DNA methylation. It was at that early stage that she realised the value of synergy between computational and “wet lab” biology. “There was no data for one critical phylogenetic group that I was studying, so I went to work in the lab at The Hebrew University to fill in the gaps,” she said. “This experience gave me a good idea of what lab work is like, and how important it is to anchor theoretical biology in to the real world.”</p>", "<p>The idea that led directly to her graduate studies, however, came from a branch of computational science that at first glance has little, if any, connection with biology: pi calculus, typically applied to problems in electronic engineering. “I was listening to a conference talk by computer scientist Robin Milner, on the application of pi calculus to dynamic communication networks, when it occurred to me that molecular networks can have similar properties,” she explains. Following this up, she developed a method for describing and understanding the dynamic relationships between entities in a biological system (such as proteins in an interaction network) using this type of “process algebra.” “Regev started out as a young graduate student applying a novel and rather obscure computational methodology to a biological problem. This type of work has only very recently been recognised as a major part of systems biology,” says Brunak. The Overton Prize is awarded in memory of former ISCB director Chris Overton, who died prematurely in 2000; his research output was always innovative and thought-provoking, and it seems particularly fitting that this prize should go to a young researcher whose graduate studies were characterised by such an unexpected—and productive—interdisciplinary leap.</p>", "<p>After graduation, Regev moved to the US to take up her first independent position, at the Bauer Center for Genomics Research at Harvard University. “As a Bauer Fellow, I had five years' guaranteed money to start my own group, with no teaching or admin responsibilities,” she says. “It was research heaven.” There, her research interests switched to the use of probabilistic graphical models to reconstruct networks based on genomic and transcription data, using yeast as a model system. Her group, in collaboration with Amos Tanay and Ron Shamir, showed that while some DNA sequences involved in gene regulation are tightly conserved across even distantly related yeast species, other such sequences diverge, and used this to infer the evolutionary history of certain regulatory elements in yeast. “Our work was vindicated when a paper from an experimental group in <italic>Molecular Cell</italic> showed the transitional event that we had predicted,” she says. “That was great motivation!”</p>", "<p>In 2006, Regev took up a position as an assistant professor at MIT, where she is also a Core Member of the Broad Institute. She has extended her network models to a range of applications including the characterisation of genes that are co-expressed in a range of cancer types but not in normal cells, and studying gene duplication. And, once again, a chance meeting sparked a productive idea. “I was returning from a conference with a colleague, Jill Mesirov, who had been trying to study variation in the gene expression of the malaria parasite in different patients' blood cells,” she explains. “Mesirov's data came from Johanna Daily and Dyann Wirth, infectious disease specialists from Harvard, who suspected that variation in gene expression might explain some of the observed variation in the clinical course of the disease. I wondered whether there might be equivalence to my own classification of yeast gene expression patterns, and so it proved: the malaria samples could be classified into three groups, similar to states characteristic of active growth, a starvation response, and a stress response in yeast.” This work was published in <italic>Nature</italic> in December 2007, where it was also featured in that journal's <italic>Making the Paper</italic> section.</p>", "<p>This is not the first time that Regev's work has been recognised by the ISCB. During the last decade, her name has appeared on a prize-winning poster or paper abstract at ISMB no fewer than four times. The Overton award is the most prestigious of her career so far, but it is unlikely to be the last.</p>", "<title>Additional Information</title>", "<p>For the full agenda and registration information for ISMB 2008, where these ISCB award winners will be joined by six other distinguished <italic>Keynote</italic> lecturers, and which will also feature a <italic>Highlights Track</italic>, <italic>Special Sessions</italic>, <italic>Technical Demonstrations</italic>, and a unique “Visual Reflections on Science” exhibition, please visit the conference Web site at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.iscb.org/ismbeccb2008\">http://www.iscb.org/ismbeccb2008</ext-link>.</p>", "<p>For a review of past ISCB Accomplishment by a Senior Scientist Award and Overton Prize winners, please see <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.iscb.org/ssaa.shtml\">http://www.iscb.org/ssaa.shtml</ext-link> and <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.iscb.org/overton.shtml\">http://www.iscb.org/overton.shtml</ext-link>, respectively.\n\n</p>" ]

[]

[ "<fig id=\"pcbi-1000101-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000101.g001</object-id><label>Image 1</label><caption><p>David Haussler</p></caption></fig>", "<fig id=\"pcbi-1000101-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000101.g002</object-id><label>Image 2</label><caption><p>Aviv Regev</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[ "<fn-group><fn fn-type=\"COI-statement\"><p>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p>CS was paid by ISCB to write this article.</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"pcbi.1000101.g001\"/>", "<graphic xlink:href=\"pcbi.1000101.g002\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Control theoretic tools have been used to model mRNA transcriptional/translational regulatory feedback mechanisms ##UREF##0##[1]##, to analyze nonlinear phenomena ##REF##17158515##[2]##,##REF##15340155##[3]##, and to control complex biological behavior ##UREF##1##[4]##,##UREF##2##[5]##. In our research, we couple systems theoretic tools (such as sensitivity analysis) with model predictive control, to better address phase resetting properties of nonlinear biological oscillators. Our work aims to alleviate circadian-related disorders (such as jet lag and advanced/delayed sleep phase syndromes) by investigating the phase resetting properties of an example circadian mathematical model. More specifically, we manipulate multiple control inputs (or target parameters) to drive the dynamic behavior of the system.</p>", "<p>Many researchers have shown that the systematic application of light pulses may reset the phase of circadian clocks. This light pulse (input) to induced phase-shift (output) mapping is most notably characterized by the phase response curve (PRC). Daan and Pittendrigh studied the PRC to establish a relationship among circadian behavior (nocturnal vs. diurnal activity), free-running period, and maximum phase advance/delay ##UREF##3##[6]##. The free-running period of an organism reflects its circadian behavior without the influence of entrainment factors such as environmental light∶dark cycles. The free-running period of nocturnal animals, for instance, is often less than 24 hours such that dusk triggers a phase delay and the onset of activity. Conversely, diurnal animals often exhibit free-running periods greater than 24 hours such that dawn triggers a phase advance and the onset of activity ##UREF##3##[6]##. Other researchers have made use of PRCs to establish light as a means to accelerate circadian entrainment ##REF##12398256##[7]##, or as a means to start, stop, and reset the phase of simplified circadian models ##REF##15363673##[8]##–##REF##17952058##[11]##.</p>", "<p>In a previous study, we develop a closed-loop nonlinear model predictive control (MPC) algorithm that minimizes the phase difference between a reference and a controlled system (each modeled as a single deterministic oscillator) through the systematic application of continuous light. Through use of MPC, circadian phase is recovered in almost half the time required by the natural open-loop sun cycles ##UREF##1##[4]##. Next, we investigated how the MPC algorithm's tuning parameters might affect the model's phase resetting dynamics ##UREF##5##[12]##. Here, we make use of sensitivity analysis to identify additional control inputs (or drug targets) that, when used by the MPC algorithm, outperform light-based circadian phase resetting. The target identification of single and multiple control inputs, coupled with the analysis of their respective performance, parallels efforts in the pharmaceutical industry to yield the greatest behavioral response with respect to the smallest system perturbation. In other words, our methodology may be used to identify optimal (and arguably non-intuitive) drug targets for therapy.</p>", "<p>To establish an upper bound relating to the time required to recover phase differences, we begin by evaluating the open-loop control algorithm in the <italic>Open-Loop Phase Recovery</italic> section. The identification and manipulation of a set of single, dual, and triple control inputs are then used to minimize phase recovery dynamics of a wild-type circadian system (as described in the <italic>Single</italic>, <italic>Dual</italic>, and <italic>Triple Target Phase Resetting</italic> sections, respectively). This case is most similar to resetting a healthy organism's phase when subject to an environmental disturbance such as jet lag. In the <italic>Short and Long Period Mutants</italic> section, we further investigate how MPC may be used to alleviate chronic circadian disorders. More specifically, we apply the algorithm to circadian oscillator models that exhibit either short or long-period mutant phenotypes. Results suggest that organisms with such syndromes may track regular 24 hour rhythms through the systematic application of light. Our findings support this unique application of systematic drug target identification coupled with model predictive control for use in medicine and pharmacology (see the <xref ref-type=\"sec\" rid=\"s3\">\n<italic>Discussion</italic>\n</xref> section). In the <xref ref-type=\"sec\" rid=\"s4\">\n<italic>Methods</italic>\n</xref> section, we describe the employed model predictive control algorithm and the state-based sensitivity analysis used to identify single and multiple parametric control inputs.</p>" ]

[ "<title>Methods</title>", "<p>A 10-state <italic>Drosophila melanogaster</italic> circadian limit cycle oscillator serves as the model system. This model consists of two coupled auto-regulatory transcription/translation negative feedback loops that characterize <italic>period</italic> and <italic>timeless</italic> gene and protein dynamics ##REF##9486845##[13]##. As demonstrated in previous work, the MPC algorithm may be applied to any stable limit cycle oscillator, including a more complex <italic>Mus musculus</italic> model ##UREF##1##[4]##. Thus, we describe the example system as a general set of nonlinear ordinary differential equations with time <italic>t</italic>, <italic>n</italic>-length state vector <bold>x</bold>(<italic>t</italic>), environmental light input <italic>l</italic>(<italic>t</italic>), additional control inputs <bold>u</bold>(<italic>t</italic>), and system dynamics <bold>f</bold>(<bold>x</bold>(<italic>t</italic>), <italic>l</italic>(<italic>t</italic>), <bold>u</bold>(<italic>t</italic>)):Given that both environmental light and additional control variables may be modeled as multiplicative inputs, the nominal wild-type (sun-cycle entrained) case requires <italic>u</italic>(<italic>t</italic>) = 1, while <italic>l</italic>(<italic>t</italic>) oscillates as a square wave with a frequency of 24 hours, between values 1 and 2. For consistency, the natural sun-cycle environment (or reference) is characterized by the nominal <italic>Drosophila melanogaster</italic> model and denoted by <bold>r</bold>(<italic>t</italic>). This reference is pre-entrained to normal 24 hour light∶dark cycles and is not subject to additional control inputs.</p>", "<title>Model Predictive Control</title>", "<p>Model predictive control ##UREF##11##[38]## is used to increase the re-synchronization or entrainment rate of circadian oscillators through the systematic application of specified control inputs. The algorithm follows a sample and hold strategy, updating the prediction and control input every <italic>t_s</italic> = 2 hours, where the discrete time index , such that a function <italic>g</italic>(<italic>kt_s</italic>) = <italic>g</italic>[<italic>k</italic>]. For simplicity, we refer to <italic>k</italic> as being equivalent to and ignore its rounding component. The manipulated control variable, <italic>u</italic>[<italic>k</italic>], optimizes an open-loop performance objective on a time interval extending from the current time to the current time plus a prediction horizon of <italic>P</italic> = 48 hours, where . This horizon allows the algorithm to take control action at the current time in response to a forecasted error. The move horizon, <italic>M</italic> = 12 hours, limits the number of control inputs within the prediction horizon such that <italic>u</italic>[<italic>k</italic>] spans a time interval . Beyond hours of simulation, the predictive model defaults to <italic>u</italic>[<italic>k</italic>] = 1. Future behaviors for a variety of control inputs are computed according to the mathematical model of the system ##REF##9486845##[13]##.</p>", "<p>The efficacy of the algorithm was evaluated with respect to a sample and hold time interval of 1, 2, and 3 hours (reflecting a move horizon of 3, 6, and 9 hours, respectively). Although shorter light pulses offer a more dynamic manipulated variable profile, it shortens the move horizon and may reduce the utility of model predictive control. Conversely, a longer pulse may reduce the possible control profiles since extended exposure to light leads to arrhythmic behavior ##REF##15746913##[39]##. Thus, we set the sampling rate to 2 hours.</p>", "<p>The fitness function penalizes the normalized predicted state error between the reference and controlled trajectories, <bold>ē</bold>[<italic>k</italic>], and the net control, <bold>ū</bold>[<italic>k</italic>], over the prediction horizon. The system output used to evaluate circadian performance (or, phase entrainment) is the trajectory defined by the total <italic>period</italic> and <italic>timeless</italic> protein complex concentrations. This state error, <bold>e</bold>[<italic>k</italic>], is normalized with respect to the nominal amplitude of oscillation while the time dependent control input, <bold>u</bold>[<italic>k</italic>], is normalized with respect to the nominal set of values, 1:where the state dynamics <bold>r</bold>[<italic>k</italic>] characterize the nominal reference. Note that the vector <bold>e</bold>[<italic>k</italic>] is , while the matrix <bold>ū</bold>[<italic>k</italic>] is <italic>m</italic>×<italic>c</italic> ( and <italic>c</italic> denotes the number of control inputs).</p>", "<p>To avoid penalizing transient effects, the state error is weighted uniformly over the move horizon (reflected in the first <italic>m</italic> diagonal values of the <italic>p</italic>×<italic>p</italic> matrix <bold>Q</bold>), and with increasing weight of slope 2 over the prediction horizon (reflected in the <italic>p</italic>−<italic>m</italic> to <italic>p</italic> diagonal values of <bold>Q</bold>). The cost of applying a light input is weighted uniformly with a magnitude of 100 as reflected in the diagonal values of the <italic>m</italic>×<italic>m</italic> matrix <bold>R</bold>. We can afford to be conservative with the cost of control in the wild-type case, since we can ensure that the lack of control (the open-loop algorithm) will eventually entrain the system. The values contained in <bold>R</bold> will be re-evaluated when the algorithm is designed to entrain mutant phenotype models. The performance of an <italic>m</italic>-length control input is measured byOnly the first move of the lowest cost control sequence evaluated at time <italic>k</italic>, , is implemented. Therefore, the sequence of actually implemented control moves may differ significantly from the sequence of control moves calculated at a particular time step. This discrepancy disappears as the prediction and move horizons near infinity. Feedback is incorporated by using the next measurement to update the optimization problem. Once the controlled state trajectories converge to within 15% of the reference state trajectories, the system is considered to have recovered its phase in <italic>T_r</italic> = min<italic>_k</italic>[|<italic>e</italic>[<italic>k</italic>]|_∞≤0.15] hours. At this point, the algorithm defaults to no control since nominal light∶dark cycles will keep the system synchronized to the new environment. Optimization of the phase synchronizing control sequences is completed through use of a genetic algorithm ##UREF##12##[40]##–##UREF##14##[42]##.</p>", "<title>Sensitivity Analysis</title>", "<p>Parametric sensitivity analysis quantifies the relative change of system behavior with respect to an isolated parametric perturbation. Parametric <italic>state</italic> sensitivity analysis assigns a value to each system parameter that defines how its perturbation affects state dynamics: . This tool is often used to identify the robustness and fragility tradeoffs of regulatory structures ##REF##15340155##[3]##, and may be tailored to evaluate specific output performance such as period, amplitude, or phase characteristics ##REF##17158515##[2]##.</p>", "<p>Assuming the model has <italic>n</italic> states and <italic>ρ</italic> parameters, the FIM is a <italic>ρ</italic>×<italic>ρ</italic> matrix describing how any two parametric perturbations might affect state dynamics. More notably, the diagonal values of the FIM describe how any single parameter may affect state dynamics. As a result, we sort the values of the FIM from greatest to least magnitude and choose the top three individual parameters (reflected by the sorted diagonal values) and top three pairs of parameters (reflected by the sorted off-diagonals) whose perturbations yield the greatest change in output.</p>", "<p>We further analyze the FIM via the singular value decomposition ##UREF##15##[43]##. Assuming FIM = <italic>F</italic>, it may be decomposed as <italic>F</italic> = <italic>UΣV^T</italic>, where Σ is an <italic>n</italic> by <italic>p</italic> diagonal matrix of non-negative singular values, <italic>σ</italic>, <italic>n</italic> is the number of states, and <italic>p</italic> is the total number of system parameters. Matrices <italic>U</italic> and <italic>V</italic> contain the eigenvectors of <italic>FF^T</italic> and <italic>F^TF</italic>, respectively. <italic>U</italic>, Σ, and <italic>V</italic> are ordered according to the magnitude of the singular values. Thus, the first column vector of <italic>U</italic> (and <italic>V</italic>) represents the output (and input) direction with largest amplification. The next most important direction is associated with the second column vector, and so forth. We determine the top three parameters associated with the three greatest input directions in <italic>ν</italic>\n₁ and <italic>ν</italic>\n₂ as ideal inputs for studying the multiple control input strategy.</p>" ]

[ "<title>Results</title>", "<p>A 10-state, 38-parameter <italic>Drosophila melanogaster</italic> (fruit fly) circadian model serves as the example system. This stable nonlinear limit cycle oscillator consists of two coupled negative feedback loops that characterize the transcriptional regulation of <italic>period</italic> and <italic>timeless</italic> mRNA and protein dynamics ##REF##9486845##[13]##. <italic>per</italic> and <italic>tim</italic> genes are transcribed in the nucleus, after which their mRNAs are transported into the cytosol where they serve as a template for protein synthesis. The doubly phosphorylated proteins form a heterodimer, PER-TIM, that enters the nucleus and inhibits gene expression, closing the feedback loop. Researchers find that environmental light increases the rate of TIM protein degradation: in this model, light targets the system by magnifying <italic>ν</italic>\n<italic>_dT</italic>, the doubly phosphorylated TIM protein degradation rate ##REF##9486845##[13]##.</p>", "<p>The phase response of this model as a function of light is shown via the dash-dotted line in ##FIG##0##Figure 1##. This curve maps the <italic>circadian time</italic> of the entraining stimulus (light pulses) against the resulting change in phase of an organism kept in a free-running environment. The circadian time index repeats every 24 hours with CT0 defining the commencement of dawn and CT12 that of dusk. It is important to note that the magnitude of light-induced phase changes (the quantitative dynamics of the PRC) may vary with respect to the intensity of light. While this model does not account for the complexity of the real network that, for instance, includes additional positive feedback loops ##REF##12581523##[14]##,##REF##11517254##[15]##, it has been experimentally validated ##REF##11517254##[15]## and is widely employed as a reference model ##REF##15340155##[3]##,##REF##11792856##[16]##.</p>", "<title>Open-Loop Phase Recovery</title>", "<p>Due to the inherent nonlinear phase response of circadian rhythms when subject to environmental/parametric perturbations, phase recovery dynamics are characterized as a function of the <italic>initial condition</italic> (IC, the circadian time at which control or entrainment begins), and <italic>initial phase difference</italic> (IP, the amount of circadian time to be recovered). To establish a phase resetting set point or upper bound (the maximum amount of time required to recover a given phase difference), we evaluate the open-loop control algorithm, where environmental light∶dark cycles serve as the only mechanism for phase re-entrainment. The phase recovery surface (##FIG##1##Figure 2##) displays the time required for the open-loop case to recover from any possible initial condition and initial phase difference. The asymmetry of the surface may be attributed to the nonlinear effects of light, as characterized by the PRC. The input (light) to output (induced phase shift) mapping of the PRC is seldom symmetric. In <italic>Drosophila melanogaster</italic>, a 15 minute pulse of light has shown to induce up to 3.6 hours of phase advance and 4.2 hours of phase delay ##REF##9486845##[13]##. Recent studies suggest that the change in phase is less sensitive to the duration of the light, and more sensitive to its time-profile ##REF##17502598##[17]##. Phase recovery times (for both open and closed-loop simulations) are evaluated with respect to initial conditions and phase differences discretized at 3 hour intervals. Thus, given the integers <italic>i</italic>,<italic>j</italic> ∈ [0,7], IC = 3<italic>i</italic> and IP = 3<italic>j</italic>.</p>", "<p>The open-loop entrainment strategy requires at most 183 hours to reset the observed states of the controlled system (cumulative protein complex concentrations) to within 15% of the reference trajectories. Mandating the convergence of state trajectories is a tighter constraint than mandating only phase trajectories, since it incorporates amplitude characteristics. The algorithm, however, may be tuned to consider only strict phase measures. The maximum open-loop recovery time refers to a 9 hour initial phase difference whose control action begins at an initial condition of 15 hours. The initial condition, or start of entrainment, is described with respect to circadian time (CT). Interestingly, there is a stark difference between resetting a 3 to 6 hour initial phase difference versus an 18 to 21 hour initial phase difference (a −6 to −3 hour phase difference). In the former, phase recovers in over 100 hours; in the latter, phase recovers in fewer than 60 hours. Additionally, the open-loop algorithm recovers 9 hour phase differences in a fraction of the time required to correct for smaller phase difference. These properties may be attributed to the nature of the phase response curve and are discussed further in the <xref ref-type=\"sec\" rid=\"s3\">\n<italic>Discussion</italic>\n</xref> section. Experimental studies in mammalian SCN cells support this asymmetry: Reddy <italic>et al.</italic> show that circadian clock resetting from a 6 hour phase advance (IP6) is accompanied by dissociation of cellular gene expression and may take up to 1 week to recover ##REF##12196553##[18]##. Conversely, resetting a 6 hour phase delay (IP18) is accompanied by coordinated gene expression and requires only 2 days of recovery ##REF##12196553##[18]##. Our simulations support these experimental conclusions as the cumulative protein concentrations in the former case diverge and require several days to converge to the nominal trajectory. In the latter, cumulative protein concentrations oscillate with smaller amplitude until they converge to the nominal trajectory within a couple days. An example of the corresponding simulations is presented in ##SUPPL##0##Figure S1##.</p>", "<title>Closed-Loop Phase Recovery</title>", "<p>The MPC algorithm (described in the <italic>Model Predictive Control</italic> section) minimizes the normalized difference between the cumulative protein complex concentration over a prediction horizon of 48 hours, by admitting control action during the first 8 hours of the simulated trajectory. This control action is multiplicative, allowing the algorithm to increase/decrease the nominal parameter by a factor of 2. The control profile defined within the move horizon is updated every 2 hours. Through use of MPC, the re-synchronization rate of the controlled system is increased nearly 3-fold through the control of light, or <italic>ν</italic>\n<italic>_dT</italic>. Although light serves as a powerful control input, we show that the manipulation of parameters such as transcription and mRNA degradation rates (<italic>ν</italic>\n<italic>_s</italic> and <italic>ν</italic>\n<italic>_m</italic>, respectively) may provide more immediate phase resetting. Since we make use of the symmetric version of the mathematical model ##REF##9486845##[13]##, we do not differentiate between <italic>per</italic> or <italic>tim</italic> specific functions. Instead, we assume that the isolated control of <italic>ν</italic>\n<italic>_sP</italic> is equivalent to the isolated control of <italic>ν</italic>\n<italic>_sT</italic>, for instance.</p>", "<p>Parametric sensitivity analysis quantifies the relative change of system behavior with respect to an isolated parametric perturbation. A large sensitivity to a parameter, for instance, suggests that the system's performance is subject to greater change with small variations in the given parameter. We make use of the Fisher Information Matrix (FIM) to evaluate the effect of parametric perturbations on the circadian system's state trajectories ##REF##15695639##[19]##. Investigation of the diagonal values, off diagonal values, and singular value decomposition of the FIM points out the relative order, or rank, of parametric sensitivity measures. This relative ordering highlights sets of control inputs whose manipulation may further reduce phase recovery times. The three greatest diagonal values, for instance, identify the most prominent individual control targets (ranked from most to least sensitive);</p>", "<p>\n<italic>ν</italic>\n<italic>_s</italic> (the mRNA transcription rate),</p>", "<p>\n<italic>ν</italic>\n<italic>_m</italic> (the mRNA degradation rate),</p>", "<p>\n<italic>k_s</italic> (the protein translation rate), and</p>", "<p>\n<italic>ν</italic>\n<italic>_d</italic> (the doubly phosphorylated protein degradation rate).</p>", "<p>Recall that <italic>ν</italic>\n<italic>_dT</italic> is the target parameter of environmental light in <italic>Drosophila</italic>. Interestingly, the rate of mRNA transcription is the target of environmental light in <italic>Mus</italic> (via <italic>per</italic> genes) ##REF##12775757##[20]##,##REF##14657377##[21]## and <italic>Neurospora</italic> (via <italic>frq</italic> genes) ##REF##16374510##[22]##. Furthermore, in our previous studies of <italic>Mus</italic> and <italic>Drosophila</italic> circadian networks, mRNA transcription rates were among the most sensitive parameters with respect to both the state- and phase-based sensitivity analysis of two independent network representations ##REF##17158515##[2]##.</p>", "<p>The greatest off diagonal values identify the most prominent pairs of control targets (ranked accordingly);</p>", "<p>\n<italic>ν</italic>\n<italic>_s</italic> and <italic>ν</italic>\n<italic>_m</italic>, and</p>", "<p>\n<italic>ν</italic>\n<italic>_s</italic> and <italic>K_I</italic> (the threshold constants for repression).</p>", "<p>Since the manipulation of more than 1 parameter voids the symmetry argument, we target <italic>tim</italic> specific parameters in the implementation of multiple control targets.</p>", "<p>The greatest input directions of the singular value decomposition identify the most prominent set of three control targets (ranked accordingly);</p>", "<p>\n<italic>ν</italic>\n<italic>_s</italic>, <italic>ν</italic>\n<italic>_m</italic>, and <italic>K_I</italic>, and</p>", "<p>\n<italic>ν</italic>\n<italic>_m</italic>, <italic>ν</italic>\n<italic>_d</italic>, and <italic>k₂</italic> (the nucleus to cytoplasm rate of transport).</p>", "<title>Single Target Phase Resetting</title>", "<p>We investigate the phase recovery dynamics corresponding to four independent isolated control inputs with respect to the initial condition and initial phase difference (##FIG##2##Figure 3##). Results show that control targets identified via sensitivity analysis (##FIG##2##Figure 3(A)–3(C)##) serve as more effective re-entrainment factors than light (##FIG##2##Figure 3(D)##). More specifically, the maximum recovery time corresponding to a control input of <italic>ν_s</italic> is 44 hours (at IC9 and IP12/IP15), <italic>ν</italic>\n<italic>_m</italic> is 50 hours (at IC21 and IP15), <italic>k_s</italic> is 59 hours (at IC12 and IP15), and <italic>ν</italic>\n<italic>_d</italic> (the light target) is 60 hours (at IC12 and IP15, or IC9 and IP12). The control profiles and state response dynamics relating to the phase recovery of IC9 and IP12 are provided in ##SUPPL##1##Figure S2## and ##SUPPL##2##Figure S3##.</p>", "<p>There is a subtle similarity among the single-input phase recovery data; namely, the sudden drop in recovery time with respect to the initial condition for initial phase differences of 0 to 15 hours. We attribute this steep recovery gradient to the PRC as it depicts a greater region of phase delay than it does a phase advance. For this reason, it is more beneficial if the organism delays its phase to recover from a 12 hour initial phase difference. Furthermore, recall that a phase delay is incurred if the organism is to receive a photic input in the late evening hours. Hence, recovering from a phase difference via a set of delaying control inputs is most efficient if control action begins around the late subjective evening. Thus, if we observe phase resetting behavior corresponding to a small phase difference (such that the subjective day of the controlled system and reference are similar), we expect it to have the shortest recovery time near an initial condition of 12 hours, or dusk (##FIG##2##Figure 3(D)##). Interestingly, each of the control inputs exhibits this property. We attribute this similarity to the unique PRC of each control input (##FIG##0##Figure 1##).</p>", "<title>Dual target phase resetting</title>", "<p>Allowing the MPC algorithm to manipulate two variables simultaneously provides more immediate phase resetting since the controller has greater flexibility. For instance, the simultaneous use of <italic>ν_s</italic> and <italic>ν</italic>\n<italic>_m</italic> requires a maximum phase recovery of 43 hours to recover a 12 hour initial phase difference entrained from an initial condition of 6 or 9 hours (##FIG##3##Figure 4(A)##). Similarly, the simultaneous control of <italic>ν</italic>\n<italic>_s</italic> and <italic>K_I</italic> requires a maximum phase recovery of 46 hours to recover a 15 hour initial phase difference (##FIG##3##Figure 4B##).</p>", "<p>Given that the common input is <italic>ν</italic>\n<italic>_s</italic>, we expect the dual control input phase recovery dynamics to be just as good (if not better) than the results generated from the single <italic>ν</italic>\n<italic>_s</italic> input algorithm. Although the dual control input strategy provides similar maximum phase recovery times, the greatest recovery time “plateau” is smaller. Therefore, the dual <italic>ν</italic>\n<italic>_s</italic> and <italic>ν_m</italic> input strategy is more effective at recovering an initial phase differences of 15 hours from IC6, while the <italic>ν</italic>\n<italic>_s</italic> and <italic>K_I</italic> pair is more effective at recovering a 12 hour initial phase difference from IC9 (compare ##FIG##2##Figure 3(A)## to ##FIG##3##Figure 4(A)–4(B)##). We would even argue that the recovery times associated with the dual input strategy may lessen if the genetic algorithm based optimizer were run over a greater number of generations. We limit the number of generations in the genetic algorithm – 15 for the single input, 75 for the dual input, and 250 for the triple input – to reflect the limited resources and time constraints evident in real world applications.</p>", "<title>Triple target phase resetting</title>", "<p>Just as the dual input case, we expect the triple input strategy to recover phase just as effectively as the single input strategy. The <italic>ν_s</italic>, <italic>ν</italic>\n<italic>_m</italic>, and <italic>K_I</italic> input strategy requires at most 39 hours to recover an initial phase difference of 12 hours at IC9 (##FIG##3##Figure 4(C)##). Interestingly, the maximum recovery time corresponding to the use of <italic>ν</italic>\n<italic>_m</italic>, <italic>ν</italic>\n<italic>_d</italic>, and <italic>k₂</italic> as simultaneous control inputs is 59 hours to recover a 3 hour phase difference at IC12 (##FIG##3##Figure 4(D)##). We attribute this abnormally high recovery time to the possibility of numerical errors associated with the optimization algorithm since <italic>ν</italic>\n<italic>_m</italic> requires no more than 30 hours to recover from a 3 hour phase difference, while <italic>ν</italic>\n<italic>_d</italic> requires no more than 54 hours. If we omit this data point as an outlier, this triple input strategy requires 42 hours to recover a 12 hour phase difference beginning at IC6. Assuming abundant computational resources and time, the triple input strategy may further outperform the dual input strategy since the MPC algorithm acquires greater flexibility (a greater number of control options) with each additional target. Each of these control inputs produces a unique PRC that allows the algorithm to further manipulate the set of targets such that the combination may yield a phase delay or advance at any time of the circadian day. In the case of a single light (or <italic>ν</italic>\n<italic>_d</italic> target) input, for instance, the algorithm must wait for the subjective morning to force a phase advance, or the subjective night to force a delay. The advantage gained through additional control targets, however, is not clear. Given the finite horizon over which the algorithm optimizes phase synchrony, in addition to the nonlinear response of the model, we can not expect a monotonic improvement of phase recovery dynamics with an increase in the number of manipulated variables. For instance, the algorithm may choose a sequence of multiple inputs that yields lower cost in the short term (as compared to a single input) with a greater cost in the long term, leading to a point of no return. This scenario may also attribute to the 59 hour recovery observed in ##FIG##3##Figure 4(D)##.</p>", "<title>Short and Long Period Mutants</title>", "<p>Mutant phenotypes of the circadian oscillator represent cases in which nominal light∶dark cycles are unable to maintain synchrony. For this reason, the MPC tuning parameters must be re-evaluated according to this phase resetting problem. In wild-type, for instance, we can afford to be more aggressive with control penalties since nominal light∶dark cycles (or, no control) will eventually entrain the system. In mutants, the weights used to penalize the state error and control inputs prove to be more influential since nominal light∶dark cycles will not entrain the system. Therefore, we set both the move and prediction horizon to 24 hours and reduce the penalty of state error and control to ones. To counter the computational expense incurred with a longer move horizon, we set the time step to 4 hours. Through MPC, we identify a more suitable light∶dark cycle that synchronizes organisms exhibiting abnormally short and long free-running periods (22 and 27 hours, respectively, as shown in ##FIG##4##Figure 5##). Determining the complete range of entrainment (which is likely wider than the 22 to 27 hour period) is non-trivial. In a previous study, we found that (i) the predicted range of entrainment may be very sensitive to the employed performance metric, and (ii) the control/light input strength may also play a dominant role in defining the bounds of this range ##UREF##6##[23]##.</p>", "<p>Given that the PRC characterizing the behavior of <italic>Drosophila melanogaster</italic> consists of phase delays during the late subjective evening, we expect short-period mutant phenotypes to require bright light after subjective dusk. Similarly, we expect long-period mutant phenotypes to require bright light in the early subjective morning to advance the cycle. Our results confirm this hypothesis. In ##FIG##1##Figure 2##, we demonstrate how bright light, admitted during the environmental night, resets the phase of short-period mutants such that it matches that of its environment. Given that the controlled system is 2 hours short, the occurrence of light during the night overlaps with the advance region of the system's PRC. Similarly, the onset of bright light at dawn overlaps with the delay region of long-period mutant PRCs (##FIG##1##Figure 2##). Our ability to maintain appropriate phase relationships between mutant phenotypes (models characterized by non-nominal parameters) and the environment (the nominal case) further proves the robustness of the algorithm despite model mismatch.</p>" ]

[ "<title>Discussion</title>", "<title>Circadian Phase Response</title>", "<p>As implied by the PRC (##FIG##0##Figure 1##), a 3 hour phase difference may be recovered immediately through admission of light at CT15. Hence, for open-loop control action to be most effective, environmental daylight should occur during the controlled system's subjective night (at CT15). In cases with small initial phase difference (such that the subject's internal time is nearly equal to environmental time), however, daylight begins entrainment once the subjective day is around CT12, by inducing small phase delays. This delay reduces the overlap between environmental daylight and the subjective night since re-entrainment of the initial phase difference began before subjective night. The opposite occurs with small <italic>negative</italic> phase differences, where an 18 hour (or −6 hour) phase difference may be recovered via a light pulse admitted at CT21. In this case, environmental daylight affects the controlled system at the start of day while it has not yet begun entrainment, maximizing the phase advancing effect of light. For this reason, open-loop entrainment via phase advances requires less recovery time despite the fact that a single pulse of light may induce a greater phase delay than advance.</p>", "<p>More generally, we find that any given initial phase difference is more readily recovered if open-loop entrainment begins between CT0 and CT9; the rate of re-entrainment depends on the initial condition. To correct initial phase differences of 0 to 9 hours (by inducing a phase delay), daylight is most effective at the end of the day, suggesting greater performance if the algorithm were to begin control action around CT6. To correct initial phase differences of 0 to −6 hours (by inducing phase advances), daylight is most effective at the start of the day, suggesting greater performance if the algorithm were to begin around CT0. In the former case, daylight overlaps with the delay region of the subject's PRC, while in the latter it overlaps with the advance region. Resetting an initial condition of 12 to 15 hours, however, presents an interesting control dilemma as environmental daylight may induce both a phase delay and phase advance. For this reason, the open-loop control algorithm requires several days to correct for such phase differences. If light were accessible to entrain the system continuously throughout the day and night (in other words, if we were to <italic>close the loop</italic>), phase recovery dynamics would be less extreme since phase resetting would rely less on the initial condition.</p>", "<p>Additional phase resetting properties may be inferred through investigation of the simulated PRCs. For instance, in the single input case, <italic>ν</italic>\n<italic>_s</italic> and <italic>k_s</italic> exhibit similar recovery dynamics with the exception that <italic>ν</italic>\n<italic>_s</italic> is more effective at resetting initial phase differences of 15 to 21 hours. This quality may be associated with the fact that manipulating <italic>k_s</italic> exhibits a strikingly similar phase response as <italic>ν</italic>\n<italic>_s</italic> where their input to output mapping is shifted by about 5 hours (##FIG##0##Figure 1##). This similarity may be attributed to the fact <italic>k_s</italic> and <italic>ν</italic>\n<italic>_s</italic> are directly involved with the irreversible production, and transcriptional/translational regulation, of clock-specific genes/proteins. Additionally, the “active” region of the <italic>ν</italic>\n<italic>_s</italic> and <italic>ν_m</italic> PRCs are wider than those of <italic>k_s</italic> and <italic>ν</italic>\n<italic>_d</italic> (or, their dead zones are shorter than those of <italic>k_s</italic> and <italic>ν</italic>\n<italic>_d</italic>), suggesting that their perturbation-induced phase shifts are accessible throughout a greater portion of the circadian day.</p>", "<title>Minimizing Control Action</title>", "<p>Of the single control input results, the manipulation of <italic>ν</italic>\n<italic>_s</italic>, identified as the most sensitive parameter, provides the shortest phase recovery times. Despite these results, <italic>ν</italic>\n<italic>_d</italic> or light-based control is most efficient. In ##FIG##5##Figure 6##, we relate the cumulative control input (a unitless measure that integrates the multiplicative control target action) to the convergence of phase via the PER-TIM complex state error. The data shown reflects the recovery of an initial phase difference of 15 hours from IC12. Analyzing this relationship may provide a basis from which the pharmaceutical industry might select one drug over another. If two different drug targets demonstrate similar response, the one that requires the least number of doses should be admitted, minimizing cost and the potential for drug related side-effects. Moreover, if the symptoms of illness are more severe than the potential for side effect, the drug that minimizes the state error may be preferred over others. The assessment of system convergence and the corresponding admitted control is key to the identification and application of control targets.</p>", "<title>Circadian Alignment and Illness</title>", "<p>In our modern “24/7” work world, social and commercial pressures often oppose our natural circadian timekeeping, causing a source of circadian stress that may lead to chronic illnesses such as cardiovascular disease and cancer ##UREF##7##[24]##. Numerous studies seem to show the effect of circadian rhythms on processes such as cell proliferation and apoptosis that eventually lead to proper growth control ##REF##14633665##[25]##–##REF##16809488##[27]##. For instance, components of the cell cycle that dictate the G1-S and G2-M transition phase have been associated with circadian transcriptional regulation ##REF##14523398##[28]##,##REF##12934012##[29]##. Also in certain conditions, cancer can be a direct consequence of the absence of the circadian regulation ##REF##14633665##[25]##,##REF##12724733##[26]##,##REF##12491517##[30]##. A review of circadian related clinical disorders describes how mutations in some clock genes are associated with alcoholism, sleeping disorders, hypertension, and morbidity ##UREF##7##[24]##,##REF##16077156##[31]##. Most commonly, poor circadian regulation leads to advanced sleep phase syndrome, delayed sleep phase syndrome, non-24-hour sleep-wake syndrome, and irregular sleep-wake pattern ##UREF##8##[32]##. In each of these cases, poor circadian phase resetting may be achieved through the systematic admission of controlled light pulses.</p>", "<p>Assuming we have access to drugs that specifically target circadian genes, we can identify the targets whose manipulation yields the most effective and immediate response through investigation of each control's phase dynamics (as shown in ##FIG##0##Figure 1##). Or, it is possible to minimize the use of control and choose targets that require the least number of doses. We may also tailor the MPC algorithm to correct phase more readily through simultaneous manipulation of multiple control targets. Even further, we may reduce the computational expense by enumerating the control solutions over a grid in the solution space (light magnitude as a function of time), and choosing the optimal control sequence via an exhaustive search. The algorithm approaches a globally optimal solution as the total possible quantization steps of the control input increases. We tested the efficacy of the algorithm with respect to a quantization of 2, 4, 8, and 16 steps ##UREF##5##[12]##. Results suggest that the shorter recovery time associated with the finer-grid enumeration may not outweigh the increase in computation time. Therefore, we may dramatically reduce computational expense by investigating control solutions for as few as 2 possible control values.</p>", "<p>Our methods show great promise for use in the pharmaceutical industry as our theoretical phase entrainment of mutant phenotypes demonstrates the robustness of the algorithm in the presence of model mismatch. This robustness alleviates concerns in the pharmaceutical industry to tailor mathematical representations of bio-chemical pathways to individual people.</p>", "<title>Mammalian Studies</title>", "<p>The study of controlled light pulses as a means of correcting phase is a common area of interest. Studies have shown that humans are much more sensitive to light than initially suspected since room light can significantly reset the phase of the human circadian clock ##UREF##9##[33]##,##REF##8596632##[34]##. Furthermore, the admission of morning light has been considered as an antidepressant by realigning the internal clock with the environment ##REF##16648247##[35]##.</p>", "<p>Additional studies suggest that the human circadian clock mechanism functions similarly to those of other mammals ##REF##8596632##[34]##. This similarity may be attributed to shape/amplitude characteristics of their respective phase response curves. Humans show phase-delay shifts of up to 3.6 hours and phase-advance shifts of up to 2.01 hours (with respect to a 6.7 hour pulse of bright light) ##REF##12717008##[36]##, which is both quantitatively and qualitatively similar to other mammalian species. This parallel motivates the experimental application of controlled light pulses for phase resetting in mammals. We have taken this first step by assessing the efficacy and computational utility of model predictive control as applied to a detailed 71-state <italic>Mus musculus</italic> circadian model ##UREF##10##[37]##. Furthermore, melatonin has proven to be a key circadian phase resetting agent for totally blind people who cannot synchronize to environmental day∶night cycles (or do so at an abnormal time) ##REF##16648247##[35]##. Therefore, melatonin may be used individually (in cases to treat the totally blind), or in combination with light to provide more effective phase resetting.</p>", "<p>Therapies designed to alleviate circadian load would have an important impact on morbidity and mortality across the developed world. Aside from correcting mutant phenotypes, phase resetting would increase performance in many healthy, or wild-type, cases such as frequent flyers avoiding jet-lag or astronauts maintaining a rigorous schedule during space exploration ##REF##17502598##[17]##. The real-time application of the proposed algorithm, however, may be a major issue; in practice, it will not be feasible to collect the corresponding protein concentration data at the molecular level. However, behavioral and/or physiological parameters that are controlled by (and correlated with) the circadian clock's dynamics are easily accessible. Such data may include actograms such as wheel running data for rodents ##UREF##3##[6]##. Hence, a missing link in the current work concerns the development of corresponding (non-linear) state estimators for reconstructing the molecular dynamics. Given the discrete nature of MPC (sampling every 4 hours), the proposed strategy is feasible in practice since sampling rates of such physiological circadian markers may be much higher.</p>" ]

[]

[ "<p>Conceived and designed the experiments: NB JS FD. Performed the experiments: NB. Analyzed the data: NB JS FD. Wrote the paper: NB JS FD. Principal Investigator: FD.</p>", "<p>Circadian entrainment is necessary for rhythmic physiological functions to be appropriately timed over the 24-hour day. Disruption of circadian rhythms has been associated with sleep and neuro-behavioral impairments as well as cancer. To date, light is widely accepted to be the most powerful circadian synchronizer, motivating its use as a key control input for phase resetting. Through sensitivity analysis, we identify additional control targets whose individual and simultaneous manipulation (via a model predictive control algorithm) out-perform the open-loop light-based phase recovery dynamics by nearly 3-fold. We further demonstrate the robustness of phase resetting by synchronizing short- and long-period mutant phenotypes to the 24-hour environment; the control algorithm is robust in the presence of model mismatch. These studies prove the efficacy and immediate application of model predictive control in experimental studies and medicine. In particular, maintaining proper circadian regulation may significantly decrease the chance of acquiring chronic illness.</p>", "<title>Author Summary</title>", "<p>The robust timing, or phase, of the circadian clock is critical in directing and synchronizing molecular, cellular, and organismal behaviors. The clock's failure to maintain precision and adaption is associated with sleeping disorders, depression, and cancer. To better study and control the timing of circadian rhythms, we make use of systems theoretic tools such as sensitivity analysis and model predictive control (MPC). Sensitivity analysis is used to identify key driving mechanisms without having to fully understand or investigate the detailed mechanistic interconnections of the large complex circadian network. Contrary to intuition, sensitivity analysis of the circadian model highlights several <italic>non</italic>-photic control inputs (such as transcriptional regulation) that outperform light-based circadian phase resetting – light is known to accelerate protein degradation. Aside from targeting individual parameters as control inputs, our methods identify combinations of control targets that may further the efficiency of entrainment. We compare the phase resetting performance of our MPC algorithm among cases involving individual and multiple simultaneous control targets (in wild-type simulations). We then tailor the algorithm to correct continuously the phase mismatch that occurs in short and long period mutant phenotypes. Through use of the presented tools, our algorithm is robust in the presence of model mismatch and outperforms the natural <italic>in silico</italic> sun-cycle–based phase recovery strategy by nearly 3-fold.</p>" ]

[ "<title>Supporting Information</title>" ]

[]

[ "<fig id=\"pcbi-1000104-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000104.g001</object-id><label>Figure 1</label><caption><title>Circadian phase response behavior.</title><p>Phase response curves traditionally characterize the light pulse to induced phase mapping of the input admitted to a free-running circadian oscillator. Here, phase response dynamics of the four system parameters exhibiting greatest state sensitivity is depicted: <italic>ν_s</italic> (mRNA transcription), <italic>ν_m</italic> (mRNA degradation), <italic>k_s</italic> (protein translation), and <italic>ν_d</italic> (protein degradation). The x-axis denotes the time at which the 2 hour pulse is given (where CT0 reflects dawn and CT12 dusk), and the y-axis describes the induced phase shift. A positive shift reflects a phase advance. Since light targets TIM specific protein degradation, <italic>ν_dT</italic>, the light-based PRC of the <italic>Drosophila</italic> model is represented via the dash-dotted line.</p></caption></fig>", "<fig id=\"pcbi-1000104-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000104.g002</object-id><label>Figure 2</label><caption><title>Natural phase entrainment.</title><p>Open-loop phase resetting dynamics are plotted as a function of the initial phase difference (x-axis) and the initial condition (y-axis). The intensity of the color reflects the amount of time required to reset a given phase via the light∶dark cycles calibrated to the initial condition: the lighter the color, the longer the recovery time. The mapping of color intensity to phase recovery times (in hours) is shown in the vertical color bar.</p></caption></fig>", "<fig id=\"pcbi-1000104-g003\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000104.g003</object-id><label>Figure 3</label><caption><title>Single input control.</title><p>Closed-loop phase resetting dynamics for single control targets (ordered according to their relative sensitivity) are described as a function of the initial phase difference (x-axis) and the initial condition (y-axis). The intensity of the color gradient reflects the amount of time required to recover from the given control conditions: the lighter the color, the longer the phase recovery. Each color bar is calibrated according to a minimum recovery time of 0 hours and maximum of 60 hours. (A) <italic>ν</italic>\n<italic>_s</italic> Single Control Target (B) <italic>ν</italic>\n<italic>_m</italic> Single Control Target (C) <italic>k_s</italic> Single Control Target (D) <italic>ν</italic>\n<italic>_d</italic> Single Control Target.</p></caption></fig>", "<fig id=\"pcbi-1000104-g004\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000104.g004</object-id><label>Figure 4</label><caption><title>Multiple input control.</title><p>Closed-loop phase resetting dynamics for dual ((A) and (B)), and triple ((C) and (D)) control targets are shown with respect to the initial phase difference (x-axis) and the initial condition (y-axis). The phase recovery time is denoted by the intensity of the color at each given data point: the lighter the color, the longer the recovery time. The mapping of color intensity to the recovery time (in hours) is reflected in the color bar. Each color bar is calibrated according to a minimum recovery time of 0 hours and maximum of 60 hours. (A) <italic>ν</italic>\n<italic>_s</italic> and <italic>ν</italic>\n<italic>_m</italic> Dual Control Targets (B) <italic>ν</italic>\n<italic>_s</italic> and <italic>K_I</italic> Dual Control Targets (C) <italic>ν</italic>\n<italic>_s</italic>, <italic>ν</italic>\n<italic>_m</italic>, and <italic>K_I</italic> Triple Control Targets (D) <italic>ν</italic>\n<italic>_m</italic>, <italic>ν</italic>\n<italic>_d</italic>, and <italic>k</italic>\n₂ Triple Control Targets.</p></caption></fig>", "<fig id=\"pcbi-1000104-g005\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000104.g005</object-id><label>Figure 5</label><caption><title>Model mis-match.</title><p>Continuous phase resetting for the short-period mutant phenotype (dotted lines) and long-period mutant phenotype (dashed lines) are depicted with respect to an initial condition of 0 and 12 hours. Upper subplots describe the observed state trajectory (cumulative PER-TIM protein complex concentrations) as a function of controlled light pulses, shown in the lower subplots. The nominal response (denoted by solid lines) is entrained via regular 24 hour light∶dark cycles. As expected, short-period mutants reset via daily light pulses that occur during the subjective night, forcing daily phase delays. Long period mutants reset via daily light pulses that occur during the subjective morning, forcing phase advances. (A) Mutant Response at IC0 (B) Mutant Response at IC12.</p></caption></fig>", "<fig id=\"pcbi-1000104-g006\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000104.g006</object-id><label>Figure 6</label><caption><title>The cost of control.</title><p>The convergence of state dynamics (from an initial phase difference of 15 hours at IC12) is plotted (top) as a function time. The corresponding cumulative control input (determined by integrating the value of each multiplicative control target over time) is described in the lower subplot. The light-induced target, <italic>ν</italic>\n<italic>_d</italic>, shown via the dash-dotted line, requires the greatest amount of multiplicative control while exhibiting the greatest amount of state error. Conversely, the <italic>k_s</italic> target corresponds to the least amount of state error and requires less admitted control.</p></caption></fig>" ]

[]

[ "<disp-formula></disp-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<disp-formula></disp-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<disp-formula></disp-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000104.s001\"><label>Figure S1</label><caption><p>Open loop phase resetting response at IC9. Phase resetting dynamics for an initial phase difference of 18 hours (or −6 hours) is shown in the blue dotted trajectory; those pertaining to IP6 are reflected in the red dashed line. The nominal protein concentration dynamics are depicted in the solid black line, while environmental sun cycles are shown in the black dotted square wave. The magnitude of the square wave oscillates between 1 and 2 and does not correspond to the y-axis of the figure.</p><p>(0.02 MB EPS)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000104.s002\"><label>Figure S2</label><caption><p>Single control input to output response for IC9. Phase resetting dynamics (upper subplot) are depicted as a function of the individual control input profiles (lower subplot). Nominal, or pre-entrained, circadian dynamics are shown in solid black. The blue dotted lines reflect phase resetting with respect to <italic>ν</italic>s, while the red dashed lines reflect those of <italic>ν</italic>m. Although phase resetting, or state convergence, among the four different control variable occurs at similar hours, both the state dynamics and control profiles for each variable are significantly different.</p><p>(0.04 MB EPS)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000104.s003\"><label>Figure S3</label><caption><p>Single control input to output response for IP12. Phase resetting dynamics (upper subplot) are depicted as a function of the individual control input profiles (lower subplot). Nominal, or pre-entrained, circadian dynamics are shown in solid black. The blue dotted lines reflect phase resetting with respect to ks, while the red dashed lines reflect those of <italic>ν</italic>d. Although phase resetting, or state convergence, among the four different control variable occurs at similar hours, both the state dynamics and control profiles for each variable are significantly different.</p><p>(0.03 MB EPS)</p></caption></supplementary-material>" ]

[ "<fn-group><fn fn-type=\"COI-statement\"><p>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p>This project was supported in part by the ICB, DAAD19-03-D-0004; NIH, GM078993; The Research Participation Program between the US DOE and AFRL/HEP.</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"pcbi.1000104.g001\"/>", "<graphic xlink:href=\"pcbi.1000104.g002\"/>", "<graphic xlink:href=\"pcbi.1000104.g003\"/>", "<graphic xlink:href=\"pcbi.1000104.g004\"/>", "<graphic xlink:href=\"pcbi.1000104.g005\"/>", "<graphic xlink:href=\"pcbi.1000104.g006\"/>", "<graphic xlink:href=\"pcbi.1000104.e001.jpg\" mimetype=\"image\" position=\"float\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e002.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e003.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e004.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e005.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e006.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e007.jpg\" mimetype=\"image\"/>", "<graphic xlink:href=\"pcbi.1000104.e008.jpg\" mimetype=\"image\" position=\"float\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e009.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e010.jpg\" mimetype=\"image\"/>", "<graphic xlink:href=\"pcbi.1000104.e011.jpg\" mimetype=\"image\" position=\"float\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e012.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pcbi.1000104.e013.jpg\" mimetype=\"image\"/>" ]

[ "<media xlink:href=\"pcbi.1000104.s001.eps\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000104.s002.eps\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000104.s003.eps\"><caption><p>Click here for additional data file.</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>The primary olfactory center of insects, the antennal lobe (AL), is a network of excitatory projection neurons (PNs) interconnected via inhibitory local neurons (LNs). Such excitatory-inhibitory architectures are known to produce network oscillations ##UREF##0##[1]##,##REF##12620157##[2]##,##REF##15218136##[3]##. Field potential oscillations have been observed in the AL of the locust ##REF##17797226##[4]##,##REF##12415296##[5]##, of the bee ##UREF##1##[6]## and the moth ##UREF##2##[7]##,##REF##12006983##[8]##. The oscillations persist after ablation of higher brain structures involved in olfaction and are thus attributable to the AL and, in particular, to the synchronization of the underlying PNs. It has been proposed that odors are encoded by distinctive synchronized neural assemblies ##REF##17797226##[4]##,##REF##12415296##[5]##,##UREF##3##[9]##. These assemblies do not only encode sensory information but also store short-term memories ##REF##9363891##[10]##. At the same time as synchronized assemblies are formed, other neurons have to desynchronize in order to avoid pathological epileptic-like hypersynchronization. In the olfactory system, it turns out that desynchronization might also be important for neural processing, as synchronized and desynchronized neurons carry qualitatively different information about the odorant ##REF##15273692##[11]##. What then are the synaptic mechanisms responsible for synchronization and desynchronization?</p>", "<p>It is now clear that PN synchrony results from the interplay with GABAergic LNs, and more specifically from ionotropic GABA_A receptors. Neural synchronization and field potential oscillations are lost when GABA_A inhibition is pharmacologically blocked by local injection of picrotoxin into the AL of the locust ##REF##8875938##[12]##, of the honeybee ##REF##9363891##[10]## and the moth ##UREF##3##[9]##. Picrotoxin desynchronizes neural assemblies and impairs discrimination of similar odors in the honeybee ##REF##9363891##[10]##,##REF##10883802##[13]##. However, picrotoxin does not affect the slow phases of inhibition observed in PNs ##REF##8875938##[12]##,##REF##16207866##[14]## and, thus, multiple inhibitory pathways are likely to be present in the insect AL. In the honeybee, a second inhibitory network has been shown to be picrotoxin-insensitive and glomerulus-specific ##REF##11826074##[15]##, and histamine has been proposed as the second inhibitory transmitter ##REF##17196109##[16]##. Experimental studies as well as computational modelling postulated the existence of slow inhibition ##REF##8875938##[12]##,##REF##11395015##[17]##. The presence of a second inhibitory network mediated by metabotropic GABA_B receptors has been shown in the <italic>Drosophila</italic> AL ##REF##16207866##[14]##. GABA_B postsynaptic potentials present a much slower decay rate than the ones produced by GABA_A inhibition. Interestingly, little evidence for oscillation and synchronization has been found in <italic>Drosophila</italic>\n##REF##16207866##[14]## (but see ##REF##11081718##[18]##). In addition, spike timing precision increases in PNs when GABA_B inhibition is pharmacologically blocked ##REF##16207866##[14]##.</p>", "<p>These observations suggest a synchronizing and desynchronizing effect of fast and slow inhibition, respectively. Without a better understanding of the role of GABAergic synapses, however, it is difficult to evaluate what such synchronization reveals about olfactory coding. Here, using computational modelling, we test the hypothesis that fast GABA_A-type inhibition synchronizes whereas slow GABA_B-type inhibition desynchronizes. Previous theoretical studies have shown that inhibitory networks synchronize (e.g. ##REF##8792237##[19]##,##REF##8815919##[20]##) and that cell heterogeneity or noise added to the input affects the synchronization properties (e.g. ##REF##9580271##[21]##,##REF##16595058##[22]##). Although synaptic transmission can be very unreliable in biological neural networks, none of the modelling studies has explored the effect of synaptic failures. The probability of synaptic failure has been shown to be 0.7 in hippocampal pyramidal neurons ##REF##7937958##[23]## and ∼0.5 for dendrodendritic synapses between Mitral and Granule Cells in the olfactory bulb ##REF##15814782##[24]##. Is there any computational advantage for this synaptic unreliability? Does it affect spike timing precision and neural synchrony? As we show both theoretically and by computer simulations, failures in synaptic transmission are especially tolerated with fast GABA_A synapses but not with slow GABA_B synapses. We also demonstrate that the relative amount of received fast and slow inhibition regulates synchrony and determines whether particular neurons engage in neural assemblies. Finally, the complementary roles of GABA_A and GABA_B synapses in the formation of neural assemblies suggest a wiring scheme that produces stimulus-specific spatial patterns of inhibition in the antennal lobe.</p>", "<p>Throughout the paper, we use computational models of increasing complexity. We first use a model of uncoupled PNs to determine whether the injection of a hyperpolarizing current step enhances spike timing precision. We then use a model of PNs coupled with GABA_A and GABA_B unreliable synapses to understand the effect of synaptic failures on neural synchrony. We finally propose a stimulus-dependent gating mechanism of lateral inhibition between PNs and use Hebbian learning to store and recall stimulus patterns in inhibitory sub-circuits.</p>" ]

[ "<title>Methods</title>", "<title>Neuron Model</title>", "<p>PNs are modelled as quadratic integrate-and-fire (QIF) neurons ##REF##8697231##[58]##. The evolution of the membrane potential <italic>V</italic> is described by:where <italic>I</italic>\n_syn(<italic>t</italic>) is the received synaptic current and <italic>I</italic>\n_ext = <italic>I</italic>+<italic>I</italic>\n_inj−<italic>I</italic>\n_th is a constant external current. <italic>I</italic> represents a driving current, <italic>I</italic>\n_inj is an injected current, and <italic>I</italic>\n_th denotes the rheobase, i.e., the minimal current required for repetitive firing. The QIF neuron fires as soon as <italic>V</italic> reaches the threshold <italic>V</italic>\n_th. Right after the spike, <italic>V</italic> is reset to the value <italic>V</italic>\n_reset.</p>", "<p>In the absence of synaptic current, the QIF neuron presents two distinct regimes depending on the sign of the external current. When <italic>I</italic>\n_ext&lt;0 there are two fixed points. The stable ones defines the resting potential\n</p>", "<p>The unstable one is the threshold above which the neuron fires a single spike. When <italic>I</italic>\n_ext&gt;0 the QIF neuron fires regularly and the firing frequency scales as , as in type 1 neurons. The QIF model represents the normal form of any type 1 neurons near the saddle-node bifurcation and is related to the so-called <italic>θ</italic>-neuron ##REF##8697231##[58]##. Since the QIF neuron is expected to reproduce the characteristics of any type 1 neuron close to bifurcation, it has been widely used as a realistic neuron model ##UREF##4##[26]##. Parameters in Equation 10 were chosen as to obtain a frequency-current response similar to the PN conductance based model by ##REF##11395015##[17]##,##REF##11395014##[46]## (see ##FIG##8##Figure 9A##): <italic>C</italic> = 0.143 nF, <italic>V_T</italic> = −41.18 mV, <italic>q</italic> = 9.29×10⁻⁴ mS V⁻¹, <italic>I</italic>\n_th = 0.527 nA, <italic>V</italic>\n_th = 30 mV and <italic>V</italic>\n_reset = −70 mV. From Equation 11, <italic>V</italic>\n_rest = −65 mV when <italic>I</italic>\n_ext = −<italic>I</italic>\n_th.</p>", "<title>Synaptic Current Model</title>", "<p>The synaptic current <italic>I</italic>\n_syn in Equation 10 results from the integration of GABAergic currents over the dendritic tree. At each synapse, <italic>I_GABA</italic> (in nA) is given bywhere <italic>E</italic> is the reversal potential of the synapse (<italic>E</italic> = −70 mV for GABA_A and −95 mV for GABA_B) and <italic>g</italic> is the peak synaptic conductance in µS. The GABA_A peak conductance <italic>g_a</italic> is in the range (0.25–1.2)×10⁻³ µS and the GABA_B peak conductance <italic>g_b</italic> is in the order of 0.06×10⁻³ µS ##UREF##8##[59]##. Conductance kinetics are modelled by decaying exponentialswhere the <italic>t_i</italic> are the times of the synaptic events and <italic>τ</italic>\n_GABA is the synaptic time decay (<italic>τ</italic>\n_GABA = 10 ms for GABA_A and 100 ms for GABA_B). The Heaviside function <italic>H</italic> ensures causality.</p>", "<p>Inhibitory interneurons may release transmitters synchronously or asynchronously ##REF##16189532##[30]##,##REF##18046390##[31]##. When synchronous release is considered, the times of the synaptic events are given by <italic>t_i</italic> = <italic>t_i^f</italic>+Δ where the <italic>t_i^f</italic> are the firing times of the presynaptic neuron and Δ = 5 ms is the propagation delay. When asynchronous release is considered, each pre-synaptic spike triggers a number of GABAergic post-synaptic events. These events are triggered asynchronously, according to an exponential distribution of standard deviation <italic>λ</italic>. The probability that a presynaptic spike at time <italic>t_i^f</italic> produces a post-synaptic event at time <italic>t_i</italic> is described by:\n</p>", "<title>Network Model</title>", "<p>Not all the PNs fire in the presence of an odor. In the Locust for example, about 100 PNs (out of 830) are activated by the presentation of an odor ##REF##8931275##[60]##. The network used in the simulation is a matrix of <italic>N</italic> = 10×10 neurons corresponding to these odor-responding PNs. We take <italic>I</italic> = 0.75 nA in Equation 10 so that, without synaptic coupling, PNs are oscillators firing at the same frequency (about 40 Hz). In the network, PNs are coupled directly via GABAergic synapses. Inhibitory LNs are not modelled explicitly because of the lack of experimental data concerning the functionning of LNs and because of LN diversity. We consider two types of inhibitory synapses, GABA_A and GABA_B, and a probability of synaptic failure (unless specified otherwise, <italic>P</italic>\n_failure = 0.5). The network was programmed in C and simulated with a fourth-order Runge-Kutta integration method with a time step of 50 µs. The initial network condition corresponds to a completely desynchronized neuronal population. This is obtained from the following procedure. The firing times <italic>T</italic> of the neurons are given by integrating Equation 10 with <italic>I</italic>\n_syn = 0 from their initial membrane potentials <italic>V</italic>(0) to the firing threshold <italic>V</italic>(<italic>I</italic>) = <italic>V</italic>\n_th\n\n</p>", "<p>The maximum firing time <italic>T</italic>\n_max is obtained when <italic>V</italic>(0) = <italic>V</italic>\n_reset. This firing time equation is then solved for <italic>V</italic>(0)\n</p>", "<p>The above equation provides initial membrane potentials <italic>V</italic>(0) for firing times <italic>T</italic> taken randomly between 0 and <italic>T</italic>\n_max. This initialization procedure of the PNs ensures firing times uniformly distributed over (0, <italic>T</italic>\n_max).</p>", "<title>Data Analysis</title>", "<p>The estimation procedure for the spike time jitter <italic>σ</italic> is similar to the one in ##REF##12843697##[61]## and is described in the caption of ##FIG##8##Figure 9B##.</p>", "<p>Estimation of the phase-locking probability closely matches the protocol in ##FIG##8##Figure 9B## to determine slots of activity. In each slot, the mean firing time <italic>T̅</italic> of the neuronal population is computed and the phase-locking probability is obtained by counting the relative number of spikes falling into a bin of ±<italic>ε</italic> ms around the mean firing time <italic>T̅</italic> (<italic>ε</italic> = 5 ms for data in ##FIG##3##Figure 4C## and <italic>ε</italic> = 1 ms for data in ##FIG##3##Figure 4D##).</p>", "<p>The critical storage capacity <italic>α</italic>\n_c is defined as the maximum number of patterns per neurons that can be stored and retrieved reliably. For the numerical estimation of <italic>α</italic>\n_c, binary patterns with <italic>fN</italic> active bits are stored using the clipped Hebbian learning rule on GABA_A synapses (Equation 7). Each individual pattern however elicits a specific sub-network of GABA_A and GABA_B coupling (as described in the <xref ref-type=\"sec\" rid=\"s2\">Results</xref> section). For each pattern, its corresponding sub-network is simulated for 3 s of biological time, starting from a completely desynchronized state. Consecutive slots of activity are determined as in ##FIG##8##Figure 9B##. The spike time jitter <italic>σ</italic> is computed for each neuron as the standard deviation of its firing times over the last activity cycles. To form a binary output, <italic>fN</italic> phase-locked neurons (with the smallest <italic>σ</italic>) are considered as active bits and the remaining (1−<italic>f</italic>)<italic>N</italic> neurons (with higher <italic>σ</italic>) are inactive bits. All the stored patterns are considered to be retrieved reliably if the mean overlap between stored and retrieved patterns exceeds 0.9. The above procedure is repeated with a larger number of stored patterns until the patterns can no longer be retrieved reliably. Each storage capacity estimated in ##FIG##7##Figure 8## has been obtained by averaging the results over five runs.</p>" ]

[ "<title>Results</title>", "<title>Enhancement of Spike Timing Precision with Somatic Injection of Hyperpolarizing Current</title>", "<p>First, we consider a population of uncoupled PNs modelled as integrate-and-fire neurons with nonlinear spike generating current (Equation 10 with <italic>I</italic>\n_syn = 0, see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). Their initial membrane potential is chosen randomly so that the PN population is completely desynchronized. To check whether inhibition synchronizes, we mimic inhibitory current injection into PNs and vary the duration of the hyperpolarizing pulse. ##FIG##0##Figure 1A## left shows representative voltage traces for hyperpolarization intervals of 6, 10, and 20 ms. Hyperpolarized PNs have a tendency to relax to their resting potential <italic>V</italic>\n_rest, given by Equation 11 (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>), and forget their initial states so that they fire synchronously when inhibition stops. The spike time jitter was calculated as the temporal dispersion of the first spikes right after inhibition. It is well fitted with a single exponential (4.1 ms time constant, ##FIG##0##Figure 1A## right). Enhancement of spike timing precision is attributable to a loss of initial conditions and can be interpreted in terms of <italic>transient resetting</italic>, as theoretically described in ##REF##16933980##[25]##. In the case of our PN model, the injected hyperpolarizing current pulse allows the integrate-and-fire neuron to jump from a repetitive spiking regime to a steady state (resting potential) across a saddle node bifurcation characteristics of type 1 excitability ##UREF##4##[26]##.</p>", "<p>To check whether transient resetting is also effective for other types of neurons, we repeated the simulations with a model of olfactory mitral cells (MCs) that displays type 2 excitability ##REF##16597718##[27]##. This MC model has two variables (membrane potential and adaptive current) which relax to their fixed point during the phase of inhibition (see ##FIG##0##Figure 1B##, left). Thus, injection of hyperpolarizing current plays a similar role in type 1 and type 2 neurons. Precise spike timing is obtained for hyperpolarization intervals of longer duration because there is enough time for variables, such as membrane potential or adaptive current, to reach their steady state and forget their initial conditions. The decay rate of the spike time jitter for the MC model is well fitted with a single exponential (time constant = 9.8 ms, ##FIG##0##Figure 1B##, right). It is also in line with the one estimated from experimental data recorded in MCs in vitro ##REF##16689623##[28]## (time constant = 6.8 ms, inset in ##FIG##0##Figure 1B##, right).</p>", "<p>Altogether these observations suggest that inhibition may play a role in enhancing spike timing precision in PNs, since it tends to eliminate the influence of initial conditions. Because long-lasting inhibition leaves more time for transient resetting, one can speculate that precise spike timing would be achieved with GABA_B-type inhibition. Evidence in favour of this hypothesis is provided by in vitro recordings in MCs ##REF##16689623##[28]## for which smaller spike time jitter is obtained with somatic current injection of longer duration (##FIG##0##Figure 1B##, inset). Therefore, one would expect higher spike time jitter in vivo when slow GABA_B inhibition is pharmacologically blocked. Application of a GABA_B antagonist in the <italic>Drosophila</italic> AL, however, shows just the opposite (see Figure 4 in ##REF##16207866##[14]##). To understand this paradox, we simulate neuron models coupled with GABA_A or GABA_B synapses in the next section.</p>", "<title>Impact of Synaptic Unreliability on Spike Timing Precision</title>", "<p>We consider two distinct networks of <italic>N</italic> = 100 neurons completely connected, one with fast GABA_A synapses (<italic>τ</italic>\n_GABA = 10 ms) and another with slow GABA_B synapses (<italic>τ</italic>\n_GABA = 100 ms). Since chemical synapses are believed to be quite unreliable ##REF##7937958##[23]##, a probability of synaptic failure <italic>P</italic>\n_failure is taken into account. Rasterplots in ##FIG##1##Figure 2A and 2B## present network oscillations in the presence of fast or slow inhibition, the frequency being higher with fast inhibition (F ∼20 Hz with GABA_A and F ∼10 Hz with GABA_B). As revealed by Equation A-1 (see ##SUPPL##0##Text S1##), the period <italic>T</italic> of the network oscillations grows as ln <italic>g</italic> where <italic>g</italic> is the peak synaptic conductance. The period is thus quite robust to changes in the strength of inhibition. However, it depends linearly on the decay time <italic>τ</italic>\n_GABA of the inhibitory synapse. This observation is in agreement with simulation results (see ##SUPPL##1##Figure S1##) and with previous studies, e.g., ##REF##9877022##[29]##. In ##FIG##1##Figure 2A and 2B##, the PN population is partially synchronized but with higher temporal dispersion in the presence of slow inhibition. We now quantify analytically the temporal dispersion of the spiking events within each cycle. As shown in ##SUPPL##0##Text S1## (Equation A-3), the spike time jitter <italic>σ</italic>\n²(<italic>n</italic>) of the PN population at the <italic>n</italic>-th cycle can be expressed as a simple linear recursive relationwhere 〈<italic>k</italic>〉 = <italic>N</italic>(1−<italic>P</italic>\n_failure) and <italic>σ_k</italic>\n²+<italic>NP</italic>\n_failure (1−<italic>P</italic>\n_failure) are the mean and variance in the number <italic>k</italic> of inhibitory synaptic events received by the PNs at each cycle. Note that the mathematical analysis in ##SUPPL##0##Text S1## did not take into account the PNs that do not receive any inhibition. Equation 1 is therefore not valid when <italic>P</italic>\n_failure = 1. ##FIG##1##Figure 2C and 2D## compares the theoretical jitter <italic>σ</italic>\n²(<italic>n</italic>) given by Equation 1 to the one obtained from simulations (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). From the figure, we see that the spike time jitter reaches a stable state in about <italic>n</italic> = 3 cycles (300 ms with GABA_B versus 150 ms with GABA_A). This stable state does not depend on initial conditions (compare ##FIG##1##Figure 2C and 2D## with ##FIG##1##Figure 2E and 2F##) but does depend on the time constant of the inhibitory synapse : <italic>σ</italic>≈1 ms for GABA_A and <italic>σ</italic>≈10 ms for GABA_B. From Equation 1, the spike time jitter obtained at convergence is given by\n</p>", "<p>\n##FIG##1##Figure 2G and 2H## compares the theoretical <italic>σ</italic> to the one obtained from simulations. From Equation 2, <italic>σ</italic> is small when 〈<italic>k</italic>〉 is large, as confirmed in ##FIG##1##Figure 2G##. Thus, variable inhibition is especially tolerated as the number of inhibitory inputs per cell is large. From Equation 2, loss of spike timing precision (high <italic>σ</italic>\n²) is attributable to variance in the amount of received inhibition <italic>σ_k</italic>\n². This variance comes from the presence of synaptic failures in our model (or from heterogeneous connectivity as we will see later). Because <italic>σ</italic> is proportional to the decay time constant of the inhibitory synapse, slow inhibition amplifies synaptic noise and leads to unpredictible firings. This finding can be noted in ##FIG##1##Figure 2H## where <italic>σ</italic>&gt;10 ms with slow GABA_B synapses for <italic>P</italic>\n_failure≥0.5. In contrast, variable inhibition is especially tolerated with fast GABA_A synapses since <italic>σ</italic>&lt;5 ms for any value of <italic>P</italic>\n_failure. Equation 2 also holds for extended AL models taking into account lateral excitation between PNs (##SUPPL##2##Figure S2##) and considering inhibitory local neurons (##SUPPL##3##Figure S3##).</p>", "<p>Our results predict that the loss of spike timing precision is attributable to variable inhibition received on slow GABA_B-type synapses. Variable inhibition may come from hererogeneous connectivity or from the presence of synaptic failures, both of them being likely to occur in vivo. Thus, blocking GABA_B inhibition leads to enhanced spike timing precision (Figure 4 in ##REF##16207866##[14]##). In contrast, in vitro injection of hyperpolarizing current pulses, as done in ##REF##16689623##[28]##, does not present such a variability. This explains the apparent contradiction between in vivo and in vitro experimental data, as noticed in the previous section.</p>", "<title>Asynchronous GABA Release Produces Long-Lasting Inhibition and Accentuates Temporal Dispersion</title>", "<p>Inhibitory cells may release transmitters synchronously or asynchronously ##REF##16189532##[30]##,##REF##18046390##[31]##. In the olfactory bulb for example, GABAergic inhibition released by Granule Cells and received by Mitral Cells is asynchronous and variable across repeated trials ##REF##9712650##[32]##,##REF##12122134##[33]##. What might be the effect of asynchronous GABA release on the spike timing precision? As shown in ##SUPPL##0##Text S1## (Equation A-4), the spike time jitter <italic>σ</italic>\n²(<italic>n</italic>) of the PN population at the <italic>n</italic>-th cycle iswhere <italic>λ</italic> is the time constant of the exponential release distribution (Equation 14 in <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). A high value of <italic>λ</italic> models the effect of asynchronous inhibition, where synaptic events may be released well after the arrival of an action potential on a synapse. On the contrary, a lower value of <italic>λ</italic> models the effect of synchronous inhibition. When <italic>λ</italic> = 0, Equation 3 becomes equivalent to Equation 1. At convergence of Equation 3, we have <italic>σ</italic>\n²(<italic>n</italic>) = <italic>σ</italic>\n²(<italic>n</italic>−1) = <italic>σ</italic>\n_asyn\n² andwhere <italic>σ</italic>\n² is the spike time jitter obtained in the case of synchronous GABA release and is simply given by Equation 2. Asynchronous release accentuates temporal dispersion by adding the extra term <italic>λ</italic>\n²/(〈<italic>k</italic>〉−1). ##FIG##2##Figure 3## compares the theoretical σ_asyn\n² to the one obtained from simulations for different values of <italic>λ</italic>\n². For the simulations, we considered a network of <italic>N</italic> = 100 neurons coupled all-to-all with fast GABA_A synapses (<italic>τ</italic>\n_GABA = 10 ms, <italic>g_a</italic> = 1 nS, <italic>P</italic>\n_failure = 0.5). For <italic>λ</italic> = 0 ms (synchronous release), we have <italic>σ</italic>\n_asyn = <italic>σ</italic> = 1 ms (temporal dispersion obtained with GABA_A synapses, see previous section). We observe that <italic>σ</italic>\n_asyn\n² increases linearly with <italic>λ</italic>\n², as predicted by Equation 4. From Equation 4, <italic>σ</italic>\n_asyn = 10 ms when <italic>λ</italic> = 70 ms, which is the same level of temporal dispersion as the one obtained with synchronous release and slow GABA_B synapses (<italic>τ</italic>\n_GABA = 100 ms, see previous section). The loss of spike-timing precision is thus achieved with asynchronous release, despite fast GABA_A synapses. Actually, the asynchronous synaptic events sum gradually over time so as to produce a resulting inhibition which decays with a time constant approximately equal to <italic>λ</italic> (when <italic>λ</italic> is large as shown previously ##UREF##5##[34]##). Asynchronous release can be seen as a way to produce long-lasting inhibition despite the fast decay time of individual events mediated by GABA_A receptors.</p>", "<title>GABA_A and GABA_B Synapses Play Opposite Roles in Synchronization</title>", "<p>A classical approach for measuring synchrony is to consider that a spike occuring at time <italic>T</italic> is phase-locked when <italic>T</italic> is within a temporal window of ±<italic>ε</italic> ms around the mean firing time <italic>T̅</italic> of the neuronal population. The relative count of these synchronous events among the population of neurons provides an estimate of the phase-locking probability. A theoretical lower bound is given by direct application of the Bienaymé-Tchebyshev inequalitywhere <italic>σ</italic>\n² depends on <italic>P</italic>\n_failure via Equation 2. ##FIG##3##Figure 4A## compares the theoretical bound given by Equations 5 and 2 to the phase-locking probability estimated from simulations (<italic>ε</italic> = 5 ms, see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). The bound has the same, monotonically decreasing, behavior as the estimated probability. For both types of inhibition, the phase-locking probability decreases with <italic>P</italic>\n_failure until it reaches a constant value (2<italic>εF</italic>, horizontal line in ##FIG##3##Figure 4A##). This desynchronized state corresponds to the case where PN firings are uniformly distributed over the duration of the oscillatory cycle (1/<italic>F</italic>). With GABA_A-type inhibition, the phase-locking probability decreases in a nonlinear way. More important is the presence of a plateau for <italic>P</italic>\n_failure&lt;0.7 which maintains a high probability of synchrony despite unreliable synapses. In contrast, the phase-locking probability decreases linearly with <italic>P</italic>\n_failure for GABA_B-type inhibition. Thus, a small amount of synaptic noise on GABA_B synapses is sufficient to degrade synchronization in homogeneous networks.</p>", "<p>In heterogeneous networks, the number of inhibitory inputs differs from one cell to another. Is the heterogeneity in connectivity sufficient to break synchrony in the absence of synaptic failure? As seen in ##FIG##3##Figure 4B##, the number of inhibitory inputs has an influence on synchrony. The neurons which receive an amount <italic>k</italic> of inhibition very different than the mean inhibition 〈<italic>k</italic>〉 fire far away from the population and, thus, are not synchronized. Synchronized neurons are those for which <italic>k</italic>≈〈<italic>k</italic>〉. In the following, we analytically quantify the conditional probability that particular neurons receiving <italic>k</italic> inhibitory synaptic events fire in synchrony with the neuronal population. A lower bound on this conditional probability was previously derived in ##REF##16212762##[35]## (Equations 3.7 and 3.8) as\n</p>", "<p>Here <italic>σ</italic>\n² is given by Equation 2. We have checked numerically that Equation 6 is a good candidate for the phase-locking probability. ##FIG##3##Figure 4C and 4D## compares the lower bound given by Equations 6 and 2 to estimated data obtained from simulations. Both for GABA_A and GABA_B type inhibition, the phase-locking probability is an inverted U-function centered on the inhibition 〈<italic>k</italic>〉 received on average by the neurons. The existence of this inverted U-function does not depend on a specific choice for <italic>ε</italic> (<italic>ε</italic> = 5 ms in ##FIG##3##Figure 4C## and 1 ms in ##FIG##3##Figure 4D##). If a cell receives either a fairly large or a fairly small amount <italic>k</italic> of inhibition relative to the mean inhibitory drive 〈<italic>k</italic>〉, then it is likely that it will fire very far away from the other cells and thus will not be synchronized. A synchronization window is defined by the values of <italic>k</italic> for which the phase-locking probability is higher than a given threshold. With GABA_B, the phase-locking probability becomes very sharp so that only neurons for which <italic>k</italic> = 〈<italic>k</italic>〉 are synchronized (very small synchronization window). Therefore, variable inhibition received on slow GABA_B synapses leads to desynchronization. In contrast, variable inhibition is especially tolerated with fast GABA_A synapses because the synchronization window is broader.</p>", "<title>The GABA_A/GABA_B Ratio Regulates Synchrony</title>", "<p>In the previous sections, the effect of GABA_A or GABA_B on synchrony has been studied in isolation. We now consider a network of <italic>N</italic> = 100 neurons coupled with both fast and slow inhibition. A probability of synaptic failure (<italic>P</italic>\n_failure = 0.5 and 0.0) is considered and two patterns of connectivity are taken into account: global (neurons are connected all-to-all) and heterogeneous (neurons are randomly connected with 0.5 probability). ##FIG##4##Figure 5## presents the spike time jitter estimated from simulations for different values of the GABA_A and GABA_B conductances <italic>g_a</italic> and <italic>g_b</italic>. In the absence of synaptic failure and network heterogeneity, the synchronized state (defined as <italic>σ</italic>&lt;5 ms, blue region in ##FIG##4##Figure 5##) extends to the entire phase space (##FIG##4##Figure 5A##). In the presence of network heterogeneity and/or synaptic failure, however, the synchronized state depends on the relative amount of received fast and slow inhibition. The dashed lines demarcating the synchronous state are similar in the case of global connectivity and <italic>P</italic>\n_failure = 0.5 (##FIG##4##Figure 5B##) as well as in the case of heterogeneous connectivity and <italic>P</italic>\n_failure = 0.0 (##FIG##4##Figure 5C##). Thus, network heterogeneity and synaptic failure play the same role in breaking synchrony. With heterogeneous connectivity and synaptic noise (<italic>P</italic>\n_failure = 0.5), the line demarcating the synchronous state in ##FIG##4##Figure 5D## is <italic>g_a</italic>/<italic>g_b</italic>≈25 (<italic>σ</italic>&lt;5 ms when <italic>g_a</italic>/<italic>g_b</italic>&gt;25).</p>", "<p>In heterogenous networks, the number of GABA_A and GABA_B inputs differs from one cell to another and thus some neurons exhibit synchronized activity while others do not. If neural assemblies do play a role in sensory representation, then the identities of the synchronized neurons would be reproducible across repeated trials and would be altered by changing the pattern of connections. To test this hypothesis, we performed repeated simulations with two different networks (A and B). ##FIG##4##Figure 5E## shows spike rasterplots obtained from network A with intact connections, and GABA_A or GABA_B blocked. The state of a PN at each oscillatory cycle is represented as a bit 1 or 0 depending on whether its firing is synchronized or not. At each oscillatory cycle, the stimulus is thus characterized as a point in a multidimensional space, where each dimension corresponds to the binary state of a given PN. ##FIG##4##Figure 5F## shows a 2D projection of these data points. Note that logistic principal component analysis (PCA) has been used for this analysis because it is better suited to modelling binary data than conventional PCA ##UREF##6##[36]##. Two clusters corresponding to networks A and B are well identified with GABA_A and GABA_B inhibition. These two clusters are almost linearly separable. They overlap, however, when GABA_A or GABA_B is blocked. These observations indicate that GABA_A and GABA_B are both needed to create specific assemblies of synchronized neurons.</p>", "<title>Storing Stimulus Patterns in Inhibitory Sub-circuits</title>", "<p>In the previous sections, we have shown that synchronized neural assemblies are triggered by GABA_A and GABA_B connectivity. In the AL of the honeybee, the GABAergic network is functionnally organized to reflect correlations between glomeruli ##REF##15673548##[37]##. In <italic>Drosophila</italic>, inhibitory LNs present specificity in their odor responses ##REF##16207866##[14]##, and this specificity results from repeated exposure to an odor ##REF##18054860##[38]##. Therefore, it seems plausible that the GABAergic network exhibits some form of Hebbian synaptic plasticity to store odor stimuli (e.g. ##REF##16354378##[39]##). To investigate the problem of learning in inhibitory networks, we use our model to store and recall representations of different input patterns. To store <italic>M</italic> binary patterns <italic>ξ_i^μ</italic> ∈ {0,1}(<italic>μ</italic> = 1···<italic>M</italic>, i = 1···<italic>N</italic>), we consider, for simplicity, that the GABA_B network is global and that the GABA_A network is trained using clipped Hebbian learning :where <italic>J_ij</italic> = 1 if presynaptic neuron <italic>j</italic> is connected to postsynaptic neuron <italic>i</italic> with a fast GABA_A type synapse and <italic>J_ij</italic> = 0 otherwise. ##FIG##5##Figure 6A## provides an example of GABA_A connectivity trained from a single pattern. The PNs in the antennal lobe do not inhibit each other directly but they do so via local neurons. Inhibitory LNs receive direct synaptic input from olfactory receptors ##REF##10761576##[40]## and show specificities in their response to odors ##REF##16207866##[14]##,##REF##18054860##[38]##. Consequently, only a sub-network of the trained connectivity may be activated by the olfactory stimulus. ##FIG##5##Figure 6B## depicts a hypothetical input-dependent gating of lateral inhibition between PNs. To develop this idea further, a GABA_A connection in our model is functionally active between neurons <italic>j</italic> and <italic>i</italic> when both <italic>J_ij</italic> = 1 (connection set by Equation 7) and <italic>ξ_j</italic> = 1 (reflecting the fact that a putative LN associated with this connection is activated by input <italic>ξ_j</italic>). A GABA_B connection is functionally active between neurons <italic>j</italic> and <italic>i</italic> only when <italic>ξ_j</italic> = 1 (GABA_B connectivity is global in the assumptions derived from our model). ##FIG##5##Figure 6C## depicts the sub-network of GABA_A and GABA_B connections activated by input pattern <italic>ξ</italic> (noisy version of training pattern <italic>ξ^μ</italic>). As seen previously, the relative number of GABA_A and GABA_B inputs modulate the degree of synchrony. In ##FIG##5##Figure 6C##, the third PN desynchronizes because it only receives GABA_B inhibition whereas the other PNs synchronize. If state 1 or 0 is assigned to synchronized or desynchronized neurons respectively, then the original training pattern <italic>ξ^μ</italic> is retrieved.</p>", "<p>To illustrate the functioning of the spiking associative memory, we used the learning rule (7) to train the GABAergic network with the three black-and-white images ‘0’, ‘1’ and ‘2’ depicted in ##FIG##6##Figure 7A##. Noisy versions of the training patterns, where 20% of the pixels are randomly flipped, are presented as test patterns. Each test pattern activates a sub-circuit of the trained connectivity and the corresponding network is simulated for 1 sec of biological time. Neurons that correspond to active and inactive bits in the original training pattern are classified as foregrounds and backgrounds, respectively. The LFP, computed as the average of the PNs' membrane potentials, oscillates at ∼25 Hz. At each cycle, particular neurons fire within a temporal window of ±5 ms around the peak of the LFP. This phase-locked activity is visualized at each LFP cyle in ##FIG##6##Figure 7B–D## (see also ##SUPPL##4##Videos S1##, ##SUPPL##5##S2##, and ##SUPPL##6##S3##). We observe that foreground neurons synchronize their activity (activity of foreground neurons in red color for both figures and videos), and fire consistently in phase with the LFP at each oscillatory cyle. These foreground neurons form a stable synchronized neural assembly that does not evolve in time. In contrast, background neurons are desynchronized (activity in blue) and fire more or less randomly. If state 1 or 0 is assigned to synchronized or desynchronized neurons respectively, then the retrieval is perfect for the three patterns.</p>", "<title>Storage Capacity Is Similar to that of Willshaw's Network</title>", "<p>An important metric of spiking associative memories is capacity. In other words, how many patterns can be stored and retrieved reliably by considering phase-locked neurons? We present a simple analysis that leads to an estimate of the capacity for a network of <italic>N</italic> neurons and provide computer simulations confirming our estimate. In the simulations of the spiking associative memory, the peak conductance values for GABA_A and GABA_B have been adjusted according to <italic>g_a</italic>/<italic>g_b</italic> = 25 (demarcating the synchronous state in ##FIG##4##Figure 5D##) so that a neuron is synchronized when the number of its GABA_A synaptic inputs exceeds that of its GABA_B inputs and is desynchronized otherwise. The final state of neuron <italic>i</italic> can therefore be written aswhere <italic>H</italic> is the heaviside function and <italic>s_i</italic> = 1 when neuron <italic>i</italic> is synchronized and 0 otherwise. The binary model defined by Equations 7 and 8 is formally equivalent to Willshaw's model of associative memory ##REF##5789326##[41]##. Interestingly, GABA_B connectivity plays the same role as the activity dependent threshold in Willshaw's model. A relatively simple analysis of the storage capacity is possible when the input patterns consist of exactly <italic>fN</italic> active bits, where <italic>f</italic> is the input activity. The storage capacity <italic>α</italic>\n_c (in terms of maximum number of patterns per neuron) obtained analytically in ##UREF##7##[42]## for the Willshaw's model is\n</p>", "<p>\n##FIG##7##Figure 8## compares the storage capacity <italic>α</italic>\n_c given by Equation 9 to the one estimated numerically for our spiking associative memory (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). As seen in the figure, the spiking network possesses storage capacities similar to those of conventional associative memories such as the Willshaw's model. The storage capacity is optimal in the sparse coding regime, where <italic>f</italic>≈ln <italic>N</italic>/<italic>N</italic>. Above this threshold, performance drops significantly.</p>" ]

[ "<title>Discussion</title>", "<p>PN synchrony has been observed in the AL of the locust ##REF##17797226##[4]##,##REF##12415296##[5]##, of the bee ##UREF##1##[6]## and the moth ##UREF##2##[7]##,##REF##12006983##[8]##. In <italic>Drosophila</italic>, PNs are inhibited via at least two distinct conductances, GABA_A and GABA_B\n##REF##16207866##[14]##. GABA_B postsynaptic potentials present a much slower decay rate than the ones produced by GABA_A inhibition. By means of computational modelling, we investigated the roles of fast and slow inhibition in spike timing precision and neuronal synchrony.</p>", "<title>Opposite Roles of Fast and Slow Inhibition</title>", "<p>We first mimicked somatic injection of hyperpolarizing current into individual cells. Our simulations show that the spike time jitter decreases with the duration of the injected current pulse (##FIG##0##Figure 1##). This observation is in agreement with in vitro experimental recordings ##REF##16689623##[28]##, because the hyperpolarizing current pulse, injected into the cells, is reproducible across repeated trials. In a network of coupled neurons, however, variable inhibition may come from heterogeneous connectivity or from the presence of synaptic failures, both being likely to occur in vivo. How does this variability affect the spike timing precision in PNs? Computer simulations and analytical results predict that the spike time jitter is proportional to the decay time constant of the inhibitory synapse (Equation 2 and ##FIG##1##Figure 2##). Hence, variable inhibition received on slow GABA_B synapses leads to unpredictible firings, whereas variable inhibition is especially tolerated with fast GABA_A synapses. Another way to produce long-lasting inhibition is by asynchronous GABA release. We demonstrate that the slow inhibition which results from the summation of many asynchronous synaptic events accentuates temporal dispersion (Equation 4 and ##FIG##2##Figure 3##).</p>", "<p>Our model predicts that fast and slow inhibition play opposite roles in PN synchrony; fast inhibition synchronizes whereas slow inhibition desynchronizes (see rasterplots in ##FIG##4##Figure 5E##). Several studies show that PN synchronization is induced by GABA_A inhibition ##REF##8875938##[12]##,##REF##9363891##[10]##,##UREF##3##[9]##. When GABA_A inhibition is pharmacologically blocked by local injection of picrotoxin into the AL, PN synchronization and field potential oscillations are lost. Evidence in favour of a desynchronization mechanism by GABA_B is provided by in vivo PN recordings : the spike time jitter decreases in PNs in the presence of the GABA_B antagonist CGP54626 (see Figure 4 in ##REF##16207866##[14]##). Additional indirect confirmation could be obtained by observing whether the oscillatory power of a recorded field potential increases when the GABA_B synapses are blocked, which would imply that more PNs are synchronized in the absence of GABA_B inhibition. A more direct confirmation would require the indexing of PN firings with respect to the common field potential, and analysis of the phase histogram in the control condition and in the presence of the GABA_B antagonist. According to our model's prediction, the PN firing phases should be more broadly distributed in the control condition.</p>", "<p>A related study on spike-time reliability was published while this manuscript was under review. In ##REF##17928562##[43]##, it is shown that fast synaptic fluctuations increase spike timing precision and synchronization, whereas slower input fluctuations have the opposite effects. This finding is in agreement with our results showing that fast, noisy GABA_A inputs improve synchrony, whereas, slow, noisy GABA_B inputs destroy it. In ##REF##17928562##[43]##, we note however that the neural response becomes unpredictible for very fast input fluctuations (time scale &lt;2 ms), a behavior neither observed in our simulations nor predicted by our theory. This discrepancy may result from differences in experimental conditions. The study in ##REF##17928562##[43]## was concerned with reliability in single neurons driven by aperiodic inputs, whereas in this article, we have focused on synchronization of coupled neurons receiving periodic GABAergic inputs.</p>", "<title>Frequency of Network Oscillations</title>", "<p>Little evidence for LFP oscillations has been found in <italic>Drosophila</italic>\n##REF##16207866##[14]##. It is possible that a coherent population oscillation hardly emerges from a network with a limited number of neurons (only 150 <italic>Drosophila</italic> PNs ##REF##12007409##[44]##,##REF##12007410##[45]##). In the case where field oscillations are observed, their frequency is less than 4 Hz ##REF##11081718##[18]##. This is low in comparison to the 20–30 Hz frequency range encountered in other insect species which include the wasp, locust, cockroach and honeybee ##UREF##1##[6]##. It is known that the decay time constant of the inhibition controls the frequency of the oscillations in inhibitory networks ##REF##9877022##[29]##,##REF##12620157##[2]##,##REF##16212762##[35]##. In agreement with this result, we found in our model that frequency is higher with fast inhibition (F ∼20 Hz with <italic>τ</italic>\n_GABA = 10 ms). The period of the network oscillation increases linearly with <italic>τ</italic>\n_GABA (see ##SUPPL##1##Figure S1##) so that a 4 Hz frequency (period = 250 ms) is obtained when <italic>τ</italic>\n_GABA = 345 ms. This observation is compatible with the time decay of the CGP54626-sensitive component observed in the <italic>Drosophila</italic> PNs ##REF##16207866##[14]##. We therefore predict that the 4 Hz LFP frequency observed in <italic>Drosophila</italic> is mainly due to strong GABA_B inhibition which masks the effects of GABA_A. This prediction could be tested experimentally by observing whether the frequency of the field oscillation reaches the 20–30 Hz frequency range in the presence of a GABA_B antagonist.</p>", "<title>Stable Neural Assemblies</title>", "<p>Our AL model converges onto assemblies of synchronized neurons triggered by the GABAergic network (##FIG##4##Figure 5##). The relative number of received GABA_A and GABA_B inputs regulates synchrony and determines whether particular neurons engage in neural assemblies. These assemblies do not evolve in time (stable synchrony). Our work differs from previous theoretical studies in which the stimuli are encoded by transient synchrony, i.e., the subset of synchronized neurons changes over time ##REF##11395014##[46]##,##REF##11395015##[17]##. In previous studies, transient synchrony is achieved by temporal variations of the fast GABA_A input. The most active LNs inhibit the others and may even suppress their activity due to strong LN-LN inhibition. These active LNs, however, increase their adaptation current, which makes subsequent firing harder. Such a fatigue mechanism leads to a complex time-varying competition between LNs that may depend on which LNs win the competition first. In contrast, the neural assemblies created by our mechanism are stable and do not depend on the initial state of the network, synchronized or not. Although LNs have not been used explicitly in our model, we propose another potential role for inhibitory local neurons (see below). Another difference with ##REF##11395014##[46]##,##REF##11395015##[17]## concerns the role of slow inhibition: in ##REF##11395015##[17]##, slow inhibition is introduced to obtain some temporal patterning associated with neural synchrony, whereas, in our study, slow inhibition is introduced to desynchronize PN activity in the presence of synaptic failure.</p>", "<title>Potential Roles for Local Neurons</title>", "<p>Modelling early olfactory systems as a network of neurons coupled with inhibition is not uncommon, see for example ##REF##16381804##[47]##. In our study, we used a simplified model of the insect AL that allows for analytic calculations. Inhibitory LNs were not considered explicitly in the mathematical derivation of the spike time jitter for the PN population (see ##SUPPL##0##Text S1##). However, the spike time jitter is not affected when our AL model is complemented with inhibitory local neurons (##SUPPL##3##Figure S3##). The inhibitory LNs in the extended model fire in synchrony, despite asynchronous PN activities. A potential role for inhibitory LNs in the antennal lobe is to produce stimulus-specific spatial patterns of inhibition. In the antennal lobe, inhibitory LNs receive direct synaptic input from olfactory receptors ##REF##10761576##[40]## and present specificities in their response to odors ##REF##16207866##[14]##,##REF##18054860##[38]##. Consequently, we hypothesized that lateral inhibition between PNs is mediated by the olfactory stimulus. We proposed an input-dependent gating mechanism of lateral inhibition between PNs so that stimulus patterns trigger specific inhibitory sub-circuits (see ##FIG##5##Figure 6##). As particular neurons synchronize or desynchronize according to the inhibition received, neural assemblies are adjusted by stimulus-induced changes in inhibitory sub-circuits. It has recently been shown that LNs are not only inhibitory. A new class of excitatory cholinergic LNs has been identified in the <italic>Drosophila</italic> AL ##REF##17289577##[48]##,##REF##17408580##[49]##. We have complemented our AL model with excitatory cholinergic synapses between PNs and show that lateral excitation redistributes activity over the ensemble of PNs so that all neurons fire, even those not receiving an external stimulation (##SUPPL##2##Figure S2##). This result is consistent with the observation that excitatory LNs in the AL form a dense network of lateral excitatory connections that may boost weak PNs above the firing threshold ##REF##17289577##[48]##.</p>", "<title>Storing Stimulus Patterns in Inhibitory Sub-Circuits</title>", "<p>To assess whether inhibitory sub-circuits are capable of memory storage, we considered that the GABA_B connectivity is fixed and global and that the GABA_A connectivity is trained according to the Hebbian axiom “cells that fire together, wire together”. We showed that lateral GABA_A connections set by Hebbian learning endow the spiking network with properties of binary associative memories (##FIG##5##Figure 6##). The activity of the spiking network converges towards fixed point attractors (assemblies of synchronized neurons) determined by the pattern of connectivity (##FIG##6##Figure 7## and ##SUPPL##4##Videos S1##, ##SUPPL##5##S2##, and ##SUPPL##6##S3##). Binary vectors are stored and retrieved as synchonized neural assemblies (as corresponding to 1 if a neuron is synchronized and to 0 otherwise). We do not claim that this model is biologically plausible or mathematically optimal, but we claim it accounts for some biological observations and allows a simple analysis of the estimation of storage capacity.</p>", "<p>A memory trace of synchronized neural activity compatible with short-term Hebbian plasticity has been revealed in the AL of honeybees ##REF##16354378##[39]##. A functionally organized inhibitory network, whose connectivity reflects correlations between glomeruli, best reproduces the experimental data ##REF##15673548##[37]##. In <italic>Drosophila</italic>, inhibitory LNs present specificity in their odor responses ##REF##16207866##[14]##, that results from repeated exposure to an odor ##REF##18054860##[38]##. It is therefore plausible that the GABAergic network exhibits some form of Hebbian synaptic plasticity enabling the storage of odor stimuli. Evidence for synaptic plasticity in inhibitory networks, however, is scarse and remains controversial. Very few research has addressed the issue of plasticity at inhibitory synapses in oscillatory networks ##REF##11224547##[50]##,##REF##15882647##[51]##. Much work in synaptic plasticity has focused on excitatory synapses. Excitatory synapses of PNs onto inhibitory LNs may also be a site for synaptic plasticity. According to our simplified model, an increase of the LN's excitatory conductance would lead to greater GABA release and thereby the “effective” inhibitory connections between PNs would be modified (##FIG##5##Figure 6B##). Such an increase of inhibitory transmitter release after long-term plasticity at excitatory synapses has been observed in cerebellar stellate cells ##REF##16957089##[52]##.</p>", "<p>The storage capacity of our simplified AL model is comparable to that of classical binary-coded models like Willshaw's network (##FIG##7##Figure 8##). Good performance in terms of stored patterns per neuron is reached when the activity in the network is sparse (very low fraction of synchronized neurons at each LFP cycle). It would be interesting to see whether odors are sparsely represented by the PN population in the AL, as experimental data about sparseness of PN activity is contradictory in <italic>Drosophila</italic>\n##REF##14684826##[53]##,##REF##17596338##[54]##. To estimate storage capacity, we deliberately considered a simplified model of the AL. The first simplification is to use binary stimulus patterns. Considering binary glomerular response (active or inactive) is not uncommon, e.g., ##REF##12408848##[55]##,##REF##17855585##[56]##. In the case of insects, however, it may be too restrictive. The dose-response curves for honeybees' glomeruli is well described by a smooth sigmoid function with estimated Hill slope parameters in the range 0.14–0.56 ##REF##14622173##[57]##. Therefore, further work is necessary to take into account graded glomerular responses in our model. The second simplification is the use of a global GABA_B network. Actually, the odor-evoked GABA_B inhibition in <italic>Drosophila</italic> has been shown to differ across glomeruli and odors ##REF##16207866##[14]##. Training both GABA_A and GABA_B connections would have the merit to convey complementary pieces of information. Fast and slow inhibition could therefore multiplex information into separate channels, in agreement with recent experimental work ##REF##15273692##[11]##.</p>" ]

[]

[ "<p>Conceived and designed the experiments: DM. Performed the experiments: DM NM. Analyzed the data: DM NM. Contributed reagents/materials/analysis tools: DM. Wrote the paper: DM.</p>", "<p>It has been proposed that synchronized neural assemblies in the antennal lobe of insects encode the identity of olfactory stimuli. In response to an odor, some projection neurons exhibit synchronous firing, phase-locked to the oscillations of the field potential, whereas others do not. Experimental data indicate that neural synchronization and field oscillations are induced by fast GABA_A-type inhibition, but it remains unclear how desynchronization occurs. We hypothesize that slow inhibition plays a key role in desynchronizing projection neurons. Because synaptic noise is believed to be the dominant factor that limits neuronal reliability, we consider a computational model of the antennal lobe in which a population of oscillatory neurons interact through unreliable GABA_A and GABA_B inhibitory synapses. From theoretical analysis and extensive computer simulations, we show that transmission failures at slow GABA_B synapses make the neural response unpredictable. Depending on the balance between GABA_A and GABA_B inputs, particular neurons may either synchronize or desynchronize. These findings suggest a wiring scheme that triggers stimulus-specific synchronized assemblies. Inhibitory connections are set by Hebbian learning and selectively activated by stimulus patterns to form a spiking associative memory whose storage capacity is comparable to that of classical binary-coded models. We conclude that fast inhibition acts in concert with slow inhibition to reformat the glomerular input into odor-specific synchronized neural assemblies.</p>", "<title>Author Summary</title>", "<p>A fundamental question in computational neuroscience is to understand how interactions between neurons underlie sensory coding and information storage. In the first relay of the insect olfactory system, odorant stimuli trigger synchronized activities in neuron populations. Synchronized assemblies may arise as a consequence of inhibitory coupling, because they are disrupted when inhibition is pharmacologically blocked. Using computational modelling, we studied the role of inhibitory, noisy interactions in producing stimulus-specific synchrony. So far, experimental data and modelling studies indicate that fast inhibition induces neural synchrony, but it remains unclear how desynchronization occurs. From theoretical analysis and computer simulations, we found that slow inhibition plays a key role in desynchronizing neurons. Depending on the balance between fast and slow inhibitory inputs, particular neurons may either synchronize or desynchronize. The complementary roles of the two synaptic time scales in the formation of neural assemblies suggest a wiring scheme that produces stimulus-specific inhibitory interactions and endows inhibitory sub-circuits with properties of binary memories.</p>" ]

[ "<title>Supporting Information</title>" ]

[ "<p>We would like to thank the three anonymous reviewers whose comments improved the paper.</p>" ]

[ "<fig id=\"pcbi-1000139-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g001</object-id><label>Figure 1</label><caption><title>Spike timing precision with somatic injection of hyperpolarizing current.</title><p>(A) is for our type 1 model of Projection Neuron (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). Left: temporal evolution of the membrane potential <italic>V</italic> with somatic injection of hyperpolarizing current pulses <italic>I</italic>\n_inj of different durations (6, 10, and 20 ms). The spike time jitter (bars above the spikes) is estimated as the temporal dispersion of the first spikes right after inhibition. Right: spike time jitter versus duration of the hyperpolarizing interval. Means and standard deviations are estimated over five runs; The solid curve is an exponential fit of the data (time constant = 4.1 ms). (B) is for a type 2 model of olfactory bulb Mitral Cell. Left: temporal evolution of the state variables (membrane potential <italic>V</italic> and adaptive current <italic>u</italic>) for different durations of the hyperpolarizing current (1, 10, and 25 ms). Right: spike time jitter versus duration of the hyperpolarizing interval. Same convention as in (A) (time constant of exponential fit = 9.8 ms). Figure inset represents the exponential fit of experimental data recorded in MCs in vitro (time constant = 6.8 ms), modified from ##REF##16689623##[28]##, ##FIG##3##Figure 4A##4.</p></caption></fig>", "<fig id=\"pcbi-1000139-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g002</object-id><label>Figure 2</label><caption><title>Spike timing precision with GABA_A or GABA_B inhibition.</title><p>In (A–G), the failure probability is <italic>P</italic>\n_failure = 0.5. (A): Spike rasterplot for GABA_A coupling. The peak GABA_A conductance is <italic>g_a</italic> = 1 nS. The frequency of the network oscillation is F ∼20 Hz. (B) Spike rasterplot for GABA_B coupling (<italic>g_b</italic> = 0.1 nS, F ∼10 Hz). (C,D) Temporal evolution of the spike time jitter <italic>σ</italic>(<italic>n</italic>), where <italic>n</italic> is the index of the oscillatory cycle. Convergence is reached in about 3 cycles, i.e., 300 ms with GABA_B and 150 ms with GABA_A. The initial condition is the desynchronized state (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). (E,F) Same conventions as in (C–D), except that the initial condition is now the synchronized state. (G) Spike time jitter <italic>σ</italic> obtained at convergence (<italic>σ</italic>(<italic>n</italic>) averaged over the last two oscillatory cycles) as a function of the mean inhibitory drive 〈<italic>k</italic>〉 received by the neurons (the number of neurons <italic>N</italic> scales from 50 to 400). (H) <italic>σ</italic> as a function of the failure probability <italic>P</italic>\n_failure. In (C–H), the stars represent the spike time jitter estimated from simulations (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>, means and standard deviations estimated over 10 runs). The solid curves are for theoretical values obtained from Equation 1 (in C–F) or from Equation 2 (in G–H).</p></caption></fig>", "<fig id=\"pcbi-1000139-g003\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g003</object-id><label>Figure 3</label><caption><title>Spike timing precision with asynchronous GABA release.</title><p>The stars represent the spike time jitter <italic>σ</italic>\n² estimated from simulations with asynchronous GABA release. For the simulations, we considered a network of <italic>N</italic> = 100 neurons coupled all-to-all with fast GABA_A synapses (<italic>τ</italic>\n_GABA = 10 ms). Each presynaptic spike triggers 10 post-synaptic events, released asynchronously according to an exponential distribution of variance <italic>λ</italic>\n² (Equation 14 in <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). The solid line is given by Equation 4.</p></caption></fig>", "<fig id=\"pcbi-1000139-g004\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g004</object-id><label>Figure 4</label><caption><title>Phase-locking probability with GABA_A or GABA_B inhibition.</title><p>(A) Phase-locking probability versus probability of synaptic failure in homogeneous networks. The stars represent data estimated from the simulations in the presence of GABA_A (blue stars) or GABA_B (red stars). The resolution at which phase-locked spikes are determined is <italic>ε</italic> = 5 ms (B). The solid curves are for the lower bounds on the phase-locking probability (Equation 5). The constant value 2<italic>εF</italic> (horizontal lines) is for the desynchronized state corresponding to the case where the firings are uniformly distributed over the duration 1/<italic>F</italic> of the oscillatory cycle. (B) Spike rasterplot over two consecutive oscillatory cyles. Synchronized spikes are those which fall within a temporal bin of ±<italic>ε</italic> around the mean firing time <italic>T̅</italic> of the PN population. Dots with the same color correspond to the spikes fired by the neurons receiving the same amount of inhibition (<italic>k</italic>/〈<italic>k</italic>〉). The number of inhibitory inputs received by a particular cell is <italic>k</italic> and the inhibition received on average by the neuronal population is 〈<italic>k</italic>〉. Synchronized neurons are those for which <italic>k</italic>≈〈<italic>k</italic>〉. (C) Phase-locking probability versus relative amount of received inhibition (<italic>k</italic>/〈<italic>k</italic>〉) in heterogeneous networks (probability of connection = 0.4 with GABA_A and 0.9 with GABA_B). The resolution at which phase-locked spikes are determined is <italic>ε</italic> = 5 ms. The lower bounds on the phase-locking probability are given by Equation 6. (D) Same conventions as in (C), except that <italic>ε</italic> = 1 ms.</p></caption></fig>", "<fig id=\"pcbi-1000139-g005\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g005</object-id><label>Figure 5</label><caption><title>Phase diagrams in the presence of network heterogeneity and/or synaptic failure.</title><p>(A–D) The synchronous stationary state (sync.) corresponding to <italic>σ</italic>&lt;5 ms is depicted as the blue region. <italic>g_a</italic> and <italic>g_b</italic> are expressed in nS and denote the values of the peak conductance <italic>g</italic> in Equation 12 for GABA_A and GABA_B, respectively. The dashed lines separating the synchronous state to the asychronous state were obtained by fitting the contour plot <italic>σ</italic> = 5 ms. The equations of the separating line are <italic>g_a</italic> = 11<italic>g_b</italic> (global, <italic>P</italic>\n_failure = 0.5, (B)), <italic>g_a</italic> = 14<italic>g_b</italic> (heterogeneous, <italic>P</italic>\n_failure = 0.0, (C) and <italic>g_a</italic> = 25<italic>g_b</italic> (heterogeneous, <italic>P</italic>\n_failure = 0.5, (D)). (E) spike rasterplots are indicated for a network (heterogeneous connectivity and <italic>P</italic>\n_failure = 0.5) with intact connections (<italic>g_a</italic> = 1 nS and <italic>g_b</italic> = 0.1 nS) and with GABA_A or GABA_B blocked. (F) Clustering of synchronized activity patterns. Two networks (A and B) of <italic>N</italic> = 100 neurons have been randomly generated with 0.5 probability of connection. At each oscillatory cycle, the network activity is represented as a binary vector in a multidimensional space (<italic>N</italic> = 100), where each dimension corresponds to the binary state of a given PN (1 if synchronized and 0 otherwise). The resolution at which synchronized neurons are determined is <italic>ε</italic> = 5 ms (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). We pooled the binary data obtained at the different oscillatory cycles (extracted between 300 to 3000 ms), for the different networks (A and B) and from repeated trials (3 runs for each network). The data were projected, using logistic PCA ##UREF##6##[36]##, onto the first two principal components (PC). Red and blue points in the PCA plane are the projected data for networks A and B, respectively. Left is for intact networks, with GABA_A and GABA_B coupling (<italic>g_a</italic> = 1 nS and <italic>g_b</italic> = 0.1 nS). Middle and right are for GABA_A or GABA_B blocked, respectively.</p></caption></fig>", "<fig id=\"pcbi-1000139-g006\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g006</object-id><label>Figure 6</label><caption><title>Storage and recall in inhibitory sub-circuits.</title><p>(A) Trained GABA_A connectivity. The spiking associative memory consists of oscillatory PNs (one PN per input component) coupled with GABA_A and GABA_B synapses. Following clipped Hebbian learning (Equation 7), GABA_A connections are created between the first, second and fourth PNs (neurons associated to active bits in the training pattern <italic>ξ^μ</italic>). For simplicity, we consider that the GABA_B network is global. (B) Hypothetical input-dependent gating of lateral inhibition in the AL. Two PNs (PN <italic>i</italic> and <italic>j</italic>) are represented as large circles. Lateral inhibition between PNs is gated by inhibitory LNs (small circles) receiving glomerular input. In the presence of an odor, the active glomerulus (black square) turns on the LN (black circle) associated to the connection <italic>j</italic> → <italic>i</italic>. The LN releases GABA that binds to GABA_A and GABA_B receptors onto the postsynaptic cell (PN <italic>i</italic>). On the contrary, the inactive glomerulus (white square) turns off the LN (white circle) thereby keeping silent the connection <italic>i</italic> → <italic>j</italic>. (C) Input-dependent gating of lateral inhibition in the spiking associative memory. The input pattern <italic>ξ</italic> (noisy version of the training pattern) activates a specific inhibitory circuit in the GABAergic network depicted in (A). The first and second PNs are associated to active bits in the input pattern <italic>ξ</italic> and their outgoing connections are thus activated. On the contrary, the third PN is associated to an inactive bit in the input pattern and its outgoing connections are turned off. PNs synchronize according to the balance between their GABA_A and GABA_B inputs (GABA_A/GABA_B ratio). Here, the first, second and fourth PNs synchronize (GABA_A/GABA_B≥1) whereas the third PN desynchronizes (GABA_A/GABA_B&lt;1) and the training pattern is retrieved (synchronized PNs are black).</p></caption></fig>", "<fig id=\"pcbi-1000139-g007\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g007</object-id><label>Figure 7</label><caption><title>Illustrative example of pattern retrieval.</title><p>(A) The learning rule Equation 7 is used to train the GABAergic network (N = 100) with the three images ‘0’, ‘1’ and ‘2’, each one having 36 black and 64 white pixels. Test patterns are noisy versions of the training patterns (20% of the pixels are randomly flipped). (B) The noisy version of ‘0’, presented as input, activates a specific sub-circuit of the trained connectivity. The corresponding network is simulated for 1 sec of biological time. Peak conductances <italic>g_a</italic> = 1 nS and <italic>g_b</italic> = 0.04 nS have been adjusted according to <italic>g_a</italic>/<italic>g_b</italic> = 25 (demarcating the synchronous state in ##FIG##4##Figure 5D##) so that a neuron is synchronized when the number of its GABA_A synaptic inputs exceeds that of its GABA_B inputs, and is desynchronized otherwise. Neurons that correspond to active and inactive bits in the original training pattern are classified as foregrounds and backgrounds, respectively. In the rasterplot, foreground neurons are artificially grouped to visualize their synchronization (spikes as red dots). Background neurons are desynchronized (spikes as blue dots). The LFP, computed as the average of the PNs' membrane potentials, oscillates at ∼25 Hz. At each cycle, particular neurons fire within a temporal window of ±5 ms around the peak of the LFP. This phase-locked activity is visualized at each LFP cyle (see ##SUPPL##4##Video S1## for its evolution). The binary retrieval is formed by assigning bit 1 or 0 to synchronized or desynchronized neurons, respectively. (C) Conventions are similar to (B), except that the noisy version of ‘1’ is presented as input (see ##SUPPL##5##Video S2## for the evolution of the phase-locked activity). (D) Conventions are similar to (B), except that the noisy version of ‘2’ is presented as input (see ##SUPPL##6##Video S3## for the evolution of the phase-locked activity).</p></caption></fig>", "<fig id=\"pcbi-1000139-g008\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g008</object-id><label>Figure 8</label><caption><title>Estimation of the storage capacity.</title><p>The storage capacity (<italic>α</italic>\n_c) is expressed in terms of maximum number of patterns stored per neurons. It is plotted as a function of the input activity (<italic>f</italic>), i.e. the input patterns consist of exactly <italic>fN</italic> active bits. The size of the network is <italic>N</italic> = 100. The plain curve is the theoretical storage capacity derived for Willshaw's model (Equation 9). Stars represent the storage capacity estimated for the spiking neural network working as a phase-locked associative memory (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>). For <italic>f</italic> = 0.1 and 0.2, the GABA_A and GABA_B peak conductances are <italic>g_a</italic> = 0.25 nS and <italic>g_b</italic> = 0.01 nS. For <italic>f</italic> = 0.05 and 0.07, <italic>g_a</italic> = 0.5 nS and <italic>g_b</italic> = 0.02 nS.</p></caption></fig>", "<fig id=\"pcbi-1000139-g009\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pcbi.1000139.g009</object-id><label>Figure 9</label><caption><title>Frequency-current response curve of the PN model (A): Firing rate <italic>F</italic> versus applied current <italic>I</italic>, and estimation of the spike time jitter (B).</title><p>(A) The curve is for our PN model (Equation 10 with <italic>I</italic>\n_inj = 0). Stars are for the simulations of the conductance-based PN model from ##REF##11395015##[17]##,##REF##11395014##[46]##. As expected, our PN model is a good approximation of the type 1 conductance-based model around the rheobase <italic>I</italic>\n_th\n##REF##8697231##[58]##. (B) The rasterplots (a) of PNs are integrated over time bins of 5 ms yielding a peri-stimulus histogram (PSTH, see b). The PSTH is further reduced by cutting above the threshold (dotted line in b) corresponding to the mean firing rate (yielding reduced PSTH, see c). From the reduced PSTH, consecutive slots of activity (in red in c) are extracted and the spike time jitter <italic>σ</italic>(<italic>n</italic>) is computed as the standard deviation of the spike times falling into each slot <italic>n</italic>. The spike time jitter at convergence <italic>σ</italic> is the one obtained at the end of the simulation.</p></caption></fig>" ]

[]

[ "<disp-formula><label>(1)</label></disp-formula>", "<disp-formula><label>(2)</label></disp-formula>", "<disp-formula><label>(3)</label></disp-formula>", "<disp-formula><label>(4)</label></disp-formula>", "<disp-formula><label>(5)</label></disp-formula>", "<disp-formula><label>(6)</label></disp-formula>", "<disp-formula><label>(7)</label></disp-formula>", "<disp-formula><label>(8)</label></disp-formula>", "<disp-formula><label>(9)</label></disp-formula>", "<disp-formula><label>(10)</label></disp-formula>", "<disp-formula><label>(11)</label></disp-formula>", "<inline-formula></inline-formula>", "<disp-formula><label>(12)</label></disp-formula>", "<disp-formula><label>(13)</label></disp-formula>", "<disp-formula><label>(14)</label></disp-formula>", "<disp-formula></disp-formula>", "<disp-formula></disp-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000139.s001\"><label>Text S1</label><caption><p>Spike time jitter of the PN population.</p><p>(0.04 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000139.s002\"><label>Figure S1</label><caption><p>The synaptic parameters control the period of the network oscillation. Period of the network oscillation versus parameters of the GABAergic synapses (time constant and synaptic conductance).</p><p>(0.02 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000139.s003\"><label>Figure S2</label><caption><p>AL model with PN-PN excitatory connections.</p><p>(0.03 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000139.s004\"><label>Figure S3</label><caption><p>AL model with inhibitory LNs.</p><p>(0.38 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000139.s005\"><label>Video S1</label><caption><p>Phase-locked activity of the spiking associative memory for noisy ‘0’ pattern. The neurons which correspond to black and white pixels in the original training pattern are classified as foregrounds and backgrounds, respectively. The network is simulated for 1 sec of biological time. The LFP, computed as the average of the PNs' membrane potentials,oscillate at ≍ 25 Hz. At each cycle, foreground and background neurons, firing within a temporal window of ? ms around the peak of the LFP, are shown as red and blue pixels, respectively.</p><p>(8.30 MB AVI)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000139.s006\"><label>Video S2</label><caption><p>Phase-locked activity of the spiking associative memory for noisy ‘1’ pattern. Conventions are similar to ##SUPPL##4##Video S1##\n</p><p>(8.30 MB AVI)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pcbi.1000139.s007\"><label>Video S3</label><caption><p>Phase-locked activity of the spiking associative memory for noisy ‘2’ pattern. Conventions are similar to ##SUPPL##4##Video S1##.</p><p>(8.30 MB AVI)</p></caption></supplementary-material>" ]

[ "<fn-group><fn fn-type=\"COI-statement\"><p>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p>This research was supported by the European network of excellence “General Olfaction and Sensing Projects on a European Level” (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.gospel-network.org\">www.gospel-network.org</ext-link>), the European research project NEUROCHEM (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.neurochem-project.org\">www.neurochem-project.org</ext-link>), and the ‘Agence Nationale de la Recherche’ for collaborative research in systems biology (grant BSYS-006-02).</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"pcbi.1000139.g001\"/>", "<graphic xlink:href=\"pcbi.1000139.e001.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e002.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.g002\"/>", "<graphic xlink:href=\"pcbi.1000139.e003.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e004.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.g003\"/>", "<graphic xlink:href=\"pcbi.1000139.e005.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.g004\"/>", "<graphic xlink:href=\"pcbi.1000139.e006.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.g005\"/>", "<graphic xlink:href=\"pcbi.1000139.e007.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.g006\"/>", "<graphic xlink:href=\"pcbi.1000139.g007\"/>", "<graphic xlink:href=\"pcbi.1000139.e008.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e009.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.g008\"/>", "<graphic xlink:href=\"pcbi.1000139.e010.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e011.jpg\" mimetype=\"image\" position=\"float\"/>", "<inline-graphic xlink:href=\"pcbi.1000139.e012.jpg\" mimetype=\"image\"/>", "<graphic xlink:href=\"pcbi.1000139.g009\"/>", "<graphic xlink:href=\"pcbi.1000139.e013.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e014.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e015.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e016.jpg\" mimetype=\"image\" position=\"float\"/>", "<graphic xlink:href=\"pcbi.1000139.e017.jpg\" mimetype=\"image\" position=\"float\"/>" ]

[ "<media xlink:href=\"pcbi.1000139.s001.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000139.s002.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000139.s003.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000139.s004.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000139.s005.avi\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000139.s006.avi\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pcbi.1000139.s007.avi\"><caption><p>Click here for additional data file.</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Dementia takes years, if not decades, to develop. It is becoming increasingly clear that effective prevention will require early intervention – that is, prior to the onset of significant neurodegeneration or detectable clinical symptoms. Thus, discovering biomarkers that predict preclinical disease is critical to successful research on lifestyle factors or pharmaceutical agents that can decrease dementia risk ##REF##14593169##[1]##. In recent years, telomere length has emerged as a potential marker for biological aging and age-related diseases. Telomeres, repetitive DNA structures that protect the ends of eukaryotic chromosomes, shorten with age. Although neurons are post-mitotic, telomere shortening has been observed in microglia, and regulation of microglial activation may play a important role in the pathogenesis of Alzheimer disease (AD)##REF##17378753##[2]##. Moreover, as a site of high metabolic activity, brain tissue is particularly vulnerable to oxidative damage, and long-term oxidative stress with aging is believed to be an initiating factor in dementia development ##REF##15031734##[3]##; since evidence indicates that both age and oxidative damage contribute to telomere shortening ##REF##12114022##[4]##, telomere length could be a risk marker for dementia in its capacity as a powerful measure of systemic imbalances in oxidative stress and antioxidant defenses. Thus, several lines of evidence indicate that telomere length might be a promising means of evaluating early risk of AD development.</p>", "<p>We used the Nurses' Health Study to conduct a pilot study to begin to explore whether telomere length in peripheral blood leukocytes (PBL) might be associated with dementia, especially the early phases marked by mild cognitive impairment (MCI) and hippocampal atrophy.</p>" ]

[ "<title>Methods</title>", "<title>Nurses' Health Study</title>", "<p>The Nurses' Health Study began in 1976, when 121,700 registered female nurses, aged 30 to 55 years, from eleven US states, responded to a mailed questionnaire. To date, 90% follow-up of the cohort has been maintained. For this pilot study, we conducted detailed neurologic evaluations among a sample of participants living in the Boston area. This sample was selected from the 525 Nurses' Health Study subjects in the Boston area who were aged 70 years or older, and who participated in telephone dementia screening ##REF##8080390##[5]## from 2005–2006. Briefly, eligible subjects for the neurologic evaluation were all women whose screening indicated possible cognitive impairments, and a random selection of those with no apparent impairments. Of the eligible women, 75% participated in the neurologic exam, and participation rates were similar in those whose screening did and did not indicate impairments. Neurologic evaluation included patient and caregiver interviews, physical exam, neurologic exam, and administration of the Weintraub Activities of Daily Living Scale, Blessed Dementia Scale, and Clinical Dementia Rating Scale. All subjects provided written, informed consent.</p>", "<p>Based on the neurologic evaluations, of the 62 women in the study sample, we diagnosed 5 with Alzheimer's disease (by NINCDS/ADRDA and DSM-IV criteria) and 8 with MCI (by Peterson criteria), and 49 who were cognitively intact ##REF##6610841##[6]##, ##REF##10190820##[7]##. We also conducted magnetic resonance imaging (MRI) in a sample of 29 of these women (3 with dementia, 7 with MCI, 19 cognitively intact). We used a 1.5 T Siemens Avanto MRI scanner (Siemens Medical Systems, Iselin, NJ) with high-resolution T₁-weighted scans. MPRAGE images were used to estimate regional atrophy in the hippocampus (TR = 7.25 msec, echo time TE = 3.0 msec, flip angle = 7°; FOV = 256 mm, matrix = 256×192, 1.33 mm sagitally acquired slices, NEX = 1). Total hippocampal volume was calculated by summing the volumes of the right and left hippocampus, dividing by the intracranial volume, and multiplying by 1000.</p>", "<p>As the time of the neurologic exam, we collected blood samples from all 62 participants. To measure PBL telomere length, we extracted genomic DNA from buffy coats using the QIAmp (Qiagen, Chatsworth, CA) 96-spin blood protocol. Relative average telomere length, expressed as the ratio of telomeres to single genes (T/S ratio), was assessed by a modified version of the real-time PCR-based telomere assay ##REF##12000852##[8]##.</p>", "<title>Statistical Analysis</title>", "<p>To quantify the relation of PBL telomere length to odds of dementia and MCI, we used logistic regression models to compute the odds ratios (OR) and 95% confidence intervals (CI), comparing women with shorter telomere length (defined as relative T/S ratio below the median) versus longer telomere length (relative T/S ratio above the median). We also used linear regression models to estimate adjusted mean differences in hippocampal volume associated with each unit increase in relative T/S ratio. In the regression models, we considered the following potential confounding factors: age (continuous years), educational attainment (college degree, master/doctoral degree), cigarette smoking (never or past, current), history of cardiovascular disease (yes, no), high blood pressure (yes, no), high cholesterol (yes, no), and type 2 diabetes (yes, no).</p>" ]

[ "<title>Results</title>", "<p>As expected, women with MCI or dementia were slightly older than controls, although the age distribution in this cohort was relatively narrow, and had less education than controls (##TAB##0##Table 1##). Blood pressure was somewhat lower in the cases than controls. On average, mean telomere length was progressively shorter in those with MCI or dementia than controls, as was mean hippocampal volume.</p>", "<p>After adjusting for age and educational attainment (##TAB##1##Table 2##), we found that women with PBL telomere length below the median had a statistically significant, higher odds of dementia or MCI (OR = 9.63, 95% CI 1.73–53.65). Second, we examined telomere length in relation to pre-clinical disease only (##TAB##0##Table 1##); the odds of MCI were 12-fold higher (OR = 12.00, 95% CI 1.24–116.5) for those with shorter telomere length compared to longer telomere length. Further adjustment for a variety of health and lifestyle factors did not appreciably change any of these results.</p>", "<p>We also examined the relation of PBL telomere length to hippocampal volume (data not shown in table). We excluded women with dementia from these analyses, so that we could assess telomere length as a possible marker for pre-clinical changes in hippocampal volume. We found that decreasing telomere length was strongly related to decreasing hippocampal volume; after adjusting for age and educational attainment, each unit decrease in relative T/S ratio resulted in a 0.25 mL decrease in hippocampal volume (p = 0.038). Adjustment for a variety of health and lifestyle factors did not alter these findings.</p>" ]

[ "<title>Discussion</title>", "<p>These preliminary data are the first to suggest that shorter PBL telomere length is related to combined dementia/MCI diagnosis, as well as to pre-clinical dementia risk, including both MCI and decreased hippocampal volume. Observed relations were independent of age, educational attainment, cigarette smoking, and various vascular factors, suggesting that PBL telomere length may be a specific marker of dementia.</p>", "<p>In particular, our findings for a relation between telomere length and hippocampal volume in those without dementia indicate the possibility that telomere shortening may be a very early indicator of dementia risk. In understanding the magnitude of association we observed between telomere length and hippocampal volume, we compared our results to other studies of hippocampal volume in older subjects. For example, the Rotterdam study reported that the apolipoproteinE e4 allele was related to a 0.21 mL decrease in hippocampal volume ##REF##12221169##[9]##, while we found that each single 0.1 unit change in the telomere to single gene ratio was related to a 0.25 mL decrease in hippocampal volume (ie, a 0.2 unit change is related to a 0.50 mL decrease; a 0.3 unit change is related to a 0.75 mL decrease, etc).</p>", "<p>Limitations of these results should be considered. First, this was a cross-sectional study, thus it cannot be determined whether telomere length predicts dementia or whether dementia leads to telomere shortening. However, in the context of a disease marker, establishing such temporal associations may not be critical. Moreover, our findings were consistent across a variety of disease states, from dementia to MCI to reduced hippocampal volume, suggesting that telomere shortening is an early sign of disease, rather than a later effect of disease onset. Second, the study sample was very small, and, although findings were statistically significant, the confidence intervals were large, indicating that the data were compatible with a wide range of associations. Thus, these should only be considered as preliminary findings, and interpreted with caution. Nonetheless, it is reassuring that we observed robust associations between telomere length and multiple cognitive outcomes. In addition, the limited literature generally supports our findings. In a small, prospective study of 195 stroke survivors, telomere length in peripheral blood mononuclear cells (PBMC) at baseline was significantly related to both risk of developing dementia over two years (OR = 0.1, 95% CI 0.0–0.8) and to decline on the MMSE, with a mean decrease of 0.77 points on the MMSE for each 1000 bp decrease in telomere length (p = 0.04) ##REF##16685698##[10]##. Among 257 older Hispanic, Caucasian and African-American individuals, in a nested case-control study, PBL telomere length was shorter in those with AD than controls (mean = 0.46 vs 0.52, p&lt;0.03, respectively) ##REF##16807921##[11]##. Finally, in a cross-sectional study of PBL telomere length in 559 older subjects, there was a significant relation between PBL telomere length and verbal fluency (p = 0.02), although not other cognitive systems ##REF##12493553##[12]##.</p>", "<p>Overall, these preliminary data and limited existing studies indicate a possible role for PBL telomere length in identifying older persons with high risk of dementia. As an easily-measured and non-invasive biomarker, further research regarding these relations in large, prospective studies is needed.</p>" ]

[]

[ "<p>Conceived and designed the experiments: FG MI HR JG ID. Performed the experiments: BH FG MI HR JG ID. Analyzed the data: FG Mv ID. Contributed reagents/materials/analysis tools: HR ID. Wrote the paper: BH FG Mv MI HR JG ID.</p>", "<title>Background</title>", "<p>Dementia takes decades to develop, and effective prevention will likely require early intervention. Thus, it is critical to identify biomarkers of preclinical disease, allowing targeting of high-risk subjects for preventive efforts. Since telomeres shorten with age and oxidative stress, both of which are important contributors to the onset of dementia, telomere length might be a valuable biomarker.</p>", "<title>Methodology/Principal Findings</title>", "<p>Among 62 participants of the Nurses' Health Study, we conducted neurologic evaluations, including patient and caregiver interviews, physical exam, neurologic exam, and neuropsychologic testing. We also conducted magnetic resonance imaging (MRI) in a sample of 29 of these women. In these preliminary data, after adjustment for numerous health and lifestyle factors, we found that truncated telomeres in peripheral blood leukocytes segregate with preclinical dementia states, including mild cognitive impairment (MCI); the odds of MCI were 12-fold higher (odds ratio = 12.00, 95% confidence interval 1.24–116.5) for those with shorter telomere length compared to longer telomere length. In addition, decreasing telomere length was strongly related to decreasing hippocampal volume (p = 0.038).</p>", "<title>Conclusions</title>", "<p>These preliminary data suggest that telomere length may be a possible early marker of dementia risk, and merits further study in large, prospective investigations.</p>" ]

[]

[]

[]

[ "<table-wrap id=\"pone-0001590-t001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0001590.t001</object-id><label>Table 1</label><caption><title>Characteristics of Controls, Cases of Mild Cognitive Impairment (MCI), and Cases of Dementia</title></caption><alternatives><table frame=\"hsides\" rules=\"groups\"><colgroup span=\"1\"><col align=\"left\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/></colgroup><thead><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"/><td align=\"left\" rowspan=\"1\" colspan=\"1\">Controls (n = 49)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">MCI cases (n = 8)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Dementia cases (n = 5)</td></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Mean age (SD), years</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">79.2 (2.2)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">79.9 (1.3)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">80.0 (1.4)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Master's or doctorate degree (%)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">8.5</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Mean systolic blood pressure (SD), mmHg</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">136.6 (17.3)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">136.3 (13.0)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">132.0 (17.9)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Mean diastolic blood pressure (SD), mmHg</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">75.6 (9.4)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">73.4 (9.5)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">71.8 (12.1)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Mean telomere length (SD), measured as relative telomere/single gene ratio</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.61 (0.14)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.51 (0.08)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.49 (0.09)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Mean right hippocampal volume (SD)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">2.13 (0.40)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1.86 (0.35)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1.37 (0.68)</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Mean left hippocampal volume (SD)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1.93 (0.36)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1.74 (0.24)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1.40 (0.42)</td></tr></tbody></table></alternatives></table-wrap>", "<table-wrap id=\"pone-0001590-t002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0001590.t002</object-id><label>Table 2</label><caption><title>Odds of Dementia or Mild Cognitive Impairment (MCI), According to Telomere Length</title></caption><alternatives><table frame=\"hsides\" rules=\"groups\"><colgroup span=\"1\"><col align=\"left\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/></colgroup><thead><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"/><td align=\"left\" rowspan=\"1\" colspan=\"1\">Odds Ratio<xref ref-type=\"table-fn\" rid=\"nt101\">*</xref> (95% confidence interval) for Dementia/MCI</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">Odds Ratio<xref ref-type=\"table-fn\" rid=\"nt101\">*</xref> (95% confidence interval) for MCI only</td></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Shorter telomere length vs. longer telomere length<xref ref-type=\"table-fn\" rid=\"nt102\">†</xref>\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">9.63 (1.73–53.65)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">12.00 (1.24–116.46)</td></tr></tbody></table></alternatives></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><fn id=\"nt101\"><label>*</label><p>Odds ratios are adjusted for age and educational attainment.</p></fn><fn id=\"nt102\"><label>†</label><p>Telomere length measured as relative telomere to single gene ratio in peripheral blood leukocytes. Shorter telomere length was defined as below the median in the population of those without any dementia or mild cognitive impairment.</p></fn></table-wrap-foot>", "<fn-group><fn fn-type=\"COI-statement\"><p><bold>Competing Interests: </bold>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p><bold>Funding: </bold>The study was supported by grants AG15424, AG05134, and CA87969 from the National Institutes of Health. The funders had no part in the design and conduct of the study, in the collection, analysis, and interpretation of the data, or in the preparation, review, or approval of the manuscript.</p></fn></fn-group>" ]

[ "<graphic id=\"pone-0001590-t001-1\" xlink:href=\"pone.0001590.t001\"/>", "<graphic id=\"pone-0001590-t002-2\" xlink:href=\"pone.0001590.t002\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Bacterial surface colonisation is a process of critical ecological and economical importance in the marine environment. Initial attachment of the cells and subsequent biofilm formation are the first, important steps in a cascade of recruitment or inhibition events of secondary settling organisms, including algal spores or invertebrate larvae. On artificial surfaces like ship hulls or aquaculture infrastructure the final surface community of bacteria as well as lower and higher eukaryotes (“the fouling community”) is clearly undesirable and often causes significant cost in clean-up or replacement ##REF##15596168##[1]##, ##UREF##0##[2]##. On living surfaces like algae or invertebrates without physical or chemical defence substantial colonisation can impair host function and ultimately lead to death. However, some marine living surfaces remain unfouled in the field and the surface-associated bacterial community has been strongly implied in providing an inhibitory effect on secondary settlers. Indeed, surface-associated bacteria have proven to be a rich source of bioactive molecules that mediate the behaviour of higher eukaryotes or possess killing activities ##REF##17497196##[3]##, ##UREF##1##[4]##. Here a mutualistic relationship can be postulated where the bacterial community protects the host from fouling, while the host surface might provide nutrients and physical protection to the bacteria. However, ecological theory would also predict gradients between mutualism, commensalism and parasitism to emerge and indeed bacterial surface communities and their members can have a pathogenic effect on the host ##UREF##2##[5]##, ##REF##17504474##[6]##.</p>", "<p>Host-associated diversity is often very distinct from the planktonic diversity of the surrounding waters ##UREF##1##[4]##, which may be reflective of differing roles for these communities. <italic>Pseudoalteromonas tunicata</italic> is a model organism for microbial-host interactions on marine living surface and the production of antifouling metabolites ##REF##18463726##[7]##. This green-pigmented bacterium has been isolated from the surface of the marine alga <italic>Ulva lactuca</italic> and the tunicate <italic>Ciona intestinalis</italic>, and related strains from the heterotrophic, gammaproteobacterial genera <italic>Pseudoalteromonas</italic> are often found in association with eukaryotic hosts such as sponges, mussels, pufferfish and a range of algae ##REF##18463726##[7]##. <italic>P. tunicata</italic> produces a range of target-specific inhibitors including a large antibacterial protein ##REF##8702270##[8]##, a small polar heat-stable anti-larval molecule ##REF##16348728##[9]##, a putative antialgal peptide ##REF##11248391##[10]##, a yellow-pigmented, antifungal tambjamine molecule ##REF##18007521##[11]##, ##REF##16957232##[12]## and a purple pigment (violacein) that inhibits protozoan grazing ##UREF##3##[13]##. This chemical arsenal has been shown to be important for the survival of <italic>P. tunicata</italic> in the highly competitive marine surface environment ##REF##17965210##[14]##, ##REF##15811995##[15]##.</p>", "<p>To obtain a better understanding of the properties of surface-associated bacteria we present here an in-depth analysis of the genome sequence of <italic>P. tunicata</italic>. Comparative genomics with related strains revealed novel insight into the mode of generating genetic variations, surface attachment and biofilm formation, stress response, nutrient acquisition and ability to degrade polymers of various eukaryotic surfaces. Our analysis also suggests mechanisms for host specificity for <italic>P. tunicata</italic> and the potential for pathogenic interactions.</p>" ]

[ "<title>Materials and Methods</title>", "<title>Genome sequencing, annotation and analysis</title>", "<p>The genome of <italic>P. tunicata</italic> strain D2 was sequenced and assembled using a hybrid sequencing strategy and further details are given in Goldberg et al. ##REF##16840556##[16]## . Annotation was performed with the pipeline implemented by the Institute of Genome Research (TIGR), which included gene identification with GLIMMER ##REF##10556321##[17]##, searches against PFAM ##REF##10592242##[18]##, TIGRFAM ##REF##11125044##[19]##, COG ##UREF##4##[20]## and PROSITE ##REF##11752303##[21]##. All open reading frames (ORF) were manually checked, and annotations were curated using Manatee software package (<ext-link ext-link-type=\"uri\" xlink:href=\"http://manatee.sourceforge.net\">http://manatee.sourceforge.net</ext-link>) following the curation guidelines outlined by Haas et al. ##REF##15784138##[22]##. Comparative analysis was performed with the Integrated Microbial Genome (IMG) system ##REF##17993666##[23]##. Instances of horizontal gene transfer were analysed by interpolated variable order motifs (IVOM) as described by Vernikos and Parkhill ##REF##16837528##[24]##. Clustered regularly interspaced short palindromic repeats (CRISPRs) were identified with CRISPR-finder ##REF##17537822##[25]##. The prediction of signal peptides (SP) was performed using SignalP v 3.0 ##REF##15223320##[26]## . The protein sequences were considered SP-positive where all the NN-scores were above the default cut-off using both softwares. The <italic>P. tunicata</italic> proteome was also tested for the presence of membrane spanning domains using TMHMM v 2.0 ##REF##11152613##[27]##. Proteins having a signal peptide were scanned for protein families against the Pfam database through global alignments (Pfam_ls) at a significance cut-off level (E-value) of 10⁻⁵. Phylogenetic analysis was performed by alignment of protein sequences using Clustal W ##REF##7984417##[28]## and neighbor-joining trees were generated with 1000 bootstraps.</p>", "<title>Physiological studies</title>", "<p>To assess the ability of <italic>P. tunicata</italic> to grow on specific substrates, overnight cultures of <italic>P. tunicata</italic> were inoculated in either marine broth (Difco 2216) or marine minimal medium (3 M) supplemented with either trehalose, chitin, chitobiose or cellulose as the sole carbon source according to Stelzer et al. ##UREF##5##[29]##. Cultures were incubated aerobically at room temperature (22°C) and growth was monitored by optical density (600 nm) for a period of 48 hours.</p>", "<p>To detect phage-particles, <italic>P. tunicata</italic> cell-free supernatant was examined under transmission electron microscopy (TEM). The sample was placed on a carbon-coated formvar copper grid and negatively stained with 2% phosphotungstic acid for 30 sec. The grid was then examined using a Hitachi H700 TEM at an accelerating voltage of 75 kV.</p>", "<p>Pigment production was measured spectrophotometrically using a Beckman DU 640 spectrophotometer at an absorbance of 575 nm and 425 nm for purple violacein and the yellow tambjamine pigment, respectively, as described in Egan et al. ##REF##12153584##[30]##.</p>" ]

[ "<title>Results and Discussion</title>", "<title>Overall comparison of <italic>Pseudoalteromonas</italic> or <italic>Alteromononas</italic> genomes</title>", "<p>Hierarchical clustering and principle component analysis of cluster of orthologous groups (COG) with other sequenced members of the <italic>Alteromonadales</italic> available in the IMG database indicate that <italic>P. tunicata</italic> is functionally most closely related to <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic>, <italic>Alteromonas macleodii</italic> and <italic>Alteromonadales</italic> sp. TW7. General features of these genomes are given in ##TAB##0##Table 1## and their phylogenetic relationship based on the 16S rRNA gene is illustrated in ##FIG##0##Figure 1##. The COG profiles between these organisms revealed however a number of COG categories that were over-represented in <italic>P. tunicata</italic> (##FIG##1##Figure 2##) including signal transduction mechanisms, defence mechanisms and cell motility. <italic>P. tunicata</italic> shows the highest proportion of genes assigned to signal transduction (10.05%) of any <italic>Alteromondales</italic> genome (second highest: <italic>Shewanella amazonensis</italic> with 8.62%; lowest: <italic>Psychromonas</italic> sp. CMPT3 with 5.10%). This trend is mainly caused by an over-representation of COG5000 (signal transduction histidine kinase involved in nitrogen fixation and metabolism regulation) and COG0664 (cAMP-binding proteins - catabolite gene activator and regulatory subunit of cAMP-dependent protein kinases). <italic>P. tunicata</italic> also has the highest value (2.4%) for the COG category defence of any <italic>Alteromondales</italic> genome and this is specifically due to more genes related to ABC transporter functions, including those predicted to be involved in transport of drugs and/ or antimicrobial peptides (COG0577, COG1131, COG1136, COG1566) and those related to the synthesis of potential bioactive compounds such as non-ribosomal peptide synthetase (NRPS) modules (COG1020). The abundance of genes involved in transport and expression of potential defence compounds are consistent with <italic>P. tunicata's</italic> successful competition on living surfaces. The major difference in the COG category cell motility is in COG0840 (methyl-accepting chemotaxis protein), where <italic>P. tunicata</italic> has 35 hits, which is at least twice as many as any other <italic>Pseudoalteromonas</italic> or <italic>Alteromononas</italic> species.</p>", "<p>We also compared the genomes of <italic>Pseudoalteromonas</italic> and <italic>Alteromononas</italic> species to the predominantly planktonic dataset of the Global Ocean Survey (GOS) ##REF##17355176##[31]##. Using recruitment plots ##REF##17355176##[31]## and a 95% nucleotide identity cut-off, which is indicative for species level similarity, ##REF##17220447##[32]## we found 31, 59, 6, 2327 and 1269 matching reads in the GOS dataset for <italic>P. tunicata</italic>, <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic>, <italic>Alteromonadales</italic> sp. TW7 and <italic>A. macleodii</italic>, respectively. The matches against the <italic>Pseudoalteromonas</italic> species are mostly in conserved regions of the chromosome (e.g. ribosomal genes) and hence reflect low gene divergence rather than true phylogenetic relatedness. The hits for <italic>Alteromonadales</italic> sp. TW7 and <italic>A. macleodii</italic> are distributed over the whole chromosome and are therefore likely due to true representation of these organisms (or close relatives) in the dataset. Over 95% of the hits in the GOS dataset come from the deeply sampled Sargasso Sea and for the same sample the abundant, free-living, phototrophic species <italic>Prochlorococcus marinus</italic> MED4 ##REF##15001713##[33]## recruits 2801 reads with the same cut-offs as above. Together this would indicate that <italic>Alteromonadales</italic> sp. TW7 and <italic>A. macleodii</italic> (or closely related organisms) constitute a significant portion of the heterotrophic population in these planktonic communities, while the three <italic>Pseudoalteromonas</italic> species are less prevalent.</p>", "<title>Mobile genetic elements</title>", "<p>The presence of mobile genetic elements is common in bacteria, however in comparison to closely related marine bacteria they are more abundant in <italic>P. tunicata</italic>, where they comprise 2% of the entire genome. The <italic>P. tunicata</italic> genome contains one 33 kb-sized P2-like prophage, which is similar to the one found in the genome of <italic>P. atlantica</italic>, <italic>Hahella chejuensis</italic>, <italic>Moritella</italic> sp PE36, <italic>P. aeruginosa</italic> PA7, <italic>Desulfovibrio vulgaris</italic>, <italic>Haemophilus influenza</italic>, <italic>Aeromonas salmonicida</italic> and <italic>Vibrio cholerae</italic> 0395. P2-like phage is a member of the myxoviridae and is one of the 5 major prophage groups commonly detected in Gammaproteobacteria ##REF##12794192##[34]##. A microscopic inspection of the <italic>P. tunicata</italic> media supernatant in late growth phase revealed phage-like particles (##SUPPL##0##Figure S1## online support material) suggesting that the <italic>P. tunicata</italic> P2-like phage is able to enter a lytic cycle. This observation together with the evident distribution of this prophage among marine and other bacteria could imply this phage in horizontal gene transfer that provide an adaptive advantage for the bacterial host in the marine environment or mediate acquisition of virulence genes.</p>", "<p>There are 33 genes encoding for transposases, of which approximately half are full-length genes. Of interest is the detection of multiple copies of the insertion sequence (IS) related to the IS492 element previously characterised in <italic>P. atlantica</italic>\n##REF##17264213##[35]##, ##REF##2537827##[36]##. Control of phase variation, related to expression of extracellular polymeric substances (EPS) in <italic>P. atlantica</italic> has been attributed to the mobilisation of IS492. The ability of <italic>P. atlantica</italic> to control EPS production in this manner may have profound effects on its lifestyle, which involves the movement of cells from the seaweed surface to the water column ##REF##2537827##[36]##, ##REF##9797294##[37]##. Adjacent to the IS492-related elements in <italic>P. tunicata</italic> are genes encoding for sensor/response regulatory proteins, a catalase enzyme and various hypothetical proteins. Other putative transposon elements (such as the IS91 and ISCps7 families) are adjacent to a relEB-like toxin/ antitoxin system, a collagenase gene and the antifungal tambjamine cluster. These transposable elements may play a role in the genetic regulation of environmentally relevant phenotypes such as environmental sensing, competition or oxidative stress resistance and their mobilisation might enable <italic>P. tunicata</italic> to switch between a surface-associated and a free-living form.</p>", "<p>Integrons play a major role in genetic transfer (and spread) of antibiotic resistant gene cassettes and toxins. The genome of <italic>P. tunicata</italic> encodes for four putative site-specific recombinases of the IntI4 type (COG0582), which are involved in the recombination and integration of class 4 integron sequences ##REF##9554855##[38]##. In contrast, <italic>Alteromonadales</italic> sp. TW7 contains two copies of this integrase, while <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic>, and <italic>A. macleodii</italic> possess only one copy each. Interestingly, one of these putative integrase genes in <italic>P. tunicata</italic> is located down-stream of the tambjamine biosynthesis cluster, which has so far not been identified in closely related strains. In addition the 30 kb sequence flanked by the gene cluster and the integrase show three regions in the IVOM analysis with signature for foreign DNA, which suggests the possibility that the antifungal characteristic of <italic>P. tunicata</italic> was originally derived via horizontal gene transfer. In fact the most similar cluster was identified in <italic>Streptomyces coelicolor</italic> A3(2), where it is involved in the production of a structural relative of the tambjamine, the compound undecylprodigiosin ##REF##17109029##[39]##.</p>", "<p>A clustered regularly interspaced short palindromic repeat (CRISPR) region of 5428 bp length consisting of 91 times 28 bp repeat regions with the consensus sequence (<named-content content-type=\"gene\">GTTCACTGCCGCACAGGCAGCTCAGAAA</named-content>) and 32 bp spacer was also identified in the <italic>P. tunicata</italic> genome. CRISPR elements are found in approximately half of the currently sequenced bacterial and archaeal genomes and are flanked by conserved CRISPR-associated (CAS) proteins ##REF##16292354##[40]##. Based on the homology of the CAS proteins to DNA replicating and processing enzymes and the likely origin of the spacer regions from mobile elements it has been speculated that CRISPRs provide ‘immunity’ to foreign DNA ##UREF##6##[41]##. This hypothesis was recently supported by studies in <italic>Streptococcus thermophilus</italic> demonstrating that the removal or addition of spacers modifies the phage-resistance phenotype of the cell ##UREF##6##[41]##. Upstream of the CRISPR element of <italic>P. tunicata</italic> is a group of five CAS proteins with three of them having homology to known CAS classes (CAS1, 3 and 4). Perfect matches to the <italic>P. tunicata</italic> repeat consensus were also found in the CRISPRs of <italic>Yersinia pestis</italic> (strains Pestoides F Nepal516, KIM, CO92, biovar Orientalis str. IP275 and Antiqua) <italic>Shewanella putrefaciens</italic> 200, <italic>Shewanella</italic> sp. W3-18-1, <italic>Psychromonas ingrahamii</italic> 37, <italic>Yersinia pseudotuberculosis</italic> IP 31758, <italic>Vibrio cholerae</italic> (strains V52 and RC385), <italic>Zymomonas mobilis</italic> subsp. mobilis ZM4, <italic>Photobacterium profundum</italic> SS9, <italic>Legionella pneumophila</italic> str. Lens and <italic>Erwinia carotovora</italic> subsp. <italic>atroseptica</italic> SCRI1043. In contrast, CRISPR elements could not be detected in the other two <italic>Pseudoalteromonas</italic> genomes and in <italic>Alteromonadales</italic> strain TW-7. One CRISPR was detected in <italic>A. macleodii</italic> (55 repeats), but its repeat sequence (<named-content content-type=\"gene\">GTGTTCCCCGTGCCCACGGGGATGAACCA</named-content>) showed no similarity to the one from <italic>P. tunicata</italic>.</p>", "<p>In summary, <italic>P. tunicata</italic> appears to have acquired the ability to protect itself from phage infection (e.g. via CRISPRs), yet shows other evidence and mechanisms of horizontal gene transfer (e.g. integrases) and genomic variation. This would suggest the requirement for a strict preference for and regulation of the movement of mobile genetic elements for the generation of phenotypic variation and niche adaptation. The abundance of signal transduction proteins and transcriptional regulator-like proteins (COG5000; see above), many of which have currently no known function, might play an additional role in this process.</p>", "<title>Surface attachment and biofilm formation</title>", "<p>Successful establishment on a host surface requires the bacterial cell to first adhere to host tissue, often followed by colonisation in the form of a biofilm. These processes have been well studied across several bacterial pathogens, however the molecular interactions occurring between marine bacteria and their hosts are not as well defined.</p>", "<p>The genome of <italic>P. tunicata</italic> encodes for several cell-surface structures and extracellular polymer components known in other organisms to be important for attachment. These include genes encoding for curli, Type IV pili, MSHA- pili and capsular polysaccharide (O-antigen). Curli are proteinaceous fibres belonging to the amyloid class and can make up a major component of the extracellular matrix of bacterial cells ##REF##16704339##[42]##. Although curli homologs can be found in several bacterial classes, the majority of studies to date have focused on their role in cell adhesion, biofilm formation, invasion and host inflammatory response in the <italic>Enterobacteriaceae</italic> (namely <italic>E. coli</italic> and <italic>Salmonella</italic>) ##REF##16704339##[42]##. <italic>P. tunicata</italic> possesses all the genes required for curli production and assembly, however the gene organization differs from that described for the <italic>Enterobacteriaceae</italic>. Rather than the two divergently transcribed operons consisting of the major structural subunits (CsgAB) and the accessory proteins (CsgDEFG) required for transcription and assembly, the <italic>csgABEFG</italic> genes in <italic>P. tunicata</italic> appear as one continuous operon with the regulator protein CsgD located downstream and in the opposite orientation. This gene arrangement is also found in <italic>Alteromonas macleodii</italic>, which may indicate a relatively recent gene rearrangement event. There was no evidence for curli genes in the other two <italic>Pseudoalteromonas</italic> species and in <italic>Alteromonas</italic> sp. TW7. Studies have shown that curli play a role in the interaction between <italic>E. coli</italic> and <italic>Salmonella</italic> with plant surfaces ##REF##16204476##[43]##, ##REF##16353558##[44]##, raising the possibility that <italic>P. tunicata</italic> produces curli to attach to algal host surfaces such as <italic>Ulva lactuca</italic>. Thus curli may be complementary to the previously described MSHA- pili mediated attachment of <italic>P. tunicata</italic> cells to marine host surfaces ##UREF##7##[45]##.</p>", "<p>The <italic>P. tunicata</italic> genome has nine separate gene clusters related to pili biogenesis (including the MSHA-pili). Five of the clusters contain a putative pre-pilin protein belonging to the type IVa pili characteristic of PilA or PilE proteins of <italic>Pseudomonas</italic> sp ##REF##7854130##[46]##, ##REF##7875574##[47]##. Some of the putative structural pilin genes are quite divergent from the characterised pili and may represent new pilus-like structures (##FIG##2##Figure 3##). One of the pili gene clusters and its flanking region is highly conserved in <italic>P. tunicata</italic>, <italic>P. haloplanktis</italic> and <italic>Alteromondales</italic> sp. TW7 and contains homologs to the two-component response regulator system <italic>algZ/algR</italic>, which are involved in the regulation of alginate synthesis and pili-mediated, twitching motility in <italic>Pseudomonas aeruginosa</italic>\n##REF##16352829##[48]##. Given that no alginate biosynthesis cluster is present in <italic>P. tunicata</italic>, we speculate that this regulatory system may only play a role in the expression of the pili cluster. Further downstream of the <italic>algZ</italic> is a homolog to <italic>mviN</italic>, a gene encoding for a membrane protein shown to be involved in the virulence in a variety of pathogens ##REF##17110975##[49]##, ##REF##2680969##[50]##, ##REF##12670987##[51]##. Upstream is a gene encoding for a homolog of ComL, a characterised lipoprotein involved in DNA uptake in <italic>Neisseria</italic> sp. (COG4105). Together the genes in this conserved cluster are potentially involved in colonisation and virulence and could reflect a role for <italic>P. tunicata</italic> as a potential pathogen.</p>", "<p>Another interesting protein identified in the <italic>P. tunicata</italic> genome is LipL32, a lipoprotein so far only found in <italic>Leptospira</italic> species and <italic>P. tunicata</italic>. LipL32 is involved in adhesion to common extracellular matrix (ECM) fibres, such as collagen and laminin in <italic>Leptospira</italic> sp. ##REF##18285490##[52]##. The same study demonstrated that the <italic>P. tunicata</italic> homolog of LipL32 is functionally similar to that of the Leptospira protein, suggesting that this protein may have an important function in mediating interactions between <italic>P. tunicata</italic> and its sessile marine hosts. The fact that the tunicate <italic>Ciona intestinalis</italic>, a known host of <italic>P. tunicata</italic>, has many of the genes for the production of the ECM components mentioned above gives further support for this hypothesis ##UREF##8##[53]##.</p>", "<p>The overall representation of different adhesive structures in <italic>P. tunicata</italic>, suggest that the bacterium can probably adhere to surfaces composed of different fibres and textures and that it possibly possesses a wider range of hosts, in addition to the previously recognised tunicates and algae.</p>", "<p>The <italic>P. tunicata</italic> genome contains several enzymes for the production of a complex extracellular biofilm matrix. A capsular polysaccharide genes cluster (<italic>cspA-D</italic>) is present as well as a gene cluster encoding for a protein with a biosynthesis/ export domain (Pfam 02563) and similarity to the exopolysaccharide production protein ExoF from <italic>Sinorhizobium meliloti</italic>\n##REF##9573151##[54]##. Downstream of the ExoF homolog lies a protein with a Wzz-like chain length determinant domain of surface polysaccharides (Pfam 02706). Upstream of the same gene is a conserved hypothetical protein that is found in the same genomic context in a range of <italic>Vibrio</italic> and <italic>Alteromondales</italic> genomes. An additional cluster of two genes with Pfam 02563 and 2706 domains is located just upstream of a Type II secretion system gene cluster and downstream of a mechano-sensitive channel protein of the MscS family. A third gene cluster contains an ExoF homolog with a Pfam 02563 domain in addition to other genes putatively involved in polysaccharide synthesis (putative ExoQ homolog, glycosyl transferase) in an arrangement unique to <italic>P. tunicata</italic> amongst all sequenced genomes. Together this data suggest that <italic>P. tunicata</italic> can produce and export a range of polysaccharides with potential role in biofilm matrix formation.</p>", "<title>Production of bioactive compounds and toxins</title>", "<p>\n<italic>P. tunicata</italic> has been recognised as a bacterium rich in bioactive secondary metabolites. Analysis of the genome has revealed the genes involved in the biosynthetic pathway of previously characterised activities including the antifungal tambjamine ##REF##18007521##[11]##, ##REF##17298379##[55]##, and the purple pigment violacein. Genome sequencing revealed that the violacein cluster of <italic>P. tunicata</italic> resembled that of other violacein-producing bacteria such as <italic>Chromobacterium violaceum</italic>, consisting of five consecutive genes <italic>vioA-D</italic> and the recently described <italic>vioE</italic>\n##REF##17176066##[56]##. However there is no indication for a recent horizontal gene transfer of the cluster in <italic>P. tunicata</italic>. Violacein has been demonstrated to have antibacterial activity and more recently to be used by biofilm-forming bacteria as a predator grazing defence strategy ##UREF##3##[13]##. In <italic>P. tunicata</italic> it has been suggested that violacein localises to the outer membrane of the cell ##UREF##3##[13]## and directly upstream from <italic>vioA</italic> is a gene encoding for a Multi-Antimicrobial and Toxic compound Extrusion (MATE) family efflux pump. These pumps are often used to protect the cell from damage by toxins and antimicrobial agents ##REF##11104814##[57]## and might provide a mechanism by which <italic>P. tunicata</italic> exports the otherwise toxic violacein compound.</p>", "<p>In addition to the previously described compounds the genome analysis revealed the potential for the production of other toxins, including a putative RTX-like toxin and a toxin/antitoxin system, which is homologous to the YoeB/YefM system in <italic>E. coli</italic> that is believed to act as a stress regulator ##REF##11717402##[58]## (see below). Noteworthy is also a 61 Kb large cluster of nine predicted non-ribosomal peptide synthetase (NRPS) genes. NRPS have been identified in many microorganisms and are responsible for the production of peptides with broad structural and biological activities ##REF##15487945##[59]##. NRPS are modular and are often composed of an adenylation (A) domain for substrate recognition, a peptidyl carrier protein (PCP) domain that holds the activated substrate, a condensation (C) domain for peptide bond formation and a thioesterase (T) domain for termination of peptide synthesis ##REF##15487945##[59]##. All nine NRPS within this cluster in the <italic>P. tunicata</italic> genome have putative AT, C and PCP domains with one having a T domain. A second T domain containing protein is located down stream (in opposite orientation to the nine NRPS) and lies in a putative operon with a two-component regulatory system, which might be involved in the expression of this terminase in response to environmental factors. The exact nature and regulation of the compound produced by this NRPS is currently under investigation.</p>", "<title>Stress response</title>", "<p>A common feature of all bacteria is their ability to sense and respond to adverse environmental conditions. Besides having all of the features for carbon and nutrient starvation as described for typical copiotrophic organism (RpoS, RelA, universal stress protein E, starvation stringent proteins and phage shock proteins) the genome of <italic>P. tunicata</italic> also encodes for a large number of proteins involved in oxidative stress and iron homeostasis. <italic>P. tunicata</italic> has four antioxidant proteins of the AhpC/Tsa family, including catalase and superoxide dismutase, in addition to an organic hydroperoxide detoxification protein and an alkyl hydroperoxide reductase, which protect against killing and DNA peroxide derived damage, respectively. Key regulators of the oxidative stress response are also present (such as SoxR). In contrast to these protection mechanisms there is an apparent absence of functions that result in the production of reactive oxygen species (ROS), except for the antibacterial protein AlpP (see below). <italic>P. tunicata</italic> is lacking the otherwise ubiquitous molybdopterin metabolism and several of the genes encoding for proteins that utilise this co-factor for the generation of ROS species (eg. xanthine oxidase). This feature has also been identified in the psychrophilic bacterium <italic>P. haloplanktis</italic> as an important strategy for reducing the effect of oxidative stress, which is increased at low temperature ##REF##16169927##[60]##. A high number of proteins involved in oxidative stress protection may be typical for bacteria associated with a eukaryotic host, as a common defence strategy used by a range of plants and animals is the production of oxidative bursts ##REF##12179964##[61]##. The oxidative stress proteins may also play an important role in protecting <italic>P. tunicata</italic> against its own antibacterial protein AlpP. AlpP functions as a lysine-oxidase resulting in the production of hydrogen peroxide that kills other bacteria, but also <italic>P. tunicata</italic> cells themselves in the centre of biofilm microcolonies ##UREF##9##[62]##. A fine-tuned and differential response to hydrogen peroxide is clearly required to prevent or facilitate cell death of kin in these situations.</p>", "<p>We also identified several genes involved in heavy metal detoxification, including <italic>cutA</italic> and <italic>czcA/czcB</italic>, which encode for a heavy metal efflux pump and are important for resistance to cobalt, zinc and cadmium ##REF##7623666##[63]##, ##REF##8991852##[64]##. Heavy metal resistance has been described to occur in some marine costal isolates ##REF##12704552##[65]## as well as in marine <italic>Alteromondales</italic> species ##REF##11927987##[66]##, ##REF##17551031##[67]## and might provide protection against sporadic or permanent influx of contaminations from land run-off.</p>", "<title>Polymer metabolism and its implication for host-bacteria interactions</title>", "<p>The <italic>P. tunicata</italic> genome analysis indicated a niche adaptation for the acquisition of substrates for growth from the extracellular digestion of surface-associated polymeric substances and uptake and utilization of their respective monomers. Degradation of organic matter in the oceans is fundamental for the cycling of elements and normally undertaken in the pelagic zone by bacteria attached to organic aggregates ##UREF##10##[68]##. Members of the Cytophaga-Flavobacterium group are amongst the most commonly found aggregate-associated organisms involved in organic matter degradation ##REF##17107561##[69]##, ##UREF##11##[70]##. According to Pfam categories the largest group of characterised proteins in the <italic>P. tunicata</italic> secretome are hydrolytic enzymes (##FIG##3##Figure 4##), suggesting that the bacterium is an efficient degrader of complex organic matter in the marine environment and thus may play a similar role as the CFB group on surfaces. In detail, the <italic>P. tunicata</italic> secretome consists of 371 predicted signal P containing proteins, and is approximately three-fold larger than that of <italic>P. atlantica</italic> and <italic>P. haloplanktis</italic> with 118 and 110 proteins, respectively. Overall the proportion of Sec-secreted proteins encoded by <italic>P. tunicata</italic> (8.5%) was similar to bacteria known to transport a large number of proteins to the extracellular environment. For example, <italic>B. subtilis</italic> is able to secrete nearly 170 proteins (∼4% of proteome) across the plasma membrane via the Sec translocation alone ##REF##10974125##[71]##,while common plant pathogens encode about 4% to 11% of signal peptide containing proteins in their genome ##REF##15808747##[72]##.</p>", "<p>The proteolytic potential of the <italic>P. tunicata</italic> secretome was high in comparison to <italic>P. haloplanktis</italic> and <italic>P. atlantica</italic>, with the identification of at least 36 peptidases (##FIG##4##Figure 5##). Two peptidases encoded in the <italic>P. tunicata</italic> genome match to collagenases in <italic>Vibrio</italic> sp. and <italic>Clostridium</italic> sp., and are not found in <italic>P. atlantica</italic> or <italic>P. haloplanktis</italic>. Collagen types I and II are the most abundant in cartilages of vertebrates, while marine invertebrates (including chordates) have a type of collagen similar to collagen type II ##REF##16280542##[73]##. Another secreted, proteolytic enzyme is a putative cyanophycinase, which degrades the amino-acid polymer cyanophycin, an important intracellular nitrogen-storage polymer predominantly found in cyanobacteria ##REF##12008968##[74]##.</p>", "<p>With respect to the saccharolytic potential of the <italic>P. tunicata</italic> secretome, extracellular digestion of the widespread carbon-storage polymer glycogen and starch was represented by sets of <italic>alpha</italic>-1,4 and 1,6-glucosidase and maltodextrinase genes, for which representative sets were also found in the other sequenced <italic>Alteromonadales</italic> genomes, but in lower abundance and degree of clustering with related gene functions, e.g. transport, intracellular metabolism, and regulation.</p>", "<p>The <italic>P. tunicata</italic> genome possesses furthermore an array of genes for the digestion of chitin (poly <italic>beta</italic>-1,4 acetylglucosamine), an important structural element of fungal cell wall and arthropod exoskeletons frequently found in the marine environment. Growth experiments confirmed the ability to utilise chitin and chitobiose (data not shown). For binding of chitin and digestion into oligosaccharides and chitobiose, a <italic>chiABC</italic> gene cluster similar to the one of <italic>Pseudoalteromonas</italic> sp. strain S91 ##REF##10220172##[75]## was found, in addition to five other candidate genes for chitinase and chitin-binding function. Chitinolytic activity is common in surface-associated bacteria, such as <italic>Vibrio</italic> species ##REF##17933912##[76]##, however no chitinases were found in the genomes of <italic>Alteromonas meacleodii</italic> and <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic> and <italic>Alteromonadales</italic> sp. TW7. One predicted <italic>P. tunicata</italic> chitinase contained the unusual catalytic motif for glycoside-hydrolase family 19 chitinases, which are primarily found in plants for defence against fungal and insect pathogens ##UREF##12##[77]##. This chitinase also is more similar to eukaryotic sequences (46% identical to <italic>Zea diploperennis</italic>) than to other prokaryotic sequences (highest identity of 39% to <italic>S. coelicolor</italic> A3).</p>", "<p>For extracellular digestion of chitooligosaccharides and of chitobiose into acetylglucosamine, at least six predicted <italic>beta</italic>-hexosaminidase (EC 3.2.1.52; chitobiase) genes were identified, which were not present in any other sequenced <italic>Alteromonadales</italic> genome. Additionally, at least three extracellular polysaccharide deacetylases (COG0726) allow <italic>P. tunicata</italic> to release acetyl units from polysaccharides such as chitin, and two of these deacetylases had no homologs in any other <italic>Alteromonadales</italic> genome. Furthermore, chitin degradation might be controlled by homologs of the CdsS/CdsR two-component regulator system, which in <italic>Pseudoalteromonas piscicida</italic> strain O-7 was characterized to modulate the expression of chitin degradation ##REF##17634925##[78]##, ##REF##10464221##[79]##.</p>", "<p>These results indicate a unique niche adaptation of <italic>P. tunicata</italic> for chitin degradation, and imply that chitin degradation is possibly a highly regulated and surface-associated trait. To further investigate this, we grew <italic>P. tunicata</italic> in liquid cultures with insoluble chitin and observed growth predominantly associated with the chitin particles. <italic>P. tunicata</italic> also demonstrated a reduction in the expression of the anti-protozoal compound violacein when chitin was available as sole carbon source (data not shown) hinting towards a complex regulatory network involving biopolymer degradation and inhibitor expression. In fact, <italic>P. tunicata</italic>'s regulated capacity of binding and degrading chitin might be regarded as a virulence trait, as chitinolytic bacteria have been associated with pathogenicity towards marine crustaceans ##REF##15574890##[80]##, ##UREF##13##[81]##. In addition, chitin degradation might accelerate the damage caused by the antifungal activity of <italic>P. tunicata</italic>, or a homolog to cyanophycinase and glycogen/ starch degradation might facilitate nutrient scavenging from bacterial cells lysed by AlpP. Together these findings imply, that <italic>P. tunicata</italic> does not only use its antimicrobial traits to out-compete other organisms for surface space, but might also effectively utilize the polymeric biomass of the carcass of competitor organisms for its own growth.</p>", "<p>Cell wall components of plant cells, in particular the <italic>beta</italic>-linked polysaccharides cellulose and xylan, are also potential substrates for an organism that preferentially lives on plant surfaces, including marine algae. Based on physiological and genetic analysis <italic>P. tunicata</italic> is not able to degrade cellulose. In particular, the putative extracellular endo-cellulases (EC 3.2.1.4; <italic>beta</italic>-1,4-endoglucan hydrolase) and endo-xylanases (EC 3.2.1.8; <italic>beta</italic>-1,4-endoxylanase) identified in <italic>P. haloplanktis</italic> TAC125 and <italic>P. atlantica</italic> T6c were absent in the <italic>P. tunicata</italic> genome. However, <italic>P. tunicata</italic> possesses an homolog to the <italic>Pseudomonas fluorescens</italic> subsp. <italic>cellulosa</italic> exo-cellohexanase (EC 3.2.1.74, <italic>beta</italic>-1,4-exoglucosidase), which catalyses the degradation of oligosaccharides up to the length of cellohexose, but not cellulose and xylan. Also absent were enzymes for the hydrolysis of agaropectin, agarose, inulin, levan and pectin. Oxygenase genes predicted for the degradation of poly-aromatic compounds, such as in plant-derived lignin or humic substances, were also underrepresented, or absent, in the <italic>P. tunicata</italic> genome. In contrast, <italic>P. tunicata</italic> has complete sets of genes for central metabolic pathways that confer interconversion of the monomers derived from extracellular polymer degradation and for supporting the anabolic pathways, including the intracellular conversion of chitin-monomer N-acetylglucosamine via fructose-6-phosphate.</p>", "<p>An obvious lack for the extracellular degradation of plant or algal associated polysaccharides is consistent with our understanding that <italic>P. tunicata</italic> has no observable, negative effect on algal host surfaces (such as <italic>U. lactuca</italic>). Also noteworthy is that the other recognised and unaffected eukaryotic host for <italic>P. tunicata</italic>, the tunicate <italic>C. intestinalis</italic>, is the only known animal that performs cellulose biosynthesis and incorporates cellulose into a protective coat ##REF##14722352##[82]##. Clearly, in the environment these hosts tissue can be damaged e.g. by other bacteria through polymer-degrading enzymes, and in this situation <italic>P. tunicata</italic> is likely to benefit from the decay of its host through some of its oligo-saccharide or mono-saccharide utilising pathways.</p>", "<p>\n<italic>P. tunicata</italic> is well equipped to convert the acquired carbon substrates into intracellular storage polymers, through intracellular starch synthesis, a trait which seems widespread in <italic>Alteromonadales</italic> and <italic>Vibrio</italic> genomes. Glycogen and starch production is conferred through a glycogen synthase gene as part of a predicted six-gene operon including glycogen-branching enzyme. This gene organization is conserved in <italic>Alteromonadales sp</italic>. TW7, and in <italic>Shewanella</italic> and <italic>Saccharophagus</italic> genomes (<italic>Alteromonadales</italic>), but not in the other <italic>Pseudoalteromonas</italic> genomes Glycogen and starch synthesis is likely to be expressed in <italic>P. tunicata</italic> during imbalanced supply of essential nutrients, which in the marine habitat is most likely caused by phosphorous-limitations (see below). Mechanisms of polyphosphate-storage are absent in the <italic>P. tunicata</italic> genome supporting the notion that stored carbon is crucial to make opportunistic use of short-term available phosphorous pools. This picture fits the general opportunistic life style of the genus <italic>Pseudoalteromonas</italic> and might explain the comparative ease with which they can be cultured ##REF##10877804##[83]##.</p>", "<title>Competition and acquisition of nutrients</title>", "<p>Low phosphorous levels have been shown to be one of the main growth-limiting factors in the marine environment. Indeed in the open ocean phospholipids and nucleic acids appear to be the primary reservoirs of low- and high-molecular weight dissolved organic phosphorus (DOP). Additional phosphorous can be available in the form of surface-associated, particulate organic phosphorus (POP), bound for example to plant or biofilm surfaces ##UREF##14##[84]##, ##UREF##15##[85]##, ##UREF##16##[86]##. The <italic>P. tunicata</italic> genome encodes high affinity phosphate transport systems (PstABC and PstS) and is well equipped to release phosphates from extracellular phosphoesters, as it encodes for at least five extracellular alkaline phosphatases as well as a number of extracellular nucleases and phospholipases. Interestingly, no gene for organophosphonate C-P lyases was identified (e.g. no Phn complex). This indicates that phospho(di)esters, rather than phosphonates (e.g. ciliatine), represent relevant sources of additional phosphorous in the habitat of <italic>P. tunicata</italic>.</p>", "<p>Despite this array of enzymes for phosphate acquisition, phosphate starvation might be still of ecological relevance for <italic>P. tunicata</italic> and appear to be under complex regulation. During growth experiments we observed a link between phosphate starvation and pigment/ bioactive production in <italic>P. tunicata</italic> i.e. phosphate starvation during logarithmic growth of <italic>P. tunicata</italic> results in the early expression of pigments and bioactive compounds. Under the same conditions a non-pigmented mutant of the ToxR-like regulator WmpR demonstrated that pigment production could be induced upon phosphate, but not carbon or nitrogen starvation ##UREF##17##[87]##. These results indicate the presence of a secondary regulator of the synthesis of bioactive compounds and pigments, which is activated by phosphate starvation. We identified in the <italic>P. tunicata</italic> genome homologs to the two-component regulatory system PhoR/PhoB, which has been shown in a number of bacteria to modulate gene expression in response to phosphate starvation ##REF##2556636##[88]##. The presence of the PhoR/PhoB proteins is likely responsible for the observed increase in pigment expression under phosphate starvation and recent reports have shown that antibiotic biosynthesis is negatively regulated by phosphate via the PhoR/PhoB system ##REF##12730372##[89]##. On marine surfaces, <italic>P. tunicata</italic> may be experiencing phosphate limitation, particularly in high-density consortia biofilms, and up-regulation of both bioactive compounds and phosphate acquisition could be an effective strategy to dominate competing organisms.</p>", "<p>Iron is another limiting nutrient in the marine environment and the <italic>P. tunicata</italic> genome shows adaptation to this situation by a range of siderophore-dependent mobilisation and uptake systems. A large range of TonB-dependent siderophore receptors (TBDR) were predicted in the <italic>P. tunicata</italic> genome, such as homologs to the well-characterised TonB-dependent ferric vibriobactin/enterobactin siderophore receptors ViuA and VuuA of <italic>V. cholerae</italic> and <italic>Vibrio vulnificus</italic>, the TonB-dependent catecholate siderophore receptor Fiu of <italic>E. coli</italic>, and the TonB-dependent ferric enterochelin siderophore receptors like IrgA of <italic>V. cholerae</italic> and CirA of <italic>E. coli</italic>. Surprisingly, no genes for the synthesis of these types of siderophores were found in the genome. <italic>P. tunicata</italic> might only utilise these TBDR to scavenge siderophores released from other organisms. Alternatively, the TBDRs might be involved in carbohydrate scavenging as recently suggested for some phytopathogenic and aquatic bacteria ##REF##17311090##[90]##. However, <italic>P. tunicata</italic> appears to produce at least one siderophore itself as indicated by the presence of a biosynthetic gene cluster for an aerobactin-like siderophore. Aerobactin synthesis proceeds from lysine and citric acid via L-lysine 6-monooxygenase (EC 1.14.13.59), N6-hydroxylysine O-acetyltransferase (EC 2.3.1.102), and aerobactin synthase (C-N ligase, EC 6.3.2.27), and these functions were located in <italic>P. tunicata</italic> in a predicted four-gene operon, which is also found in <italic>Alteromonadales</italic> TW7, but not in other available <italic>Alteromonadales</italic> genome sequences. More specifically, a bifunctional enzyme in <italic>P. tunicata</italic> is predicted to account for both the N6-hydroxylysine O-acetyltransferase (N-terminal half) and aerobactin synthetase (C-terminal half) activities. A homolog of this bifunctional enzyme is also present in <italic>Alteromonadales</italic> sp. TW7 (ATW7_00745), <italic>Photorhabdus luminescens</italic> (CAE17002) and <italic>Chromohalobacter salexigens</italic> (YP_573109). The gene cluster also encodes a multidrug resistance efflux pump (COG0477), which might play a role in siderophore export.</p>", "<p>Urea is ubiquitous in nature and many microorganisms utilise urea as nitrogen source. The <italic>P. tunicata</italic> genome encodes for a putative urea transporter permease of the Yut protein class, which has been characterised in <italic>Yersinia</italic> species, however does not possess a urease. Instead, urea is converted to ammonium and carbonate through a urea carboxylase and an allophanate hydrolase ##REF##15090492##[91]##. This is similar to recently described <italic>Roseobacter</italic> genomes that are also often found in association with host surfaces ##REF##17526795##[92]##.</p>", "<title>Conclusion</title>", "<p>The <italic>P. tunicata</italic> genome reflects an adaptation to successful persistence and competition on marine surfaces. The potential for <italic>P. tunicata</italic> to benefit from the decay of host tissue without causing direct harm, together with the production of inhibitory compounds against other colonisers, can give <italic>P. tunicata</italic> a selective advantage within the highly competitive surface environment. In addition, the capacity of <italic>P. tunicata</italic> to bind and degrade chitin-based oligosaccharides may suggest an expansion of its host range beyond that of the currently recognised algal and tunicate hosts. Interaction with a variety of hosts might also be facilitated by the range of surface structures (e.g. pili, curli) available to <italic>P. tunicata</italic>.</p>", "<p>In contrast to mutualistic host associations, the presence of genes homologous to virulence traits of characterised pathogens raised the interesting speculation that <italic>P. tunicata</italic> could act as an opportunistic pathogen. Microbial diseases of marine organisms are increasingly being identified and there is now strong evidence that many of these disease progressions are induced by environmental factors ##REF##12077394##[93]##. Characteristic traits of pathogens such as the production of toxins, pili, capsule polysaccharides and siderophores have been suggested to also improve bacterial fitness toward typical environmental stress conditions ##REF##17951515##[94]##, ##REF##16341015##[95]##. Therefore the study of such “dual function” traits in model non-pathogenic host associated microbes such as <italic>P. tunicata</italic> will play an important role in our overall understanding of the emergence of new microbial diseases.</p>" ]

[ "<title>Results and Discussion</title>", "<title>Overall comparison of <italic>Pseudoalteromonas</italic> or <italic>Alteromononas</italic> genomes</title>", "<p>Hierarchical clustering and principle component analysis of cluster of orthologous groups (COG) with other sequenced members of the <italic>Alteromonadales</italic> available in the IMG database indicate that <italic>P. tunicata</italic> is functionally most closely related to <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic>, <italic>Alteromonas macleodii</italic> and <italic>Alteromonadales</italic> sp. TW7. General features of these genomes are given in ##TAB##0##Table 1## and their phylogenetic relationship based on the 16S rRNA gene is illustrated in ##FIG##0##Figure 1##. The COG profiles between these organisms revealed however a number of COG categories that were over-represented in <italic>P. tunicata</italic> (##FIG##1##Figure 2##) including signal transduction mechanisms, defence mechanisms and cell motility. <italic>P. tunicata</italic> shows the highest proportion of genes assigned to signal transduction (10.05%) of any <italic>Alteromondales</italic> genome (second highest: <italic>Shewanella amazonensis</italic> with 8.62%; lowest: <italic>Psychromonas</italic> sp. CMPT3 with 5.10%). This trend is mainly caused by an over-representation of COG5000 (signal transduction histidine kinase involved in nitrogen fixation and metabolism regulation) and COG0664 (cAMP-binding proteins - catabolite gene activator and regulatory subunit of cAMP-dependent protein kinases). <italic>P. tunicata</italic> also has the highest value (2.4%) for the COG category defence of any <italic>Alteromondales</italic> genome and this is specifically due to more genes related to ABC transporter functions, including those predicted to be involved in transport of drugs and/ or antimicrobial peptides (COG0577, COG1131, COG1136, COG1566) and those related to the synthesis of potential bioactive compounds such as non-ribosomal peptide synthetase (NRPS) modules (COG1020). The abundance of genes involved in transport and expression of potential defence compounds are consistent with <italic>P. tunicata's</italic> successful competition on living surfaces. The major difference in the COG category cell motility is in COG0840 (methyl-accepting chemotaxis protein), where <italic>P. tunicata</italic> has 35 hits, which is at least twice as many as any other <italic>Pseudoalteromonas</italic> or <italic>Alteromononas</italic> species.</p>", "<p>We also compared the genomes of <italic>Pseudoalteromonas</italic> and <italic>Alteromononas</italic> species to the predominantly planktonic dataset of the Global Ocean Survey (GOS) ##REF##17355176##[31]##. Using recruitment plots ##REF##17355176##[31]## and a 95% nucleotide identity cut-off, which is indicative for species level similarity, ##REF##17220447##[32]## we found 31, 59, 6, 2327 and 1269 matching reads in the GOS dataset for <italic>P. tunicata</italic>, <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic>, <italic>Alteromonadales</italic> sp. TW7 and <italic>A. macleodii</italic>, respectively. The matches against the <italic>Pseudoalteromonas</italic> species are mostly in conserved regions of the chromosome (e.g. ribosomal genes) and hence reflect low gene divergence rather than true phylogenetic relatedness. The hits for <italic>Alteromonadales</italic> sp. TW7 and <italic>A. macleodii</italic> are distributed over the whole chromosome and are therefore likely due to true representation of these organisms (or close relatives) in the dataset. Over 95% of the hits in the GOS dataset come from the deeply sampled Sargasso Sea and for the same sample the abundant, free-living, phototrophic species <italic>Prochlorococcus marinus</italic> MED4 ##REF##15001713##[33]## recruits 2801 reads with the same cut-offs as above. Together this would indicate that <italic>Alteromonadales</italic> sp. TW7 and <italic>A. macleodii</italic> (or closely related organisms) constitute a significant portion of the heterotrophic population in these planktonic communities, while the three <italic>Pseudoalteromonas</italic> species are less prevalent.</p>", "<title>Mobile genetic elements</title>", "<p>The presence of mobile genetic elements is common in bacteria, however in comparison to closely related marine bacteria they are more abundant in <italic>P. tunicata</italic>, where they comprise 2% of the entire genome. The <italic>P. tunicata</italic> genome contains one 33 kb-sized P2-like prophage, which is similar to the one found in the genome of <italic>P. atlantica</italic>, <italic>Hahella chejuensis</italic>, <italic>Moritella</italic> sp PE36, <italic>P. aeruginosa</italic> PA7, <italic>Desulfovibrio vulgaris</italic>, <italic>Haemophilus influenza</italic>, <italic>Aeromonas salmonicida</italic> and <italic>Vibrio cholerae</italic> 0395. P2-like phage is a member of the myxoviridae and is one of the 5 major prophage groups commonly detected in Gammaproteobacteria ##REF##12794192##[34]##. A microscopic inspection of the <italic>P. tunicata</italic> media supernatant in late growth phase revealed phage-like particles (##SUPPL##0##Figure S1## online support material) suggesting that the <italic>P. tunicata</italic> P2-like phage is able to enter a lytic cycle. This observation together with the evident distribution of this prophage among marine and other bacteria could imply this phage in horizontal gene transfer that provide an adaptive advantage for the bacterial host in the marine environment or mediate acquisition of virulence genes.</p>", "<p>There are 33 genes encoding for transposases, of which approximately half are full-length genes. Of interest is the detection of multiple copies of the insertion sequence (IS) related to the IS492 element previously characterised in <italic>P. atlantica</italic>\n##REF##17264213##[35]##, ##REF##2537827##[36]##. Control of phase variation, related to expression of extracellular polymeric substances (EPS) in <italic>P. atlantica</italic> has been attributed to the mobilisation of IS492. The ability of <italic>P. atlantica</italic> to control EPS production in this manner may have profound effects on its lifestyle, which involves the movement of cells from the seaweed surface to the water column ##REF##2537827##[36]##, ##REF##9797294##[37]##. Adjacent to the IS492-related elements in <italic>P. tunicata</italic> are genes encoding for sensor/response regulatory proteins, a catalase enzyme and various hypothetical proteins. Other putative transposon elements (such as the IS91 and ISCps7 families) are adjacent to a relEB-like toxin/ antitoxin system, a collagenase gene and the antifungal tambjamine cluster. These transposable elements may play a role in the genetic regulation of environmentally relevant phenotypes such as environmental sensing, competition or oxidative stress resistance and their mobilisation might enable <italic>P. tunicata</italic> to switch between a surface-associated and a free-living form.</p>", "<p>Integrons play a major role in genetic transfer (and spread) of antibiotic resistant gene cassettes and toxins. The genome of <italic>P. tunicata</italic> encodes for four putative site-specific recombinases of the IntI4 type (COG0582), which are involved in the recombination and integration of class 4 integron sequences ##REF##9554855##[38]##. In contrast, <italic>Alteromonadales</italic> sp. TW7 contains two copies of this integrase, while <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic>, and <italic>A. macleodii</italic> possess only one copy each. Interestingly, one of these putative integrase genes in <italic>P. tunicata</italic> is located down-stream of the tambjamine biosynthesis cluster, which has so far not been identified in closely related strains. In addition the 30 kb sequence flanked by the gene cluster and the integrase show three regions in the IVOM analysis with signature for foreign DNA, which suggests the possibility that the antifungal characteristic of <italic>P. tunicata</italic> was originally derived via horizontal gene transfer. In fact the most similar cluster was identified in <italic>Streptomyces coelicolor</italic> A3(2), where it is involved in the production of a structural relative of the tambjamine, the compound undecylprodigiosin ##REF##17109029##[39]##.</p>", "<p>A clustered regularly interspaced short palindromic repeat (CRISPR) region of 5428 bp length consisting of 91 times 28 bp repeat regions with the consensus sequence (<named-content content-type=\"gene\">GTTCACTGCCGCACAGGCAGCTCAGAAA</named-content>) and 32 bp spacer was also identified in the <italic>P. tunicata</italic> genome. CRISPR elements are found in approximately half of the currently sequenced bacterial and archaeal genomes and are flanked by conserved CRISPR-associated (CAS) proteins ##REF##16292354##[40]##. Based on the homology of the CAS proteins to DNA replicating and processing enzymes and the likely origin of the spacer regions from mobile elements it has been speculated that CRISPRs provide ‘immunity’ to foreign DNA ##UREF##6##[41]##. This hypothesis was recently supported by studies in <italic>Streptococcus thermophilus</italic> demonstrating that the removal or addition of spacers modifies the phage-resistance phenotype of the cell ##UREF##6##[41]##. Upstream of the CRISPR element of <italic>P. tunicata</italic> is a group of five CAS proteins with three of them having homology to known CAS classes (CAS1, 3 and 4). Perfect matches to the <italic>P. tunicata</italic> repeat consensus were also found in the CRISPRs of <italic>Yersinia pestis</italic> (strains Pestoides F Nepal516, KIM, CO92, biovar Orientalis str. IP275 and Antiqua) <italic>Shewanella putrefaciens</italic> 200, <italic>Shewanella</italic> sp. W3-18-1, <italic>Psychromonas ingrahamii</italic> 37, <italic>Yersinia pseudotuberculosis</italic> IP 31758, <italic>Vibrio cholerae</italic> (strains V52 and RC385), <italic>Zymomonas mobilis</italic> subsp. mobilis ZM4, <italic>Photobacterium profundum</italic> SS9, <italic>Legionella pneumophila</italic> str. Lens and <italic>Erwinia carotovora</italic> subsp. <italic>atroseptica</italic> SCRI1043. In contrast, CRISPR elements could not be detected in the other two <italic>Pseudoalteromonas</italic> genomes and in <italic>Alteromonadales</italic> strain TW-7. One CRISPR was detected in <italic>A. macleodii</italic> (55 repeats), but its repeat sequence (<named-content content-type=\"gene\">GTGTTCCCCGTGCCCACGGGGATGAACCA</named-content>) showed no similarity to the one from <italic>P. tunicata</italic>.</p>", "<p>In summary, <italic>P. tunicata</italic> appears to have acquired the ability to protect itself from phage infection (e.g. via CRISPRs), yet shows other evidence and mechanisms of horizontal gene transfer (e.g. integrases) and genomic variation. This would suggest the requirement for a strict preference for and regulation of the movement of mobile genetic elements for the generation of phenotypic variation and niche adaptation. The abundance of signal transduction proteins and transcriptional regulator-like proteins (COG5000; see above), many of which have currently no known function, might play an additional role in this process.</p>", "<title>Surface attachment and biofilm formation</title>", "<p>Successful establishment on a host surface requires the bacterial cell to first adhere to host tissue, often followed by colonisation in the form of a biofilm. These processes have been well studied across several bacterial pathogens, however the molecular interactions occurring between marine bacteria and their hosts are not as well defined.</p>", "<p>The genome of <italic>P. tunicata</italic> encodes for several cell-surface structures and extracellular polymer components known in other organisms to be important for attachment. These include genes encoding for curli, Type IV pili, MSHA- pili and capsular polysaccharide (O-antigen). Curli are proteinaceous fibres belonging to the amyloid class and can make up a major component of the extracellular matrix of bacterial cells ##REF##16704339##[42]##. Although curli homologs can be found in several bacterial classes, the majority of studies to date have focused on their role in cell adhesion, biofilm formation, invasion and host inflammatory response in the <italic>Enterobacteriaceae</italic> (namely <italic>E. coli</italic> and <italic>Salmonella</italic>) ##REF##16704339##[42]##. <italic>P. tunicata</italic> possesses all the genes required for curli production and assembly, however the gene organization differs from that described for the <italic>Enterobacteriaceae</italic>. Rather than the two divergently transcribed operons consisting of the major structural subunits (CsgAB) and the accessory proteins (CsgDEFG) required for transcription and assembly, the <italic>csgABEFG</italic> genes in <italic>P. tunicata</italic> appear as one continuous operon with the regulator protein CsgD located downstream and in the opposite orientation. This gene arrangement is also found in <italic>Alteromonas macleodii</italic>, which may indicate a relatively recent gene rearrangement event. There was no evidence for curli genes in the other two <italic>Pseudoalteromonas</italic> species and in <italic>Alteromonas</italic> sp. TW7. Studies have shown that curli play a role in the interaction between <italic>E. coli</italic> and <italic>Salmonella</italic> with plant surfaces ##REF##16204476##[43]##, ##REF##16353558##[44]##, raising the possibility that <italic>P. tunicata</italic> produces curli to attach to algal host surfaces such as <italic>Ulva lactuca</italic>. Thus curli may be complementary to the previously described MSHA- pili mediated attachment of <italic>P. tunicata</italic> cells to marine host surfaces ##UREF##7##[45]##.</p>", "<p>The <italic>P. tunicata</italic> genome has nine separate gene clusters related to pili biogenesis (including the MSHA-pili). Five of the clusters contain a putative pre-pilin protein belonging to the type IVa pili characteristic of PilA or PilE proteins of <italic>Pseudomonas</italic> sp ##REF##7854130##[46]##, ##REF##7875574##[47]##. Some of the putative structural pilin genes are quite divergent from the characterised pili and may represent new pilus-like structures (##FIG##2##Figure 3##). One of the pili gene clusters and its flanking region is highly conserved in <italic>P. tunicata</italic>, <italic>P. haloplanktis</italic> and <italic>Alteromondales</italic> sp. TW7 and contains homologs to the two-component response regulator system <italic>algZ/algR</italic>, which are involved in the regulation of alginate synthesis and pili-mediated, twitching motility in <italic>Pseudomonas aeruginosa</italic>\n##REF##16352829##[48]##. Given that no alginate biosynthesis cluster is present in <italic>P. tunicata</italic>, we speculate that this regulatory system may only play a role in the expression of the pili cluster. Further downstream of the <italic>algZ</italic> is a homolog to <italic>mviN</italic>, a gene encoding for a membrane protein shown to be involved in the virulence in a variety of pathogens ##REF##17110975##[49]##, ##REF##2680969##[50]##, ##REF##12670987##[51]##. Upstream is a gene encoding for a homolog of ComL, a characterised lipoprotein involved in DNA uptake in <italic>Neisseria</italic> sp. (COG4105). Together the genes in this conserved cluster are potentially involved in colonisation and virulence and could reflect a role for <italic>P. tunicata</italic> as a potential pathogen.</p>", "<p>Another interesting protein identified in the <italic>P. tunicata</italic> genome is LipL32, a lipoprotein so far only found in <italic>Leptospira</italic> species and <italic>P. tunicata</italic>. LipL32 is involved in adhesion to common extracellular matrix (ECM) fibres, such as collagen and laminin in <italic>Leptospira</italic> sp. ##REF##18285490##[52]##. The same study demonstrated that the <italic>P. tunicata</italic> homolog of LipL32 is functionally similar to that of the Leptospira protein, suggesting that this protein may have an important function in mediating interactions between <italic>P. tunicata</italic> and its sessile marine hosts. The fact that the tunicate <italic>Ciona intestinalis</italic>, a known host of <italic>P. tunicata</italic>, has many of the genes for the production of the ECM components mentioned above gives further support for this hypothesis ##UREF##8##[53]##.</p>", "<p>The overall representation of different adhesive structures in <italic>P. tunicata</italic>, suggest that the bacterium can probably adhere to surfaces composed of different fibres and textures and that it possibly possesses a wider range of hosts, in addition to the previously recognised tunicates and algae.</p>", "<p>The <italic>P. tunicata</italic> genome contains several enzymes for the production of a complex extracellular biofilm matrix. A capsular polysaccharide genes cluster (<italic>cspA-D</italic>) is present as well as a gene cluster encoding for a protein with a biosynthesis/ export domain (Pfam 02563) and similarity to the exopolysaccharide production protein ExoF from <italic>Sinorhizobium meliloti</italic>\n##REF##9573151##[54]##. Downstream of the ExoF homolog lies a protein with a Wzz-like chain length determinant domain of surface polysaccharides (Pfam 02706). Upstream of the same gene is a conserved hypothetical protein that is found in the same genomic context in a range of <italic>Vibrio</italic> and <italic>Alteromondales</italic> genomes. An additional cluster of two genes with Pfam 02563 and 2706 domains is located just upstream of a Type II secretion system gene cluster and downstream of a mechano-sensitive channel protein of the MscS family. A third gene cluster contains an ExoF homolog with a Pfam 02563 domain in addition to other genes putatively involved in polysaccharide synthesis (putative ExoQ homolog, glycosyl transferase) in an arrangement unique to <italic>P. tunicata</italic> amongst all sequenced genomes. Together this data suggest that <italic>P. tunicata</italic> can produce and export a range of polysaccharides with potential role in biofilm matrix formation.</p>", "<title>Production of bioactive compounds and toxins</title>", "<p>\n<italic>P. tunicata</italic> has been recognised as a bacterium rich in bioactive secondary metabolites. Analysis of the genome has revealed the genes involved in the biosynthetic pathway of previously characterised activities including the antifungal tambjamine ##REF##18007521##[11]##, ##REF##17298379##[55]##, and the purple pigment violacein. Genome sequencing revealed that the violacein cluster of <italic>P. tunicata</italic> resembled that of other violacein-producing bacteria such as <italic>Chromobacterium violaceum</italic>, consisting of five consecutive genes <italic>vioA-D</italic> and the recently described <italic>vioE</italic>\n##REF##17176066##[56]##. However there is no indication for a recent horizontal gene transfer of the cluster in <italic>P. tunicata</italic>. Violacein has been demonstrated to have antibacterial activity and more recently to be used by biofilm-forming bacteria as a predator grazing defence strategy ##UREF##3##[13]##. In <italic>P. tunicata</italic> it has been suggested that violacein localises to the outer membrane of the cell ##UREF##3##[13]## and directly upstream from <italic>vioA</italic> is a gene encoding for a Multi-Antimicrobial and Toxic compound Extrusion (MATE) family efflux pump. These pumps are often used to protect the cell from damage by toxins and antimicrobial agents ##REF##11104814##[57]## and might provide a mechanism by which <italic>P. tunicata</italic> exports the otherwise toxic violacein compound.</p>", "<p>In addition to the previously described compounds the genome analysis revealed the potential for the production of other toxins, including a putative RTX-like toxin and a toxin/antitoxin system, which is homologous to the YoeB/YefM system in <italic>E. coli</italic> that is believed to act as a stress regulator ##REF##11717402##[58]## (see below). Noteworthy is also a 61 Kb large cluster of nine predicted non-ribosomal peptide synthetase (NRPS) genes. NRPS have been identified in many microorganisms and are responsible for the production of peptides with broad structural and biological activities ##REF##15487945##[59]##. NRPS are modular and are often composed of an adenylation (A) domain for substrate recognition, a peptidyl carrier protein (PCP) domain that holds the activated substrate, a condensation (C) domain for peptide bond formation and a thioesterase (T) domain for termination of peptide synthesis ##REF##15487945##[59]##. All nine NRPS within this cluster in the <italic>P. tunicata</italic> genome have putative AT, C and PCP domains with one having a T domain. A second T domain containing protein is located down stream (in opposite orientation to the nine NRPS) and lies in a putative operon with a two-component regulatory system, which might be involved in the expression of this terminase in response to environmental factors. The exact nature and regulation of the compound produced by this NRPS is currently under investigation.</p>", "<title>Stress response</title>", "<p>A common feature of all bacteria is their ability to sense and respond to adverse environmental conditions. Besides having all of the features for carbon and nutrient starvation as described for typical copiotrophic organism (RpoS, RelA, universal stress protein E, starvation stringent proteins and phage shock proteins) the genome of <italic>P. tunicata</italic> also encodes for a large number of proteins involved in oxidative stress and iron homeostasis. <italic>P. tunicata</italic> has four antioxidant proteins of the AhpC/Tsa family, including catalase and superoxide dismutase, in addition to an organic hydroperoxide detoxification protein and an alkyl hydroperoxide reductase, which protect against killing and DNA peroxide derived damage, respectively. Key regulators of the oxidative stress response are also present (such as SoxR). In contrast to these protection mechanisms there is an apparent absence of functions that result in the production of reactive oxygen species (ROS), except for the antibacterial protein AlpP (see below). <italic>P. tunicata</italic> is lacking the otherwise ubiquitous molybdopterin metabolism and several of the genes encoding for proteins that utilise this co-factor for the generation of ROS species (eg. xanthine oxidase). This feature has also been identified in the psychrophilic bacterium <italic>P. haloplanktis</italic> as an important strategy for reducing the effect of oxidative stress, which is increased at low temperature ##REF##16169927##[60]##. A high number of proteins involved in oxidative stress protection may be typical for bacteria associated with a eukaryotic host, as a common defence strategy used by a range of plants and animals is the production of oxidative bursts ##REF##12179964##[61]##. The oxidative stress proteins may also play an important role in protecting <italic>P. tunicata</italic> against its own antibacterial protein AlpP. AlpP functions as a lysine-oxidase resulting in the production of hydrogen peroxide that kills other bacteria, but also <italic>P. tunicata</italic> cells themselves in the centre of biofilm microcolonies ##UREF##9##[62]##. A fine-tuned and differential response to hydrogen peroxide is clearly required to prevent or facilitate cell death of kin in these situations.</p>", "<p>We also identified several genes involved in heavy metal detoxification, including <italic>cutA</italic> and <italic>czcA/czcB</italic>, which encode for a heavy metal efflux pump and are important for resistance to cobalt, zinc and cadmium ##REF##7623666##[63]##, ##REF##8991852##[64]##. Heavy metal resistance has been described to occur in some marine costal isolates ##REF##12704552##[65]## as well as in marine <italic>Alteromondales</italic> species ##REF##11927987##[66]##, ##REF##17551031##[67]## and might provide protection against sporadic or permanent influx of contaminations from land run-off.</p>", "<title>Polymer metabolism and its implication for host-bacteria interactions</title>", "<p>The <italic>P. tunicata</italic> genome analysis indicated a niche adaptation for the acquisition of substrates for growth from the extracellular digestion of surface-associated polymeric substances and uptake and utilization of their respective monomers. Degradation of organic matter in the oceans is fundamental for the cycling of elements and normally undertaken in the pelagic zone by bacteria attached to organic aggregates ##UREF##10##[68]##. Members of the Cytophaga-Flavobacterium group are amongst the most commonly found aggregate-associated organisms involved in organic matter degradation ##REF##17107561##[69]##, ##UREF##11##[70]##. According to Pfam categories the largest group of characterised proteins in the <italic>P. tunicata</italic> secretome are hydrolytic enzymes (##FIG##3##Figure 4##), suggesting that the bacterium is an efficient degrader of complex organic matter in the marine environment and thus may play a similar role as the CFB group on surfaces. In detail, the <italic>P. tunicata</italic> secretome consists of 371 predicted signal P containing proteins, and is approximately three-fold larger than that of <italic>P. atlantica</italic> and <italic>P. haloplanktis</italic> with 118 and 110 proteins, respectively. Overall the proportion of Sec-secreted proteins encoded by <italic>P. tunicata</italic> (8.5%) was similar to bacteria known to transport a large number of proteins to the extracellular environment. For example, <italic>B. subtilis</italic> is able to secrete nearly 170 proteins (∼4% of proteome) across the plasma membrane via the Sec translocation alone ##REF##10974125##[71]##,while common plant pathogens encode about 4% to 11% of signal peptide containing proteins in their genome ##REF##15808747##[72]##.</p>", "<p>The proteolytic potential of the <italic>P. tunicata</italic> secretome was high in comparison to <italic>P. haloplanktis</italic> and <italic>P. atlantica</italic>, with the identification of at least 36 peptidases (##FIG##4##Figure 5##). Two peptidases encoded in the <italic>P. tunicata</italic> genome match to collagenases in <italic>Vibrio</italic> sp. and <italic>Clostridium</italic> sp., and are not found in <italic>P. atlantica</italic> or <italic>P. haloplanktis</italic>. Collagen types I and II are the most abundant in cartilages of vertebrates, while marine invertebrates (including chordates) have a type of collagen similar to collagen type II ##REF##16280542##[73]##. Another secreted, proteolytic enzyme is a putative cyanophycinase, which degrades the amino-acid polymer cyanophycin, an important intracellular nitrogen-storage polymer predominantly found in cyanobacteria ##REF##12008968##[74]##.</p>", "<p>With respect to the saccharolytic potential of the <italic>P. tunicata</italic> secretome, extracellular digestion of the widespread carbon-storage polymer glycogen and starch was represented by sets of <italic>alpha</italic>-1,4 and 1,6-glucosidase and maltodextrinase genes, for which representative sets were also found in the other sequenced <italic>Alteromonadales</italic> genomes, but in lower abundance and degree of clustering with related gene functions, e.g. transport, intracellular metabolism, and regulation.</p>", "<p>The <italic>P. tunicata</italic> genome possesses furthermore an array of genes for the digestion of chitin (poly <italic>beta</italic>-1,4 acetylglucosamine), an important structural element of fungal cell wall and arthropod exoskeletons frequently found in the marine environment. Growth experiments confirmed the ability to utilise chitin and chitobiose (data not shown). For binding of chitin and digestion into oligosaccharides and chitobiose, a <italic>chiABC</italic> gene cluster similar to the one of <italic>Pseudoalteromonas</italic> sp. strain S91 ##REF##10220172##[75]## was found, in addition to five other candidate genes for chitinase and chitin-binding function. Chitinolytic activity is common in surface-associated bacteria, such as <italic>Vibrio</italic> species ##REF##17933912##[76]##, however no chitinases were found in the genomes of <italic>Alteromonas meacleodii</italic> and <italic>P. haloplanktis</italic>, <italic>P. atlantica</italic> and <italic>Alteromonadales</italic> sp. TW7. One predicted <italic>P. tunicata</italic> chitinase contained the unusual catalytic motif for glycoside-hydrolase family 19 chitinases, which are primarily found in plants for defence against fungal and insect pathogens ##UREF##12##[77]##. This chitinase also is more similar to eukaryotic sequences (46% identical to <italic>Zea diploperennis</italic>) than to other prokaryotic sequences (highest identity of 39% to <italic>S. coelicolor</italic> A3).</p>", "<p>For extracellular digestion of chitooligosaccharides and of chitobiose into acetylglucosamine, at least six predicted <italic>beta</italic>-hexosaminidase (EC 3.2.1.52; chitobiase) genes were identified, which were not present in any other sequenced <italic>Alteromonadales</italic> genome. Additionally, at least three extracellular polysaccharide deacetylases (COG0726) allow <italic>P. tunicata</italic> to release acetyl units from polysaccharides such as chitin, and two of these deacetylases had no homologs in any other <italic>Alteromonadales</italic> genome. Furthermore, chitin degradation might be controlled by homologs of the CdsS/CdsR two-component regulator system, which in <italic>Pseudoalteromonas piscicida</italic> strain O-7 was characterized to modulate the expression of chitin degradation ##REF##17634925##[78]##, ##REF##10464221##[79]##.</p>", "<p>These results indicate a unique niche adaptation of <italic>P. tunicata</italic> for chitin degradation, and imply that chitin degradation is possibly a highly regulated and surface-associated trait. To further investigate this, we grew <italic>P. tunicata</italic> in liquid cultures with insoluble chitin and observed growth predominantly associated with the chitin particles. <italic>P. tunicata</italic> also demonstrated a reduction in the expression of the anti-protozoal compound violacein when chitin was available as sole carbon source (data not shown) hinting towards a complex regulatory network involving biopolymer degradation and inhibitor expression. In fact, <italic>P. tunicata</italic>'s regulated capacity of binding and degrading chitin might be regarded as a virulence trait, as chitinolytic bacteria have been associated with pathogenicity towards marine crustaceans ##REF##15574890##[80]##, ##UREF##13##[81]##. In addition, chitin degradation might accelerate the damage caused by the antifungal activity of <italic>P. tunicata</italic>, or a homolog to cyanophycinase and glycogen/ starch degradation might facilitate nutrient scavenging from bacterial cells lysed by AlpP. Together these findings imply, that <italic>P. tunicata</italic> does not only use its antimicrobial traits to out-compete other organisms for surface space, but might also effectively utilize the polymeric biomass of the carcass of competitor organisms for its own growth.</p>", "<p>Cell wall components of plant cells, in particular the <italic>beta</italic>-linked polysaccharides cellulose and xylan, are also potential substrates for an organism that preferentially lives on plant surfaces, including marine algae. Based on physiological and genetic analysis <italic>P. tunicata</italic> is not able to degrade cellulose. In particular, the putative extracellular endo-cellulases (EC 3.2.1.4; <italic>beta</italic>-1,4-endoglucan hydrolase) and endo-xylanases (EC 3.2.1.8; <italic>beta</italic>-1,4-endoxylanase) identified in <italic>P. haloplanktis</italic> TAC125 and <italic>P. atlantica</italic> T6c were absent in the <italic>P. tunicata</italic> genome. However, <italic>P. tunicata</italic> possesses an homolog to the <italic>Pseudomonas fluorescens</italic> subsp. <italic>cellulosa</italic> exo-cellohexanase (EC 3.2.1.74, <italic>beta</italic>-1,4-exoglucosidase), which catalyses the degradation of oligosaccharides up to the length of cellohexose, but not cellulose and xylan. Also absent were enzymes for the hydrolysis of agaropectin, agarose, inulin, levan and pectin. Oxygenase genes predicted for the degradation of poly-aromatic compounds, such as in plant-derived lignin or humic substances, were also underrepresented, or absent, in the <italic>P. tunicata</italic> genome. In contrast, <italic>P. tunicata</italic> has complete sets of genes for central metabolic pathways that confer interconversion of the monomers derived from extracellular polymer degradation and for supporting the anabolic pathways, including the intracellular conversion of chitin-monomer N-acetylglucosamine via fructose-6-phosphate.</p>", "<p>An obvious lack for the extracellular degradation of plant or algal associated polysaccharides is consistent with our understanding that <italic>P. tunicata</italic> has no observable, negative effect on algal host surfaces (such as <italic>U. lactuca</italic>). Also noteworthy is that the other recognised and unaffected eukaryotic host for <italic>P. tunicata</italic>, the tunicate <italic>C. intestinalis</italic>, is the only known animal that performs cellulose biosynthesis and incorporates cellulose into a protective coat ##REF##14722352##[82]##. Clearly, in the environment these hosts tissue can be damaged e.g. by other bacteria through polymer-degrading enzymes, and in this situation <italic>P. tunicata</italic> is likely to benefit from the decay of its host through some of its oligo-saccharide or mono-saccharide utilising pathways.</p>", "<p>\n<italic>P. tunicata</italic> is well equipped to convert the acquired carbon substrates into intracellular storage polymers, through intracellular starch synthesis, a trait which seems widespread in <italic>Alteromonadales</italic> and <italic>Vibrio</italic> genomes. Glycogen and starch production is conferred through a glycogen synthase gene as part of a predicted six-gene operon including glycogen-branching enzyme. This gene organization is conserved in <italic>Alteromonadales sp</italic>. TW7, and in <italic>Shewanella</italic> and <italic>Saccharophagus</italic> genomes (<italic>Alteromonadales</italic>), but not in the other <italic>Pseudoalteromonas</italic> genomes Glycogen and starch synthesis is likely to be expressed in <italic>P. tunicata</italic> during imbalanced supply of essential nutrients, which in the marine habitat is most likely caused by phosphorous-limitations (see below). Mechanisms of polyphosphate-storage are absent in the <italic>P. tunicata</italic> genome supporting the notion that stored carbon is crucial to make opportunistic use of short-term available phosphorous pools. This picture fits the general opportunistic life style of the genus <italic>Pseudoalteromonas</italic> and might explain the comparative ease with which they can be cultured ##REF##10877804##[83]##.</p>", "<title>Competition and acquisition of nutrients</title>", "<p>Low phosphorous levels have been shown to be one of the main growth-limiting factors in the marine environment. Indeed in the open ocean phospholipids and nucleic acids appear to be the primary reservoirs of low- and high-molecular weight dissolved organic phosphorus (DOP). Additional phosphorous can be available in the form of surface-associated, particulate organic phosphorus (POP), bound for example to plant or biofilm surfaces ##UREF##14##[84]##, ##UREF##15##[85]##, ##UREF##16##[86]##. The <italic>P. tunicata</italic> genome encodes high affinity phosphate transport systems (PstABC and PstS) and is well equipped to release phosphates from extracellular phosphoesters, as it encodes for at least five extracellular alkaline phosphatases as well as a number of extracellular nucleases and phospholipases. Interestingly, no gene for organophosphonate C-P lyases was identified (e.g. no Phn complex). This indicates that phospho(di)esters, rather than phosphonates (e.g. ciliatine), represent relevant sources of additional phosphorous in the habitat of <italic>P. tunicata</italic>.</p>", "<p>Despite this array of enzymes for phosphate acquisition, phosphate starvation might be still of ecological relevance for <italic>P. tunicata</italic> and appear to be under complex regulation. During growth experiments we observed a link between phosphate starvation and pigment/ bioactive production in <italic>P. tunicata</italic> i.e. phosphate starvation during logarithmic growth of <italic>P. tunicata</italic> results in the early expression of pigments and bioactive compounds. Under the same conditions a non-pigmented mutant of the ToxR-like regulator WmpR demonstrated that pigment production could be induced upon phosphate, but not carbon or nitrogen starvation ##UREF##17##[87]##. These results indicate the presence of a secondary regulator of the synthesis of bioactive compounds and pigments, which is activated by phosphate starvation. We identified in the <italic>P. tunicata</italic> genome homologs to the two-component regulatory system PhoR/PhoB, which has been shown in a number of bacteria to modulate gene expression in response to phosphate starvation ##REF##2556636##[88]##. The presence of the PhoR/PhoB proteins is likely responsible for the observed increase in pigment expression under phosphate starvation and recent reports have shown that antibiotic biosynthesis is negatively regulated by phosphate via the PhoR/PhoB system ##REF##12730372##[89]##. On marine surfaces, <italic>P. tunicata</italic> may be experiencing phosphate limitation, particularly in high-density consortia biofilms, and up-regulation of both bioactive compounds and phosphate acquisition could be an effective strategy to dominate competing organisms.</p>", "<p>Iron is another limiting nutrient in the marine environment and the <italic>P. tunicata</italic> genome shows adaptation to this situation by a range of siderophore-dependent mobilisation and uptake systems. A large range of TonB-dependent siderophore receptors (TBDR) were predicted in the <italic>P. tunicata</italic> genome, such as homologs to the well-characterised TonB-dependent ferric vibriobactin/enterobactin siderophore receptors ViuA and VuuA of <italic>V. cholerae</italic> and <italic>Vibrio vulnificus</italic>, the TonB-dependent catecholate siderophore receptor Fiu of <italic>E. coli</italic>, and the TonB-dependent ferric enterochelin siderophore receptors like IrgA of <italic>V. cholerae</italic> and CirA of <italic>E. coli</italic>. Surprisingly, no genes for the synthesis of these types of siderophores were found in the genome. <italic>P. tunicata</italic> might only utilise these TBDR to scavenge siderophores released from other organisms. Alternatively, the TBDRs might be involved in carbohydrate scavenging as recently suggested for some phytopathogenic and aquatic bacteria ##REF##17311090##[90]##. However, <italic>P. tunicata</italic> appears to produce at least one siderophore itself as indicated by the presence of a biosynthetic gene cluster for an aerobactin-like siderophore. Aerobactin synthesis proceeds from lysine and citric acid via L-lysine 6-monooxygenase (EC 1.14.13.59), N6-hydroxylysine O-acetyltransferase (EC 2.3.1.102), and aerobactin synthase (C-N ligase, EC 6.3.2.27), and these functions were located in <italic>P. tunicata</italic> in a predicted four-gene operon, which is also found in <italic>Alteromonadales</italic> TW7, but not in other available <italic>Alteromonadales</italic> genome sequences. More specifically, a bifunctional enzyme in <italic>P. tunicata</italic> is predicted to account for both the N6-hydroxylysine O-acetyltransferase (N-terminal half) and aerobactin synthetase (C-terminal half) activities. A homolog of this bifunctional enzyme is also present in <italic>Alteromonadales</italic> sp. TW7 (ATW7_00745), <italic>Photorhabdus luminescens</italic> (CAE17002) and <italic>Chromohalobacter salexigens</italic> (YP_573109). The gene cluster also encodes a multidrug resistance efflux pump (COG0477), which might play a role in siderophore export.</p>", "<p>Urea is ubiquitous in nature and many microorganisms utilise urea as nitrogen source. The <italic>P. tunicata</italic> genome encodes for a putative urea transporter permease of the Yut protein class, which has been characterised in <italic>Yersinia</italic> species, however does not possess a urease. Instead, urea is converted to ammonium and carbonate through a urea carboxylase and an allophanate hydrolase ##REF##15090492##[91]##. This is similar to recently described <italic>Roseobacter</italic> genomes that are also often found in association with host surfaces ##REF##17526795##[92]##.</p>", "<title>Conclusion</title>", "<p>The <italic>P. tunicata</italic> genome reflects an adaptation to successful persistence and competition on marine surfaces. The potential for <italic>P. tunicata</italic> to benefit from the decay of host tissue without causing direct harm, together with the production of inhibitory compounds against other colonisers, can give <italic>P. tunicata</italic> a selective advantage within the highly competitive surface environment. In addition, the capacity of <italic>P. tunicata</italic> to bind and degrade chitin-based oligosaccharides may suggest an expansion of its host range beyond that of the currently recognised algal and tunicate hosts. Interaction with a variety of hosts might also be facilitated by the range of surface structures (e.g. pili, curli) available to <italic>P. tunicata</italic>.</p>", "<p>In contrast to mutualistic host associations, the presence of genes homologous to virulence traits of characterised pathogens raised the interesting speculation that <italic>P. tunicata</italic> could act as an opportunistic pathogen. Microbial diseases of marine organisms are increasingly being identified and there is now strong evidence that many of these disease progressions are induced by environmental factors ##REF##12077394##[93]##. Characteristic traits of pathogens such as the production of toxins, pili, capsule polysaccharides and siderophores have been suggested to also improve bacterial fitness toward typical environmental stress conditions ##REF##17951515##[94]##, ##REF##16341015##[95]##. Therefore the study of such “dual function” traits in model non-pathogenic host associated microbes such as <italic>P. tunicata</italic> will play an important role in our overall understanding of the emergence of new microbial diseases.</p>" ]

[ "<title>Conclusion</title>", "<p>The <italic>P. tunicata</italic> genome reflects an adaptation to successful persistence and competition on marine surfaces. The potential for <italic>P. tunicata</italic> to benefit from the decay of host tissue without causing direct harm, together with the production of inhibitory compounds against other colonisers, can give <italic>P. tunicata</italic> a selective advantage within the highly competitive surface environment. In addition, the capacity of <italic>P. tunicata</italic> to bind and degrade chitin-based oligosaccharides may suggest an expansion of its host range beyond that of the currently recognised algal and tunicate hosts. Interaction with a variety of hosts might also be facilitated by the range of surface structures (e.g. pili, curli) available to <italic>P. tunicata</italic>.</p>", "<p>In contrast to mutualistic host associations, the presence of genes homologous to virulence traits of characterised pathogens raised the interesting speculation that <italic>P. tunicata</italic> could act as an opportunistic pathogen. Microbial diseases of marine organisms are increasingly being identified and there is now strong evidence that many of these disease progressions are induced by environmental factors ##REF##12077394##[93]##. Characteristic traits of pathogens such as the production of toxins, pili, capsule polysaccharides and siderophores have been suggested to also improve bacterial fitness toward typical environmental stress conditions ##REF##17951515##[94]##, ##REF##16341015##[95]##. Therefore the study of such “dual function” traits in model non-pathogenic host associated microbes such as <italic>P. tunicata</italic> will play an important role in our overall understanding of the emergence of new microbial diseases.</p>" ]

[ "<p>Conceived and designed the experiments: TT FFE SE. Performed the experiments: TT FFE DSD SE. Analyzed the data: TT FFE DS AMP CB AP SSB SE. Contributed reagents/materials/analysis tools: NFS JJ SF. Wrote the paper: TT FFE DS SK SE.</p>", "<title>Background</title>", "<p>Colonisation of sessile eukaryotic host surfaces (e.g. invertebrates and seaweeds) by bacteria is common in the marine environment and is expected to create significant inter-species competition and other interactions. The bacterium <italic>Pseudoalteromonas tunicata</italic> is a successful competitor on marine surfaces owing primarily to its ability to produce a number of inhibitory molecules. As such <italic>P. tunicata</italic> has become a model organism for the studies into processes of surface colonisation and eukaryotic host-bacteria interactions.</p>", "<title>Methodology/Principal Findings</title>", "<p>To gain a broader understanding into the adaptation to a surface-associated life-style, we have sequenced and analysed the genome of <italic>P. tunicata</italic> and compared it to the genomes of closely related strains. We found that the <italic>P. tunicata</italic> genome contains several genes and gene clusters that are involved in the production of inhibitory compounds against surface competitors and secondary colonisers. Features of <italic>P. tunicata</italic>'s oxidative stress response, iron scavenging and nutrient acquisition show that the organism is well adapted to high-density communities on surfaces. Variation of the <italic>P. tunicata</italic> genome is suggested by several landmarks of genetic rearrangements and mobile genetic elements (e.g. transposons, CRISPRs, phage). Surface attachment is likely to be mediated by curli, novel pili, a number of extracellular polymers and potentially other unexpected cell surface proteins. The <italic>P. tunicata</italic> genome also shows a utilisation pattern of extracellular polymers that would avoid a degradation of its recognised hosts, while potentially causing detrimental effects on other host types. In addition, the prevalence of recognised virulence genes suggests that <italic>P. tunicata</italic> has the potential for pathogenic interactions.</p>", "<title>Conclusions/Significance</title>", "<p>The genome analysis has revealed several physiological features that would provide <italic>P. tunciata</italic> with competitive advantage against other members of the surface-associated community. We have also identified properties that could mediate interactions with surfaces other than its currently recognised hosts. This together with the detection of known virulence genes leads to the hypothesis that <italic>P. tunicata</italic> maintains a carefully regulated balance between beneficial and detrimental interactions with a range of host surfaces.</p>" ]

[ "<title>Supporting Information</title>" ]

[ "<p>We thank Granger Sutton and his team for the ongoing development and maintenance of the Celera Assembler. We thank Robert Friedman and acknowledge the J. Craig Venter Institute (JCVI) Joint Technology Center, under the leadership of Yu-Hui Rogers, for producing the genomic libraries and the sequence data.</p>" ]

[ "<fig id=\"pone-0003252-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0003252.g001</object-id><label>Figure 1</label><caption><title>Phylogenetic tree based on 16S rRNA gene sequence of <italic>Pseudoalteromonas</italic> and <italic>Alteromonas</italic> species compared in the study.</title><p>Tree was generated by maximum likelihood analysis of 1276 nucleotide positions. The sequence alignment and phylogenetic calculations were performed and manually checked with the ARB software package [97]. The 16S rRNA gene sequence of <italic>Silicibacter pomeroyi</italic> DSS-3 was used as an outgroup.</p></caption></fig>", "<fig id=\"pone-0003252-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0003252.g002</object-id><label>Figure 2</label><caption><title>Functional comparison of <italic>Pseudoalteromonas</italic> and <italic>Alteromonas</italic> genomes.</title><p>Relative abundance compared to all COGs (panel A) and absolute number of categories (panel B) for selected <italic>Pseudoalteromonas</italic> and <italic>Alteromonas</italic> species. COGs were extracted from IMG using greater than 30% identity and expectancy values of less than 10⁻⁵ cut-offs and assigned to functional categories.</p></caption></fig>", "<fig id=\"pone-0003252-g003\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0003252.g003</object-id><label>Figure 3</label><caption><title>Phylogeny of pili proteins.</title><p>Phylogenetic tree of pili proteins found in the <italic>P. tunicata</italic> genome. The pili proteins of <italic>P. tunicata</italic> are highlighted in bold and a specific name is given, if characterised. Other characterised pili proteins are shown in bold. The ten best blast-hits in NCBI's non-redundant database of the putative pili protein of <italic>P. tunicata</italic> and characterised pili proteins were used to construct the tree. Accession number and taxonomic source are shown. Bootstraps values over 750 are not shown.</p></caption></fig>", "<fig id=\"pone-0003252-g004\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0003252.g004</object-id><label>Figure 4</label><caption><title>Functional properties of predicted <italic>P. tunicata</italic> secretome.</title><p>Major protein families identified in the secretome of <italic>P. tunicata</italic> according to the Pfam database. Hypothetical proteins (141 entries) did not match to any HMM in the Pfam database; 20 proteins have the TonB-dependent receptor domain (TonB_dep_Rec); the glycolytic hydrolases (‘glyco_hydro’) group contains 11 proteins in total; 36 proteins have HMM domains that are found in different peptidases; ‘a/amido-hydrolases’ refers to alpha- and amido-hydrolases with 6 proteins in total; ‘MipA’ refers to Mlt-interacting protein like sequences with 5 sequences; ‘SBP_bac’ for extracellular solute binding protein includes SBP_bac_1 (1) and SBP_bac_3 (5); ‘Porin-like’ refers to the HMMs for OmpA (2), OmpH (1), OmpW (1), Porin_O_P (1) and Porin_1 (1); ‘MCP_signal’ refers to methyl-accepting chemotaxis protein (5); and ‘Others’ refers to all the other sequences that matched to different protein families in the Pfam database (135 entries).</p></caption></fig>", "<fig id=\"pone-0003252-g005\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0003252.g005</object-id><label>Figure 5</label><caption><title>Peptidases in predicted secretome of <italic>P. tunicata.</italic>\n</title><p>Comparison of the peptidase profiles in the secretome of the three sequenced <italic>Pseudoalteromonas</italic> species. The total number of proteins that belong to the peptidase group is 36 in <italic>P. tunicata</italic>, 12 in <italic>P. atlantica</italic> and 15 in <italic>P. haloplanktis</italic>. The colour bars represent the percentage of each subgroup of peptidases. Details of each peptidase family type can be found at <ext-link ext-link-type=\"uri\" xlink:href=\"http://merops.sanger.ac.uk\">http://merops.sanger.ac.uk</ext-link>.</p></caption></fig>" ]

[ "<table-wrap id=\"pone-0003252-t001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pone.0003252.t001</object-id><label>Table 1</label><caption><title>General feature of <italic>Pseudoalteromonas</italic> and <italic>Alteromonas</italic> genomes compared in this study.</title></caption><alternatives><table frame=\"hsides\" rules=\"groups\"><colgroup span=\"1\"><col align=\"left\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/></colgroup><thead><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"/><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>P. tunicata</italic> D2</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>P. haloplanktis</italic> TAC125</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>P. atlantica</italic> T6c</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>A. macleodii</italic> Deep ecotype</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>Alteromonas sp</italic>. TW-7</td></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Basespairs</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">4982425</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">3850272</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">5187005</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">4413342</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">4104952</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">GC percentage</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">40</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">40</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">45</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">45</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">40</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">Coding sequences (CDS)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">4504</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">3487</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">4313</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">4163</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">3783</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">average length CDS</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">984</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">978</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1051</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">935</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">972</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">CDS assigned to COG</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">3096</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">2639</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">3279</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">2926</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">2797</td></tr></tbody></table></alternatives></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"pone.0003252.s001\"><label>Figure S1</label><caption><p>Phage particle of P. tunicata. Transmission electron micrograph of phage-like structures observed in the spent medium of Pseudoalteromonas tunicata. Bar = 200 nm</p><p>(4.80 MB TIF)</p></caption></supplementary-material>" ]

[ "<fn-group><fn fn-type=\"COI-statement\"><p><bold>Competing Interests: </bold>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p><bold>Funding: </bold>Sequencing and assembly were supported by the Betty and Gordon Moore Foundation as part of its Marine Microbial Sequencing Project (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.moore.org/marinemicro\">www.moore.org/marinemicro</ext-link>). The JCVI software team, under the leadership of Saul A. Kravitz, manages assembly and Web delivery of data for the GBMF-funded project (<ext-link ext-link-type=\"uri\" xlink:href=\"http://moore.jcvi.org\">http://moore.jcvi.org</ext-link>). This research was also supported by funds provided by the Australian Research Council Discovery Project scheme. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"pone.0003252.g001\"/>", "<graphic xlink:href=\"pone.0003252.g002\"/>", "<graphic id=\"pone-0003252-t001-1\" xlink:href=\"pone.0003252.t001\"/>", "<graphic xlink:href=\"pone.0003252.g003\"/>", "<graphic xlink:href=\"pone.0003252.g004\"/>", "<graphic xlink:href=\"pone.0003252.g005\"/>" ]

[ "<media xlink:href=\"pone.0003252.s001.tif\"><caption><p>Click here for additional data file.</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>The rabbit, because of its large eyes, ease of handling, and cost effectiveness, has become a standard ophthalmic animal model for many surgical experiments, such as glaucoma filtration surgery (GFS), as well as for the development of new devices and medical therapies. Over the last five years, microarrays, which can simultaneously evaluate changes in gene expression of thousands of genes and are unparalleled in their utility as a discovery tool, have been extensively used to identify the molecular mechanisms of disease. Microarrays are commercially available for the most common animal models. However, there is no microarray for the rabbit. Furthermore, there is little rabbit sequence in public databases from which a microarray could be developed. In this paper we report the sequencing of rabbit tissues and the development of two rabbit microarrays. To test the biological validity of these microarrays as a research tool, we performed GFS on rabbits and compared the results of this study to other GFS and wound-healing studies.</p>" ]

[ "<title>Methods</title>", "<title>Rabbit eye cDNA library construction</title>", "<p>Eight- to nine-week-old, New Zealand white rabbits were obtained from Myrtle's Rabbitry (Thompsons Station, TN). Animals were treated according to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and in compliance with institutional animal use and care guidelines. Eyes from two adult rabbits were dissected and separated into two tissue groups: anterior (cornea, conjunctiva, and iris) and posterior (retina and sclera). Two separate cDNA libraries were constructed using a procedure descibed in reference [##REF##12107413##1##]. Briefly, total RNA was extracted from tissue using RNAzol (Tel-Test Inc., Friendswood, TX) and Poly (A)+ RNA was isolated using an oligo-dT cellulose affinity column. An aliquot of the RNA was run on a denaturing gel, and quality was assessed by the presence of a smooth, mRNA-shaped curve of an appropriate range of fragment size. Oligo-dT primed cDNA was synthesized at Bioserve Biotechnology (Laurel, MD) using the Superscript II system (Invitrogen, Carlsbad CA). The cDNA was run over a Sephacryl S-500 HR column (Invitrogen) to fractionate cDNA larger than 500 bp prior to being directionally cloned in <italic>Not</italic>I/<italic>Sal</italic>I sites in the pCMVSPORT6 vector (Invitrogen).</p>", "<title>cDNA sequencing</title>", "<p>Methods for sequencing and bioinformatic analysis are described in detail elsewhere [##REF##12107413##1##,##REF##12107414##2##]. Briefly, randomly picked clones were sequenced from the 5' and 3' ends at the NIH Intramural Sequencing Center (NISC). Grouping and identification of sequence tags (GRIST) were used to analyze and assemble the data [##REF##12107414##2##]. Clusters of sequences were also examined using SeqMan II (DNAstar, Madison, WI) to check assembly of clusters and to examine alternative transcripts. Sequences are available through <ext-link ext-link-type=\"uri\" xlink:href=\"http://neibank.nei.nih.gov\">NEIBank</ext-link>.</p>", "<title>Chip fabrication</title>", "<p>cDNA sequences from the new rabbit eye libraries were combined with available rabbit sequences from the National Center for Biotechnology Information (NCBI) dbEST database. Paracel Transcript Assembler (PTA; Paracel Inc., Pasadena, CA), which performs a series of sequence cleaning, chimera sequence identification, sequence clustering, and sequence assembly steps, was used to generate a set of nonredundant sequences (contigs) [##UREF##0##3##]. For each contig, a homology search was performed using the BLASTX and BLASTN application of Paracel BLAST version 1.5.6 (Paracel Inc.) against the NCBI NR and NT databases, respectively. The e-value threshold was set at e-4.</p>", "<p>BLAST results were parsed and stored in BlastQuest [##UREF##1##4##], a SQL database, developed by ICBR that facilitates the management of BLAST results and GeneOntology Consortium (GO) [##REF##11483584##5##] term browsing. AssemblyFilter software, also developed by ICBR, was used to query the top 100 BLAST hits for each contig against the NCBI Gene database, which contains annotation information, including gene function, based on GO terms and metabolic pathway association based on GenMAPP and KEGG pathway database maps [##REF##16381885##6##]. The GO terms and pathway information associated with the lowest e-value and consistent between NR and NT search were assigned to the query assembly. In cases where two contigs mapped to the same gene, the contig assembled from the smaller number of sequences was eliminated to minimize gene redundancy within the entire set of contigs. Finally, AssemblyFilter and ESTScan [##REF##10786296##7##] were used to determine the sequence orientation.</p>", "<p>Contigs were submitted to Agilent Technologies (Palo Alto, CA), where a number of probes were designed for each contig with the company's software. Multiple probes were designed for each contig and quality scores, ranging from 1-4, accompanied each designed probe. Due to limited space, we chose to only include one probe per contig and probes with one of the two highest quality scores on the array. Unfortunately, not all the contigs with high quality probes that were designed could fit on the array. All probes with homology to a sequence in the NR database were placed on the arrays. The remainder of the space on the arrays was filled with probes for which we could not find a homologous sequence. Custom arrays were manufactured on glass slides on which 60-base oligonucleotide probes were synthesized in situ with a non-contact printer. Two separate rabbit arrays were manufactured in the eight-pack format (eight individually hybridizable 1.9K arrays per piece of glass). The first set covered 1,577 genes obtained from anterior tissue, while the second set covered 1,577 genes expressed in the posterior tissue. No probe was common to both arrays. Each set of the 1,577 elements included 10 positive and five negative rabbit controls. The remaining elements were Agilent probes for scanner alignment and evaluating dye bias.</p>", "<title>Glaucoma filtration surgery</title>", "<p>Six eight- to nine-week-old male New Zealand white rabbits were obtained from Charles River Laboratories (Wilmington, MA). Animals were treated according to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and in compliance with institutional animal use and care guidelines. GFS was performed using guidelines described in reference [##REF##14744889##8##]. Briefly, the eyelids were retracted using an eyelid speculum. A partial thickness, corneal traction suture was placed in the superior cornea and used to rotate the eye inferiorly. A limbus-based conjunctival flap was fashioned in the superior lateral quadrant of the eye, approximately 8 mm from the limbus. The conjunctiva and Tenon's capsule were undermined by blunt dissection until the limbus was reached. A clear corneal paracentesis tract was made between the 5 and 7 o'clock positions using a Beaver blade (Becton Dickinson &amp; Co., Franklin Lakes, NJ) and a viscoelastic material (Healon® 10 mg/ml, Pharmacia &amp; Upjohn) was injected to maintain the anterior chamber.</p>", "<p>Starting close to the limbus, a needle tract tunnel was created through the sclera and into the anterior chamber using a beveled 22G, IV cannula (Insyte®; Becton Dickinson Vascular Access, Sandy, UT). To reduce the risk of later iris obstruction the cannula was positioned so that its orifice was beyond the pupillary margin, following withdrawal of the metal cannula stylus. The distal end of the cannula was then truncated so that its extrascleral portion was about 1mm in length. The cannula was then tethered to the sclera, to prevent dislodgement, with a single, encircling 10-0 nylon suture (Ethicon Inc., Somerville, NJ). The conjunctiva and Tenon's capsule were finally closed in separate layers, in a watertight fashion, using an 8-0 polygalactan (Vicryl®; Ethicon Inc.) attached to a BV needle and a combined neomycin and dexamethasone ointment instilled into the cul-de-sac.</p>", "<title>RNA isolation and target labeling</title>", "<p>Rabbits were sacrificed 14 days after surgery. An approximate 4x4 mm section of bleb tissue, consisting of conjunctiva and Tenon's capsule was harvested, immediately placed in a 1.5 mL microcentrifuge tube, snap frozen in liquid nitrogen, and stored at -80 °C. Conjunctiva and Tenon's capsule tissue from the control eye was harvested, frozen, and stored in the same manner.</p>", "<p>Total RNA was extracted from tissue with an RNeasy® Mini column (Qiagen, Valencia, CA). The quality of each sample was evaluated from a 200 ng aliquot with a 2100 Bioanalyzer (Agilent Technologies). Quality was assessed based on the relative abundance of the 18 and 28s ribosomal bands and on the presence of baseline rise, both of which revealed no RNA degradation. A 400 ng aliquot of total RNA was used as template for complementary DNA (cDNA) synthesis with the Low RNA Input Fluorescent Linear Amplification Kit (Agilent Technologies) according to the manufactures protocol. The subsequent cDNA product served as a template for in vitro transcription (IVT), during which one of two cyanine-labeled nucleotides (Perkin Elmer, Wellesley, MA) was incorporated into the synthesized cRNA. All RNA samples from the control eyes were labeled with cyanine 3-CTP (Cy3) while bleb samples were labeled with cyanine 5-CTP (Cy5). IVT reactions were cleaned with RNeasy® Mini columns (Qiagen), and both cRNA concentration and specific activity were measured with a ND-1000 spectrophotometer (NanoDrop Technologies Wilmington, Delaware). The quality of each cRNA sample was evaluated from a 200 ng aliquot with a 2100 Bioanalyzer (Agilent Technologies). Quality was assessed by total yield, specific activity of product, and by the presence of a smooth, mRNA-shaped curve of an appropriate range in fragment size.</p>", "<title>Array hybridization and generation of expression values</title>", "<p>Labeled cRNA samples were processed by the University of Florida's Interdisciplinary Center for Biotechnology Research (ICBR) Gene Expression Core Facility (Gainesville, FL). For each rabbit a 100 ng aliquot of the Cy5-labeled sample was combined with a 100 ng aliquot from its paired Cy3-labeled control. This mixture was incubated at 60 °C for 30 min in a high salt buffer to fragment the labeled cRNA into 30-200 base strands. Arrays were hybridized at 60 °C for 17 h in a rotating oven. After hybridization, arrays passed through both a low and high stringency wash according to the manufacture's protocols. Arrays were dried with filtered nitrogen gas and scanned in each of two wavelengths (green: 570, red: 670) with an Agilent G2505 B Scanner (Agilent Technologies). Signal values were corrected for both local background and potential differences in hybridization intensity across the array. A red:green signal ratio was calculated for each element, and ratios were normalized with a lowess transformation. All corrections and transformations of signal values were performed with Feature Extraction software version 8.1 (Agilent Technologies).</p>", "<p>A one-group Student's t-test was performed on log2 transformed signal ratios of each probe individually. The null hypothesis, that surgery does not affect probe transcript level, was rejected if the ratio were significantly different from zero. Statistical tests were performed with <ext-link ext-link-type=\"uri\" xlink:href=\"http://genomics3.biotech.ufl.edu/AnalyzeIt/AnalyzeIt.html\">AnalyzeIt Tools</ext-link> software developed by ICBR.</p>", "<title>Real-time polymerase chain reaction</title>", "<p>Quantitative real-time polymerase chain reaction (PCR) was performed on bleb and control RNA samples for five rabbit genes: interleukin 1-β (<italic>IL1β</italic>), matrix metalloproteinase-9 (<italic>MMP9</italic>), transforming growth factor-β1 (<italic>TGFβ1</italic>), transforming growth factor-β2 (<italic>TGFβ2</italic>) and fibronectin (<italic>FN1</italic>). Real-time PCR was simultaneously performed for 18s ribosomal RNA, and its expression served as an internal control. All primers and TaqMan® probes were designed and synthesized by Applied Biosystem (Foster City, CA; ##TAB##0##Table 1##). Reverse transcription was primed with random hexamers. Real-time reactions were performed in a 25 μl volume containing a 1X solution of TaqMan® Universal PCR Master Mix, 200 nM of forward primer, 200 nM of reverse primer, 50 nM of probe, and 20 ng of template cDNA. The reaction was initially heated to 50 °C for 2 min, then the temperature was raised to 95 °C for 10 min, and followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. All real-time reactions were performed with the 7900HT sequence detection system (Applied Biosystems, Foster City, CA). Both bleb and control samples were assayed in triplicate. Relative quantitative of expression levels was determined for each gene. All results are expressed as an expression ratio of the bleb to control tissue, normalized against 18s expression levels.</p>" ]

[ "<title>Results</title>", "<title>Sequencing</title>", "<p>For each library, 3,840 clones were sequenced from both 5' and 3' ends. After removal of empty vector, <italic>E.coli</italic> and mitochondrial contaminants, high quality sequences were obtained for 3,118 and 3,418 clones from the anterior tissue library (library identifier: naf) and posterior tissue library (library identifier: nag), respectively. The average quality read length was 599 bases for naf and 626 bases for nag. After GRIST analysis, the naf data yielded 1,929 potentially unique gene clusters, while nag produced 1870 clusters. Annotated sequence information is available at the NEIBank <ext-link ext-link-type=\"uri\" xlink:href=\"http://neibank.nei.nih.gov/cgi-bin/libList.cgi?org=rabbit\">rabbit library</ext-link> website. In addition, names and annotation of microarray probes for all contigs and those represented on each microarray can be found at the University of Florida <ext-link ext-link-type=\"uri\" xlink:href=\"http://vision.biotech.ufl.edu/eye/summary.html\">Ophthalmic Gene Microarray Project</ext-link>.</p>", "<title>Treatment effects</title>", "<p>An example of a scanned slide is shown in ##FIG##0##Figure 1##. Of the 3,154 total probes present on the two arrays, 2,522 had a signal value above the background, in either the red or green channel, on at least two experimental arrays. Genes represented by these probes were considered present. The remainder were considered absent (no expression under any experimental condition). However, some of the absent calls may be attributed to \"bad\" probes which have suboptimal hybridization characteristics and may not work under any experimental conditions. Only future hybridizations under different experimental conditions will allow us to determine which probes need to be redesigned.</p>", "<p>A one-group Student's t-test was performed on log2 transformed signal ratios (red versus green) to identify genes whose expression was significantly effected by GFS. The null hypothesis was rejected if the ratio was significantly different form zero. The expression of 71 and 315 (##TAB##1##Table 2##) genes was significantly altered by GFS at the p=0.01 and 0.05 level, respectively. The expression of 26 genes changed by more than two-fold and the expression of five by more than four-fold (##TAB##2##Table 3##). Of the 26 genes that changed by more than two fold, only 16 had an assigned GO annotation for biological process (##TAB##2##Table 3##). The biological process connected with all but two of these genes is characteristic of those associated with tissue injury and healing.</p>", "<title>Real-time polymerase chain reaction</title>", "<p>Sufficient quantities of RNA were only available to perform real-time PCR on four of the six original rabbit ocular samples. The changes in expression level of <italic>IL1B</italic>, <italic>MMP9</italic>, <italic>TGFB1</italic>, <italic>TGFB2</italic>, and <italic>FN1</italic> are listed in ##TAB##3##Table 4##. Of these five genes, the expression of two, <italic>IL1B</italic> and <italic>MMP9</italic>, increased dramatically (&gt;100 fold). The expression of two, <italic>FN1</italic> and <italic>TGFB1</italic>, increased moderately (three- to six-fold) and the expression of one, <italic>TGFB2</italic>, was unchanged. The array-based expression changes in <italic>IL1B</italic>, <italic>MMP9</italic>, and <italic>FN1</italic> were generally consistent with those of real-time PCR in that the expression of both <italic>IL1B</italic> and <italic>MMP9</italic> increased significantly and dramatically. For example, of the 3,155 genes present on the array, <italic>IL1B</italic> and <italic>MMP9</italic> had the second and fifth largest change in gene expression, respectively. Additionally, the expression of <italic>FN1</italic> increased significantly, but to a lesser degree than <italic>IL1B</italic> and <italic>MMP9</italic>.</p>", "<p>Real-time results for <italic>TGFB1</italic> and <italic>TGFB2</italic> were more difficult to interpret. For <italic>TGFB2</italic>, the signal values in the green channel were inordinately high. Only on the fourth array, where there was a small increase in the red signal, were the real-time and array results consistent. For <italic>TGFB1</italic>, the results were difficult to interpret because all the signal values were close to background levels. On two of the arrays, gene expression goes from present to absent, but the remaining two, gene expression goes from absent to present. Therefore, an accurate estimate of fold change was not possible.</p>" ]

[ "<title>Discussion</title>", "<p>Rabbits, because of their large eyes, ease of handling, and cost effectiveness of their use, have become an important and standard ophthalmic animal model for many surgical experiments as well as for the development of new devices and medical therapies. To date, the research community has been forced to use commercially available chips for similar model systems (rat) due to the absence of rabbit arrays. When this project was initiated, only a small amount of rabbit sequence was available from which microarray probes could be designed. To this end, we have successfully sequenced cDNA inserts from clones originating from rabbit ocular tissues. These sequences are available to the public. In addition, we developed two rabbit arrays containing a total of 3,154 unique contigs.</p>", "<p>To the best of our knowledge, this study represents the first report using a rabbit microarray. Because so little rabbit sequence was available prior to this study, it was difficult to make direct comparisons with other rabbit studies, which have focused only on the expression of one or a few genes. Therefore, we compared our results with those from other model systems. Because direct comparisons were made between species and with nonhomologous, studies we compared the general biological interpretation of results between studies.</p>", "<p>In a generalized model of wound healing, initial tissue injury stimulates various signaling events including the release of local cytokines and growth factors that eventually led to the production of structural proteins that are components of the extracellular matrix. To evaluate the biological validity of our rabbit microarrays we compared the results with those of our previous rat GFS study [##REF##15557454##9##]. In our rat microarray study, we examined changes in gene expression 2, 5, and 12 days after GFS using the Affymetrix rat 230A GeneChip®. In our current study we took samples only 14 days after surgery. This, in general, mimics the day 12 sample in the rat study. Both the 12- and 14-day sampling times correlate with the latter stages of the wounding response. Therefore, we would not only expect to find similarities between our rabbit 14-day and rat 12-day results, but we would also expect to find gene expression differences characteristic of a late wound stage, principally in genes associated with the structural process of wound healing.</p>", "<p><italic>TGFB2</italic> and connective tissue growth factor (<italic>CTGF</italic>) are two growth factors that mediate the wound-healing process. Their expression generally peaks early in the wound-healing response. For both transcript levels in the rat [##REF##15557454##9##] and protein levels in rabbit [##REF##14744889##8##], <italic>CTGF</italic> was found to reach peak expression levels 5-7 days after surgery. In both cases <italic>CTGF</italic> expression levels returned to presurgical levels by days 12-14. Like the results from the rat microarray, we found no significant difference in <italic>CTGF</italic> expression on day 14 (##TAB##1##Table 2##). In our rat GFS microarray study, we found the expression level of <italic>TGFB2</italic> decreased 2.5 fold by day 12 [##REF##15557454##9##]. In this study, the <italic>TGFB2</italic> expression level decreased 71%, but this difference was not significant.</p>", "<p>In our rat GFS model, scar tissue began to form 7-14 days after GFS [##REF##15557454##9##]. Consistent with this scarring, we saw an increase in the expression of genes which act as structural components of the extracellular matrix. In particular, we found the expression of various collagens, procollagens, biglycan, fibronectin, lumican and vimentin to increase five days after surgery [##REF##15557454##9##]. Expression level of these genes was still elevated on day 12. The results of our current study are, in general, consistent with these results. On day 14, we found both lumican and fibronectin had significantly increased (##TAB##1##Table 2##). Additionally, lysozyme, a gene thought to be directly involved with the defense against bacteria [##REF##11067934##10##,##REF##16861647##11##], was significantly increased in both the rat and rabbit microarray studies (##TAB##1##Table 2##). Finally, in our current study <italic>MMP9</italic> was one of the genes that increased the most (##TAB##2##Table 3##). In the rat microarray study, however, <italic>MMP9</italic> increased only on day 2 and 5 and was back to control levels on day 12 [##REF##15557454##9##]. A number of genes that changed the most on the rabbit array (serum amyloid protein A-1 [<italic>SAA1</italic>], serum amyloid protein A-3 [<italic>SAA3</italic>], cystine-rich secretory protein-3 [<italic>CRISP3</italic>], and alpha-1-acid glycoprotein [<italic>AGP</italic>]) were not present on the rat array, thus, no comparisons could be made with these results.</p>", "<p>The acute-phase response is an immediate set of nonspecific host inflammatory responses to tissue injury, surgical trauma, or infection. The response is characterized by the release of proinflammatory cytokines, particularly interleukin-6 (<italic>IL6</italic>), tumor necrosis factor-alpha (<italic>TNFA</italic>), and <italic>IL1B</italic>, which stimulate and mediate the hepatic synthesis and subsequent release of acute-phase proteins (APP) into the blood steam [##REF##10504381##12##, ####REF##9971870##13##, ##REF##15158206##14####15158206##14##]. An APP has been defined as one whose plasma concentration changes by at least 25% during inflammation [##REF##15158206##14##]. The concentration of some APP, however, is known to increase by as much as 1,000 fold [##REF##9971870##13##]. The systemic nature of this response constitutes an innate immune response and the APP function to restore homeostasis, neutralize pathogens and promote conditions necessary for tissue repair. APP production in extrahepatic tissues has also been observed in many mammal species and constitutes a local, rather than systemic, response [##REF##10504381##12##, ####REF##9971870##13##, ##REF##15158206##14##, ##REF##16541013##15####16541013##15##].</p>", "<p>As further biological validation of our rabbit microarrays we would expect to see changes in the expression of genes associated with an acute-phase response, the repair of tissue, and the defense against pathogens. We did find significant changes in the expression of genes involved in the acute-phase response. Of the three primary cytokines that helped mediate this response, only <italic>IL1B, IL6</italic> are represented on our rabbit arrays. The expression of both increased significantly; <italic>ILB1</italic> by 8.6 fold and <italic>IL6</italic> by 13% (##TAB##1##Table 2##). The gene with the largest increase in expression (22 fold), <italic>SAA3</italic> and its family member, <italic>SAA1</italic>, which increased 2.3 fold, are classified as acute-phase reactants [##REF##9971870##13##] (##TAB##2##Table 3##). Serum amyloid A consists of a family of apolipoproteins that are highly conserved across all vertebrates and are sensitive markers for inflammation [##REF##10504381##12##,##REF##15158206##14##,##REF##16837143##16##]. <italic>SAA3</italic> is the predominant extrahepatic form in rats, mice, and rabbits [##REF##10504381##12##,##REF##8235444##17##] and was required for effective stimulation of collagenase by <italic>IL1B</italic> in rabbit corneal fibroblasts [##REF##9434623##18##].</p>", "<p>The transcript level of three other APP genes, <italic>AGP</italic>, ceruloplasmin (<italic>CP</italic>), and <italic>FN1</italic> was significantly increased following GFS. Both <italic>AGP</italic>, whose expression increased 7 fold, and <italic>CP</italic>, whose expression increased 3.3 fold, are believed to act as antiinflammatory and immunomodulatory agents (##TAB##2##Table 3##). Therefore, it has been hypothesized that their extrahepatic expression functions to reduce inflammatory induced tissue damage [##REF##15158206##14##,##REF##11058758##19##,##REF##9374699##20##].</p>", "<p><italic>FN1</italic> gene expression increased 2.5 fold following GFS (##TAB##2##Table 3##). In humans, it appears that a single gene codes for two distinct forms, cellular and plasma. The plasma form is classified as an APP. The cellular form, however, is the major cell surface glycoprotein of many fibroblast cell lines. A major <italic>FN1</italic> function is in the adhesion of cells to extracellular matrix (ECM) materials, particularly collagen, and its presence is therefore instrumental in tissue repair [##REF##14718567##21##]. We observed the transcript expression level of two other ECM-associated genes, lumican (<italic>LUM</italic>) and <italic>MMP9,</italic> to be increased after GFS. <italic>LUM</italic> increased 2.5 fold (##TAB##2##Table 3##). LUM is present in large quantities in the corneal stroma, where it not only interacts with collagen molecules to limit fibril growth, but also plays a critical role in the regular spacing of fibrils and acquisition of corneal transparency [##REF##9606218##22##]. <italic>MMP9</italic> expression, increased 3.6 fold. <italic>MMP9</italic> is an endoprotease that cleaves matrix substrates, such as gelatin and collagen types IV, V, and VII, and therefore, plays a major role in the alteration of the ECM after tissue injury [##REF##15809093##23##,##REF##14704547##24##]. <italic>MMP9</italic> was shown to be upregulated at both the transcriptional and translational levels by SAA in human THP1 cells [##REF##15809093##23##]. Also of note is that MMP9 was found to be upregulated at both the mRNA and protein level in mononuclear blood cells of normal-tension glaucoma patients [##REF##14704547##24##].</p>", "<p>As with the increase in the expression of genes associated with the acute-phase response and tissue repair, we found the expression of defense-related genes to increase. The expression of lysozyme C and <italic>CRISP3</italic>, which are principal enzymes of the innate immune system, increased 2.5- and 3.6 fold, respectively (##TAB##2##Table 3##). Lysozyme is a protein that degrades bacterial cell walls. It is a component of granules of neutrophils and the major secretory product of macrophages [##REF##11067934##10##,##REF##16861647##11##]. Similarly, <italic>CRISP3</italic>, which was originally purified from human neutrophils, is present in the gelatinase granules of human neutrophils, along with lysozyme, collagenase, and gelatinase [##REF##12223513##25##].</p>", "<p>It is clear that our rabbit microarrays are providing gene expression results that are compatible with those found in similar tissue injury, wound healing and GFS surgery studies. For example, the genes that changed the most, <italic>SAA1</italic>, <italic>SAA3 AGP</italic>, and <italic>FN1</italic> are acute-phase reactants. If not directly involved, they are at least markers for inflammation. Other genes showing large changes include <italic>IL1B</italic>, which is known to mediate inflammatory responses, <italic>LUM</italic> and <italic>MMP9</italic>, which are associated with degradation and/or remodeling of the ECM, and both lysozyme and <italic>CRISP3</italic> which are implicated as a direct defense against pathogenic challenge. Our microarray results are also generally consistent with the magnitude and direction of real-time PCR results when microarray signal values are well above background levels.</p>", "<p>The availability of rabbit microarrays affords a new and unique opportunity for identifying molecular controls for ocular processes. An addition benefit is that these microarrays have been successfully used for a nonophthalmic study involving vascular tissue (Dr. Scott Berceli, personal communication). Fortunately, relatively few genes exhibit tissue-specific expression, and therefore, fulfill a similar function in different cell types. Thus, gene-expression and associated cellular behavior and responses across multiple tissue types, including epithelial, connective, neurological, and immunologic, may potentially be investigated using this technology, implying a utility for more than ocular research alone. The greatest limitation of these microarrays is that they only represent approximately 3000 unique genes. Further sequencing is underway to obtain a greater coverage of the expressed rabbit genome. This will no doubt, enhance the utility of the rabbit microarrays.</p>" ]

[]

[ "<p>This is an open-access article distributed under the terms of the\n Creative Commons Attribution License, which permits unrestricted use,\n distribution, and reproduction in any medium, provided the original\n work is properly cited.</p>", "<title>Purpose</title>", "<p>To develop a microarray for the rabbit that can be used for ocular gene expression research.</p>", "<title>Methods</title>", "<p>Messenger RNA was isolated from anterior segment tissues (cornea, conjunctiva, and iris) and posterior segment tissues (lens, retina, and sclera) of rabbit eyes and used to create two independent cDNA libraries through the NEIBank project. Clones from each of these libraries were sequenced from both the 5' and 3' ends. These sequences and those from the National Center for Biotechnology Information (NCBI) taxonomy database for rabbit were combined and electronically assembled into a set of unique nonoverlapping continuous sequences (contigs). For each contig, a homology search was performed using BLASTX and BLASTN against both the NCBI NR and NT databases to provide gene annotation. Unique contigs were sent to Agilent Technologies, where 60 base oligonucleotide probes were designed and synthesized, in situ, on two different arrays in an 8 array x 1900 element format. Glaucoma filtration surgery was performed on one eye of six rabbits. After 14 days, tissue was harvested from the conjunctiva and Tenon's capsule of both the surgically treated and untreated control eyes. Total RNA from each sample was labeled with cyanine dyes and hybridized to our custom microarrays.</p>", "<title>Results</title>", "<p>Of the 3,154 total probes present on the two arrays, 2,522 had a signal value above the background. The expression of 315 genes was significantly altered by glaucoma filtration surgery. Genes whose expression was altered included proteins associated with inflammatory response, defense response, and proteins involved in synthesis of the extracellular matrix.</p>", "<title>Conclusions</title>", "<p>The results of this rabbit microarray study are consistent with those from other wound healing studies, indicating that this array can provide valid information on broad patterns of gene expression. This is the first microarray available for rabbit studies and is a valuable tool that can be used to study molecular events in the eye.</p>" ]

[]

[ "<title>Acknowledgements</title>", "<p>This work was supported by a Research Opportunity Fund from the University of Florida's Division of Sponsored Research and through the NEIBank project of the National Eye Institute. The authors thank Dr W. Clay Smith for reviewing an earlier version of this manuscript.</p>" ]

[ "<fig id=\"f1\" fig-type=\"figure\" position=\"float\"><label>Figure 1</label><caption><p>Examples of images generated from scanned microarrays. Example of an image generated from a scanned eight-pack microarray (<bold>A</bold>) and a single 1,900 element (<bold>B</bold>) microarray. Each circle represents a unique spotted probe. Red probes indicate that gene expression in the surgically treated eye is higher than in the control, while green marks higher gene expression in the control. Yellow indicates no difference in gene expression between the surgically treated and control eyes. Black denotes absence of detectable signal.</p></caption></fig>" ]

[ "<table-wrap id=\"t1\" position=\"float\"><label>Table 1</label><caption><title>Primer and TaqMan® probe sequences for selected rabbit genes.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"100\" span=\"1\"/><col width=\"214\" span=\"1\"/><col width=\"246\" span=\"1\"/><col width=\"204\" span=\"1\"/><tbody><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Gene</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Forward primer</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>TaqMan probe</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Reverse primer</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Matrix metalloproteinase-9<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CCGGCATTCAGGGAGATG<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CTGGGCAAGGGCGTCGTGGTT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TCGGCGTTTCCAAAGTACGT<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Interleukin-1 beta<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TTGCTGAGCCAGCCTCTCTT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CTGCCATTCAGGCAAGGCCAGC<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CTGGGTACCAAGGTTCTTTGAACT<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Transforming growth factor beta-2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CGCCAAGGAGGTCTACAAGATAG<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CATGCCGTCCTACTTCCCCTCCGA<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GGTGGGTGGGATGGCATT<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Transforming growth factor beta-1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AAGGGCTACCACGCCAACT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AGTACAGCAAGGTCCTGGCCCTG(7Gs7Cs)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CCGGGTTGTGCTGGTTGT<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Fibronectin</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GTGGAATACGTGGTCAGTGTCTATG</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CCGTTCCGGTTTTGTG</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TGGTGGTTACTGCAGTCTGAAC</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t2\" position=\"float\"><label>Table 2</label><caption><title>Altered gene expression following glaucoma filtration surgery.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"97\" span=\"1\"/><col width=\"79\" span=\"1\"/><col width=\"68\" span=\"1\"/><col width=\"156\" span=\"1\"/><col width=\"40\" span=\"1\"/><tbody><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Probe name</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>p value</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>mean log2 difference</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Hit definition</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>E-value</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.44<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">13kDa differentiation-associated protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31359<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.49<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2310004I24Rik protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31415<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.79<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2810480G15Rik protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30841<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.36<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2-aminomuconic acid semialdehyde dehydrogenase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40335<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.64<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3-methyl-2-oxobutanoate dehydrogenase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41433<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.027<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.51<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5-HT3 receptor<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40101<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.044<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.52<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Acetyl-CoA acetyltransferase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31324<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.017<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Acid sphingomyelinase-like phosphodiesterase 3a<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40551<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Adipose differentiation-related protein; adipophilin<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41499<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.45<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ADP/ATP translocase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30414<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.51<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ADP-ribosylation factor-like 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41261<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.034<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.30<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Aggrecan core protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31524<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.005<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.67<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Aggrecanase-2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30810<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.14<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AgrB<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.7<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41308<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.81<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Alpha-1-acid glycoprotein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40520<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.044<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.77<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Alpha-1-glycoprotein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30067<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.52<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Alpha-actinin-2-associated LIM protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40392<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.92<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Anterior gradient 2 homolog<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40535<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.045<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.21<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Apolipoprotein D<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30343<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.033<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ATP synthase, H+ transporting, mitochondrial F1 complex, beta polypeptide<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30203<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.22<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">BAP31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41216<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.93<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Beta casein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41423<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.010<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.87<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Beta tropomyosin<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41440<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.57<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Beta-arrestin 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31084<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Bifunctional cbiH protein and precorrin-3B C17-methyltransferase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40296<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.12<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">BMS1-like, ribosome assembly protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40651<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.040<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.44<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Bromo-adjacent homology domain-containing protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31039<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.76<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">C. elegans SRA-13 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40046<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.69<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ca2+-transporting ATPase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_40223<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.36<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Calmodulin 1 (phosphorylase kinase, delta)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30531<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.11<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cathepsin B<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41428<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.05<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cathepsin E<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40013<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.000<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.07<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cathepsin K<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41131<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.49<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CD36 antigen<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31102<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.035<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.87<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cell cycle control protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_41026<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.48<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CG11023-PA<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">7.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40760<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CG6137-PA<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.7<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_30175<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.013<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.29<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Chaperonin containing TCP1, subunit 2 (beta)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_40241<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.29<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Chaperonin containing TCP1, subunit 2 (beta)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40835<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.029<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Chloroquine resistance marker protein, putative<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41489<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-2.06<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Chondromodulin-I<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_y_31574<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.040<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Chromosome 17 open reading frame 37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30335<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.52<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Complement C4A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30932<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.74<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Conserved hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40995<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.013<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Conserved hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40346<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.040<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Corneal endothelium specific protein 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40478<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.66<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Corneal endothelium specific protein 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30192<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.72<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CRTAC1-B protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31242<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.027<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.67<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cryptochrome 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41539<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.009<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.04<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cystatin B<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41097<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.47<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cystic fibrosis transmembrane conductance regulator<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_40228<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.045<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cytochrome b<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30452<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cytochrome c oxidase subunit I<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30490<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.47<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cytochrome c oxidase subunit III<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41352<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cytochrome P450 2A10<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41474<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.15<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cytochrome P450 8B1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30439<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cytoskeleton-associated protein 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40165<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.040<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.56<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">DEAD (Asp-Glu-Ala-Asp) box polypeptide 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40175<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.79<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Dihydrolipoamide dehydrogenase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30308<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.028<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.72<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Dimeric dihydrodiol dehydrogenase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.036<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.46<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">DKFZP564B167 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31202<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.91<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">DNA segment, Chr 6, ERATO Doi 253<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30647<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">DNA topoisomerase, type I<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30660<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.039<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.43<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">DNA-directed RNA polymerase II largest subunit<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30059<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.025<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">DnaJ homolog, subfamily A, member 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30305<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.23<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">DSIF p160<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30277<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.49<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Dynactin 1 0.0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_30346<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.82<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ENO1 protein 0.<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30721<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.10<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ENSANGP00000019839<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.3<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30587<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.68<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">EPAS1 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40498<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.92<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Epithelial chloride channel protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30788<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Eukaryotic translation initiation factor 2B<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31113<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.62<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Eukaryotic translation initiation factor 5<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_y_41571<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">EWS/ZSG fusion protein short isoform<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30768<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.73<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Expressed sequence AI462446<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30758<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.028<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.36<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Fibrillar collagen<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31501<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.28<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Fibronectin<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41550<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">FK506-binding protein 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30610<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.017<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Foocen-m2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30129<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.62<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">FUS/TLS protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30813<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.035<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.68<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GABA/noradrenaline transporter<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.7<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40112<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.50<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GL004 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30658<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GL014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40973<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.035<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.78<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GLP_680_59866_66603<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40723<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.97<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GLP_82_33920_36064<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.3<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31557<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.77<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Glucocorticoid receptor<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_30398<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.62<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Glucocorticoid-induced leucine zipper<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_40359<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.69<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Glucocorticoid-induced leucine zipper<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31290<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.27<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Glutamine fructose-6-phosphate transaminase 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41492<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.033<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.79<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Glyceraldehyde 3-phosphate dehydrogenase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41246<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.80<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Glyceraldehyde 3-phosphate dehydrogenase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41208<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.71<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Glycogen debranching enzyme<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41424<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GPI-linked NAD(P+)--arginine ADP-ribosyltransferase 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40796<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.047<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.39<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Gro-1 Operon gene GOP-1, GOP-1 2.2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41431<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.34<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">GTPase regulator associated with the focal adhesion kinase pp125<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40925<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.22<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Heat shock protein 17.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41231<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Heat shock protein 8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40953<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.19<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Heme maturase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31240<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.47<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Heparan sulfate glucosaminyl N-deacetylase/N-sulfotransferase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40160<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.000<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.46<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Heterogeneous nuclear ribonucleoprotein C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41472<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.21<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Histocompatibility antigen DM Heterodimer heavy chain<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30332<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.64<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">HRMT1L2 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30569<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical class II basic helix-loop-helix protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30551<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.74<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30950<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.69<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.2<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40650<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.031<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.65<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40582<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.49<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40929<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.49<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30852<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.034<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.42<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31105<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.045<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40704<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.028<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40776<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.24<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_41030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.15<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">7.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30710<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.049<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.22<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.2<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40581<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.043<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30692<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.47<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.2<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40956<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.73<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.6<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40928<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.75<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30565<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.70<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30056<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.043<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.63<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein BC009732<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_41054<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.043<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.45<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein Cj0447<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">8.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30796<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein Daro140001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.6<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31033<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.005<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.91<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein Daro170901<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.031<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.52<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein DKFZp761A052.1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40882<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.025<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.68<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein DKFZp761H221.1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31126<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.034<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.16<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein Exigu022705<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">7.3<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40979<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.047<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.27<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein FG08398.1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.2<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40111<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.013<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.44<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein MGC14353; thioredoxin (Trx)-related protein, 14 kDa<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_y_31570<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein MGC33212<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.45<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein MGC4825<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40649<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein OB1070<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31461<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.42<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein RL076<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.6<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30623<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein UL126<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40933<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.031<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.29<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein UM01639.1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40842<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.42<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein UM01797.1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41124<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.44<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypoxanthine phosphoribosyltransferase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41230<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.77<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypoxia inducible factor 1 alpha subunit<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31343<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.42<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypoxia up-regulated 1; calcium binding protein, 140 kDa<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41458<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.30<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Immunoglobulin heavy chain variable region<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41415<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.41<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Immunoglobulin heavy chain VDJ region<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31509<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.046<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Importin beta-3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40458<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.017<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.29<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Inositol(myo)-1(or 4)-monophosphatase 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31512<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.029<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Integrin beta1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41542<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.18<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Interleukin 60.0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41565<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.11<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Interleukin-1 beta<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30860<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Iron (III) ABC transporter, ATPase component<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31347<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.033<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Junction adhesion molecule 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31262<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.30<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Karyopherin alpha 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30101<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">KIAA0077<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30624<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.044<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.58<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">KIAA0653 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30281<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">KIAA0735 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40209<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">KIAA1289 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41561<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.045<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.10<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lactase-glycosylceramidase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31425<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.41<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lanosterol synthase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40685<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.56<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">LD11664p<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41277<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.60<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Low density lipoprotein-related protein 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40419<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lumican<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40213<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.05<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lumican<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40194<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.010<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">LysozymeC(1,4-beta-N-acetylmuramidase C)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30224<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.035<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.45<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lysyl oxidase-like protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.034<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.23<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Macrolide-efflux protein 3.9<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40724<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.19<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">MADS box protein TDR4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.3<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40715<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.17<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Major facilitator family transporter<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.2<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_40252<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Mammary tumor integration site 6 oncogene protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41311<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.021<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">MAPK-activated protein kinase 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41197<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.58<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Matrix metalloproteinase-1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41283<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.85<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Matrix metalloproteinase-9<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40923<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.009<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Maturase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.7<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30559<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.24<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Membrane protein TGN38 long form<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31527<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.80<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">MGC5309 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40323<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.87<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">MHC class II histocompatibility antigen RLA-DQ alpha chain<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40271<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.005<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.50<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Mitochondrial ATP synthase, O subunit<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40449<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.021<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.16<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Mitochondrial isoleucine tRNA synthetase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41237<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.06<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Mono (ADP-ribosyl)transferase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41350<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Monocyte differentiation antigen CD14<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40183<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Multi-PDZ-domain-containing protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40471<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.78<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Mutant alpha-1 collagen type 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40493<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NADH dehydrogenase subunit 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40622<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.010<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.00<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NADH dehydrogenase subunit 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.3<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31429<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.22<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NADH dehydrogenase, Fe-S protein 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40140<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.29<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NADH dehydrogenase, Fe-S protein 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40409<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.70<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NADH dehydrogenase1 alpha subcomplex 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31543<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.033<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.45<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NADP(H)-oxidase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31371<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.14<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Nedd4 WW binding protein 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40416<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.83<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Neutrophil granules matrix glycoprotein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30451<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.021<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NIR2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41335<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.23<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Nitric oxide synthase, inducible<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30229<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.55<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">N-myc downstream-regulated gene 2 isoform b<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30979<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Non-phototropic hypocotyl 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.7<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31505<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.013<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.27<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Nuclear receptor subfamily 5 group A member 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31360<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.70<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Nudix (nucleoside diphosphate linked moiety X)-type motif 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30432<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.046<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.26<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">OK/SW-CL.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Hs_31471<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.62<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Optineurin<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41275<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.26<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ORM1-like 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40096<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.08<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Peptidyl prolyl isomerase H<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30217<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.013<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.47<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Peptidyl-prolyl cis/trans isomerase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31468<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Permeases of the drug/metabolite transporter superfamily<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41377<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.76<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Phosphatidylethanolamine-binding protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_30183<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.043<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.48<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Phosphoglycerate kinase 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41437<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.16<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Phospholipase A2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41418<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.018<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Potassium inwardly-rectifying channelJ10<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_31072<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.021<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.53<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Prespore-specific protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.7<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31207<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.021<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.17<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Procollagen, type VI, alpha 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40554<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Proline-rich protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30983<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.034<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.30<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Prolyl endopeptidase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30326<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.21<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Proteasome 26S subunit, non-ATPase, 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31331<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.63<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Protein inhibitor of activated STAT PIASy<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31303<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.56<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Protein phosphatase 1D magnesium-dependent, delta isoform<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40192<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Protein phosphatase 2, catalytic subunit, alpha isoform<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40376<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Protein transport protein SEC61 beta subunit<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30690<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Putative gamma-glutamyltranspeptidase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.2<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40531<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.028<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.72<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Putative nuclear protein (1H963)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40972<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.047<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.17<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Putative signal transducer<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40587<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.61<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Putative yir4 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40146<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.56<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Pyruvate dehydrogenase E1 component beta subunit<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30286<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Pyruvate kinase, M1 isozyme<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31406<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.15<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Quininoid dihydropteridine reductase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30365<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.029<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RAB5-interacting protein isoform a<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40043<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.21<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ribosomal protein L27<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40300<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ribosomal protein L31; 60S ribosomal protein L31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40110<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.70<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ribosomal protein L5<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_30176<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.39<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ribosomal protein S12<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40373<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ribosomal protein S26<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_40235<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.34<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ribosomal protein S3a; 40S ribosomal protein S3a<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31352<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.91<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ribosome binding protein 1 isoform mRRp61<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40391<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.040<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RIKEN cDNA 1110020P15<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31177<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.010<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.39<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RIKEN cDNA 1300017E09<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31399<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.017<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.75<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RIKEN cDNA 2510025F08<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31403<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.16<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RIKEN cDNA 4930556P03<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30988<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.039<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.66<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RIKEN cDNA 4933437K13<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30495<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.76<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RIKEN cDNA A530046H20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40331<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.27<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ring-box 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30892<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.005<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.54<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">RNA polymerase beta chain<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40403<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.19<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">SECP43 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31341<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Secreted modular calcium-binding protein 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30486<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Seipin<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41531<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Serum amyloid A-1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41285<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.48<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Serum amyloid A-3 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40486<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.36<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to 40S ribosomal protein S6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40398<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.021<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to 60S ribosomal protein L34<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30580<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.028<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to Ac2-210<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30431<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.041<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.48<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to CG5987-PA<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41235<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.57<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to cyclin I<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31214<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.044<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.60<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to ganglioside-induced differentiation associated protein 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30968<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.28<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40524<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.040<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to hypothetical protein 9630041N07<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30389<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.87<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to Ldb1a<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31272<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.22<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to NNX3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41147<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.53<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to Nucleolar RNA helicase II<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40138<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.044<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to ribosomal protein L30<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30322<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.036<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.29<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to RIKEN cDNA 0610043B10<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31394<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.034<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.11<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to RIKEN cDNA 5730403E06<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40614<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.70<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to T cell receptor V delta 8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.3<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40202<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.51<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to ubiquinol-cytochrome c reductase binding protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.039<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.51<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Single-stranded DNA binding protein 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41236<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.017<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">SLC26A7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31395<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.000<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.87<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Sloan-Kettering viral oncogene homolog<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40348<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.33<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Small nuclear ribonucleoprotein polypeptide E<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31313<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.68<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">SMC1 structural maintenance of chromosomes 1-like 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41221<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.44<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Sodium/glucose cotransporter 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30466<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.030<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.63<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Sodium/potassium/calcium exchanger 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31523<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.44<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Sperm-binding glycoprotein ZP2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31221<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Splicing factor, arginine/serine-rich 5<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41289<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.027<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Succinate dehydrogenase 0.0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.047<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.31<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">SUMO-1 activating enzyme subunit 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40388<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.57<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Synovial sarcoma translocation gene on chromosome 18-like 2; kiaa-iso protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41132<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.049<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.28<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">T-cell receptor beta-chain precursor<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31311<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.035<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.49<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Telomerase binding protein, p23<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_41090<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Thymosin, beta 4, X chromosome; prothymosin beta 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41317<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.89<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Tissue inhibitor of metalloproteinase-2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_41076<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Transcriptional regulator9.0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31168<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.17<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Transforming growth factor alpha<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41523<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.29<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Translation initiation factor eIF-2B-delta<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.26<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Translokin<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30842<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Tryptophanyl-tRNA synthetase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40794<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.035<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.46<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Tubulin beta chain<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31328<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.79<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Tweety homolog 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.70<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ubiquinol-cytochrome c reductase, Rieske iron-sulfur polypeptide 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Mm_31210<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.66<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ubiquitin ligase E3 alpha-II<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40115<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.71<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Uncharacterized hematopoietic stem/progenitor cells protein MDS029<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30772<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.76<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31071<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.71<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.6<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30995<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.50<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.1<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30628<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.014<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.44<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_41047<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.40<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">7.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40641<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.017<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.39<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_41089<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.034<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.30<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">9.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_41084<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.026<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.26<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">9.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30901<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.037<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.18<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.6<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30563<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.047<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.12<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31141<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.14<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">8.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40874<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.23<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40868<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.005<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.3<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.39<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30622<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.044<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.41<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40813<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.044<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.47<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.5<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40821<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.60<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.6<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40282<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.82<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40775<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.94<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unknown<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40341<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.045<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unnamed protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40699<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.021<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unnamed protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30370<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unnamed protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40420<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.010<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.45<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unnamed protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_40305<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.019<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.50<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unnamed protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_m_30347<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.031<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.61<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unnamed protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30055<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.032<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.71<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Unnamed protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30237<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.56<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Vacuolar ATP synthase 16 kDa proteolipid subunit<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41218<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.033<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.22<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Vacuolar proton-translocating ATPase a2 isoform<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41254<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.023<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.28<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Vascular endothelial growth factor receptor 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41151<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-0.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">V-erb-a erythroblastic leukemia viral oncogene homolog 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41502<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.05<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Voltage-dependent anion channel 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30382<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.74<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">X-ray crystal structure of human ceruloplasmin at 3.0 Angstroms<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.0<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40769<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.050<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.42<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Zn-dependent carboxypeptidase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.8<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_31024</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.027</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.49</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Zonadhesin</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.6</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t3\" position=\"float\"><label>Table 3</label><caption><title>Genes whose expression was significantly (p less than or equal to 0.05) altered by more than two-fold.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"94\" span=\"1\"/><col width=\"38\" span=\"1\"/><col width=\"62\" span=\"1\"/><col width=\"100\" span=\"1\"/><col width=\"67\" span=\"1\"/><col width=\"83\" span=\"1\"/><tbody><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Probe name</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>p-value</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Mean log2 difference</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Hit definition</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>E-value</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>GO biological process term</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41285<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.48<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Serum amyloid A-3 protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Acute-phase response<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41565<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.007<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.11<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Interleukin-1 beta (IL-1 beta)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Antimicrobial humoral response, inflammatory response<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41308<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.81<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Alpha-1-acid glycoprotein (Orosomucoid)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Acute-phase response, inflammatory response<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40013<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.000<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.07<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cathepsin K (EC 3.4.22.-)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lysosome, proteolysis and peptidolysis<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30389<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.87<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Similar to Ldb1a<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">8.46E-37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41283<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.85<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">92 kDa type IV collagenase (Matrix metalloproteinase-9)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Collagen catabolism, collagenase activity<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40416<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.83<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Neutrophil granules matrix glycoprotein SGP28<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.40E-20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Defense response, innate immune response, cell-cell adhesion<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30382<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.74<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Ceruloplasmin At 3.0 angstrom<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.04E-38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">oxidoreductase activity<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40419<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.37<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lumican<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.93E-20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Collagen binding, collagen fibril organization<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40194<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.010<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.32<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lysozyme C (1,4-beta-N-acetylmuramidase C)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lysozyme activity, response to bacteria, cell wall catabolism<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_31501<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.011<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.28<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Fibronectin<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Acute phase response, inflammatory response, cell adhesion and migration, extracellular matrix structural constituent<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41531<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.016<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Serum amyloid A-1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Acute phase response<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40715<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.038<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.17<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">major facilitator family transporter<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.18908<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30531<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.003<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.11<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cathepsin B<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.13E-06<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Negative regulation of inflammatory response, proteolysis and peptidolysis<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41428<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.05<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cathepsin E (EC 3.4.23.34)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Positive regulation of cytokine secretion, proteolysis and peptidolysis<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40213<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.048<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.05<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Lumican<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Collagen binding, collagen fibril organization<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41539<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.009<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.04<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cystatin B<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40622<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.010<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.00<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">NADH dehydrogenase subunit 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.335226<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41502<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.05<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Voltage-dependent anion channel 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Mitochondrial outer membrane<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41237<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.06<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Mono (ADP-ribosyl) transferase<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40346<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.040<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Corneal endothelium specific protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.09E-38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41352<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.020<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cytochrome P450 2A10 (CYPIIA10)<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Electron transport, oxidoreductase activity on paired donors, oxygen binding<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30067<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.004<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.52<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Alpha-actinin-2-associated LIM protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_c_40478<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.66<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Corneal endothelium specific protein 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.32E-09<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_r_30565<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.006<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.70<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Hypothetical protein<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.00204271<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">UF_Oc_n_41489</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.008</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-2.06</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Chondromodulin-I (Leukocytecell-derived chemotaxin 1)</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Cell differentiation, extracellular space</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t4\" position=\"float\"><label>Table 4</label><caption><title>Real-time PCR results.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"92\" span=\"1\"/><col width=\"88\" span=\"1\"/><col width=\"87\" span=\"1\"/><col width=\"87\" span=\"1\"/><col width=\"87\" span=\"1\"/><tbody><tr><td colspan=\"5\" valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\"><bold>Transforming growth factor beta 2</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Sample</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Green Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Red Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Array</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Real-Time</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">73.9**<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">45.3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.38<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.77<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">21.2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">9*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-2.35<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.26<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">850.6**<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">83.2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-10.20<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.04<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">7A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">263.9**<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">43<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-96.06<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.15<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Mean<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-5.00<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.31<hr/></td></tr><tr><td colspan=\"5\" valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\"><bold>Matrix metalloproteinase 9</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Sample</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Green Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Red Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Array</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Real-Time</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">29.7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">106.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.60<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">727.2<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">18.4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">158.9<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">8.64<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1928.9<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">26.2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">69.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.67<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">574.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">7A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">30.9<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">110.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.59<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1020.4<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Mean<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.62<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1062.7<hr/></td></tr><tr><td colspan=\"5\" valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\"><bold>Interleukin-1 beta</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Sample</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Green Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Red Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Array</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Real-Time</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">14<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">254.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">18.25<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">86.03<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.3*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">50.4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">11.69<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">258.92<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">11.1*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.43<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">67.75<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">7A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">17.2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">306.1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">17.82<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">146.83<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Mean<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">13.3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">139.9<hr/></td></tr><tr><td colspan=\"5\" valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\"><bold>Fibronectin</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Sample</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Green Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Red Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Array</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Real-Time</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">9.9*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">24.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.50<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.68<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.9*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.1*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.08<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.65<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">4.3*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">10.2*<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.36<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.36<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">7A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">15.3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">29.7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.94<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.42<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Mean<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.22<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">5.53<hr/></td></tr><tr><td colspan=\"5\" valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\"><bold>Transforming growth factor beta 1</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><bold>Sample</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Green Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Red Signal</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Array</bold><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><bold>Real-Time</bold><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">8.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">11.5<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.30<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">2.43<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">10.3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">18.2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.77<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.64<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">14.9<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">12.8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.16<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1.46<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">7A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">12<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">11.3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-1.06<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">6.91<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Mean</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"/><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"/><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">0.21</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">3.11</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Nucleotide sequence of the TaqMan® probe and both forward and reverse primers used in real-time PCR reactions. The probe and primer sequences used in real-time PCR reactions and the microarray probe for individual rabbit genes were designed from the same contig.</p></table-wrap-foot>", "<table-wrap-foot><p>Genes whose expression was significantly (p less than or equal to 0.05) altered by glaucoma filtration surgery. Mean log2 difference is the level of gene expression in tissue surrounding the surgical site compared to similar tissue from control eye. The p value represents the level of significance in treatment means, the hit definition is the name of the most homologous gene, and E-value is the level of confidence that the matched name is incorrect.</p></table-wrap-foot>", "<table-wrap-foot><p>Mean log2 difference is the level of gene expression in tissue surrounding the surgical site compared to similar tissue from the nonsurgically treated control eye. The p-value represents the level of significance in treatment means, hit definition the name of the most homologous gene and E-value the level of confidence that the matched name is incorrect. Also listed are the GeneOntology Consortium terms for biological process associated with the most homologous gene.</p></table-wrap-foot>", "<table-wrap-foot><p>Change in gene expression in tissue surrounding the surgical site compared to similar tissue from the non-surgically treated control eye based on real-time PCR (real-time) and microarrays (Array). Fold change values are arithmetic and not logarithmic. Also listed are the corresponding processed red and green signal values from the microarrays. The asterisk denotes signal values that are indistinguishable from background values and the double asterisk indicates signal values that are inordinately high.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"mv-v13-164-f1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Since the beginning of the AIDS pandemic, opportunistic infections (O.I.'s) have been recognized as common complications of HIV infection. The spectrum of O.I.'s in the HIV infected subjects varies from one region to another [##REF##14556060##1##]. Of these, opportunistic protozoan infections are the most serious ones causing severe morbidity and mortality. Reports indicate that diarrhoea occurs in 30–60% of patients in developed countries and in about 90% of AIDS patients in Haiti and Africa [##REF##9093236##2##]. Given the similar background of poverty and malnutrition, the clinical picture of the disease in India parallels that of Africa.</p>", "<p>There have been reports regarding frequency of various pathogens causing diarrhea from different parts of India [##REF##12044119##3##, ####REF##8713519##4##, ##REF##12195048##5####12195048##5##]. However, there is paucity of data on correlations of CD4 levels and the etiology of diarrhea among the HIV patients of Eastern part of Uttar Pradesh, India. As most of the protozoan infections are treatable, it is important that an early and accurate diagnosis be made [##REF##2013618##6##].</p>", "<p>Thus, the present study was conducted to isolate and identify the protozoans causing diarrhoea in HIV patients so as to give an accurate diagnosis to avoid empirical treatment. An attempt was made to elucidate the associations between diarrhoea and CD4 counts and to study the effect of HAART as well as anti diarrhoeal treatments in these patients. For the first time an analysis of the seasonal variations in the occurrence of intestinal protozoan infections was also made as it holds epidemiological significance.</p>" ]

[ "<title>Methods</title>", "<p>This study was conducted from January 2006 to October 2007 in the Department of Microbiology and ART centre of S.S. Hospital, I.M.S., B.H.U., Varanasi, India. Being a tertiary care hospital, it caters to the patients from the neighbouring areas of U.P., M.P., Bihar, Jharkhand, Chattisgarh and Nepal. The study was a part of PhD work. For all assignments of this type institute ethical committee first review the protocol and approves it.</p>", "<title>Study cases</title>", "<p>The stool samples were collected from 366 HIV positive patients attending the Anti Retroviral Therapy (ART) centre, inpatient department and the ICTC (Integrated Counselling and Testing Centre). All these patients came with the presenting complaint of diarrhoea and were investigated for the enteric protozoan as and when they reported. These patients were mostly promiscuous by habit.</p>", "<p>The subjects who were HIV negative and without diarrhoea were not included in the study.</p>", "<title>Controls</title>", "<p>Stool samples from 200 cases were collected from the non-HIV positive family members of the patients who had diarrhoea and were obviously from similar environmental, social and economic background.</p>", "<p>A questionnaire was prepared to document the age, sex, route of transmission of infection and the demographic profile.</p>", "<p>We also studied the effect of HAART and anti diarrhoeal preparations in these patients, as they were potential culprits for HAART and anti diarrhoeal treatment. All the patients on HAART were being treated with Ziduvudine, Lamivudine, Efavirenz and Nevirapine. The antiparasitic drugs prescribed were ornidazole, metrogyl, nitazoxamide and albendazole.</p>", "<p>The patients were defined HIV seropositive if they tested positive for HIV infection by ELISA test (Micro Lisa – HIV, New Delhi, India) and a rapid test (HIV-Comb, New Delhi) [##UREF##0##7##].</p>", "<title>CD4 cell estimation</title>", "<p>The CD4 cell count estimations were done by FACS Count (Becton Dickinson, Singapore). Each time the patients provided their stool samples, their most recent CD4 count was recorded for analysis. The CD4 cell counts were taken post therapy also in order to assess the immune restoration for case management.</p>", "<p>The duration and episodes of recurrent diarrhoea were recorded thus classifying it as acute if it lasted for less than a month and chronic if it lasted for more than a month.</p>", "<title>Parasitological Examinations</title>", "<p>Stool samples were collected in wide-mouthed disposable containers and processed immediately. If there was delay in processing the samples, they were preserved at 4°C. The consistency of the stool samples was recorded. A small portion of the sample was emulsified in a drop of saline and lugol's iodine on the slide and observed under the microscope. The samples were concentrated by Modified formol ether technique [##UREF##1##8##]. Thereafter, the samples were stained by modified acid fast and modified safranin technique [##UREF##1##8##]. The smears were microscopically examined. Screening for <italic>Microsporidia </italic>spores was done with the help of Calcoflour White staining method and identified on the basis of their size [##UREF##1##8##].</p>", "<p>During parasitological analysis we correlated the occurrence of the parasites with the route of HIV transmission in the patients. The study was divided into three parts depending on the seasons viz. summer, rainfall and winter to correlate and study the possible role of seasonal variation in protozoan diarrhoea.</p>" ]

[ "<title>Results</title>", "<p>In the study we observed that there was a male preponderance 271/366, (74.0%) in the age group 31–40 years. Maximum patients (95%) belonged to Varanasi and the rural areas of Eastern UP.</p>", "<p>A total of 366 patients were screened and of these 134(36.6%) patients, showed immune restoration, which clinically indicated positive response to HAART.</p>", "<p>Of the 366 HIV positive patients with diarrhoea, 112 patients had episodes of acute and 254 had chronic diarrhoea. The details are given in a flowchart (Figure ##FIG##0##1##).</p>", "<p>The macroscopic examination of the stool samples revealed that positivity of finding a pathogen in the sample was four times more in the case of watery samples as compared to the semi formed samples. <italic>Cryptosporidium </italic>spp. (39.8%) was the most commonly found parasite in the HIV positive patients followed by <italic>Microsporidia </italic>spp. (26.7%). There were (25.1%) cases of mixed infections. Of the mixed infections 7.65% cases showed presence of helminths like Hookworm, <italic>H. nana </italic>and <italic>Trichuris trichiura </italic>along with the enteric coccidian. The remaining 17.45% were mixed infections of <italic>Cryptosporidium </italic>spp., <italic>Cyclospora </italic>spp. and <italic>Microsporidia </italic>spp. The samples taken from the controls showed a predominance of helminthic infestation with <italic>Ascaris lumbricoides </italic>(22.0%) leading the list followed by Hookworm (20%). The data is shown in the ensuing table. (Table ##TAB##0##1##).</p>", "<p>The study revealed that Giardiasis was more commonly found in cases that had a history of homosexual practice. Of all the cases examined, 25/366 (6.8%) had acquired the infection through homosexual route and 315/366 (86.0%) through heterosexual route. Sixteen out of 25 (64%) homosexual men had diarrhea due to <italic>Giardia </italic>spp.</p>", "<title>Correlation between CD4 count, type and duration of diarrhoea</title>", "<p>The CD4 levels were inversely proportional to the duration of diarrhoea. Patients with chronic diarrhoea had lower CD4 counts than those who had acute diarrhoea. Table ##TAB##1##2## shows the association between types of diarrhoea, parasites isolated and CD4 counts of 366 AIDS patients.</p>", "<p>The incidence of diarrhea showed a distinct seasonality. Diarrhoea caused by <italic>Cyclospora </italic>spp. was found to be associated with increasing ambient temperatures. Of all the <italic>Cyclospora </italic>spp. isolated 63.6% were obtained from the patients in the summer season. A direct correlation was found between increased rainfall and isolation of <italic>Cryptosporidium </italic>oocysts (76.0%) and <italic>Giardia </italic>cysts (56.7%) in the stool samples collected from the patients having diarrhoea. However, no seasonal variation was observed in the occurrence of diarrhoea caused by <italic>Microsporidia </italic>spp. and other parasites. (Table ##TAB##2##3##).</p>" ]

[ "<title>Discussion</title>", "<p>With the emergence of AIDS, parasitic diarrhea has gained significance, as it is one of the important causes of morbidity and mortality. The line of treatment being different for diverse parasites necessitates a definitive diagnosis and study of the etiological agents causing diarrhea, especially when it can be fatal in this vulnerable group of individuals.</p>", "<p>Our study showed prevalence of more males than females (p &lt; 0.05). Predominance of male cases may be due to their migration to the metropolitan cities in search of work. Staying away from the families for longer periods and males being promiscuous by habit resulted in them, acquiring HIV infection.</p>", "<p>The drugs of choice for diarrhoea for practising clinician were ornidazole, metrogyl, nitazoxamide and albendazole. However, Sengupta et al observed that paramomycin was effective against cryptosporidiosis [##REF##12044119##9##]. However, HAART added to the efficacy of the aforementioned anti protozoals and 36.6% patients were found to have a rebound in their CD4 cell counts. A study conducted by Guadalupe et al, showed that viral suppression is more effective in GALT (gut-associated lymphoid tissue) of patients with primary HIV infection than patients having chronic HIV infection during HAART [##REF##16873279##10##]. They found delay in restoration of gut mucosal immune system of patients with chronic infection as gut acts as a viral reservoir and keeps from eradicating the virus.</p>", "<p>In this study we came across patients ranging from initial to advanced stages of the disease. There were 69.3% patients with chronic and 30.6% patients with acute diarrhea (p &lt; 0.05). Recurrent episodes and presence of diarrhea even at higher CD4 levels can be attributed to reduced intestinal mucosal immunity [##REF##7489940##11##].</p>", "<p>The percentage of parasite isolation in our study was 78.5% in acute and 50.7% in chronic cases. A similar study conducted in the same set up by Attili et al showed lower isolation rates [##REF##16509972##12##]. This discordance in the results could be due to more than one technique used in this study for identification, which might have increased the sensitivity. Lower isolation rate of parasites in chronic cases was because most of the patients with chronic diarrhoea were on empirical antidiarrheal treatment. Moreover, in this country anti-diarrhoeal drugs are freely available across the counter in the drug store even without prescription.</p>", "<p>In this study it was observed that probability of finding a pathogen from watery and semi formed stools was four times greater as compared to formed stools [##REF##16509972##12##]. This can be attributed to greater shedding, more inflammatory response and greater virulence of the pathogens causing watery diarrhea.</p>", "<p>Samples collected from the controls coming from the same environmental background helped in tracing the source of infection. Parasite like <italic>Cryptosporidium </italic>spp. isolated from both the groups indicated water as the main source of infection, which highlighted poor sewage disposal practices and sewage spills. Presence of Hookworm indicated lack of sanitation and low socio-economic status of the cases coming from rural areas.</p>", "<p><italic>Cryptosporidium </italic>spp. (39.8%) was the most commonly isolated protozoan followed by <italic>Microsporidia </italic>spp. (26.7%). As compared to the controls, the observed incidence of these organisms in HIV patients was significantly higher (p &lt; 0.05). Another study conducted by Samantaray et al, also showed similar isolation rates in HIV patients [##UREF##2##13##] whereas, in a study of Southern India lesser number of <italic>Cryptosporidium </italic>spp. (9%) were isolated [##UREF##3##14##]. A study conducted in Mumbai showed the infection rate of <italic>Microsporidia </italic>spp. in HIV patients as 17.18% [##UREF##4##15##]. On the contrary our study detected 26.7% of <italic>Microsporidia </italic>spp. This increase in isolation rates could be due to the fact that the numbers of cases studied were much higher in our study as compared to the study of Siddhartha et al [##UREF##4##15##]. When compared to other studies of Southern India, isolation rate of <italic>Isospora belli </italic>(0.5%) was lower in our study [##UREF##3##14##]. This discrepancy in the findings maybe attributed to geographical variation. Calcoflour White staining technique, which is a screening method, identified the Microsporidia spp. However, its presence will be confirmed later by Chromotrope 2R staining method in order to avoid any false positive results, if any. The clinical picture and the microscopic examination of the dysenteric stools revealed haematophagous trophozoites suggesting that the <italic>Entamoeba </italic>spp. isolated were presumably that of <italic>Entamoeba histolytica</italic>. This screening method was adopted due to lack of facilities for isoenzyme analysis and other tests to differentiate it from <italic>Entamoeba dispar </italic>[##REF##15610563##16##]. Although the study was conducted to screen for the enteric protozoan, we came across 7.65% cases where helminths like Hookworm, <italic>H. nana </italic>and <italic>Trichuris trichiura </italic>co-existed with protozoa. These were probably flushed from the intestine because of diarrhoea or expelled after treatment. Therefore, we reported their presence as and when we came across the helminths during stool examination.</p>", "<p>It was found that 6.8% patients had acquired the HIV through homosexual route. <italic>Giardia </italic>spp. (64%) was present more in this group of people as compared to those who had heterosexual practice. Our findings are in accordance with Curry et al [##REF##2013618##6##].</p>", "<p>The maximum parasitic isolation was in the group of patients who had CD4 cell counts below 200 cells/μl and <italic>Cryptosporidium </italic>spp. was found to be the most commonly acquired protozoa causing chronic diarrhoea. The isolation rates decreased with the increase in the CD4 cell counts. This finding is in accordance with the study conducted by Attili et al. They also found an inverse correlation between CD4 counts and isolation rates of parasites from diarrhoea patients [##REF##16509972##12##].</p>", "<p>Key climatic variables, particularly humidity and temperatures have always had a relationship to waterborne diseases. The Milwaukee episode of 1993, which affected 403,000 people, is a commonly quoted waterborne Cryptosporidiosis outbreak [##REF##11359688##17##]. The advent of rainy season marks the beginning of many infectious diseases. In our work too, the maximum parasitic isolation was during rainfall, with <italic>Cryptosporidium </italic>spp. at the top of the list followed by <italic>Giardia </italic>spp. (p &lt; 0.05). We observed that isolation rate of <italic>Cyclospora </italic>spp. peaked during the summer and was significantly higher as compared to in the other seasons (p &lt; 0.05). This is because sporulation or maturation of the immature oocysts excreted in the faeces depends on warm temperatures [##REF##11359688##17##].</p>", "<title>Limitations of the study</title>", "<p>1. Immune restoration as detected by CD4 counts was clinically assessed.</p>", "<p>2. Calcoflour White staining for <italic>Microsporidia </italic>spp. is merely a screening method. However, the authors have plans to carry out identification techniques like Chromotrope 2R staining to confirm the results.</p>", "<p>3. The study was done as and when the symptoms of diarrhoea appeared and accordingly they were categorised season wise. It is difficult to establish the time period of initiation of infection.</p>" ]

[ "<title>Conclusion</title>", "<p>From this study it appears that the CD4 counts and prevalence of the protozoal infection in a particular geographic area should be considered before instituting empirical therapy to the AIDS patients attending the ART centre.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Protozoan infections are the most serious among all the superimposed infections in HIV patients and claim a number of lives every year. The line of treatment being different for diverse parasites necessitates a definitive diagnosis of the etiological agents to avoid empirical treatment. Thus, the present study has been aimed to elucidate the associations between diarrhoea and CD4 counts and to study the effect of HAART along with management of diarrhoea in HIV positive patients. This study is the first of its kind in this area where an attempt was made to correlate seasonal variation and intestinal protozoan infestations.</p>", "<title>Methods</title>", "<p>The study period was from January 2006 to October 2007 wherein stool samples were collected from 366 HIV positive patients with diarrhea attending the ART centre, inpatient department and ICTC of S.S. hospital, I.M.S., B.H.U., Varanasi. Simultaneously, CD4 counts were recorded to assess the status of HIV infection vis-à-vis parasitic infection. The identification of pathogens was done on the basis of direct microscopy and different staining techniques.</p>", "<title>Results</title>", "<p>Of the 366 patients, 112 had acute and 254 had chronic diarrhea. The percentages of intestinal protozoa detected were 78.5% in acute and 50.7% in chronic cases respectively. Immune restoration was observed in 36.6% patients after treatment on the basis of clinical observation and CD4 counts. In 39.8% of HIV positive cases <italic>Cryptosporidium </italic>spp. was detected followed by <italic>Microsporidia </italic>spp. (26.7%). The highest incidence of intestinal infection was in the rainy season. However, infection with <italic>Cyclospora </italic>spp. was at its peak in the summer. Patients with chronic diarrhea had lower CD4 cell counts. The maximum parasitic isolation was in the patients whose CD4 cell counts were below 200 cells/μl.</p>", "<title>Conclusion</title>", "<p>There was an inverse relation between the CD4 counts and duration of diarrhea. <italic>Cryptosporidium </italic>spp. was isolated maximum among all the parasites in the HIV patients. The highest incidence of infection was seen in the rainy season.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>All the authors read and approved the final manuscript. LT designed the study, performed the experimental work, conceived and drafted the manuscript. AKG provided the CD4 cell count data and helped to edit the manuscript, SS participated in coordination of the study and provided the clinical data and TMM supervised the study design, coordination of the study and helped to edit the manuscript.</p>", "<title>Pre-publication history</title>", "<p>The pre-publication history for this paper can be accessed here:</p>", "<p><ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-230X/8/36/prepub\"/></p>" ]

[ "<title>Acknowledgements</title>", "<p>The authors would like to acknowledge Prof. Gajendra Singh, Director, IMS, BHU for his support in carrying out the study.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Flowchart showing the symptoms, diarrhoeal episodes and percentage of pathogen positivity among 366 HIV positive patients.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Comparison of the parasites isolated from the stool samples of AIDS patients and normal controls.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Parasites isolated</td><td align=\"center\">Percentage of parasites isolated in HIV positive patients with diarrhoea. (n = 366)</td><td align=\"center\">Percentage of parasites isolated in HIV negative persons with diarrhoea. (n = 200)</td></tr></thead><tbody><tr><td align=\"left\"><italic>Cryptosporidium </italic>spp.</td><td align=\"center\">146(39.8%)</td><td align=\"center\">42(21.0%)</td></tr><tr><td align=\"left\"><italic>Microsporidia </italic>spp.</td><td align=\"center\">98(26.7%)</td><td align=\"center\">--</td></tr><tr><td align=\"left\"><italic>Cyclospora </italic>spp.</td><td align=\"center\">88(24.0%)</td><td align=\"center\">3(1.3%)</td></tr><tr><td align=\"left\"><italic>Giardia </italic>spp.</td><td align=\"center\">37(10.1%)</td><td align=\"center\">--</td></tr><tr><td align=\"left\"><italic>Entamoeba </italic>spp.</td><td align=\"center\">11(3.0%)</td><td align=\"center\">4(2.0%)</td></tr><tr><td align=\"left\"><italic>Isospora belli</italic></td><td align=\"center\">2(0.5%)</td><td align=\"center\">--</td></tr><tr><td align=\"left\">Hookworm</td><td align=\"center\">17(4.6%)</td><td align=\"center\">40(20.0%)</td></tr><tr><td align=\"left\"><italic>Trichuris trichiura</italic></td><td align=\"center\">9(2.4%)</td><td align=\"center\">--</td></tr><tr><td align=\"left\"><italic>Hymenolepsis nana</italic></td><td align=\"center\">2(0.5%)</td><td align=\"center\">6(3%)</td></tr><tr><td align=\"left\"><italic>Ascaris lumbricoides</italic></td><td align=\"center\">--</td><td align=\"center\">44(22.0%)</td></tr><tr><td align=\"left\">Mixed infections</td><td align=\"center\">92(25.1%)</td><td align=\"center\">--</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>The associations between type of diarrhoea, parasites isolated and CD4 counts of 366 AIDS patients.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Parasites isolated</td><td align=\"center\" colspan=\"2\">CD4 cells <break/>&lt; 200 cells/μl</td><td align=\"center\" colspan=\"2\">CD4 cells <break/>200–350 cells/μl</td><td align=\"center\" colspan=\"2\">CD4 cells <break/>350–500 cells/μl</td><td align=\"left\">Total</td></tr></thead><tbody><tr><td/><td align=\"left\">Acute cases</td><td align=\"left\">Chronic cases</td><td align=\"left\">Acute cases</td><td align=\"left\">Chronic cases</td><td align=\"left\">Acute cases</td><td align=\"left\">Chronic cases</td><td/></tr><tr><td/><td>n = 57</td><td>n = 179</td><td>n = 34</td><td>n = 42</td><td>n = 21</td><td>n = 33</td><td/></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\"><italic>Cryptosporidium </italic>spp.</td><td align=\"left\">38/57 (66.6%)</td><td align=\"left\">56/179 (31.2%)</td><td align=\"left\">11/34 (32.3%)</td><td align=\"left\">20/42 (47.6%)</td><td align=\"left\">8/21 (38%)</td><td align=\"left\">13/33 (39.3%)</td><td align=\"left\">146</td></tr><tr><td align=\"left\"><italic>Microsporidia </italic>spp.</td><td align=\"left\">15/57 (26.3%)</td><td align=\"left\">68/179 (37.9%)</td><td align=\"left\">5/34 (14.7%)</td><td align=\"left\">7/42 (16.6%)</td><td align=\"left\">3/21 (14.2%)</td><td align=\"left\">--</td><td align=\"left\">98</td></tr><tr><td align=\"left\"><italic>Cyclospora </italic>spp.</td><td align=\"left\">4/57 (7.0%)</td><td align=\"left\">43/179 (24.0%)</td><td align=\"left\">3/34 (8.8%)</td><td align=\"left\">28/42 (66.6%)</td><td align=\"left\">5/21 (23.8%)</td><td align=\"left\">5/33 (15.5%)</td><td align=\"left\">88</td></tr><tr><td align=\"left\"><italic>Giardia </italic>spp.</td><td align=\"left\">11/57 (19.3%)</td><td align=\"left\">7/179 (3.91%)</td><td align=\"left\">8/34 (23.5%)</td><td align=\"left\">4/42 (9.5%)</td><td align=\"left\">4/21 (19.5%)</td><td align=\"left\">3/33 (9.5%)</td><td align=\"left\">37</td></tr><tr><td align=\"left\"><italic>Entamoeba </italic>spp.</td><td align=\"left\">4/57 (7%)</td><td align=\"left\">1/179 (0.5%)</td><td align=\"left\">5/34 (14.7%)</td><td align=\"left\">--</td><td align=\"left\">1/21 (4.7%)</td><td align=\"left\">--</td><td align=\"left\">11</td></tr><tr><td align=\"left\"><italic>Isospora belli</italic></td><td align=\"left\">--</td><td align=\"left\">2/179 (1.1%)</td><td align=\"left\">--</td><td align=\"left\">--</td><td align=\"left\">--</td><td align=\"left\">--</td><td align=\"left\">2</td></tr><tr><td align=\"left\">Hookworm</td><td align=\"left\">1/57 (1.7%)</td><td align=\"left\">6/179 (3.3%)</td><td align=\"left\">--</td><td align=\"left\">5/42 (11.9%)</td><td align=\"left\">2/21 (9.5%)</td><td align=\"left\">3/33 (9%)</td><td align=\"left\">17</td></tr><tr><td align=\"left\"><italic>Trichuris trichiura</italic></td><td align=\"left\">--</td><td align=\"left\">3/179 (1.6%)</td><td align=\"left\">4/34 (11.7%)</td><td align=\"left\">--</td><td align=\"left\">2/21 (9.5%)</td><td align=\"left\">--</td><td align=\"left\">9</td></tr><tr><td align=\"left\"><italic>H. nana</italic></td><td align=\"left\">--</td><td align=\"left\">1/179 (0.5%)</td><td align=\"left\">--</td><td align=\"left\">1/42 (2.3%)</td><td align=\"left\">--</td><td align=\"left\">--</td><td align=\"left\">2</td></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\">Total</td><td align=\"left\">73</td><td align=\"left\">187</td><td align=\"left\">36</td><td align=\"left\">65</td><td align=\"left\">25</td><td align=\"left\">24</td><td align=\"left\">410</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Correlation between seasonal variation and parasitic diarrhoea</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Parasites</td><td align=\"center\" colspan=\"2\">Summer</td><td align=\"center\" colspan=\"2\">Rainfall</td><td align=\"center\" colspan=\"2\">Winter</td></tr><tr><td>Isolated</td><td align=\"center\" colspan=\"2\">(March–June)</td><td align=\"center\" colspan=\"2\">(July–October)</td><td align=\"center\" colspan=\"2\">(Nov–Feb)</td></tr><tr><td/><td colspan=\"6\"><hr/></td></tr><tr><td/><td align=\"center\">Cases</td><td align=\"center\">Controls</td><td align=\"center\">Cases</td><td align=\"center\">Controls</td><td align=\"center\">Cases</td><td align=\"left\">Controls</td></tr></thead><tbody><tr><td align=\"left\"><italic>Cryptosporidium </italic>spp.</td><td align=\"center\">19 (5.1%)</td><td align=\"center\">13 (6.5%)</td><td align=\"center\">111 (30.3%)</td><td align=\"center\">23 (11.5%)</td><td align=\"center\">16 (4.3%)</td><td align=\"left\">6 (3%)</td></tr><tr><td align=\"left\"><italic>Microsporidia </italic>spp.</td><td align=\"center\">32 (8.7%)</td><td align=\"center\">--</td><td align=\"center\">38 (10.3%)</td><td align=\"center\">--</td><td align=\"center\">28 (7.6%)</td><td align=\"left\">--</td></tr><tr><td align=\"left\"><italic>Cyclospora </italic>spp.</td><td align=\"center\">56 (15.3%)</td><td align=\"center\">2 (1%)</td><td align=\"center\">25 (6.8%)</td><td align=\"center\">1 (0.5%)</td><td align=\"center\">7 (1.1%)</td><td align=\"left\">--</td></tr><tr><td align=\"left\"><italic>Giardia </italic>spp.</td><td align=\"center\">9(2.4%)</td><td align=\"center\">--</td><td align=\"center\">21(5.7%)</td><td align=\"center\">--</td><td align=\"center\">7(1.1%)</td><td align=\"left\">--</td></tr><tr><td align=\"left\"><italic>Entamoeba </italic>spp.</td><td align=\"center\">5(1.3%)</td><td align=\"center\">1(0.5%)</td><td align=\"center\">4(1.0%)</td><td align=\"center\">3(1.5%)</td><td align=\"center\">2(0.5%)</td><td/></tr><tr><td align=\"left\"><italic>Isospora belli</italic></td><td align=\"center\">2(0.5%)</td><td align=\"center\">--</td><td align=\"center\">--</td><td align=\"center\">--</td><td align=\"center\">--</td><td align=\"left\">--</td></tr><tr><td align=\"left\">Hookworm</td><td align=\"center\">5 (1.3%)</td><td align=\"center\">10 (5%)</td><td align=\"center\">12 (3.2%)</td><td align=\"center\">27 (13.5%)</td><td align=\"center\">--</td><td align=\"left\">s3 (1.5%)</td></tr><tr><td align=\"left\"><italic>Trichuris trichiura</italic></td><td align=\"center\">4(1.0%)</td><td align=\"center\">--</td><td align=\"center\">3(0.8%)</td><td align=\"center\">--</td><td align=\"center\">2(0.5%)</td><td align=\"left\">--</td></tr><tr><td align=\"left\"><italic>H. nana</italic></td><td align=\"center\">--</td><td align=\"center\">--</td><td align=\"center\">2(0.5%)</td><td align=\"center\">4(2%)</td><td align=\"center\">--</td><td align=\"left\">2(1%)</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Figures in parentheses express the percentage (%) of parasites isolated.</p></table-wrap-foot>", "<table-wrap-foot><p>Figures in parentheses express the percentage (%) of parasites isolated.</p></table-wrap-foot>", "<table-wrap-foot><p>Figures in parentheses express the percentage (%) of parasites isolated.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-230X-8-36-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>About 16% of middle-aged men and 7% of women snore habitually [##REF##3385458##1##,##REF##12853524##2##]. They suffer from daytime sleepiness and run an increased risk of cardiovascular diseases [##REF##9623692##3##, ####REF##9485530##4##, ##REF##10573247##5##, ##REF##10676674##6####10676674##6##]. Snoring is a sign of increased upper airway resistance, usually due to a compromised upper airway during sleep. Snoring and daytime sleepiness are also symptoms of obstructive sleep apnea.</p>", "<p>Enlargement of the lymphatic system with hypertrophy of the tonsils and the tongue are common causes of a reduction in the size of the upper airways. Infants snore during respiratory infections, and school children snore and suffer from sleep apnea when their tonsils are enlarged [##REF##11694695##7##, ####REF##12801107##8##, ##REF##9082788##9##, ##REF##1523035##10####1523035##10##]. Mandibular retrognathia and narrowing of the lateral pharyngeal area reduce the upper airway size, with snoring and sleep apnea as a result in adults [##REF##17503232##11##, ####REF##7962567##12##, ##REF##12746251##13####12746251##13##]. Obesity, age, smoking and chronic bronchitis are other risk factors for snoring among adults [##REF##15242843##14##,##REF##11392586##15##].</p>", "<p>There has been great attention to research focusing on the early life origins of adult disease during the last two decades [##REF##2252919##16##]. Increasing evidence show that early life environment may influence health throughout life. Risk factors for adult cardiovascular diseases and diabetes mellitus for example, include maternal smoking, low birth weight and socio-economic class [##REF##2570282##17##, ####REF##11032153##18##, ##REF##12876166##19####12876166##19##]. Exposure to pets and growing up on a farm appear to be protective for allergy, while severe respiratory infections in childhood and living in a large family increase the risk of asthma [##REF##15131565##20##, ####REF##11590373##21##, ##REF##11597666##22##, ##REF##12403876##23####12403876##23##].</p>", "<p>To our knowledge there are no studies investigating whether the susceptibility to adult snoring and sleep apnea could be partly determined by early life environment. In the present paper we aimed to investigate whether environmental factors in childhood are associated with snoring later in life.</p>" ]

[ "<title>Methods</title>", "<p>A postal questionnaire was sent to 21,802 men and women aged 25–54 years in 1999–2001, with two reminders to non-responders [##REF##15242843##14##]. Altogether 16,190 subjects (74%) responded. The responders were more likely being women (52.9 vs. 47.7%, p &lt; 0.001) and slightly older (40 ± 7 vs. 39 ± 7 years, p &lt; 0.001) than the non-responder. This sample was enrolled in the Respiratory Health in Northern Europe (RHINE) survey, which is a follow-up of subjects who participated in the European Community Respiratory Health Survey (ECHRS) in 1990–1994 [##REF##8050554##24##]. The subjects were randomly selected from population registers in Reykjavik in Iceland, Bergen in Norway, Umeå, Uppsala and Göteborg in Sweden, Aarhus in Denmark and Tartu in Estonia. Ethics committees in Aarhus, Bergen, Göteborg, Reykjavik, Tartu, Umeå and Uppsala approved the study protocol. All the subjects gave their written informed consent for participation. The full protocols are available on the internet [##UREF##0##25##,##UREF##1##26##].</p>", "<title>Questions on snoring and daytime sleepiness</title>", "<p>Loud and disturbing snoring, and daytime sleepiness during the last few months was assessed using a five-point scale according to the Basic Nordic Sleep Questionnaire: never, less than once a week, 1–2 days or nights a week, 3–5 days or nights a week and almost every day or night [##REF##10607192##27##]. Habitual snoring was defined as loud and disturbing snoring at least three nights a week. A total of 15,556 subjects answered the questions on snoring. Daytime sleepiness was defined as feeling sleepy during the daytime at least one to two days a week</p>", "<title>Questions on childhood environment</title>", "<p>The following questions concerning early life were included in the questionnaire:</p>", "<p>\"How old was your mother when you were born?\"</p>", "<p>\"Was there any pet in your home at the time when you were born?\" Alternative responses: Dog, cat, other pet.</p>", "<p>\"Was there any pet in your home when you were a child?\" Alternative responses: Dog, cat, other pet.</p>", "<p>\"Were you hospitalized for a respiratory infection at any time before the age of 2 years?\" Alternative responses: yes or no.</p>", "<p>\"Did you suffer from recurrent otitis in childhood?\" Alternative responses: yes or no.</p>", "<p>\"What education did your mother have?\" \"What education did your father have?\" Alternative responses: primary school, high school, university, other.</p>", "<p>\"How many people lived in your home, when you were five years old?\"</p>", "<p>In addition, questions on maternal smoking history during pregnancy and when the subjects were younger than 5 years of age were included in the questionnaire used in Bergen, Norway.</p>", "<title>Status in adulthood</title>", "<p>Asthma was defined as answering yes to the question: \"Have you had an attack of asthma in the last 12 months?\" and/or \"Are you currently taking any medicine (including inhalers, aerosols or tablets) for asthma?\"</p>", "<p>Allergic rhinitis was defined as answering yes to the question: \"Do you have any nasal allergies including hay fever?\"</p>", "<p>Chronic bronchitis was defined as a negative answer to: \"Have you ever had asthma?\" and positive answers to all the following three questions: \"Do you usually bring up phlegm or do you have phlegm, which you have difficulty bringing up?\", \"Do you bring up phlegm in this way almost every day for at least three months every year?\" and \"Have you had episodes of this kind for at least two years in a row?\".</p>", "<p>A pack-year of smoking corresponded to 20 cigarettes a day/year. Body mass index (BMI) in kg/m²was calculated from self-reported height and weight. Type of dwelling was used as a proxy for socio-economic status.</p>", "<title>Statistical methods</title>", "<p>Chi-square test was used to test for differences between proportions and the Mann-Whitney U-test was used to test for differences between continuous variables. The associations between early life factors and adult snoring were analyzed with multivariable logistic regression in one model, adjusting for childhood environmental factors and possible confounders in adulthood such as chronic bronchitis, asthma, allergic rhinitis, smoking, BMI, age, gender, type of dwelling and centre. In these models we included variables that were related to habitual snoring with a p-value &lt; 0.1 in univariate analysis.</p>", "<p>Interactions by gender and BMI were tested for early life factors significantly associated with snoring (p &lt; 0.05) in the adjusted model. Potential heterogeneity between centers was addressed by meta-analysis [##REF##3802833##28##]. Data are presented as odds ratio (OR) and 95% confidence interval (CI). The adjusted proportion of snoring that could be explained by different risk factors was calculated as the population attributable fraction (PAF). All analysis were performed using Stata 8.0.</p>" ]

[ "<title>Results</title>", "<p>The characteristics of the study population according to centre are presented in Table ##TAB##0##1##. Habitual snoring was reported by 2,851 subjects (18%). Habitual snorers were more often men, more obese, older, had a higher prevalence of asthma and chronic bronchitis and had smoked more than non-snorers (Table ##TAB##1##2##). The prevalence of exposure to different environmental factors in early life among snorers and non-snorers is given in Table ##TAB##2##3##. Snoring subjects came from homes with lower parental education, larger household size and more pets when newborn, they had more often been hospitalized for respiratory infection before the age of 2 years, and they had more often had recurrent otitis in childhood (Table ##TAB##2##3##).</p>", "<p>The associations of childhood factors with adult snoring when adjusting for potential confounding factors are presented in table ##TAB##3##4##. Being hospitalized for a respiratory infection before the age of two years, suffering from recurrent otitis as a child, being born by a younger mother, growing up in a large family and being exposed to a dog at home as a newborn were significantly associated with adult snoring, independent of childhood exposure to cats or other pets, parents' education, adult chronic bronchitis, asthma, allergic rhinitis, active smoking, BMI, age, gender, current type of dwelling and centre (Table ##TAB##3##4##). The same childhood factors except family size were also associated with snoring accompanied by daytime sleepiness (Table ##TAB##3##4##).</p>", "<p>Stratifying by BMI, an association between hospitalization for respiratory infection before age 2 years and adult snoring was only observed among overweight subjects (6,401 subjects with BMI &gt; 25 kg/m²). Among the overweight, early hospitalization for respiratory infection was associated with adult snoring with an OR = 1.67; 95% CI 1.26–2.20, while among subjects with normal weight the OR was 0.82; 95% CI 0.55–1.23. This difference in associations of childhood respiratory infections with adult shoring according to BMI was highly significant (p _interaction= 0.006). There was no significant interaction by gender, and there was no heterogeneity between centers.</p>", "<p>The adjusted population attributable fraction for snoring of having been exposed to a dog when newborn was 3.4% while the corresponding figures were 2.5% for otitis, 1.4% for growing up in family of more than 5 persons and 0.7% for being hospitalized for a respiratory disease before the age of 2. Of the adult risk factors, the adjusted population attributable fraction for snoring was 3.0%, for rhinitis 4.5%, for chronic bronchitis 9.1%, for obesity (BMI ≥ 30 kg/m²) 14.1% and for ever smoking (Figure ##FIG##0##1##).</p>", "<p>Eighteen percent of subjects from Bergen reported that their mother had smoked during pregnancy and 33% reported that she had smoked when they were younger than 5 years of age. Neither maternal smoking during pregnancy (adjusted OR = 1.06; 95% CI 0.63–1.79), nor maternal smoking in childhood (adjusted OR = 1.05; 95% CI 0.66–1.67) were related to adult snoring in this sub-sample (n = 2,506).</p>" ]

[ "<title>Discussion</title>", "<p>Being hospitalized for a respiratory infection before the age of two years, having had recurrent otitis in childhood, having been exposed to a dog as a newborn, having grown up in a large family were associated with habitual snoring later in life. These findings were independent of other childhood exposures and adult risk factors for snoring. When considering habitual snoring with daytime sleepiness combined, the same childhood factors were associated with increased adult risk. Our observations were demonstrated in a large population study in Northern Europe and were consistent across the seven centers. The findings are new and indicate that a predisposition for adult snoring and possibly also for obstructive sleep apnea could be established early in life.</p>", "<p>Obesity is a major cause of snoring and sleep apnea. It is, however, important to increase knowledge about other preventable causes of habitual snoring, since a large number of snorers suffer from daytime sleepiness and an increased risk of cardiovascular disease and even early death [##REF##3385458##1##,##REF##9623692##3##, ####REF##9485530##4##, ##REF##10573247##5##, ##REF##10676674##6####10676674##6##,##REF##9828847##29##]. The present study showed that early life environment may be of importance for snoring later in life. Further knowledge of this subject could contribute to primary prevention of adult snoring.</p>", "<p>Our results indicate an association between early life environment and snoring later in life. It is, however, only possible to speculate about causal relationships and mechanisms based on the present data. Previous studies have shown that children with large tonsils develop retrognathia and posteriorly inclined mandibles as a result of changes in tongue posture and mouth breathing [##REF##2401330##30##,##REF##11883823##31##]. Studies on growing monkeys has also shown that induced oral respiration leads to a lowering of the chin, a steeper mandibular plane angle, and an increase in gonial angle as compared with control animals [##REF##6961782##32##]. It is possible that subjects reporting otitis, severe respiratory infections or living in a large family in childhood more frequently had infections in the upper airways with hypertrophy of the tonsils and subsequent narrowing of the adult upper airways. Further, endotoxins are proinflammatory cell wall components from gram-negative bacteria and airborne endotoxins that are prevalent especially in homes with dogs [##REF##11564624##33##]. We hypothesize that infections in childhood and exposure to airborne endotoxins in infancy stimulate the lymphatic system with subsequent enlargement of the tonsils. Remaining large tonsils or retrognathia due to large tonsils in childhood may compromise the upper airways, and could explain the associations between early life factors and snoring in adulthood as observed in this study. Unfortunately, we do not have information about history of tonsillectomy and/or adenoidectomy which might have been valuable for further understanding of these mechanisms.</p>", "<p>A severe infection in childhood was only related to snoring later in life among overweight subjects, indicating that subjects who suffered from severe infections in childhood run a higher risk of habitual snoring if they become obese later in life. It is difficult to speculate on this relationship, but it seems reasonable that obesity, which is a common cause of snoring, increase potential negative consequences related to severe airways infections early in life.</p>", "<p>The strengths of the present study are the large number of subjects, the multi-centre structure, the detailed analysis of childhood environmental factors and the high response rate to the questionnaire. The response rate analysis showed that men and younger subjects were slightly underrepresented. As the absolute differences between non-responders and responders were relatively small and we do not think that this has affected our results substantially.</p>", "<p>The present analysis is limited by recall bias in assessments of childhood environment based on information in adulthood. A recent analysis, based on a multi-cultural study of childhood pet keeping, indicated that adults report important childhood events like having a dog or cat very consistently [##UREF##2##34##]. We therefore assume that the reporting of pets and household size in this study is fairly reliable. Reports of childhood hospitalization could possibly be biased with regard to childhood social class and subsequent respiratory infection; however, the analyses were adjusted for parental education. It is unlikely that the misclassification of any of these childhood factors would be differential with regard to adult snoring and we believe that the misclassification in this study is non-differential and may have attenuated the effects.</p>", "<p>Other limitations include residual confounding from variables not included in the present study, such as current pet keeping, current household size, seasonal allergies, mouth breathing in sleep and childhood snoring. It is for example possible that persons exposed to pets during childhood are more likely to keep pets as adults, and that the association with current snoring and dogs is explained by current exposure rather than by previous exposure.</p>", "<p>Snoring was based on subjective reports, which is a common limitation in epidemiological studies. Subjective reports are, however, the most commonly used instrument for measuring snoring, in part because of the technical problems involved with microphone recordings as well as the ability of subjective reports to give an average of the subject's degree of snoring, whereas the result of a single night's recording may be misleading. Objective recordings using microphones correlate well with subjective snoring in young adults [##REF##9551753##35##].</p>" ]

[ "<title>Conclusion</title>", "<p>The predisposition for adult snoring and possibly also for obstructive sleep apnea may be partly established early in life. Having had severe airway infections or recurrent otitis in childhood, being exposure to a dog as a newborn, and growing up in a large family appear as possible risk factors for snoring in adulthood. We speculate that these factors may enhance inflammatory processes and thereby alter upper airway anatomy early in life causing an increased susceptibility for adult snoring. The presented findings are new and suggest that further knowledge about early life environment might contribute to primary prevention of snoring.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>To our knowledge, no studies of the possible association of early life environment with snoring in adulthood have been published. We aimed to investigate whether early life environment is associated with snoring later in life.</p>", "<title>Methods</title>", "<p>A questionnaire including snoring frequency in adulthood and environmental factors in early life was obtained from 16,190 randomly selected men and women, aged 25–54 years, in Sweden, Norway, Iceland, Denmark and Estonia (response rate 74%).</p>", "<title>Results</title>", "<p>A total of 15,556 subjects answered the questions on snoring. Habitual snoring, defined as loud and disturbing snoring at least 3 nights a week, was reported by 18%. Being hospitalized for a respiratory infection before the age of two years (adjusted odds ratio (OR) = 1.27; 95% confidence interval (CI) 1.01–1.59), suffering from recurrent otitis as a child (OR = 1.18; 95%CI 1.05–1.33), growing up in a large family (OR = 1.04; 95%CI 1.002–1.07) and being exposed to a dog at home as a newborn (OR = 1.26; 95%CI 1.12–1.42) were independently related to snoring later in life and independent of a number of possible confounders in adulthood. The same childhood environmental factors except household size were also related with snoring and daytime sleepiness combined.</p>", "<title>Conclusion</title>", "<p>The predisposition for adult snoring may be partly established early in life. Having had severe airway infections or recurrent otitis in childhood, being exposed to a dog as a newborn and growing up in a large family are environmental factors associated with snoring in adulthood.</p>" ]

[ "<title>Abbreviations</title>", "<p>BMI: Body mass index; CI: Confidence interval; OR: Odds ratio.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>KAF participated in the design, analysis, interpretation and drafted the manuscript. CJ participated in the design and coordination of the study, acquisition of data and to critically draft the manuscript. CS conceived the study, performed the analysis and helped to draft the manuscript. TG, AG, MG, BL, ENL, EN, LN, EO and KT participated in the design, acquisition of data and to critically draft the manuscript. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>The following scientists in the RHINE study group are acknowledged for helpful contributions: Balder B, Bodman G-M, Boman G, Björnsson E, Dahlman-Höglund A, Farkhooy A, Forsberg B, Gislason D, Hellgren J, Jensen EJ, Jõgi R, Lillienberg L, Lundbäck B, Norbäck D, Olin A-C, Real F, Tunsäter A, Wenzel Larsen T and Wieslander G.</p>", "<p>The study was supported financially by the Swedish Heart and Lung Foundation, the Vårdal Foundation for Health Care Science and Allergy Research, the Swedish Asthma and Allergy Association, the Icelandic Research Council, the Norwegian Research Council, project 135773/330, the Norwegian Asthma and Allergy Association, the Danish Lung Association and the Estonian Science Foundation, grant no 4350. Karl A Franklin and Christer Janson are recipients of science awards from the Swedish Heart and Lung Foundation.</p>", "<p>The funding organizations had no role in study design, in the collection of data, analysis and interpretation of data, in writing the manuscript or in the decision to submit the manuscript for publication.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Adjusted population attributable fraction for childhood (black bars) and adult risk factors (white bars) for snoring.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Characteristics of the population.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\">Reykjavik <break/>(n = 1,969)</td><td align=\"center\">Bergen<break/> (n = 2,506)</td><td align=\"center\">Umeå <break/>(n = 2,640)</td><td align=\"center\">Uppsala <break/>(n = 2,572)</td><td align=\"center\">Göteborg <break/>(n = 2,188)</td><td align=\"center\">Aarhus <break/>(n = 2,607)</td><td align=\"center\">Tartu <break/>(n = 1,708)</td><td align=\"center\">All subjects <break/>(n = 16,190)</td></tr></thead><tbody><tr><td align=\"left\">Women (%)</td><td align=\"center\">54.6</td><td align=\"center\">51.9</td><td align=\"center\">51.5</td><td align=\"center\">52.5</td><td align=\"center\">54.2</td><td align=\"center\">52.2</td><td align=\"center\">56.1</td><td align=\"center\">52.9</td></tr><tr><td align=\"left\">Age (years)</td><td align=\"center\">41 ± 7</td><td align=\"center\">41 ± 7</td><td align=\"center\">41 ± 7</td><td align=\"center\">40 ± 7</td><td align=\"center\">40 ± 7</td><td align=\"center\">39 ± 7</td><td align=\"center\">36 ± 7</td><td align=\"center\">40 ± 7</td></tr><tr><td align=\"left\">BMI (kg/m²)</td><td align=\"center\">25.3 ± 4.0</td><td align=\"center\">24.7 ± 4.0</td><td align=\"center\">25.2 ± 3.9</td><td align=\"center\">24.6 ± 3.9</td><td align=\"center\">25.0 ± 3.9</td><td align=\"center\">24.3 ± 4.2</td><td align=\"center\">24.2 ± 4.2</td><td align=\"center\">24.8 ± 4.1</td></tr><tr><td align=\"left\">Habitual snoring (%)</td><td align=\"center\">20.6</td><td align=\"center\">16.9</td><td align=\"center\">20.7</td><td align=\"center\">18.6</td><td align=\"center\">20.4</td><td align=\"center\">17.7</td><td align=\"center\">12.0</td><td align=\"center\">18.3</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Characteristics of habitual and non-habitual snorers in adulthood.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\">Snorers <break/>n = 2,851</td><td align=\"center\">Non-snorers <break/>n = 12,705</td><td align=\"center\">p-value</td></tr></thead><tbody><tr><td align=\"left\">Women n (%)</td><td align=\"center\">956 (34)</td><td align=\"center\">7,267 (57)</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Age (years)</td><td align=\"center\">42 ± 7</td><td align=\"center\">39 ± 7</td><td align=\"center\">&lt;0.001</td></tr><tr><td align=\"left\">BMI (kg/m²)</td><td align=\"center\">27 ± 5</td><td align=\"center\">24 ± 4</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Smoking (pack-years)</td><td align=\"center\">7.1 ± 11.5</td><td align=\"center\">3.8 ± 7.7</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Asthma n (%)</td><td align=\"center\">256 (9.0)</td><td align=\"center\">774 (6.1)</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Allergic rhinitis n (%)</td><td align=\"center\">686 (25)</td><td align=\"center\">2,875 (23)</td><td align=\"center\">0.07</td></tr><tr><td align=\"left\">Chronic bronchitis n (%)</td><td align=\"center\">299 (11)</td><td align=\"center\">531 (4.2)</td><td align=\"center\">&lt; 0.001</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Early life characteristics according to adult habitual snoring</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\">Snorers <break/>n = 2,851</td><td align=\"center\">Non-snorers <break/>n = 12,705</td><td align=\"center\">p-value</td></tr></thead><tbody><tr><td align=\"left\">Hospitalized for respiratory infection before 2 years of age n (%)</td><td align=\"center\">132 (4.7)</td><td align=\"center\">479 (3.8)</td><td align=\"center\">0.03</td></tr><tr><td align=\"left\">Otitis in childhood n (%)</td><td align=\"center\">617 (22)</td><td align=\"center\">2,490 (20)</td><td align=\"center\">0.006</td></tr><tr><td align=\"left\">Dog at home when newborn n (%)</td><td align=\"center\">617 (22)</td><td align=\"center\">2,173 (17)</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Dog at home in childhood n (%)</td><td align=\"center\">1,156 (41)</td><td align=\"center\">4,890 (39)</td><td align=\"center\">0.033</td></tr><tr><td align=\"left\">Cat at home when newborn n (%)</td><td align=\"center\">565 (24)</td><td align=\"center\">2,095 (19)</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Cat at home in childhood n (%)</td><td align=\"center\">1,036 (43)</td><td align=\"center\">4,690 (43)</td><td align=\"center\">0.779</td></tr><tr><td align=\"left\">Other pet at home when newborn n (%)</td><td align=\"center\">164 (6.8)</td><td align=\"center\">533 (4.9)</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Other pet at home in childhood n (%)</td><td align=\"center\">356 (15)</td><td align=\"center\">1,557 (14)</td><td align=\"center\">0.435</td></tr><tr><td align=\"left\">Household size &gt; 5 n (%)</td><td align=\"center\">673 (24)</td><td align=\"center\">2,529 (20)</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Mother's age at delivery (years)</td><td align=\"center\">27.8 ± 6.3</td><td align=\"center\">28.1 ± 6.1</td><td align=\"center\">0.096</td></tr><tr><td align=\"left\">Mother university educated n (%)</td><td align=\"center\">205 (7.4)</td><td align=\"center\">1,376 (11)</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\">Father university educated n (%)</td><td align=\"center\">373 (14)</td><td align=\"center\">2,146 (17)</td><td align=\"center\">&lt; 0.001</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Adjusted odds ratios* for the associations between early life factors and adult habitual snoring, and habitual snoring with daytime sleepiness combined (n = 13,484).</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"2\">Habitual snoring</td><td align=\"center\" colspan=\"2\">Habitual snoring with daytime sleepiness</td></tr><tr><td/><td align=\"left\">Unadjusted OR<break/> (95% CI)</td><td align=\"left\">Adjusted OR <break/>(95% CI)</td><td align=\"left\">Unadjusted OR <break/>(95% CI)</td><td align=\"left\">Adjusted OR<break/> (95% CI)</td></tr></thead><tbody><tr><td align=\"left\">Hospitalized for respiratory infection before 2 years of age</td><td align=\"left\">1.27 (1.04–1.54)</td><td align=\"left\">1.27 (1.01–1.59)</td><td align=\"left\">1.46 (1.16–1.83)</td><td align=\"left\">1.40 (1.07–1.81)</td></tr><tr><td align=\"left\">Otitis in childhood</td><td align=\"left\">1.15 (1.04–1.27)</td><td align=\"left\">1.18 (1.05–1.33)</td><td align=\"left\">1.37 (1.22–1.56)</td><td align=\"left\">1.34 (1.18–1.54)</td></tr><tr><td align=\"left\">Dog at home when newborn</td><td align=\"left\">1.34 (1.21–1.48)</td><td align=\"left\">1.26 (1.12–1.42)</td><td align=\"left\">1.38 (1.22–1.56)</td><td align=\"left\">1.35 (1.17–1.54)</td></tr><tr><td align=\"left\">Cat at home when newborn</td><td align=\"left\">1.30 (1.17–1.44)</td><td align=\"left\">0.99 (0.86–1.14)</td><td align=\"left\">1.29 (1.13–1.47)</td><td align=\"left\">1.01 (0.85–1.20)</td></tr><tr><td align=\"left\">Other pet at home when newborn</td><td align=\"left\">1.43 (1.19–1.71)</td><td align=\"left\">1.13 (0.90–1.42)</td><td align=\"left\">1.51 (1.21–1.87)</td><td align=\"left\">1.20 (0.92–1.56)</td></tr><tr><td align=\"left\">Household size (one more person)</td><td align=\"left\">1.07 (1.04–1.10)</td><td align=\"left\">1.04 (1.002–1.07)</td><td align=\"left\">1.05 (1.02–1.09)</td><td align=\"left\">1.03 (0.99–1.07)</td></tr><tr><td align=\"left\">Mother's age at delivery (per 5 years' increase)</td><td align=\"left\">0.98 (0.95–1.01)</td><td align=\"left\">0.96 (0.93–0.999)</td><td align=\"left\">0.96 (0.92–0.997)</td><td align=\"left\">0.95 (0.90–0.99)</td></tr><tr><td align=\"left\">Asthma</td><td align=\"left\">1.52 (1.31–1.77)</td><td align=\"left\">1.14 (0.94–1.37)</td><td align=\"left\">1.85 (1.56–2.19)</td><td align=\"left\">1.28 (1.03–1.57)</td></tr><tr><td align=\"left\">Allergic rhinitis</td><td align=\"left\">1.09 (0.99–1.20)</td><td align=\"left\">1.22 (1.08–1.36)</td><td align=\"left\">1.31 (1.16–1.47)</td><td align=\"left\">1.38 (1.20–1.58)</td></tr><tr><td align=\"left\">Chronic bronchitis</td><td align=\"left\">2.72 (2.34–3.15)</td><td align=\"left\">2.33 (1.95–2.80)</td><td align=\"left\">3.34 (2.84–3.93)</td><td align=\"left\">2.76 (2.27–3.35)</td></tr><tr><td align=\"left\">Smoking (per 5 pack years increase)</td><td align=\"left\">1.20 (1.17–1.22)</td><td align=\"left\">1.15 (1.12–1.18)</td><td align=\"left\">1.18 (1.15–1.21)</td><td align=\"left\">1.13 (1.10–1.17)</td></tr><tr><td align=\"left\">Body mass index (per 5 kg/m²increase)</td><td align=\"left\">2.04 (1.94–2.15)</td><td align=\"left\">1.82 (1.72–1.93)</td><td align=\"left\">1.80 (1.70–1.90)</td><td align=\"left\">1.69 (1.58–1.80)</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>BMI = body mass index. Means are expressed ± SD.</p></table-wrap-foot>", "<table-wrap-foot><p>BMI = body mass index. Means are expressed ± SD.</p></table-wrap-foot>", "<table-wrap-foot><p>Means are expressed ± SD.</p></table-wrap-foot>", "<table-wrap-foot><p>CI = confidence interval, *The logistic regression also included parents' education age, gender, type of dwelling and centre.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1465-9921-9-63-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Chronic Fatigue Syndrome (CFS) is characterized by persistent or relapsing unexplained fatigue that lasts for at least six months and results in substantial reduction in previous levels of daily functioning [##UREF##0##1##]. Causes of CFS have not been found and most patients do not recover spontaneously [##UREF##1##2##]. Based on the CDC-94 criteria, CFS prevalence figures of 112 and 420 per 100.000 were found [##REF##10625135##3##,##REF##12860574##4##].</p>", "<p>Cognitive behavior therapy (CBT) has proven to be an effective treatment for CFS [##REF##11560542##5##,##REF##12562565##6##]. Since the treatment of CFS with CBT has been available only in a few specialized university medical centers in The Netherlands, just a small minority of CFS patients can benefit from it. Nationwide implementation is needed to realize access to CBT treatment for all CFS patients. However, when decision makers have to judge whether such implementation is worthwhile and should be paid for, they need information about its costs and benefits for individual patients, the healthcare system and society.</p>", "<p>The number of cost effectiveness analyses (CEA) of CBT for CFS and chronic fatigue (CF) are few compared to clinical evaluations. One study performed a cost consequence analysis of CBT for CF in general practice compared to regular counseling by a GP. It reported that counselling was a less costly intervention than CBT, and that both interventions led to reductions in fatigue. But no overall cost-effectiveness advantage was found for either form of therapy [##REF##11271867##7##]. Another study, concerning a CEA of CBT for CF, [##REF##15554570##8##] found similar cost effectiveness for CBT and graded exercise for CF. It also reported a high probability that these therapies are cost-effective compared to usual care. A third study reported a CEA of CBT for CFS and found, although with some statistical uncertainties, that regarding a time horizon of 14 months, total costs to society were lower for (ex) CFS patients that had followed CBT treatment than for those who had received usual care or guided support groups [##UREF##2##9##]. Taken together these studies indicate that CBT for CFS or CF might be cost effective for society compared to usual care.</p>", "<p>Until now nothing is known about the costs and efficiency of implementing CBT for CFS in a clinical practice setting. It might be possible that the efficiency of CBT for CFS reduces if the implementation costs are high or if the treatment effectiveness reduces. The present study therefore evaluated the broader so-called <italic>policy </italic>costs and effects of a pilot implementation project in which CBT for CFS was made available in a mental health center (MHC). In a policy study all extra costs of implementing the treatment (like training therapists, informing GPs, organizing and meetings) are being included as fixed costs in the analysis, in addition to the costs and effects of just performing the treatment [##UREF##3##10##,##REF##11743840##11##].</p>", "<p>The MHC of this study was a regional middle-sized institution located in the East of The Netherlands, covering mostly rural and some urbanized areas. It had locations in four separate sub-regions and the CBT for CFS treatment was offered at two of them. This MHC was the main provider of mental health care in this area, offering outpatient and inpatient services for the full range of problems and patients.</p>" ]

[ "<title>Methods</title>", "<title>Design</title>", "<p>The evaluation was a prospective, non-controlled before and after comparison in a MHC with an observation period of 8 months.</p>", "<title>Implementation interventions</title>", "<p>The implementation program contained four major interventions. First, six behavior therapists who were working in the concerning MHC were trained at the Nijmegen Expertcenter for Chronic Fatigue. They were selected on bases of their prior education in CBT and on their willingness and possibility to participate in this implementation project. None of these therapists had previous experience with CFS patients. Their number of years working as a behavioral therapist varied from two to 13 years. Second, because GPs in the region were not familiar with this new treatment setting for CFS, announcements were made in the media and information brochures were distributed to GPs. GPs could also order copies of these brochures for their waiting rooms. Third, informational interventions were performed that were directed at the patient population. These consisted of several media announcements and distribution of patient brochures. Fourth, employees of the mental health care institution were informed and, if applicable, settled into the project.</p>", "<title>Patients and treatment procedure</title>", "<p>Patients who attended the treatment were all diagnosed as CFS and referred to the MHC by their GP or a medical specialist. Inclusion criteria were as follows: a GPs diagnosis of CFS (based on the CDC-94 criteria), not enrolled in a new claim for disability-related benefits, and 18 years or older. After the first session the patient had to fill in several fatigue related paper and pencil questionnaires. At 8 months follow up, when treatment was finished, the questionnaire had to be filled in again. Before starting this study it was judged by the Nijmegen Medical Hospital Ethical Commission, who indicated no need for informed consent.</p>", "<p>To measure fatigue we used the Checklist Individual Strength (CIS20), which is a self-report measure on a 7-point Likert scale for fatigue severity over the last two weeks. The CIS has good reliability (Cronbach's alpha varying from 0,83 to 0,92) and discriminative validity [##REF##7965927##12##]. Physical functioning in daily life was measured with the 'physical functioning' subscale of the SF-36 [##REF##3393032##13##]. This subscale is a validated 10-item scale with a score varying from 0 (maximum of limitations) to 100 (no limitations). The Euroqol-5d was used to measure QALYs [##REF##9366889##14##].</p>", "<p>In some instances this questionnaire results contradicted the diagnosis of CFS. For example, when a psychiatric co-morbidity was found that could explain the severe fatigue. In such occasions treatment was not started and the patient was referred to another treatment program in the organization.</p>", "<p>The CBT treatment protocol prescribes 16 sessions in a period of 8 months [##UREF##4##15##]. In this treatment, first the model of psychological and behavioral perpetuating factors of fatigue is explained to the patient. Then the patient formulates his or her goals for therapy. Afterwards the patient starts a structured graded activity program beginning with some daily minutes of walking or bicycling, which is tailored to their base line daily activity level. Subsequently, dysfunctional fatigue related cognitions are being challenged to diminish somatic attributions of fatigue, to improve a sense of control over symptoms and to facilitate behavior change. Finally a plan for work rehabilitation is outlined and worked out. Patients without a paid job focus on rehabilitation in other personal activities. The last session deals with relapse prevention and further improvement of self-control.</p>", "<title>Measurement and valuation</title>", "<title>1. Treatment implementation costs</title>", "<p>Personnel costs, for therapists' trainings, coordinating activities and monthly working group assemblies, were calculated by counting the total amount of hours that concerned people had invested and by multiplying these hours with personnel's gross salary per hour, including 39% employers' charges. For training and supervision only the hours that were actually attended were calculated, per person. The hours that people had spent on the implementation were counted retrospectively by interviewing concerned people. Traveling costs related to these activities were calculated by summing up the total amount of kilometers by car and counting €0.16 per kilometer. Material costs for informing GPs and patients were determined by summing up al printing, copying and distributing costs of used materials. Accommodation costs were calculated as 10% of personnel costs [##UREF##5##16##].</p>", "<title>2. CBT integral treatment costs</title>", "<p>For the CBT for CFS treatments integral prices were calculated, implying that all direct (executing) and indirect (overhead) costs of the MHC for offering the treatment were included in the calculation. Total costs of performed treatment sessions were determined by first summing up all therapists, diagnostic assistants and secretaries invested time per treatment. For each patient the total number of attended therapy sessions was registered. Planned sessions that were cancelled less than 24 hours before the session were also calculated. Per session one hour of work was counted for a therapist. Per treated patient 15 minutes secretary work was counted. Per intake and per post treatment session 30 minutes work for a diagnostic assistant was counted. The personnel costs for secretaries and diagnostic assistants were also based on gross salary plus39% charges. Then, for overhead costs and building use, 20% and 10% respectively of personnel costs were added to the personnel costs [##UREF##5##16##]. Treatment material costs were too small to count.</p>", "<title>3. Direct medical costs (apart from CBT treatment)</title>", "<p>Volumes of medical consumption were measured with a paper and pencil questionnaire that was filled in by the patient at base line and after treatment. Patients were asked how many visits that they had made in the previous six months to a GP, medical specialist, physiotherapist, psychologist, psychiatrist and alternative medical practitioner. Use of home care support (average hours per week in the last 6 months) hospitalization (number of nights in 6 months) and use of (prescribed and not prescribed) medication were also asked. To value patients' medical consumption, cost prices were used as given in the Dutch cost analyses manual [##UREF##5##16##] after recalculating them to the 2004 price level (Table ##TAB##0##1##). Costs of prescribed medication were calculated based on the Dutch indicated market prices per month based on 'defined daily doses'. Six percent taxes and €6,51 pharmacy costs per client using medication were added to this market costs. Patients were asked to give a price indication per month of their costs incurred in purchasing over-the-counter medication.</p>", "<title>4. Direct non-medical costs</title>", "<p>For each CFS patient traveling costs for attending the CFS treatment sessions were applied. Distances from patients' homes to the MHC's treatment location were found at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.routenet.nl\"/>. This distance was multiplied by each patient's total number of attended sessions. Again € 0.16 per kilometer was calculated.</p>", "<title>5. Indirect non-medical costs</title>", "<p>Patients' lost productivity costs due to absenteeism from paid work were also measured with the paper and pencil questionnaire. The questionnaire contained questions about work and daily activities, based on the 'Health and Labour Questionnaire' [##REF##8840661##17##]. The number of hours of paid work in the last two weeks was filled in. We valued the days of absenteeism from paid work with Dutch standard productivity costs specified for age, sex and education level [##UREF##5##16##] and using the human capital method. Transfer payments related to occupational disability insurances were not included since these are neither a gain nor a cost to society [##UREF##3##10##]. The productivity costs per two weeks were then multiplied by 13 to provide the costs per 6 months.</p>", "<p>Informal care measured at baseline and after treatment with a paper and pencil questionnaire about the number of hours per week that patients had received informal care. This was costed at € 8.38 per hour [##UREF##5##16##], the wage rate for a cleaner. Time costs for patients attending the treatment sessions were excluded.</p>", "<title>Economic evaluation method</title>", "<title>Perspective</title>", "<p>Total costs of implementing CBT for CFS were analyzed both from a societal perspective (including also non medical costs such as travelling expenses and productivity costs, regardless of who carried them) and from a health care perspective (indicating that only medical costs were relevant) [##UREF##3##10##]. For the societal perspective we calculated costs per gained QALY. For the health care perspective instead we calculated costs per recovered patient, being a measure of health rather than a measure of general welfare, which corresponds better to the more limited scope of the health care perspective [##UREF##6##18##]</p>", "<title>Calculation methods</title>", "<p>Total costs were divided into costs for implementing the new treatment (the fixed, so called 'organizing' costs), and costs for facilitating and using the CBT for CFS treatments (the variable, so called 'executing' costs) [##REF##14532369##19##]. Fixed costs were related to assembly- and organizing activities of the working group, informational interventions towards GPs and the public and training and supervising the initial therapists. Variable costs comprised of: 1. Costs for continuing the treatment facility, comprising of repeatedly providing training and supervision for new therapists (we assumed that because of personnel turnover every two years two new therapist need to be trained and supervised) and continuing PR activities; 2. Costs for clients attending the treatment sessions (e.g. traveling costs); 3. Costs for organizing and facilitating the treatments (mainly labour costs); 4. Societal costs (including use of healthcare services other than CBT and lost productivity costs due to absenteeism from paid work) and 5. Costs for performing the treatment program (time for therapist performing the treatment sessions, costs for building use, etc).</p>", "<p>Because the time period between costs and effects was less then 12 months, we did not apply the principle of discounting. All costs were recalculated to 2004 by using the 2004 'derivative cost-of-living index figures' [##UREF##7##20##]. All cost prices included in the analyses were valued in terms of integral cost prices [##UREF##5##16##].</p>", "<title>Data analysis</title>", "<title>Missing data</title>", "<p>in the original database an average of 0.5 cases per item were randomly missing because some patients had failed to answer all questions of a particular questionnaire. These missing data were filled in with the median value for the particular item. In the cases of missing data due to loss of follow up the method of last observation carried forward was used [##REF##15729743##21##], indicating that intake measurements were used as post treatment. Analyses were performed on basis of intention to treat; patients who attended an intake but did not start treatment and patients who dropped out of treatment were all included in the cost analyses.</p>", "<title>Cost and outcome calculations</title>", "<p>Given the non-controlled design of the present study, it did not fulfil the criteria of a 'full economic evaluation' [##UREF##3##10##], and hence the usually calculated incremental costs effectiveness ratios (ICER) could not be analysed. In stead we calculated cost outcome ratios (CORs). Cost outcome ratios (COR) are concerned with the joint difference in costs and outcomes before and after (implementing and) performing a certain treatment [##UREF##3##10##]. This ratio thus indicates the financial investment that is needed to gain a certain treatment effect, based on the assumption that autonomous change regarding the patients is negligible.</p>", "<p>The COR was calculated in two ways. First by defining treatment effect as 'percentage of recovered patients' (health care perspective), and second by using quality adjusted life-years (QALYs) as a measure for treatment effect (societal perspective). The recovery rate was analysed by calculating the percentage of patients experiencing significant clinical improvement (CSI). Patients were defined as being CSI at post treatment if they had a reliable change index &gt; 1.96 on the CIS fatigue severity subscale [##REF##2002127##22##], a fatigue severity score &lt;= 35 and a Rand-36 physical functioning score &gt; = 65 [##REF##7965927##12##]. Quality of life was measured using utility scores of the Euroqol [##REF##9366889##14##]. This utility score, lying between 0 (health state equal to death) and 1 (perfect health state), represents the QALY due to some intervention.</p>", "<p>Since the health care costs were measured over a period of 6 months, while the individual durations of treatment differed between 2.2 and 16.2 months, all medical and non-medical costs at follow up were extrapolated to the individually defined treatment period, before including them in the cost outcome analyses.</p>", "<p>Utility scores were measured two times, at intake and at follow up. Since the difference in utility scores between the two measurements was presumably reached gradually instead of at once, and because the duration of treatment differed per patient, the gained QALYs at post treatment were calculated as: 0.5 * (utility score post treatment – utility score intake)/12 * individual number of months of treatment [##UREF##3##10##].</p>", "<title>Analysis of uncertainty</title>", "<p>Because it was presumed that, as usually, the measured medical costs would follow a skewed distribution, a normality assumption would be problematic when estimating confidence intervals. Therefore the non-parametric bootstrap method [##REF##9285227##23##] was used to quantify the uncertainty of the calculated COR. In the bootstrap method this uncertainty is quantified by plotting cost-effectiveness acceptability (CEA) curves by means of repeated re-sampling of the costs and outcome data (the bootstrapping), which generates a distribution of mean costs and outcomes of two situations [##UREF##8##24##]. These distributions are then used to calculate the probability that one of the situations is the optimal choice, given a range of possible maximum values (ceiling ratio) that a decision maker might be willing to pay for a unit of improvement in outcome. Because the present study did not calculate cost effectiveness, we used the term 'COR acceptability curve' instead of 'ICER acceptability curve'.</p>", "<title>Scenario calculation</title>", "<p>For both the societal and the healthcare perspective, the COR of implementing CBT for CFS in a MHC was also calculated for a period of 5 years.</p>" ]

[ "<title>Results</title>", "<title>Descriptives</title>", "<p>Figure ##FIG##0##1## presents the patient flow. From the 143 patients that entered the MHC during the observation period, 18 'no show' patients never showed up at the intake session. Since they only caused negligible costs they were excluded from this study. The remaining 125 patients were included. At intake 13 patients appeared not to fulfil the diagnostic criteria for CFS, these patients did not start treatment. Of the 112 patients that started treatment, 28 dropped out of treatment quickly after the intake session. Of the 84 patients that followed treatment 12 dropped out half way or later and 72 finished treatment.</p>", "<title>Missing data</title>", "<p>At 8 months follow up 74 of the 84 treated patients filled in the questionnaire and 10 patients failed to do this (7 drop out patients and 3 treatment completers). Of the 13 'non CFS' and the 28 'non starting' patients their intake measurements were used as post treatment, since no treatment effect was to be expected within less than 2 sessions.</p>", "<title>Sample characteristics</title>", "<p>Patients' characteristics are shown in Table ##TAB##1##2##. Of the 77 patients (62%) that had a paid job (42 fulltime and 35 part time) 54 were actually working and 23 were on sickness benefit.</p>", "<title>Treatment characteristics</title>", "<p>The mean duration of the 84 performed treatments was 8.4 months (SD 3.3) and varied from 2.2 to 16.2 months. No relations were found between duration of treatment and several other variables, like treatment effect, decrease in medical consumption or lost productivity costs after treatment. The mean number of treatment sessions was 14.5 (SD 5.6) and varied from 6 to 23 sessions.</p>", "<title>Treatment effects</title>", "<p>Effect based on fatigue severity: after treatment 46 of 125 patients (37%) were recovered. Effect based on Euroqol: the mean utility score at intake was 0.57 (SD 0.27) and post treatment 0.65 (SD 0.30) (table ##TAB##2##3##).</p>", "<title>Costs results</title>", "<p>Table ##TAB##3##4## shows the total fixed and variable costs of preparing and introducing the implementation of CBT for CFS, with a total of € 90.765 and € 59.300 respectively. The costs results of performing and using CBT (table ##TAB##4##5##) reveal that per patient a mean of € 597 were spent per CBT treatment.</p>", "<p>Table ##TAB##5##6## gives the amounts of medical care other than CBT treatment. These results were used for calculating all (non) medical costs (table ##TAB##6##7##). As can be seen, total medical costs decreased from € 1.112 per six months before treatment to € 810 after treatment (95% CI -€ 784 to -€ 26). Total non-medical costs also decreased, from € 1.249 per six months before treatment to € 1.012 after treatment (95% CI -€ 813 to € 271).</p>", "<p>In table ##TAB##7##8## the figures on work and absenteeism are given, showing that the mean number of working hours according to contract had fallen from 16.4 per week before treatment to 14.9 after treatment (95% CI -5.4 to 3.2). The number of real worked hours however had risen from 9.3 before treatment to 11.4 hours per week after treatment (95% CI – 2.6 to 5.5), implying that the number of lost productivity hours and its costs decreased, from € 218 per patient per week before treatment tot € 122 after treatment (95% CI -€ 173 to -€ 6).</p>", "<title>Cost outcome ratios</title>", "<p>From the societal perspective the mean societal costs per patient per six months were € 8.030 before implementing CBT for CFS and € 6.869 after it (95% CI -€ 3.489 to € 1.083). The mean gained QALY per patient was 0,03. Given the lower cost level and a higher health state of patients, the COR-estimate indicates dominancy. The five years scenario calculation analysis, in which the total amount of treated patients was up-scaled to 3.33 times the amount of patients that were treated in the implementation period of 1,5 years (also figure ##FIG##1##2##), revealed a greater than 90% probability for a favorable COR for all acceptability thresholds.</p>", "<p>From the health care perspective it was found that mean costs per patient per six months were € 1.117 before implementation and treatment and € 2.586 after it (95% CI € 958 to € 1.876). Given the recovery rate of 37% the COR of implementing CBT for CFS was € 5.320 per recovered CFS patient. The COR acceptability curve (figure ##FIG##2##3##) shows that the probability that implementing CBT for CFS has a favorable COR is 100% when the decision maker values a recovered CFS patient at least € 6.500.</p>", "<p>The 5 years scenario calculation (also figure ##FIG##2##3##) showed that the 100% guarantee for an acceptable COR was reached at the willingness to pay threshold of € 4.500.</p>", "<title>Sensitivity analysis</title>", "<p>In sensitivity analysis applied for the societal perspective, the costs for (informal) home care and productivity costs were varied. As also has been found in other studies [##REF##11271867##7##,##REF##15554570##8##] (informal) home care appeared to cause major costs. Besides that questions may be raised about the accuracy of the measured amounts of home care. It is a difficult aspect to measure, for example the distinction between informal care and normal household activities is not clearly defined, both for researchers and for patients, especially if the informal caregiver shares a household with the patient [##REF##15386676##25##]. Since patients have a tendency to overestimate their hours of informal care, we performed a calculation reducing informal home care to 50% and leaving it out at all. In a third calculation both informal and formal home care were omitted from the analyses. These calculations showed that if informal home care was omitted from the analysis, and when both informal and formal homecare were omitted, the probability that implementing CBT for CFS has an acceptable COR remained above 80% for all acceptability thresholds.</p>", "<p>In addition, two extra analyses were done, in which productivity costs were set to 70% and to 30% of the original base case level. This revealed a drop in cost savings of CBT to -€ 16.800 and -€18.730 respectively. It appeared that implementing CBT for CFS remained dominant at both the 70% and the 30% level.</p>", "<p>Finally, to get an impression of this study's results when compensating for spontaneous recovery, an additional analysis was performed. This was done from the health care perspective, presuming a spontaneous recovering rate of 5% [##UREF##1##2##], implying a recovery rate due to treatment of 32%. It revealed that the COR would rise from € 5.320 to about € 5.969 per recovered patient.</p>" ]

[ "<title>Discussion</title>", "<p>This study has shown that from a societal perspective the cost outcome ratio (COR) after implementing CBT for CFS in a MHC was dominant compared to before. From a healthcare perspective the COR after implementation was more costly but also more effective than before, and the 100% probability that the COR is acceptable was reached at the willingness to pay threshold of € 4.500 is positive. Given that CBT is the only effective treatment for CFS and has been scarcely available until now, this is relevant information in favor of nationwide implementation. Although some studies have already examined the cost effectiveness of behavioral treatments for chronic fatigue (CF) [##REF##11271867##7##,##REF##15554570##8##] and for CFS [##UREF##2##9##], there has been no research into the cost effectiveness of such a treatment that also took into account the costs of designing the implementation interventions needed for implementing the treatment and the costs of actually implementing the treatment in a non-academic setting. Such a study implies a less homogenous patient population and less control over the content of performed treatment sessions than an academic setting can guarantee.</p>", "<p>Concerning age and gender, the patient population was fully representative of the CFS population. Compared to other trials in the area of CFS, the baseline fatigue severity was a little lower and relatively many patients had a paid job [##UREF##9##26##,##REF##15585538##27##]. These differences could be explained by the fact that the treatment facility at the mental health care institution was more easily accessible. Patients may be recognized as CFS by their GP and referred to CBT in an earlier phase than patients referred (mostly by a medical specialist) to a specialized hospital setting.</p>", "<p>As was also found in earlier cost effectiveness studies, [##REF##15554570##8##,##UREF##2##9##] an overall lower use of health care facilities was measured after CBT for CFS than before it. This may be explained by the fact that during treatment with CBT patients are instructed not to use other treatments or medication and by the fact that when starting treatment all patients were diagnosed as CFS. Looking for a diagnosis and a lack of affective treatment are the main reasons for CFS patients' high use of health care facilities [##REF##8792781##28##]. Concerning work productivity, fewer patients had a paid job after treatment than before, but the mean hours of paid work per week had increased after treatment. Given the short time horizon (8 months) the full influence of CBT for CFS on work productivity might be revealed to be larger and the impact on cost-effectiveness more pronounced.</p>", "<p>In this study we used a conservative method, last observation carried forward, in cases of missing data. This imputation method might have influenced the results in a conservative, negative direction. However the proportion of missing data was in our opinion rather small (&lt; 12%) thus the chance that significantly different results were obtained is small.</p>", "<p>A serious limitation of this study is it's non-controlled before and after design, which implies that incremental cost effectiveness compared to a natural course control group, or compared to a guided support group controlling for any placebo effect, could not be analysed. However, the incremental cost effectiveness ratio (ICER) of CBT for CFS compared to usual care was recently reported by Severens et al. [##UREF##2##9##]. The focus and contribution of the present study was primarily to investigate costs and consequences of implementing this evidence based treatment in a clinical practice setting. This is a relevant issue in bridging the gap between science and research, since proven (cost) effectiveness under laboratory conditions of RCTs does not guarantee the same in the practice field of health care. Both smaller treatment effects due to the less controlled situation and accompanying costs of including costs for implementing the treatment might change the cost-outcome ratio.</p>", "<p>Another weak point in this study is the variable follow up time. Although the mean time period between intake and post treatment was 8.4 months, and analyses were done using this time horizon for all patients, the real time interval varied considerable. The problem hereby is that in fact we do not know what this implies for the results that were found.</p>", "<p>A strong point though is the fact that, besides the usual included medical-, productivity-, and patient related costs also protocol driven- and implementation related costs were included [##REF##17645685##29##], giving a more complete and more relevant view on the cost and outcomes of providing nationwide CBT for CFS.</p>" ]

[ "<title>Conclusion</title>", "<p>To conclude, the results of this study suggest that implementing cognitive behavioral therapy for chronic fatigue syndrome in a mental health center is feasible and advisable. This strategy appeared to be dominant (resulting in lower costs and higher health states) compared to the starting situation from a societal perspective. From a health care perspective the implementation also implied better health states, but also higher costs, and the probability of a positive cost outcome ratio depended on how much value is placed on a recovered CFS patient. The outcomes of this study might facilitate the decision for health care providers whether or not to adopt CBT for CFS in their institution.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>This study investigated the costs and outcomes of implementing cognitive behavior therapy (CBT) for chronic fatigue syndrome (CFS) in a mental health center (MHC). CBT is an evidence-based treatment for CFS that was scarcely available until now. To investigate the possibilities for wider implementation, a pilot implementation project was set up.</p>", "<title>Method</title>", "<p>Costs and effects were evaluated in a non-controlled before- and after study with an eight months time-horizon. Both the costs of performing the treatments and the costs of implementing the treatment program were included in the analysis. The implementation interventions included: informing general practitioners (GPs) and CFS patients, training therapists, and instructing the MHC employees. Given the non-controlled design, cost outcome ratios (CORs) and their acceptability curves were analyzed. Analyses were done from a health care perspective and from a societal perspective. Bootstrap analyses were performed to estimate the uncertainty around the cost and outcome results.</p>", "<title>Results</title>", "<p>125 CFS patients were included in the study. After treatment 37% had recovered from CFS and the mean gained QALY was 0.03. Costs of patients' health care and productivity losses had decreased significantly. From the societal perspective the implementation led to cost savings and to higher health states for patients, indicating dominancy. From the health care perspective the implementation revealed overall costs of €5.320 per recovered patient, with an acceptability curve showing a 100% probability for a positive COR at a willingness to pay threshold of €6.500 per recovered patient.</p>", "<title>Conclusion</title>", "<p>Implementing CBT for CFS in a MHC appeared to have a favorable cost outcome ratio (COR) from a societal perspective. From a health care perspective the COR depended on how much a recovered CFS patient is being valued. The strength of the evidence was limited by the non-controlled design. The outcomes of this study might facilitate health care providers when confronted with the decision whether or not to adopt CBT for CFS in their institution.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>KS collected all data, performed the statistical analysis together with JLS and wrote manuscript. JLS contributed to the development of the study design concentrating on the costs aspects, advised about the performance of the statistical analysis, checked the analysis and results and provided the Bootstrap program. MW contributed to the development of the study design concentrating on the implementation outcome aspects and revised the manuscript critically several times. GB delivered the treatment outcome measurement scales, helped with interpreting the results and helped to draft the manuscript. All authors read and approved the final manuscript.</p>", "<title>Pre-publication history</title>", "<p>The pre-publication history for this paper can be accessed here:</p>", "<p><ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1472-6963/8/175/prepub\"/></p>" ]

[ "<title>Acknowledgements</title>", "<p>None</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Patient flow.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Acceptability curve showing the probability that implementing CBT for CFS has a favorable cost outcome ratio over a range of willingness to pay regarding societal costs per QALY</bold>. Societal willingness to pay for a CFS patient's gained QALY.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Acceptability curve showing the probability that implementing CBT for CFS has a favorable cost outcome ratio over a range of willingness to pay regarding health care costs per recovered patient</bold>. Willingness to pay for recovering a CFS patient.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Cost-prices used to value the different health care volumes, measured at patient level before and after treatment.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><italic>Health care volume</italic></td><td align=\"right\"><italic>Cost price</italic></td></tr></thead><tbody><tr><td align=\"left\"> General practitioner (per visit)</td><td align=\"right\">€ 20.39</td></tr><tr><td align=\"left\"> Medical specialist (per visit)</td><td align=\"right\">€ 63.99</td></tr><tr><td align=\"left\"> Physiotherapist (per visit)</td><td align=\"right\">€ 22.96</td></tr><tr><td align=\"left\"> Psychologist (per visit)</td><td align=\"right\">€ 125.14</td></tr><tr><td align=\"left\"> Psychiatrist (per visit)</td><td align=\"right\">€ 88.81</td></tr><tr><td align=\"left\"> Non-physician alternative medicine practitioner (per visit)</td><td align=\"right\">€ 48.87</td></tr><tr><td align=\"left\"> Home care (per hour)</td><td align=\"right\">€ 21.90</td></tr><tr><td align=\"left\"> Informal home support (per hour)</td><td align=\"right\">€ 8.38</td></tr><tr><td align=\"left\"> Hospitalization (per night)</td><td align=\"right\">€ 333.40</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Patients' characteristics (N = 125)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Categorical variables</td><td align=\"center\">(N/%)</td></tr></thead><tbody><tr><td align=\"left\"> Sex (man/women)</td><td align=\"center\">42 (34%)/83 (66%)</td></tr><tr><td align=\"left\"> Higher education</td><td align=\"center\">51 (41%)</td></tr><tr><td align=\"left\"> Having a paid job</td><td align=\"center\">77 (62%)</td></tr><tr><td align=\"left\"> Married/living together/living with parents</td><td align=\"center\">98 (78%)</td></tr><tr><td/><td/></tr><tr><td align=\"left\">Continuos variables</td><td align=\"center\">M (SD)</td></tr><tr><td align=\"left\"> Age</td><td align=\"center\">38.7 (10.2)</td></tr><tr><td align=\"left\"> Duration of fatigue (years)</td><td align=\"center\">6.3 (7.0)</td></tr><tr><td align=\"left\"> Fatigue severity (Cis20)</td><td align=\"center\">48.3 (8.0)</td></tr><tr><td align=\"left\"> Physical impairment (Rand 36)</td><td align=\"center\">54.0 (23.4)</td></tr><tr><td align=\"left\"> Social impairment (Rand 36)</td><td align=\"center\">41.5 (23.7)</td></tr><tr><td align=\"left\"> Psychosocial well-being (SCL-90)</td><td align=\"center\">165.1 (42.1)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Mean utility scores at intake and 8 months follow up (N = 125)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><italic>Intake</italic></td><td align=\"center\"><italic>Follow up</italic></td><td align=\"center\"><italic>Δ follow up – intake</italic></td><td align=\"center\"><italic>95% CI</italic></td><td align=\"center\"><italic>P</italic></td></tr></thead><tbody><tr><td align=\"left\">Mean (SD)</td><td align=\"center\">0.57 (0.27)</td><td align=\"center\">0.65 (0.30)</td><td align=\"center\">0.078 (0.028)</td><td align=\"center\">0.03 to 0.09</td><td align=\"center\">&lt; 0.001</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Costs of developing and introducing the implementation of CBT for CFS divided in fixed and variable costs.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"right\"><italic>Volume</italic></td><td align=\"right\"><italic>Calculated costs value per volume</italic></td><td align=\"right\"><italic>Costs</italic></td></tr></thead><tbody><tr><td align=\"left\">Personnel costs</td><td/><td/><td/></tr><tr><td align=\"left\">Fixed</td><td/><td/><td/></tr><tr><td align=\"left\"> Therapists</td><td align=\"right\">647 hours</td><td align=\"right\">€ 55.50/hour</td><td align=\"right\">€ 35.909</td></tr><tr><td align=\"left\"> Management</td><td align=\"right\">312 hours</td><td align=\"right\">€ 73.17/hour</td><td align=\"right\">€ 22.829</td></tr><tr><td align=\"left\"> Others</td><td align=\"right\">793 hours</td><td align=\"right\">€ 35.87/hour</td><td align=\"right\">€ 28.444</td></tr><tr><td align=\"left\"> Total fixed</td><td/><td/><td align=\"right\">€ 87.182</td></tr><tr><td align=\"left\">Variable</td><td/><td/><td/></tr><tr><td align=\"left\"> Therapists</td><td align=\"right\">460 hours</td><td align=\"right\">€ 55.50/hour</td><td align=\"right\">€ 25.530</td></tr><tr><td align=\"left\"> Management</td><td align=\"right\">20 hours</td><td align=\"right\">€ 73.17/hour</td><td align=\"right\">€ 1.463</td></tr><tr><td align=\"left\"> Supervisor</td><td align=\"right\">266 hours</td><td align=\"right\">€ 62.52/hour</td><td align=\"right\">€ 16.625</td></tr><tr><td align=\"left\"> Others</td><td align=\"right\">6 hours</td><td align=\"right\">€ 35.87/hour</td><td align=\"right\">€ 215</td></tr><tr><td align=\"left\"> Total variable</td><td/><td/><td align=\"right\">€ 43.833</td></tr><tr><td align=\"left\">TOTAL personnel costs</td><td/><td/><td align=\"right\">€ 131.015</td></tr><tr><td/><td/><td/><td/></tr><tr><td align=\"left\"><italic>Material costs</italic></td><td/><td/><td/></tr><tr><td align=\"left\">Fixed</td><td/><td/><td/></tr><tr><td align=\"left\"> Building use</td><td align=\"center\" colspan=\"2\">10% of personal costs</td><td align=\"right\">€ 8.718</td></tr><tr><td align=\"left\"> PR activities</td><td align=\"center\" colspan=\"2\">2000 information letters and brochures</td><td align=\"right\">€ 1.705</td></tr><tr><td align=\"left\"> Total fixed</td><td/><td/><td align=\"right\">€ 10.423</td></tr><tr><td align=\"left\">Variable</td><td/><td/><td/></tr><tr><td align=\"left\"> Building use</td><td align=\"center\" colspan=\"2\">10% of personal costs</td><td align=\"right\">€ 4383</td></tr><tr><td align=\"left\"> PR activities</td><td align=\"center\" colspan=\"2\">2500 information letters and brochures</td><td align=\"right\">€ 1.983</td></tr><tr><td align=\"left\"> Total variable</td><td/><td/><td align=\"right\">€ 6.366</td></tr><tr><td align=\"left\">TOTAL material costs</td><td/><td/><td align=\"right\">€ 16.789</td></tr><tr><td/><td/><td/><td/></tr><tr><td align=\"left\">Traveling costs</td><td/><td/><td/></tr><tr><td align=\"left\">Fixed</td><td/><td/><td/></tr><tr><td align=\"left\"> Therapists</td><td align=\"right\">3605 km</td><td align=\"right\">€ 0.18/km</td><td align=\"right\">€ 649</td></tr><tr><td align=\"left\"> Management</td><td align=\"right\">3328 km</td><td align=\"right\">€ 0.18/km</td><td align=\"right\">€ 599</td></tr><tr><td align=\"left\"> Others</td><td align=\"right\">1411 km</td><td align=\"right\">€ 0.18/km</td><td align=\"right\">€ 254</td></tr><tr><td align=\"left\"> Total fixed</td><td/><td/><td align=\"right\">€ 1.502</td></tr><tr><td align=\"left\">Variable</td><td/><td/><td/></tr><tr><td align=\"left\"> Supervisor</td><td align=\"right\">5300 km</td><td align=\"right\">€ 0.18/km</td><td align=\"right\">€ 954</td></tr><tr><td align=\"left\"> Therapists</td><td align=\"right\">1275 km</td><td align=\"right\">€ 0.18/km</td><td align=\"right\">€ 230</td></tr><tr><td align=\"left\"> Management</td><td align=\"right\">40 km</td><td align=\"right\">€ 0.18/km</td><td align=\"right\">€ 7</td></tr><tr><td align=\"left\"> Total variable</td><td/><td/><td align=\"right\">€ 1.191</td></tr><tr><td align=\"left\">TOTAL traveling costs</td><td/><td/><td align=\"right\">€ 2.693</td></tr><tr><td/><td/><td/><td/></tr><tr><td align=\"left\">TOTAL fixed costs</td><td/><td/><td align=\"right\">€ 90.675</td></tr><tr><td align=\"left\">TOTAL variable costs</td><td/><td/><td align=\"right\">€ 59.300</td></tr><tr><td align=\"left\">TOTAL 'developing and introducing' costs</td><td/><td/><td align=\"right\">€ 149.975</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Mean costs per patient (in €) of using and performing CBT for CFS in mental health care (N = 125)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"right\"><italic>Mean</italic></td><td align=\"right\"><italic>SD</italic></td><td align=\"right\"><italic>Median</italic></td><td align=\"right\"><italic>Max</italic></td></tr></thead><tbody><tr><td align=\"left\">Personal costs of CBT treatment (only therapist costs)</td><td align=\"right\">€ 417</td><td align=\"right\">€ 314</td><td align=\"right\">€ 435</td><td align=\"right\">€ 1349</td></tr><tr><td align=\"left\">Personal costs of CBT treatment (secretary and test assistants costs)</td><td align=\"right\">€ 28</td><td align=\"right\">€ 7</td><td align=\"right\">€ 34</td><td align=\"right\">€ 34</td></tr><tr><td align=\"left\">Overhead costs and costs for building facilities</td><td align=\"right\">€ 143</td><td align=\"right\">€ 101</td><td align=\"right\">€ 146</td><td align=\"right\">€ 442</td></tr><tr><td align=\"left\">Patients travelling costs (return price)</td><td align=\"right\">€ 9</td><td align=\"right\">€ 6</td><td align=\"right\">€ 7</td><td align=\"right\">€ 43</td></tr><tr><td align=\"left\">TOTAL mean costs per patient of using and performing CBT for CFS</td><td align=\"right\">€ 597</td><td align=\"right\">€ 424</td><td align=\"right\">€ 628</td><td align=\"right\">€ 17892</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T6\"><label>Table 6</label><caption><p>Volumes of medical consumption (except CGT for CFS treatment) over a period of 6 months measured at patients level at intake and follow up (N = 125).</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"3\">Intake</td><td align=\"center\" colspan=\"3\">Follow up</td></tr></thead><tbody><tr><td/><td align=\"center\">N</td><td align=\"center\"><italic>Mean (SD)</italic></td><td align=\"center\"><italic>Median</italic></td><td align=\"center\">N</td><td align=\"center\"><italic>Mean (SD)</italic></td><td align=\"center\"><italic>Median</italic></td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"left\"><italic>Medical care</italic></td><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">GP (number of visits)</td><td align=\"center\">111</td><td align=\"center\">3.1 (3.7)</td><td align=\"center\">2</td><td align=\"center\">93</td><td align=\"center\">2.0 (1.9)</td><td align=\"center\">2</td></tr><tr><td align=\"left\">Medical specialist (n.o. visits)</td><td align=\"center\">55</td><td align=\"center\">1.2 (1.8)</td><td align=\"center\">0</td><td align=\"center\">38</td><td align=\"center\">0.9 (1.5)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Physiotherapist (n.o. visits)</td><td align=\"center\">29</td><td align=\"center\">3.8 (9.0)</td><td align=\"center\">0</td><td align=\"center\">23</td><td align=\"center\">2.7 (7.4)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Psychologist, other than CBT for CFS</td><td align=\"center\">22</td><td align=\"center\">1.0 (3.1)</td><td align=\"center\">0</td><td align=\"center\">13</td><td align=\"center\">0.4 (1.4)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Psychiatrist (number of visits)</td><td align=\"center\">9</td><td align=\"center\">0.2 (1.2)</td><td align=\"center\">0</td><td align=\"center\">7</td><td align=\"center\">0.2 (1.2)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Home care (hours per 6 months)</td><td align=\"center\">21</td><td align=\"center\">26.9 (104.3)</td><td align=\"center\">0</td><td align=\"center\">23</td><td align=\"center\">22.7 (88.4)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Hospitalisation (nights)</td><td align=\"center\">13</td><td align=\"center\">0.4 (2.0)</td><td align=\"center\">0</td><td align=\"center\">9</td><td align=\"center\">0.2 (1.5)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Prescribed medication (yes/no)</td><td align=\"center\">92</td><td align=\"center\">77%</td><td/><td align=\"center\">87</td><td align=\"center\">72%</td><td/></tr><tr><td align=\"left\"><italic>Non-medical care</italic></td><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Informal home care (hrs per 6 months)</td><td align=\"center\">37</td><td align=\"center\">132.3 (268.3)</td><td align=\"center\">0</td><td align=\"center\">33</td><td align=\"center\">110.6 (182.8)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Altern. med. practitioner (n.o. visits)</td><td align=\"center\">30</td><td align=\"center\">1.0 (2.3)</td><td align=\"center\">0</td><td align=\"center\">21</td><td align=\"center\">0.8 (2.0)</td><td align=\"center\">0</td></tr><tr><td align=\"left\">Non prescribed medication (yes/no)</td><td align=\"center\">56</td><td align=\"center\">53%</td><td/><td align=\"center\">27</td><td align=\"center\">35%</td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T7\"><label>Table 7</label><caption><p>Mean medical and non-medical costs per 6 months measured at patient level at intake and follow up (N = 125).</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"3\">Intake</td><td align=\"center\" colspan=\"3\">Follow up</td></tr></thead><tbody><tr><td/><td align=\"right\"><italic>Mean</italic></td><td align=\"right\"><italic>SD</italic></td><td align=\"right\"><italic>Median</italic></td><td align=\"right\"><italic>Mean</italic></td><td align=\"right\"><italic>SD</italic></td><td align=\"right\"><italic>Median</italic></td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"left\">Medical costs</td><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">GP</td><td align=\"right\">€ 63</td><td align=\"right\">€ 67</td><td align=\"right\">€ 41</td><td align=\"right\">€ 41</td><td align=\"right\">€ 40</td><td align=\"right\">€ 41</td></tr><tr><td align=\"left\">Medical specialist</td><td align=\"right\">€ 77</td><td align=\"right\">€ 103</td><td align=\"right\">€ 0</td><td align=\"right\">€ 58</td><td align=\"right\">€ 95</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Physiotherapist</td><td align=\"right\">€ 87</td><td align=\"right\">€ 202</td><td align=\"right\">€ 0</td><td align=\"right\">€ 62</td><td align=\"right\">€ 181</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Psychologist</td><td align=\"right\">€ 125</td><td align=\"right\">€ 377</td><td align=\"right\">€ 0</td><td align=\"right\">€ 50</td><td align=\"right\">€ 172</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Psychiatrist</td><td align=\"right\">€ 18</td><td align=\"right\">€ 107</td><td align=\"right\">€ 0</td><td align=\"right\">€ 18</td><td align=\"right\">€ 95</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Home care</td><td align=\"right\">€ 589</td><td align=\"right\">€ 1629</td><td align=\"right\">€ 0</td><td align=\"right\">€ 498</td><td align=\"right\">€ 1235</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Hospitalisation</td><td align=\"right\">€ 134</td><td align=\"right\">€ 720</td><td align=\"right\">€ 0</td><td align=\"right\">€ 67</td><td align=\"right\">€ 509</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Prescribed medicine</td><td align=\"right\">€ 19</td><td align=\"right\">€ 50</td><td align=\"right\">€ 3</td><td align=\"right\">€ 16</td><td align=\"right\">€ 52</td><td align=\"right\">€ 3</td></tr><tr><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"><italic>Non medical costs</italic></td><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Alternative med. pr.</td><td align=\"right\">€ 56</td><td align=\"right\">€ 114</td><td align=\"right\">€ 0</td><td align=\"right\">€ 39</td><td align=\"right\">€ 94</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Non prescr. medicine</td><td align=\"right\">€ 52</td><td align=\"right\">€ 121</td><td align=\"right\">€ 0</td><td align=\"right\">€ 25</td><td align=\"right\">€ 53</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Informal homecare</td><td align=\"right\">€ 1109</td><td align=\"right\">€ 2322</td><td align=\"right\">€ 0</td><td align=\"right\">€ 927</td><td align=\"right\">€ 1573</td><td align=\"right\">€ 0</td></tr><tr><td align=\"left\">Travelling costs</td><td align=\"right\">€ 32</td><td align=\"right\">€ 32</td><td align=\"right\">€ 23</td><td align=\"right\">€ 21</td><td align=\"right\">€ 28</td><td align=\"right\">€ 11</td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"left\">Total medical costs</td><td align=\"right\">€ 1112</td><td align=\"right\">€ 2258</td><td align=\"right\">€ 362</td><td align=\"right\">€ 810</td><td align=\"right\">€ 1350</td><td align=\"right\">€ 241</td></tr><tr><td align=\"left\">Total non-medical costs</td><td align=\"right\">€ 1249</td><td align=\"right\">€ 2396</td><td align=\"right\">€ 112</td><td align=\"right\">€ 1012</td><td align=\"right\">€ 1822</td><td align=\"right\">€ 72</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T8\"><label>Table 8</label><caption><p>Patient volumes of work and lost productivity costs per week, measured at patient level before and after treatment (N = 125).</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"4\">Intake</td><td align=\"center\" colspan=\"4\">Follow up</td></tr></thead><tbody><tr><td/><td align=\"right\"><italic>Mean</italic></td><td align=\"right\"><italic>SD</italic></td><td align=\"right\"><italic>Median</italic></td><td align=\"right\"><italic>Max</italic></td><td align=\"right\"><italic>Mean</italic></td><td align=\"right\"><italic>SD</italic></td><td align=\"right\"><italic>Median</italic></td><td align=\"right\"><italic>Max</italic></td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td align=\"left\">Number of contract hours</td><td align=\"right\">16.2</td><td align=\"right\">16.3</td><td align=\"right\">10</td><td align=\"right\">40</td><td align=\"right\">14.9</td><td align=\"right\">16.2</td><td align=\"right\">7</td><td align=\"right\">40</td></tr><tr><td align=\"left\">Number of worked hours</td><td align=\"right\">9.4</td><td align=\"right\">13.5</td><td align=\"right\">0</td><td align=\"right\">45</td><td align=\"right\">11.4</td><td align=\"right\">14.7</td><td align=\"right\">0</td><td align=\"right\">46</td></tr><tr><td align=\"left\">Absenteeism in hours</td><td align=\"right\">7.4</td><td align=\"right\">12.3</td><td align=\"right\">0</td><td align=\"right\">40</td><td align=\"right\">4.1</td><td align=\"right\">8.8</td><td align=\"right\">0</td><td align=\"right\">40</td></tr><tr><td align=\"left\">Lost productivity costs</td><td align=\"right\">€ 218</td><td align=\"right\">€ 392</td><td align=\"right\">€ 0</td><td align=\"right\">€ 1544</td><td align=\"right\">€ 122</td><td align=\"right\">€ 292</td><td align=\"right\">€ 0</td><td align=\"right\">€ 1544</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>All volumes were valued in terms of Dutch integral cost prices at the price level of 2004 (Oostenbrink et al., 2004).</p></table-wrap-foot>", "<table-wrap-foot><p>N = the number of patients that had been using this form of healthcare</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1472-6963-8-175-1\"/>", "<graphic xlink:href=\"1472-6963-8-175-2\"/>", "<graphic xlink:href=\"1472-6963-8-175-3\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The measurement of volatile gases in breath for the purpose of the diagnosis, screening and monitoring of disease is an attractive proposition due to the inherently non-invasive nature of such methodology [##REF##17305586##1##]. For example, the volatile biomarker nitric oxide has been successfully developed for the monitoring of airway inflammation while other putative markers for a conditions including cancer have been identified [##REF##17986377##2##]. To be adopted clinically the methods used for the detection and quantification of breath biomarkers likely need to be rapid, simple to use, reproducible and able to detect the presence of disease early in its course. The latter feature can reduce mortality, morbidity, patient distress and decrease associated costs for the healthcare system [##REF##16482730##3##]. As such more sensitive detection methods may offer the advantage that volatile biomarkers can be detected and acted upon at earlier stages in the disease process.</p>", "<p>A technique which shows promise for diagnostic breath analysis is Selected Ion Flow Tube Mass Spectrometry (SIFT-MS). SIFT-MS is suited to this role due to its linear response, reproducible absoloute concentration measurements, rapid analysis and lack of interference by the major constituents of expired air, in particular water vapour [##UREF##0##4##]. SIFT-MS is a form of chemical ionization in which precursor ions (usually H₃O⁺, NO⁺or O₂⁺) are reacted with gas mixtures to produce ionized products characteristic of the volatile chemicals present [##UREF##0##4##]. The specific precursor ions used in the reaction, formed using a microwave water vapour ion source, are selected using a quadrupole mass filter. The precursor ions are then introduced into a fast flowing stream of helium and reacted with the air samples containing the trace gases of interest in a short flow tube. The ionized products along with remaining precursor ions are then quantified using a downstream quadrupole mass analyser. Once the rate constant for the reaction is known absolute concentrations of the trace gases present can be calculated without the need for calibration standards.</p>", "<p>The sensitivity of SIFT-MS instruments has been a limiting factor since many potential biomarkers are present at concentrations in the low parts per billion by volume (PPBV) range or lower. This abundance is currently considered to be at or beneath the limits of detection for the technique, reported to be in the single digit parts per billion range for a single breath (10 second measurement time) [##UREF##0##4##, ####REF##369632##5##, ##REF##16075884##6####16075884##6##]. The sensitivity of SIFT-MS instruments is partly dependent on the rate constant for the reaction between the precursor ion and the trace gas but also on several modifiable factors: the rate of precursor ion production, the amount of trace gas introduced in a given time, the measurement time and the background 'noise' signal produced in the absence of analyte. Recently it has been reported [##REF##17302391##7##] that doubling the sampling flow rate allowed the detection of phosphine gas at concentrations of approximately 200 parts per trillion by volume (PPTV). Whether such sensitivities can be achieved for breath analysis by this modification is unclear since the authors used dry nitrogen as a carrier gas rather than the humid air mixture characteristic of human breath [##REF##17302391##7##]. This difference is of importance since increasing the sampling flow rate of expired breath elevates the proportion of precursor ions reacting with the abundant water vapour thereby potentially antagonising the beneficial effect of the increased sampling rate [##UREF##0##4##]. Alternatively, a recent report suggests that optimizing the water content of the gas mixture used in the instrument's ion source can markedly elevates the rate of precursor ion generation. Using this modification the effect of enhanced precursor rates upon instrument sensitivity for the analysis of humid air mixtures has been investigated.</p>", "<title>SIFT-MS analysis</title>", "<p>SIFT-MS analysis was performed using a Profile 3 instrument (Instrument Science, UK) modified to allow a variable water vapour abundance to enter the microwave source as described [##UREF##1##8##]. By this means H₃O⁺count rates of approximately 2.2 × 10⁶counts per second (cps) were achieved while introducing ambient air in the SIFT-MS, with the total count rate of H₃O⁺plus it's hydrates (H₃O⁺.(H₂O)_nwhere n = 1, 2 or 3) being approximately 2.5 × 10⁶cps; typical maximal O₂⁺count rates were approximately 3 × 10⁶cps while NO⁺were approximately 1.8 × 10⁶cps.</p>", "<title>Gas standards prepared with humidified bottled gas</title>", "<p>For these experiments the compounds xylene and toluene (Sigma Aldrich, USA) were chosen since both are present in low levels in the breath of healthy non-smokers (BRoss, unpublished observations). A 5 L Tedlar bag (SKC Inc, USA) was filled and evacuated 3 times with bottled air to reduce the concentration of bag derived volatile compounds. The bag was then inflated with bottled air before followed by the introduction of 20 mL of water into the bag via the sampling valve port. The bag was then heated to 40°C for 30 minutes to produce humidified air which was used in subsequent experiments. A measured quantity (260 mL) of the humidified air was then transferred to empty 0.5 L Tedlar bags by means of a heated syringe. Up to 500 μL of a commercially available gas standard containing approximately 5 parts per million by volume xylene and toluene in helium (Standard and Technical Gases, United Kingdom) was then introduced into the 0.5 L bags using 50, 250 or 500 μl capacity gas syringes (Fisher Scientific, Canada) via a septum to produce known concentrations of each gas. After further incubation at 40°C for 15 minutes to allow mixing, the standards were introduced into the SIFT-MS by negative pressure via a transfer line heated to 65°C at a flow rate of 0.21 Torr L s^-1. Measurements were made by opening the inlet valve and waiting 10 seconds before quantifying gas levels over the period 10 to 20 seconds.</p>", "<p>The water content of the gas samples prepared using humidified bottled air (5.8 ± 0.1% (mean ± SD) was similar to that of human breath. The concentration of xylene and toluene (measured by their reaction with the H₃O⁺precursor as previously described in detail [##REF##17939161##9##]) in humidified bottled air prior to introduction of these chemicals was not differentiable from the instrument background i.e. that derived from the count rate determined in the absence of sampled air. Using a 10 second sampling period both xylene and toluene could be differentiated from background at concentrations of approximately 500 PPTV (Figure ##FIG##0##1##) with instrument blanks (sampling line closed) being approximately 0.5 cps for both compounds, while 500 PPTV resulted in approximately 2 cps product ions. Increasing concentrations of both compounds produced a linear rise in measured concentration over the concentration investigated (approximately 0 – 8 PPBV).</p>", "<title>Gas standards prepared using human breath</title>", "<p>A breath sample was obtained from a 40 year old male in good health followed by the construction of gas standards using a similar method as for bottled air with the exception that no humidification was required. Specifically, the subject exhaled into a 3 L Tedlar bag and 260 mL was transferred into 0.5 L Tedlar bags, followed by the addition of known quantities of the 5 PPMV gas standard in helium by means of a gas syringe. The subject gave informed consent for the procedure. Detectable concentrations of xylene (product ion m/z 107, <italic>k </italic>= 2.3 × 10^-9cm³s^-1) and toluene (product ion m/z 93, <italic>k </italic>= 2.1 × 10^-9cm³s^-1) product ions, measured using the H₃O⁺precursor, were present in the absence of added standards (approximately 1 PPBV for xylene and toluene). These ions could well be derived from other compounds, however levels measured were significantly above instrument backgrounds (levels recorded in the absence of sampled air) and of similar concentrations (approximately 1 – 2 PPBV) when xylene or toluene levels were measured using reaction with the NO⁺or O₂⁺precursors (xylene: product ion m/z 106, <italic>k = </italic>1.4 × 10^-9cm³s^-1; toluene: product ion m/z 92, <italic>k </italic>= 1.4 × 10^-9cm³s^-1). It should be noted that the proton bound dimer of ethanol has the same m/z as the product of the reaction between toluene and H₃O⁺. As such, in the presence of high ethanol concentrations ethanol may contribute to the apparent toluene signal making the use of the other precursor ions advisable for the analysis of this compound. It is presently unclear as to whether these xylene and toluene are compounds produced metabolically or are due to ambient levels of these gases which were present in the environment at similar concentrations. Using a 10 second sampling period, 500 PPTV exogenous concentrations of both compounds were differentiable from the exhaled breath levels (Figure ##FIG##0##1##).</p>", "<title>Measurement of trace malodorous compounds in human breath</title>", "<p>In order to further investigate the ability of SIFT-MS to detect compounds present at concentrations of approximately 1 PPBV or less alveolar breath analysis was performed using three human subjects (a 40 year old female, a 41 year old male and a 26 year old female all of whom gave informed consent). Alveolar breath and static oral air were sampled as described [##UREF##2##10##]. Briefly, to sample alveolar breath the subject expired via a heated stainless steel tube into which was inserted the SIFT-MS sampling line so that a small fraction of the flow was drawn into the instrument using negative pressure at a flow rate of 0.21 Torr L s^-1. For oral air the stainless steel sampling tube was sealed at one end and an adapter attached at the other end which allows a piece of PTFE tubing (length 10 cm, OD 1/4\", ID 3/8\") to be attached. The subject was asked to close their mouth and breathe though their nose for 3 minutes to concentrate volatile compounds in the mouth. A piece of tubing is then inserted 2.5 – 5 cm into a nearly closed mouth. The subject is asked to not touch any mouth surface with the tube, and must not blow into, or inhale through, the tubing. Levels of the test compounds were measured using the H₃O⁺precursor by their following product ions and utilising the following rate constants: hydrogen sulphide: m/z 35, <italic>k </italic>= 2.0 × 10^-9cm³s^-1; methylmercaptan: m/z 49, <italic>k </italic>= 2.5 × 10^-9cm³s^-1; indole: m/z 118 and 136, <italic>k </italic>= 3.3 × 10^-9cm³s^-1; and methylindole: m/z 132 and 150, <italic>k </italic>= 3.3 × 10^-9cm³s^-1[##UREF##2##10##, ####REF##10867689##11##, ##UREF##3##12####3##12##]. As expected [##UREF##2##10##] (Table ##TAB##0##1##) oral air contained significant levels of the sulphur compounds hydrogen sulphide and methylmercaptan while levels in alveolar air were very low. For methylmercaptan the actual alveolar concentrations of the gas may be lower given that the reaction of H₃O⁺with the ¹⁸O isotopologue of ethanol will result in the production of the same product ion as that with methylmercaptan. As such approximately 0.6% of the ethanol concentration would contribute to the apparent methylmercaptan level measured using the above procedure e.g. 1 PPMV ethanol would contribute 6 PPB to the methylmercaptan concentration. Given the ubiquity of ethanol in both laboratory and clinical settings this mimicking effect of the compound should always be taken into account. The enhanced sensitivity of the SIFT-MS also allowed the detection ion product signatures of the compounds indole and methylindole. The presence of these compounds in human breath was supported by the generation of product ions of m/z 117 and 131 at rates approximately 2 cps above background for the reaction of human breath with O₂⁺, which are consistent with products generated by charge transfer reactions with indole and methylindole respectively [##UREF##3##12##]. Such rates of product ion generation equate with similar concentration levels of the compounds with that detected using H₃O⁺. The possibility that these ion products could be due to isopotopologues must be considered although significant generation of ions one or two mass units lower that the putative product ions for methylindole and indole was not apparent in the subjects studied. Nevertheless, as the sensitivity SIFT-MS increases the possibility that the presence of isopotopologues will become a significant factor in any particular measurement is also increased. In addition, the introduction of humidified lab air possessing a similar water content to that of human breath (6% v/v) did not change the apparent ambient levels of these compounds, suggesting that the product ions concerned are not due to a hydrate of a reaction product derived from another compound. The presented data therefore are suggestive of both methylindole and indole being present in human breath alveolar breath at low levels. Both compounds are characteristic of faecal odour and their detection in breath at low levels may derive from either the gut or from an oral bacterial source. Further investigation is however required to conclusively determine the existence of the indoles in human breath, particularly as the construction of accurate spectra is made difficult due to their apparent low abundance.</p>", "<title>Limits of detection and quantification</title>", "<p>The presented data suggest that recent advances in the design of the ion source intrinsic to SIFT-MS instruments can be utilized to achieve significant sensitivity gains for breath trace gas analysis. The rate of precursor ion generation reported in this and another recent study [##UREF##1##8##] are in the 2 – 3 million counts per second range for H₃O⁺, a rate approaching the 10 million counts per second generated by a related technique, proton transfer mass spectrometry (PTR-MS) [##REF##12619749##13##]. The ion generation method used by SIFT-MS allows the selection of the alternative NO⁺or O₂⁺precursor ions to increase the chemical resolution of SIFT-MS compared to PTR-MS, while avoiding the quantification errors encountered by PTR-MS occurring due to the variably elevated ion energies typical of the technique [##REF##17471381##14##]. By optimising the water vapour content of the ion source gas mixture the precursor count rates were sufficient to allow the detection of analyte concentrations of approximately 500 PPTV using the H₃O⁺precursor. This was achieved using humidified air samples similar to human breath with a sampling time typical of a single expiration (10 seconds).</p>", "<p>The determination of the limit of detection (LOD) can also be estimated using various parameters associated with the SIFT-MS analysis [##REF##17302391##7##], these being the sensitivity of the measurement (how many product ions are produced for a given concentration of analyte in a particular time) and the instrument background counts at the product m/z, that is the counts produced in the absence of analyte. For the compounds measured in this study sensitivity was in the range 2 – 4 cps per PPBV, while the background count rates for the product(s) ions were in the range 0.3 to 3 counts per second. The estimated LOD for 10 second measurements were calculated as 170, 130, 300, 440, 320 and 350 PPTV for indole, methylindole, hydrogen sulphide, methylmercaptan, xylene and toluene respectively. For the latter two compounds such computed values are in agreement with empirically derived limits of detection (see Figure ##FIG##0##1##). It is also possible to calculate the limit of quantification (LOQ), the concentration which can be determined with a particular level of precision. For a standard deviation of 20% using a 10 second measurement time the LOQs are 0.8, 0.7, 1.0, 1.3, 1.4 and 1.5 PPBV for the presumed analysis of indole, methylindole, hydrogen sulphide, methylmercaptan, xylene and toluene respectively. As such the presented data indicate that the modified instrument used in this study can produce reasonably precise measurements of each gas present at a concentration of approximately 1 PPBV. Clearly the technique is capable of lower LOD and LOQ when longer measurement times are used, but this will necessarily result in an inability to quantify trace gases present at these concentrations in a single expiration. The use of a breath storage device e.g. Tedlar bags, can allow longer measurement times to be used although this complicates the sampling methodology and can be the source of artefact.</p>" ]

[]

[]

[]

[ "<title>Conclusion</title>", "<p>In conclusion, advances in ion source design have resulted in significant improvements in the sensitivity of the SIFT-MS procedure resulting in the potential for additional biomarkers to be readily investigated. As the ion generation process becomes better understood it is likely that additional improvements in the rate of precursor ion generation will be achieved with a concomitant enhancement of sensitivity. It should be recognized, however, that background count rates also put a limit on the trace gas concentration that can be detected in a given time period. Background counts are influenced by a variety of factors including electrical noise in the detector, impurities in the carrier helium gas, the presence of isotopologues and hydrates of other compounds, and by the production of low abundance ions in the source which possess the same m/z ratio as the reaction product of the compound of interest [##UREF##4##15##]. The two latter factors may ultimately limit the detection sensitivity of SIFT-MS for specific trace gases rather than the rate of precursor ion generation. Optimisation of the ion source and/or the upstream mass filter to reduce the introduction of unwanted ions into the flow tube may, however, alleviate the problem of unwanted low abundance source ions. Nevertheless, for many trace gases sub-PPBV detection in single breath samples by SIFT-MS is readily achievable with the likelihood of further sensitivity enhancements in the future.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Selected ion flow tube mass spectrometry (SIFT-MS) allows the real time quantification of trace gases in air. Due to its tolerance of high humidity levels the technique is particularly suited to the chemical analysis of breath. The detection limit of SIFT-MS has previously reported to be approximately 5 – 10 PPBV which is insufficient for the measurement of some low abundance constituents of breath. Recent developments in the design of SIFT-MS instruments have increased the ion precursor count rates. It is, however, unclear as to how these advances will affect instrument sensitivity for breath analysis.</p>", "<title>Findings</title>", "<p>Standard gases were prepared by adding known quantities of compounds present at zero or very low levels in breath (xylene and toluene) to either humidified bottled air or actual human breath. These were then analysed by SIFT-MS to calculate the limits of detection for each compound under conditions which mimic a single breath exhalation. For xylene and toluene the limits of detection was approximately 0.5 PPBV per 10 seconds of analysis time. Results gained using this level of sensitivity suggested the presence of low levels of the compounds indole and methylindole in human alveolar and static oral air, although further studies are necessary to confirm these findings.</p>", "<title>Conclusion</title>", "<p>Recent advances in SIFT-MS have increased the techniques sensitivity for breath analysis into the sub PPBV range enabling the real time quantification of low level trace gases in human breath.</p>" ]

[ "<title>Competing interests</title>", "<p>The author declares that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>All experimental work and production of the manuscript was by BMR.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This work was aided by a grant from the Northern Cancer Research Foundation to BMR.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Detection of xylene and toluene added to humidified air and human breath</bold>. Xylene (A) or toluene (B) vapour were added to either humidified bottled air (circles) or human breath (triangles) contained in Tedlar bags to produce a range of concentrations in duplicate. The standards were introduced into the sampling line of the SIFT-MS and after a delay of 10 seconds xylene and toluene levels were measured by reaction with the H₃O⁺precursor for a period of 10 seconds. These values were then plotted against the calculated standard concentrations. The best fit linear regression line is shown for both bottled air (solid line) and human breath (dashed line).</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Levels of trace volatile compounds in alveolar and oral air.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Sample</bold></td><td align=\"left\"><bold>Compartment</bold></td><td align=\"left\"><bold>Indole</bold></td><td align=\"left\"><bold>Methylindole</bold></td><td align=\"left\"><bold>Hydrogen Sulphide</bold></td><td align=\"left\"><bold>Methylmercaptan</bold></td></tr></thead><tbody><tr><td align=\"left\">Ambient</td><td align=\"left\">NA</td><td align=\"left\">0.4</td><td align=\"left\">0</td><td align=\"left\">0.4</td><td align=\"left\">4</td></tr><tr><td align=\"left\">Subject 1</td><td align=\"left\">Alveolar</td><td align=\"left\">1.7</td><td align=\"left\">0.4</td><td align=\"left\">10</td><td align=\"left\">10</td></tr><tr><td/><td align=\"left\">Oral</td><td align=\"left\">2.4</td><td align=\"left\">0.8</td><td align=\"left\">48</td><td align=\"left\">71</td></tr><tr><td align=\"left\">Subject 2</td><td align=\"left\">Alveolar</td><td align=\"left\">2.2</td><td align=\"left\">0.4</td><td align=\"left\">9.2</td><td align=\"left\">9.4</td></tr><tr><td/><td align=\"left\">Oral</td><td align=\"left\">3.2</td><td align=\"left\">0.9</td><td align=\"left\">23</td><td align=\"left\">44</td></tr><tr><td align=\"left\">Subject 3</td><td align=\"left\">Alveolar</td><td align=\"left\">1.1</td><td align=\"left\">0.9</td><td align=\"left\">11</td><td align=\"left\">4.3</td></tr><tr><td/><td align=\"left\">Oral</td><td align=\"left\">1.7</td><td align=\"left\">1.2</td><td align=\"left\">41</td><td align=\"left\">38</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Levels of indole, methylindole (all isomers), hydrogen sulphide and methylmercaptan were measured in ambient air, or in a single alveolar breath or static oral air (approximately 10 second measurement duration) of 3 healthy human subjects by SIFT-MS using characteristic products ions of the reaction with H₃O⁺precursors. Values are in PPBV. NA: not applicable.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1756-0500-1-41-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Access to prophylactic cotrimoxazole and to timely initiation of antiretroviral therapy can tremendously improve the survival of HIV infected children [##REF##15555666##1##, ####REF##14593035##2##, ##REF##17806062##3##, ##REF##14551485##4##, ##REF##17443476##5####17443476##5##]. HIV testing is the \"gateway\" or first step to introducing infected children to these interventions[##UREF##0##6##]. Unfortunately, the coverage of HIV testing still remains low in most resource constrained countries, particularly for children, and existing pediatric clinical algorithms have low sensitivity for detecting HIV, particularly among the severely malnourished [##REF##17617274##7##,##REF##16257023##8##]. Thus, most HIV-infected children remain undiagnosed, even in households with an adult already on antiretroviral treatment (ART)[##REF##16885775##9##].</p>", "<p>Reasons for low coverage of pediatric testing are varied but include high postnatal dropout from Prevention of Mother-to-Child Transmission (PMTCT) programmes, and relatively few testing facilities, especially in rural areas [##REF##16452554##10##,##REF##16359404##11##]. In 2005, Malawi had an estimated 91,000 HIV infected children (≤ 14 years) [##REF##17197006##12##]. Among them, the cumulative number on ART by March 2006 was 2718, representing only 3.0% of all infected children and 5.8% of patients on ART in Malawi to date [##REF##17197006##12##].</p>", "<p>The HIV epidemic is contributing to increased child mortality and severe malnutrition throughout Africa [##REF##10824214##13##,##REF##15824540##14##]. In Malawi and elsewhere where HIV is highly prevalent, studies suggest that 30% or more of all children admitted to inpatient nutrition rehabilitation units are HIV-infected [##REF##10824214##13##,##REF##15251404##15##, ####UREF##1##16##, ##REF##9299820##17####9299820##17##]. A prior study in Malawi demonstrated that many HIV-infected children with severe acute malnutrition (SAM) can achieve an adequate Weight-For-Height (WFH) with appropriate therapeutic feeding, although recovery times are significantly longer and mortality is much higher compared to HIV-uninfected children [##REF##15981758##18##]. Unfortunately in Africa, the low coverage of both HIV testing and therapeutic feeding and the multiple factors preventing caregivers bringing children to hospital and remaining with them for extended periods [##REF##17617274##7##,##REF##17141707##19##,##UREF##2##20##] means that the majority of all HIV-infected African children do not receive any nutritional care. Thus, it is important to find innovative ways of identifying HIV-infected children, especially those from remote areas, to allow them to benefit from cotrimoxizole prophylaxis, nutritional care, and ARV treatment in developing countries.</p>", "<p>Community-based Therapeutic Care (CTC) is an approach for managing SAM in children &lt; 5 years that provides care close to where people live and focuses on early detection of SAM through community mobilization [##UREF##3##21##]. Children with appetite and without complications are managed directly in the community in an Outpatient Therapeutic Programme (OTP) and provided with routine medical care and nutrient-dense, pathogen resistant Ready-to-Use Therapeutic Foods (RUTF) for nutritional rehabilitation [##REF##10347999##22##] on a weekly basis. Children with generalized severe oedema, anorexia and medical complications are treated at inpatient facilities according to national protocols, until they are stabilized and appetite has returned and they are able to rejoin the OTP [##REF##12885498##23##,##UREF##4##24##]. Children are discharged from the OTP when they achieve WFH ≥ 80%. Previous evaluations show that CTC achieves high coverage and has lower default and mortality rates compared to traditional therapeutic feeding programs [##REF##12480359##25##,##REF##15817865##26##]. Throughout Africa, home-based care programs are providing support for HIV-affected households. We undertook this study to identify appropriate ways of diagnosing HIV among malnourished children and to assess the feasibility and outcome of treating severely malnourished HIV-infected children in a community based program. This paper presents the uptake of HIV testing when integrated within an ongoing CTC programme, the comparative outcomes of HIV-infected and uninfected malnourished children enrolled in the programme and the diagnostic value of anthropometric and other admission characteristics for identifying children at risk of HIV where diagnostic testing is not available.</p>" ]

[ "<title>Methods</title>", "<p>The study was carried out from December 2002 to May 2005 in the Dowa district of central Malawi, where antenatal and adult HIV prevalence were estimated to be 9.8% and 6.4%, respectively [##REF##14600523##27##,##UREF##5##28##] Voluntary HIV Counselling and Testing (VCT) was offered to caregivers and children who were enrolled or had recently graduated from a CTC programme run by the Ministry of Health and the non-governmental organization Concern Worldwide. Clinical records were reviewed to assess our primary outcomes: HIV testing uptake and test results and indices of nutritional recovery in HIV-infected and uninfected children. Two groups of caregivers were invited to participate: 1) those who were discharged from CTC prior to VCT introduction (retrospective cohort), and 2) those who entered CTC after testing was introduced (prospective cohort).</p>", "<title>Retrospective Cohort (RC)</title>", "<p>Traditional leaders, Health Surveillance Assistants (HSAs), and community volunteers were responsible for locating recent programme graduates using CTC discharge records. Caregivers were visited at home to explain the study objectives and procedures and to invite them to bring their children to the next recruitment day. A verbal autopsy was carried out for children who had died since leaving the programme [##REF##14600523##27##]. Only families of children presently residing in the geographic catchment areas of the 17 health centres providing CTC in Dowa district were contacted. At recruitment, VCT was offered to children and their caregivers. Pre and post-test counselling was carried out by trained nurses and HSAs in accordance with guidelines from the Malawi Ministry of Health for HIV testing and counselling of adults and children &lt; 13 years [##UREF##5##28##].</p>", "<p>Outcomes of interest from the RC include: 1) HIV testing uptake and outcome; 2) nutritional recovery whilst in the CTC programme (weight gain, change in Mid Upper Arm Circumference, time to recovery); and 3) mortality and change in nutritional status from CTC discharge to recruitment. Clinical data were extracted from an electronic database and individual programme monitoring cards.</p>", "<title>Prospective cohort (PC)</title>", "<p>Procedures were similar for the PC children except that VCT was offered at admission to the programme. All PC children were treated according to standard CTC protocols and children whose caregivers chose not to join the study were not treated any differently. Children were recruited into the study from 29/11/2004 to 09/04/2005 corresponding to the 2004/2005 hunger period (~70% of annual admissions are usually observed during the hunger season). The outcomes of interest of the PC were the same than that of the RC with 2 exceptions: 1) we could calculate mortality and programme default (children who left the programme prior to achieving &gt; 85% of median WFH); 2) data on follow up after discharge were unavailable.</p>", "<title>Data collection</title>", "<p>At admission to the study, a clinical assessment and history was obtained for each child by one of three trained nurses using a structured data collection tool. The history included ascertainment of the signs and symptoms used for clinical diagnosis of paediatric HIV in algorithms used by WHO [##REF##14997238##29##,##UREF##6##30##] and Action against Hunger (AAH) [##UREF##7##31##] [Table ##TAB##0##1##] and identification of the presence of any proxy indicator commonly used in Malawi to indirectly identify individuals infected or affected by HIV/AIDS when blood testing it is not possible or appropriate [##UREF##8##32##]. Baseline nutritional assessments included weight and height, MUAC, and presence or absence of bilateral, pitting oedema [##UREF##3##21##,##UREF##9##33##, ####REF##1262113##34##, ##REF##4132501##35####4132501##35##]. Information on the parents' vital status was also obtained.</p>", "<title>Standardisation and quality control of clinical data</title>", "<p>A pilot study to train and standardize the nurses in the collection of clinical data was conducted prior to commencing the RC. The principal researcher (PB), a paediatrician with experience in managing HIV infected children, carried out the training and the standardization process. He also provided daily supervision of the 3 nurses during the first month and twice weekly thereafter. The supervision included examination of 2 randomly selected children per nurse and comparing the paediatrician findings and the nurse findings. Finally, all the data collected by the nurses were checked in the office by the paediatrician for consistency prior to data entry. The 3 trained nurses collected data at enrolment in the RC and admission and follow up data for the PC.</p>", "<title>HIV testing procedures</title>", "<p>HIV sero-status in adults and children &gt; 12 months was ascertained via antibody testing from finger-prick blood samples using a serial algorithm with Determine^®as first test and Unigold^®as a confirmatory test [##REF##15642990##36##,##REF##8823855##37##]. For discordant results, Bioline^®rapid test was used as the tie-breaker. For children &lt; 12 months, whole blood HIV DNA polymerase chain reaction (PCR) testing was used (Roche Amplicor Version 1.5). Children between 12 and 18 months were tested using the same algorithm as for older children, but positive antibody results were confirmed by PCR since maternal antibodies may persist in infant blood for up to 18 months [##REF##18162933##38##,##REF##17825652##39##]. Rapid test results were given to caregivers within one hour of the test. PCR test results were provided within 2 weeks. Caregivers' uptake and HIV test results presented in the present papers concern only biological parents of the child.</p>", "<title>Treatment provided</title>", "<p>Children were treated using standard CTC protocol. Children with SAM without complications were directly admitted into OTP. Children in both cohorts were provided with Vitamin A, de-worming, anaemia treatment, antibiotics for bacterial infections, and malaria prophylaxis according to standard CTC protocols [##UREF##9##33##]. For the RC, children found to be anaemic (haemoglobin &lt; 11 g/dl) received the standard Malawian IMCI anaemia treatment and were referred to the nearest health centre for continuation of treatment and follow up. Children with severe anaemia (Haemoglobin &lt; 6 g/dl) were referred to the district hospital for appropriate management. RUTF (200 kcal/kg/day) was provided as weekly take-home rations. During the RC recruitment a protection ration was given to households of admitted children. No protection ration was given during the PC recruitment. All HIV-positive children were referred to the Lighthouse Project in Lilongwe, which provides comprehensive paediatric HIV care following national guidelines. HIV-positive adults were referred to the Dowa District Hospital Antiretroviral Therapy (ART) clinic.</p>", "<title>Statistical analyses</title>", "<p>Data analyses were carried out using Epi-Info 6.04 [##UREF##10##40##] and SPSS for Windows Version 12 [##UREF##11##41##]. Daily weight gain (g/kg/day) was calculated as the difference between the weight (in grams) at discharge and the weight (in grams) at admission (RC) or lowest weight recorded during participation in CTC (PC), divided by weight at admission (in kilograms) multiplied by number of days in the programme. Changes in MUAC and WFH were determined by calculating the difference between anthropometric measurements at admission and discharge. Means, medians and inter-quartile ranges (IQR) are used to describe and compare continuous parameters. Differences in means were ascertained using Student's t-test and analysis of variance; Kruskall Wallis tests were used to compare medians. Dichotomous variables were compared using χ²analyses and Fisher's exact test. The sensitivity, specificity, positive predictive value and negative predictive value of different proxy indicators and algorithms for diagnosing paediatric HIV infection were also determined using the PC data. The proxy indicators researched are: presence of a chronically ill adult in the household, recent premature death of an adult on the household, female headed household, widow headed household, elderly headed household, child headed household and presence of a tuberculosis infected child or adult[##UREF##8##32##]. Stepwise logistic regression with backward elimination was used to identify predictors of HIV infection (α = 0.05).</p>", "<p>Written informed consent was obtained from all study caregivers, usually the mother. The study protocol was approved by the College of Medicine Research and Ethics Committee in Malawi.</p>" ]

[ "<title>Results</title>", "<p>A total of 2,592 children under 5 years of age had participated in the CTC programme prior to the study. Of these, 809 (31.2%) resided outside the study area and were not eligible for the RC. Of the 1783 possible participants, 180 had moved, 69 had died (according to key informants), 113 could not be located due to an improper address and 1,421 were invited to participate [Figure ##FIG##0##1##]. The cause of death was ascertained for 27 (39%) of the 69 children that died with the main causes being malaria (n = 8), HIV (n = 7), malnutrition (n = 7), and pneumonia (n = 3).</p>", "<p>Of the 1421 children invited to participated, 148 (10.4%) did not turn up. Amongst the returnees, the median time between discharge from the programme and the invitation to participate in the study was 15.6 months (IQR: 10.5–23.3), the average age at CTC admission was 30.0 (17.2) months (median: 26.9; IQR: 17.6 – 37.8) and at study enrolment was 47.2 months (median: 44.3; IQR: 34.4–57. 2). VCT uptake was 92.2% (1174/1273) for children and 58.4% (743/1273) for caregivers. The reasons for refusing the HIV test were: need to consult with husband in 15.2% (15/99) of cases, not psychologically ready in 13.1% (13/99) of cases, fear of the results in 8.1% (8/99), no authority for allowing testing in 7.1% (7/99), not needed because the child is healthy in 7.1%(7/99), already tested in 5.1% (5/99) and others (believes already infected or fear of aggravating child anaemia or not willing to cause pain) in 2.0% of cases (3/99). Thirty-eight caregivers (38.4% of refusals) did not disclose the reasons for refusal.</p>", "<p>735 children were eligible for the PC. The average age at admission was 26.5 (13.7) months (median: 23.0; IQR: 16.9–34.1). VCT uptake was 97% (714/735) for children and 64.1% (471/735) for parents in this cohort. The reasons for refusal were not collected by study design.</p>", "<p>There was no difference in socio-demographic and clinical characteristics at admission between tested and not tested children in either the RC or the PC (data not shown).</p>", "<p>Other baseline characteristics of children in the RC and PC groups are shown in Table ##TAB##1##2##. Family history of tuberculosis (TB) was more common RC (p &lt; 0.01). Only 22 PC children (3.1%) and 29 RC children (2.5%) were HIV-infected (p = 0.45). HIV prevalence was similar amongst parents in both cohorts. Amongst caregivers tested, 5.4% (58/1081) of mothers and 2.9% (3/133) of fathers were HIV-infected giving an overall prevalence amongst caregivers of 5.0% (61/1214).</p>", "<p>In both RC and PC cohorts, there was no significant difference in the sex ratio, age distribution and in frequency of family history of tuberculosis among HIV-infected and HIV-negative children (Table ##TAB##2##3##). HIV-infected children were more likely to be orphaned and to come from a household with at least one HIV proxy indicator (Table ##TAB##2##3##). Nutritional status at admission differed between HIV-infected and HIV-uninfected children (Table ##TAB##3##4##). HIV-infected children were more likely to be admitted with MUAC &lt; 110 mm and less likely to have oedema than uninfected children in both cohorts. Oedematous malnutrition, however, was the most frequent admission characteristic in both infected and uninfected children. An equal proportion of HIV-positive (8/22 = 36.4%) and HIV-negative children (249/692 = 36.0%) required inpatient stabilisation prior to referral to OTP (p = 0.971). About one-third of all HIV-positive children were orphaned (at least one parent dead) compared with &lt; 10% of HIV-negative children (p &lt; 0.001 in both cohorts).</p>", "<p>Nutritional recovery also varied by HIV status (Table ##TAB##3##4##). In the PC, 13 HIV-infected children (59.1%) and 523 HIV-uninfected (83.4%) achieved discharge WFH (p = 0.003). Average rate of weight gain was 4.7 g/kg/day in uninfected children and 2.8 g/kg/d in HIV-positive PC children. Median recovery time to discharge was 42 days for uninfected children and 56 days for HIV-positive children. Five HIV-positive children (22.7%) and 89 HIV-negative children (14.2%) defaulted from the programme (p &lt; 0.001). Four HIV-positive children died during the CTC treatment (18.2%); mortality amongst uninfected children was 1.8% (p &lt; 0.001). Estimated daily weight gain was lower in the RC children, in part due to the high prevalence of oedema at admission and use of admission weight to estimate this parameter.</p>", "<p>Nutritional status at follow-up varied according to HIV status. Approximately 15 months after discharge, 24 out of 28 (85.7%) HIV-infected RC children were not malnourished (WFH &gt; 80% reference median and no bilateral pitting oedema). However, six of these children had a MUAC below 125 mm, including one with a MUAC &lt; 110 mm, giving an overall malnutrition rate of 35.7% (10 out of 28) in HIV-infected children compared with a malnutrition rate of 2.0% (22/1094) in HIV-negative children (p = 0.001).</p>", "<p>The predictive characteristics for variables included in clinical algorithms, for individual proxy indicators, and for the 3 algorithms presently used to diagnose paediatric HIV are shown in Table ##TAB##4##5##. All three of the clinical algorithms had positive likelihood ratio lower than 10 and negative likelihood ratio higher than 0.1 suggesting limited utility for ruling-out and ruling in paediatric HIV. The presence of one or more HIV proxy indicators in combination with MUAC &lt; 110 mm was also a poor predictor of the presence of an HIV infection. In contrast, however, severely malnourished children with a MUAC &gt; 110 mm and no proxy indicators were 10-fold less likely to be HIV-infected, suggesting that this combination may be useful for ruling out HIV where confirmatory paediatric testing is not widely available. The other criteria with high positive likelihood for predicting HIV (i.e., PLR &gt; 10) are having a deceased father or living in a widow-headed household.</p>" ]

[ "<title>Discussion</title>", "<p>The study aims were to assess whether a CTC programme can be used as an entry point for HIV services, including HIV testing and treatment of malnutrition in HIV-positive children, and to compare outcomes of HIV-positive and HIV-negative children within the programme. More than half of the HIV-infected children in the PC (59.1%) recovered to a satisfactory nutritional status using CTC protocols, suggesting that SAM can be managed in the community for many HIV-infected children. Furthermore, about two-thirds of the infected children identified after discharge were still adequately nourished. This nutritional recovery occurred without use of antiretroviral therapy (ART), suggesting that severe malnutrition was primarily the result of semi-starvation among recovering study children. These findings are different from those normally observed in developed and middle income countries where severe malnutrition in HIV-positive children is typically caused by HIV-related metabolic disturbances, which do not improve without ART [##REF##9785358##42##]. Indeed, the poor response to nutritional intervention of cachexia, a form of malnutrition that is primarily due to chronic systemic inflammation is a well-known phenomenon [##REF##9915917##43##,##REF##11033592##44##].</p>", "<p>The 59.1% recovery rate for HIV-infected children observed in the PC arm of our study includes deaths in both the inpatient-based stabilisation phase and the outpatient-based recovery phase of care. This figure is similar to the 56% recovery rate reported in a study in southern Malawi where RUTF was used for Home Therapy (HT) in the recovery phase of care, after patients had been discharged from a hospital Nutrition Rehabilitation Unit (NRU) [##REF##15981758##18##]. As mortality amongst severely malnourished HIV-positive children is usually highest during the initial phase of treatment and may exceed 30% [##UREF##12##45##], the results from our study are encouraging.</p>", "<p>It is possible that our improved recovery rates arise from the decentralised nature of the CTC model of care that is designed to remove barriers to access and promote early presentation before serious complications develop. However, the numbers of HIV subjects in our PC are small and these findings need to be confirmed by larger studies. The results of the present study are also encouraging when compared with the 3-month mortality rate of 42.9% among severely wasted (WFH &lt; 70%) children started on ART, recently reported in Malawi [##REF##17439677##46##]. Mortality in this group was &gt; 10-fold higher than among children starting on ART who were not acutely malnourished (WFH &gt; 80%) [##REF##17439677##46##].</p>", "<p>As observed in other nutritional studies carried out in Malawi, the HIV-positive children in our cohorts recovered more slowly than the HIV-negative children [##REF##15981758##18##,##UREF##12##45##]. The possible reasons for slower weight gain include reduced intake due to poor appetite, nutrient malabsorption, increased incidence of infections that were unresponsive to the broad-spectrum antibiotics used, and increased nutrient requirements due to HIV [##UREF##13##47##]. Despite the possibility of reduced appetite especially at the beginning of treatment, we believe that HIV-positive children may need more RUTF than HIV-negative children to achieve similar growth rates and improvements in other nutritional indices. Increasing the amount of daily energy offered to HIV infected children may improve their weight gain and reduce the length of stay in the program, and further testing of this hypothesis is needed. Continued nutritional surveillance and supplementation after discharge may also help HIV-infected children to remain well-nourished. Reducing recovery time and subsequent length and cost of program participation will reduce default rates, which occurred, on average, after 56 days by families with all HIV-positive children but at 70 days for those who finally defaulted. Similarly, adapting CTC routine antibiotic treatment to the epidemiology of HIV-associated infections and inclusion of routine prophylactic cotrimoxazole for HIV-positive children, as currently recommended by WHO, may improve recovery in this group [##REF##15555666##1##,##REF##11741168##48##].</p>", "<p>The low relapse rates following CTC is in contrast to NRU and other outpatient SAM treatment programmes where morbidity and mortality after discharge are high [##REF##1376587##49##, ####REF##9364128##50##, ##REF##17076212##51####17076212##51##]. Although survival bias cannot be ruled out as an explanation for this finding, it is possible that the CTC design, which uses community mobilization and referral for early identification and treatment of SAM also improves long-term recovery compared to hospital-based treatment programmes. SAM is a progressive condition and prognosis is directly associated with the lead time to presentation and treatment. Initiating nutritional intervention as soon as SAM presents may be especially important for HIV-infected and exposed children.</p>", "<p>The high VCT uptake for adults and children, the low HIV-prevalence amongst SAM children, and the low nutritional relapse rate amongst surviving HIV-positive and uninfected children over a year after discharge are all noteworthy in our study. The high VCT uptake is comparable to that observed recently in some NRUs in Malawi [##UREF##1##16##] but contrasts with anecdotal reports of reluctance to come forward for HIV testing offered by clinics and therapeutic feeding programmes. We believe that the \"opt-out\" approach used in this study, with HIV testing offered to everyone with the right to refuse rather than the standard \"opt-in\" approach where people have to specifically request HIV testing, contributed to the high uptake seen here [##REF##16359404##11##,##REF##17825654##52##,##REF##18090395##53##]. We also believe that offering testing through a programme such as CTC that is well established in the community improves trust and reduces the fear of stigmatization. Caregivers were informed about the opportunity for HIV testing one week prior to travelling to the health centre and no substantial compensation was offered, ruling out the likelihood of coercion in the RC. The high uptake observed here suggests that CTC is a potentially innovative way to increase access to and coverage of HIV testing, particularly in rural areas [##UREF##14##54##]. It is important to note however that CTC would have to be combined with other community based HIV testing and counselling approaches like the home based and mobile VCT in order to obtain good coverage; in prior studies we observed that only 16% of HIV-affected households has a malnourished child treated in the CTC programme in the past 18 months [##REF##16885775##9##,##REF##17943913##55##,##REF##15746219##56##].</p>", "<p>There are a number of factors that are likely to have contributed to the low prevalence of HIV amongst severely malnourished children in this study. Chronic food insecurity, frequent common childhood illnesses, poor access to modern health care and suboptimal complementary feeding practices all cause SAM in the absence of HIV in Malawi [##REF##14629321##57##, ####REF##17414147##58##, ##REF##16306929##59####16306929##59##]. The decentralised nature and the high coverage rates obtained by CTC programmes means that a higher proportion of admissions live in remote rural areas far from towns and main roads [##REF##17466097##60##] in contrast to admissions in more centrally located urban hospital units. People from rural areas are all subsistence farmers, have poor income and low educational level and have no possibility for travelling within or outside the country. These factors are known to increase the prevalence of malnutrition and lower that of HIV [##REF##14629321##57##,##REF##12891068##61##, ####REF##8418181##62##, ##REF##11600832##63##, ##REF##12380885##64####12380885##64##]. Lastly, the low HIV prevalence might also be explained by the high mortality of HIV infected infants biasing our data to include those children who have survive infancy. Without effective treatment, it is estimated that over 50% of infants who acquired HIV infection through mother-to-child transmission will die by two years of age (compared to 8% of uninfected children) while in Malawi kwashiorkor, the most common form of SAM in children, peaks between 18 to 23 months of age [##REF##15464184##65##, ####REF##8924243##66##, ##REF##11099620##67##, ##REF##8534041##68####8534041##68##]. The relatively older average ages of children in both the PC (26.4 months) and RC cohorts (47.2 months), suggests the possibility of survival bias.</p>", "<p>Our estimated adult HIV prevalence of 5.0% is predictably lower than the antenatal HIV prevalence rate in the central region (9.8%) but similar to comparable adult prevalence rates of 6.4% and 4.1% for the region and for an adjacent district, respectively, as reported in the 2004 Malawi Demographic and Health Survey [##UREF##15##69##,##UREF##16##70##]. Our study suggests that targeting adult caretakers of malnourished children for HIV testing is feasible but additional outreach and counseling efforts may be needed to increase uptake.</p>", "<p>Our analysis confirms that clinical algorithms designed to diagnose paediatric HIV are neither sufficiently sensitive nor specific in severely malnourished children and that blood tests are therefore required to confirm the diagnosis [##REF##16257023##71##]. In context where blood tests are not available, the combination of MUAC &gt; 110 mm and absence of proxy indicators can be used to rule out the presence of HIV. These family history variables could be incorporated into future CTC protocols for confirmed or suspected paediatric HIV in setting without possibility of blood tests [##REF##11964422##72##].</p>", "<p>SAM is one of several criteria for initiating ART in HIV-positive children [##UREF##17##73##]. The fact that some previously undiagnosed HIV-positive children recovered from malnutrition, and were still healthy and asymptomatic an average of 15 months after discharge from CTC without ART, suggests that the presence of malnutrition should not be the sole criteria for initiating ART in food insecure settings. One possibility is that initiation of ART could be reserved for children who do not respond to CTC or at least could be delayed until nutrition improvement to minimize antiretroviral side effects. Several studies have reported that HIV-infected children tend to develop marasmus rather than kwashiorkor and that CD4 count remains higher in HIV-infected children with kwashiorkor compared to those with marasmus [##REF##9299820##17##,##REF##17042940##74##,##REF##10477405##75##]. Therefore, delayed initiation of ART could be considered for children with kwashiorkor. This approach could help to prevent unnecessary exposure to ARV drugs that have side effects and toxicity and to reduce the risk of developing resistance [##REF##17693883##76##]. Further research, probably in the form of randomized trials, is urgently needed to strengthen this evidence base before any change of practice is recommended.</p>", "<p>Several limitations of our study deserve mention. This study was carried out in conjunction with an ongoing CTC programme and clinical records were reviewed in order to obtain data on nutritional recovery. Although programme procedures were standardized, we were unable to verify nutritional measurements for accuracy. As noted previously, the RC may be subject to survival bias, and therefore we have limited our use and interpretation of the RC data. The statistical power of these analyses is also limited by the small number of HIV-positive children in the study and by further reduction of the sample size due to missing data for age and nutritional status. Nevertheless, the data from both cohorts paint a consistent picture with regard to the potential positive impact of Community-based Therapeutic Care for managing SAM in HIV-positive and uninfected children in rural Africa.</p>" ]

[ "<title>Conclusion</title>", "<p>The results of the present study demonstrate that CTC is a valuable entry point for HIV testing for severely malnourished children and that good recovery rates can be achieved in HIV-infected severely malnourished children admitted to the program. These results indicate that CTC can be used to improve the coverage of HIV services, especially in rural areas. The approach has several important advantages over traditional inpatient therapeutic care, including earlier intervention, greater coverage, and increased accessibility. All of these characteristics are particularly important for providing timely care to HIV-exposed and vulnerable children. Additional research on feeding protocols for HIV-infected children and on timing of ART initiation are needed to refine CTC protocols in settings where HIV is common.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>In Malawi and other high HIV prevalence countries, studies suggest that more than 30% of all severely malnourished children admitted to inpatient nutrition rehabilitation units are HIV-infected. However, clinical algorithms designed to diagnose paediatric HIV are neither sensitive nor specific in severely malnourished children. The present study was conducted to assess : i) whether HIV testing can be integrated into Community-based Therapeutic Care (CTC); ii) to determine if CTC can improve the identification of HIV infected children; and iii) to assess the impact of CTC programmes on the rehabilitation of HIV-infected children with Severe Acute Malnutrition (SAM).</p>", "<title>Methods</title>", "<p>This community-based cohort study was conducted in Dowa District, Central Malawi, a rural area 50 km from the capital, Lilongwe. Caregivers and children admitted in the Dowa CTC programme were prospectively (Prospective Cohort = PC) and retrospectively (Retrospective Cohort = RC) admitted into the study and offered HIV testing and counseling. Basic medical care and community nutrition rehabilitation was provided for children with SAM. The outcomes of interest were uptake of HIV testing, and recovery, relapse, and growth rates of HIV-positive and uninfected children in the CTC programme. Student's t-test and analysis of variance were used to compare means and Kruskall Wallis tests were used to compare medians. Dichotomous variables were compared using Chi²analyses and Fisher's exact test. Stepwise logistic regression with backward elimination was used to identify predictors of HIV infection (α = 0.05).</p>", "<title>Results</title>", "<p>1273 and 735 children were enrolled in the RC and PC. For the RC, the average age (SD) at CTC admission was 30.0 (17.2) months. For the PC, the average age at admission was 26.5 (13.7) months. Overall uptake of HIV testing was 60.7% for parents and 94% for children. HIV prevalence in severely malnourished children was 3%, much lower than anticipated. 59% of HIV-positive and 83% of HIV-negative children achieved discharge Weight-For-Height (WFH) ≥ 80% of the NCHS reference median (p = 0.003). Clinical algorithms for diagnosing HIV in SAM children had poor sensitivity and specificity.</p>", "<title>Conclusion</title>", "<p>CTC is a potentially valuable entry point for providing HIV testing and care in the community to HIV infected children with SAM.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors Paluku Bahwere, Kate Sadler, Saul Guerrero and Steve Collins work for Valid International, an organisation that has been engaged in the research and development of Community-based Therapeutic Care. Dr Steve Collins is also unpaid director of Valid Nutrition, a not-for-profit company established to research and manufacture ready-to-use therapeutic food in developing countries. There is no conflicting interest for all the other authors.</p>", "<title>Authors' contributions</title>", "<p>PB: concept and design of the research, data collection, data analysis and interpretation, drafting of the paper. EP: concept and design of the research, critical revision of paper content. MCJ: data collection, analysis of data and revision of the manusdcript. KS: concept and design of the research, data analysis and interpretation, revision of the draft of the paper. CHGT: concept and design of the research, critical revision of paper content. SG: design and concept of the research, data collection and revision of the manuscript. SC: concept and design of the research, data analysis and interpretation, revision of the draft of the paper. All authors read and approved the final manuscript.</p>", "<title>Pre-publication history</title>", "<p>The pre-publication history for this paper can be accessed here:</p>", "<p><ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2334/8/106/prepub\"/></p>" ]

[ "<title>Acknowledgements</title>", "<p>The authors would like to acknowledge the invaluable assistance of the Ministry of Health clinic staff in Dowa District and all the CTC programme beneficiaries and their families. Funding for this paper was provided by the Bureau for Africa, Office of Sustainable Development of the United States Agency for International Development (USAID) under the terms of Contract AOT-C-00-99-00237-00 and Food and Nutrition Technical Assistance (FANTA).</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Description of the retrospective cohort.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Clinical algorithms for diagnosing paediatric HIV</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Algorithm</bold></td><td align=\"center\" colspan=\"2\"><bold>Variables</bold></td><td align=\"left\"><bold>Diagnostic Criteria</bold></td></tr></thead><tbody><tr><td/><td align=\"left\"><bold>History</bold></td><td align=\"left\"><bold>Physical findings</bold></td><td/></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><bold>Original IMCI† Algorithm for Paediatric HIV</bold></td><td align=\"left\">• Chest infection requiring hospital admission in the past 3 months</td><td align=\"left\">• Weight below 3^rdcentile</td><td align=\"left\">• Classify as suspected symptomatic HIV infection if 3 positive findings</td></tr><tr><td/><td align=\"left\">• &gt; = 2 episodes of diarrhea in the past 3 months</td><td align=\"left\">• Poor weight gain (growth monitoring card)</td><td/></tr><tr><td/><td align=\"left\">• Episode of persistent diarrhea (lasting 14 days) in the past 3 months</td><td align=\"left\">• Any enlarged lymph glands in more than one of the following sites: neck, axillary or groin</td><td/></tr><tr><td/><td align=\"left\">• Fever &gt; = 1 month</td><td align=\"left\">• Oral thrush extending to the back of the mouth or throat</td><td/></tr><tr><td/><td align=\"left\">• Poor appetite</td><td/><td/></tr><tr><td/><td align=\"left\">• Chronic ear infection (14 days)</td><td/><td/></tr><tr><td/><td align=\"left\">• History or evidence of past or present herpes zoster</td><td/><td/></tr><tr><td/><td align=\"left\">• History or evidence of severe seborrheic dermatitis</td><td/><td/></tr><tr><td/><td align=\"left\">• History of past or present TB</td><td/><td/></tr><tr><td/><td align=\"left\">• Parent or sibling known to have TB</td><td/><td/></tr><tr><td/><td align=\"left\">• Parent or sibling known to be HIV-positive</td><td/><td/></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><bold>Improved IMCI Algorithm (South Africa adaptation)</bold></td><td align=\"left\">• Pneumonia today§</td><td align=\"left\">• Weight below 3^rdcentile</td><td align=\"left\">• Classify as suspected symptomatic HIV infection if 3 positive findings</td></tr><tr><td/><td align=\"left\">• History of weight loss¶</td><td align=\"left\">• Poor weight gain (history or RTH card)</td><td/></tr><tr><td/><td align=\"left\">• Persistent diarrhea now or in past 3 months</td><td align=\"left\">• Any enlarged lymph glands in more than one of the following; neck, axillae or groin</td><td/></tr><tr><td/><td align=\"left\">• Ear discharge now or in the past</td><td align=\"left\">• Oral thrush</td><td/></tr><tr><td/><td/><td align=\"left\">• Parotid swelling</td><td/></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><bold>Action Against Hunger Algorithm</bold></td><td align=\"left\">• Ear discharge now or in the past</td><td/><td align=\"left\">• When child presents with &gt; = 3 criteria refer for further HIV support/care and health education</td></tr><tr><td/><td align=\"left\">• Enlarged lymph glands now</td><td/><td/></tr><tr><td/><td align=\"left\">• Pneumonia today or persistent cough &gt; 1 month</td><td/><td/></tr><tr><td/><td align=\"left\">• Persistent diarrhea</td><td/><td/></tr><tr><td/><td align=\"left\">• Low weight gain¶</td><td/><td/></tr><tr><td/><td align=\"left\">• Oral thrush</td><td/><td/></tr><tr><td/><td align=\"left\">• Marasmus or Marasmus/Kwashiokor</td><td/><td/></tr><tr><td/><td align=\"left\">• Fever for more than one month</td><td/><td/></tr><tr><td/><td align=\"left\">• Child is an orphan (one or both parents)</td><td/><td/></tr><tr><td/><td align=\"left\">• Child's parents are sick or one of siblings has died</td><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Baseline characteristics of participants in the retrospective and prospective cohorts</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"3\"><bold>Prospective</bold></td><td align=\"center\" colspan=\"3\"><bold>Retrospective</bold></td><td align=\"center\" colspan=\"2\"><bold>Total</bold></td></tr></thead><tbody><tr><td/><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td align=\"left\"><bold>mean (SD)</bold></td><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td align=\"left\"><bold>mean(SD)</bold></td><td align=\"right\"><bold>N</bold></td><td align=\"right\"><bold>%</bold></td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td align=\"left\"><bold>VCT uptake</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Accepted VCT</td><td align=\"right\">714</td><td align=\"right\">97.1</td><td/><td align=\"right\">1174</td><td align=\"right\">92.2</td><td/><td align=\"right\">1888</td><td align=\"right\">94.0</td></tr><tr><td align=\"left\">Refused VCT</td><td align=\"right\">21</td><td align=\"right\">2.9</td><td/><td align=\"right\">99</td><td align=\"right\">7.8</td><td/><td align=\"right\">120</td><td align=\"right\">6.0</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">735</td><td align=\"right\">100.0</td><td/><td align=\"right\">1273</td><td align=\"right\">100.0</td><td/><td align=\"right\">2008</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>Parents testing uptake</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Parent and child tested</td><td align=\"right\">471</td><td align=\"right\">66.0</td><td/><td align=\"right\">743</td><td align=\"right\">63.3</td><td/><td align=\"right\">1214</td><td align=\"right\">64.3</td></tr><tr><td align=\"left\">Only child tested</td><td align=\"right\">243</td><td align=\"right\">34.0</td><td/><td align=\"right\">431</td><td align=\"right\">36.7</td><td/><td align=\"right\">674</td><td align=\"right\">35.7</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\"><bold>714</bold></td><td align=\"right\"><bold>100.0</bold></td><td/><td align=\"right\"><bold>1174</bold></td><td align=\"right\"><bold>100.0</bold></td><td/><td align=\"right\"><bold>1888</bold></td><td align=\"right\"><bold>100.0</bold></td></tr><tr><td align=\"left\"><bold>Parents' vital status</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Orphan (1 or both parents dead)</td><td align=\"right\">35</td><td align=\"right\">5.0</td><td/><td align=\"right\">100</td><td align=\"right\">8.6</td><td/><td align=\"right\">135</td><td align=\"right\">7.3</td></tr><tr><td align=\"left\">Both parents alive</td><td align=\"right\">661</td><td align=\"right\">95.0</td><td/><td align=\"right\">1065</td><td align=\"right\">91.4</td><td/><td align=\"right\">1726</td><td align=\"right\">92.7</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">696</td><td align=\"right\">100.0</td><td/><td align=\"right\">1165</td><td align=\"right\">100.0</td><td/><td align=\"right\">1861</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>Family history of TB</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Yes</td><td align=\"right\">65</td><td align=\"right\">9.1</td><td/><td align=\"right\">127</td><td align=\"right\">19.1</td><td/><td align=\"right\">192</td><td align=\"right\">13.9</td></tr><tr><td align=\"left\">No</td><td align=\"right\">648</td><td align=\"right\">90.9</td><td/><td align=\"right\">539</td><td align=\"right\">80.9</td><td/><td align=\"right\">1187</td><td align=\"right\">86.1</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">713</td><td align=\"right\">100.0</td><td/><td align=\"right\">666</td><td align=\"right\">100.0</td><td/><td align=\"right\">1379</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>Sex</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Female</td><td align=\"right\">353</td><td align=\"right\">49.4</td><td/><td align=\"right\">595</td><td align=\"right\">52.4</td><td/><td align=\"right\">948</td><td align=\"right\">51.3</td></tr><tr><td align=\"left\">Male</td><td align=\"right\">361</td><td align=\"right\">50.6</td><td/><td align=\"right\">540</td><td align=\"right\">47.6</td><td/><td align=\"right\">901</td><td align=\"right\">48.7</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">714</td><td align=\"right\">100.0</td><td/><td align=\"right\">1135</td><td align=\"right\">100.0</td><td/><td align=\"right\">1849</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>Age (months)</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">&lt; 12</td><td align=\"right\">62</td><td align=\"right\">10.5</td><td/><td align=\"right\">157</td><td align=\"right\">16.8</td><td/><td align=\"right\">219</td><td align=\"right\">14.4</td></tr><tr><td align=\"left\">12-&lt;24</td><td align=\"right\">272</td><td align=\"right\">46.1</td><td/><td align=\"right\">354</td><td align=\"right\">37.9</td><td/><td align=\"right\">626</td><td align=\"right\">41.1</td></tr><tr><td align=\"left\">24-&lt;36</td><td align=\"right\">153</td><td align=\"right\">25.9</td><td/><td align=\"right\">275</td><td align=\"right\">29.5</td><td/><td align=\"right\">428</td><td align=\"right\">28.1</td></tr><tr><td align=\"left\">&gt; = 36</td><td align=\"right\">103</td><td align=\"right\">17.5</td><td/><td align=\"right\">147</td><td align=\"right\">15.8</td><td/><td align=\"right\">250</td><td align=\"right\">16.4</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">590</td><td align=\"right\">100.0</td><td/><td align=\"right\">933</td><td align=\"right\">100.0</td><td/><td align=\"right\">1523</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>Admission category</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Oedema</td><td align=\"right\">597</td><td align=\"right\">83.7</td><td/><td align=\"right\">659</td><td align=\"right\">67.4</td><td/><td align=\"right\">1256</td><td align=\"right\">74.3</td></tr><tr><td align=\"left\">Maramus</td><td align=\"right\">51</td><td align=\"right\">7.2</td><td/><td align=\"right\">160</td><td align=\"right\">16.4</td><td/><td align=\"right\">211</td><td align=\"right\">12.5</td></tr><tr><td align=\"left\">MUAC &lt; 110 mm</td><td align=\"right\">46</td><td align=\"right\">6.5</td><td/><td align=\"right\">75</td><td align=\"right\">7.7</td><td/><td align=\"right\">121</td><td align=\"right\">7.2</td></tr><tr><td align=\"left\">Other criteria</td><td align=\"right\">19</td><td align=\"right\">2.7</td><td/><td align=\"right\">84</td><td align=\"right\">8.6</td><td/><td align=\"right\">103</td><td align=\"right\">6.1</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">713</td><td align=\"right\">100.0</td><td/><td align=\"right\">978</td><td align=\"right\">100.0</td><td/><td align=\"right\">1691</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>Presence of oedema</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Yes</td><td align=\"right\">597</td><td align=\"right\">83.7</td><td/><td align=\"right\">659</td><td align=\"right\">67.4</td><td/><td align=\"right\">1256</td><td align=\"right\">74.3</td></tr><tr><td align=\"left\">No</td><td align=\"right\">116</td><td align=\"right\">16.3</td><td/><td align=\"right\">319</td><td align=\"right\">32.6</td><td/><td align=\"right\">435</td><td align=\"right\">25.7</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">713</td><td align=\"right\">100.0</td><td/><td align=\"right\">978</td><td align=\"right\">100.0</td><td/><td align=\"right\">1691</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>WHM % (SD)</bold></td><td/><td/><td align=\"left\">83.3(13.0)</td><td/><td/><td align=\"left\">84.2(14.4)</td><td/><td/></tr><tr><td align=\"left\">&gt; = 80</td><td align=\"right\">397</td><td align=\"right\">58.7</td><td/><td align=\"right\">552</td><td align=\"right\">59.9</td><td/><td align=\"right\">949</td><td align=\"right\">59.4</td></tr><tr><td align=\"left\">70 to &lt; 80%</td><td align=\"right\">187</td><td align=\"right\">27.6</td><td/><td align=\"right\">186</td><td align=\"right\">20.2</td><td/><td align=\"right\">373</td><td align=\"right\">23.3</td></tr><tr><td align=\"left\">&lt; 70%</td><td align=\"right\">93</td><td align=\"right\">13.7</td><td/><td align=\"right\">183</td><td align=\"right\">19.9</td><td/><td align=\"right\">276</td><td align=\"right\">17.3</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">677</td><td align=\"right\">100.0</td><td/><td align=\"right\">921</td><td align=\"right\">100.0</td><td/><td align=\"right\">1598</td><td align=\"right\">100.0</td></tr><tr><td align=\"left\"><bold>MUAC (SD)</bold></td><td/><td/><td align=\"left\">118.0(16.9)</td><td/><td/><td align=\"left\">119.2(18.9)</td><td/><td/></tr><tr><td align=\"left\">&gt; = 125 mm</td><td align=\"right\">219</td><td align=\"right\">34.9</td><td/><td align=\"right\">326</td><td align=\"right\">42.0</td><td/><td align=\"right\">545</td><td align=\"right\">38.8</td></tr><tr><td align=\"left\">110 to &lt; 125 mm</td><td align=\"right\">223</td><td align=\"right\">35.5</td><td/><td align=\"right\">208</td><td align=\"right\">26.8</td><td/><td align=\"right\">431</td><td align=\"right\">30.7</td></tr><tr><td align=\"left\">&lt; 110 mm</td><td align=\"right\">186</td><td align=\"right\">29.6</td><td/><td align=\"right\">242</td><td align=\"right\">31.2</td><td/><td align=\"right\">428</td><td align=\"right\">30.5</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">628</td><td align=\"right\">100.0</td><td/><td align=\"right\">776</td><td align=\"right\">100.0</td><td/><td align=\"right\">1404</td><td align=\"right\">100.0</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Demographic characteristics, tuberculosis history and nutrition admission criteria according to the HIV status of children from retrospective and prospective cohorts.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"4\"><bold>Prospective</bold></td><td align=\"left\"><bold>p-value</bold></td><td align=\"center\" colspan=\"4\"><bold>Retrospective</bold></td><td align=\"center\"><bold>p-value</bold></td></tr><tr><td/><td colspan=\"4\"><hr/></td><td/><td colspan=\"4\"><hr/></td><td/></tr><tr><td/><td align=\"center\" colspan=\"2\"><bold>HIV+ve</bold></td><td align=\"center\" colspan=\"2\"><bold>HIV-ve</bold></td><td/><td align=\"center\" colspan=\"2\"><bold>HIV+ve</bold></td><td align=\"center\" colspan=\"2\"><bold>HIV-ve</bold></td><td/></tr><tr><td align=\"left\"><bold>Variable</bold></td><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td/><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td/></tr></thead><tbody><tr><td align=\"left\"><bold>Sex</bold></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Female</td><td align=\"right\">12</td><td align=\"right\">54.5</td><td align=\"right\">341</td><td align=\"right\">49.3</td><td/><td align=\"right\">15</td><td align=\"right\">53.6</td><td align=\"right\">580</td><td align=\"right\">52.4</td><td/></tr><tr><td align=\"left\">Male</td><td align=\"right\">10</td><td align=\"right\">45.5</td><td align=\"right\">351</td><td align=\"right\">50.7</td><td align=\"left\">0.626</td><td align=\"right\">13</td><td align=\"right\">46.4</td><td align=\"right\">527</td><td align=\"right\">47.6</td><td align=\"center\">0.902</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">22</td><td align=\"right\">100.0</td><td align=\"right\">692</td><td align=\"right\">100.0</td><td/><td align=\"right\">28</td><td align=\"right\">100.0</td><td align=\"right\">1107</td><td align=\"right\">100.0</td><td/></tr><tr><td align=\"left\"><bold>Age</bold></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">&lt; 12</td><td align=\"right\">5</td><td align=\"right\">25.0</td><td align=\"right\">57</td><td align=\"right\">10.0</td><td/><td align=\"right\">4</td><td align=\"right\">16.0</td><td align=\"right\">153</td><td align=\"right\">16.9</td><td/></tr><tr><td align=\"left\">12-&lt;24</td><td align=\"right\">6</td><td align=\"right\">30.0</td><td align=\"right\">266</td><td align=\"right\">46.7</td><td/><td align=\"right\">9</td><td align=\"right\">36.0</td><td align=\"right\">345</td><td align=\"right\">38.0</td><td/></tr><tr><td align=\"left\">24-&lt;36</td><td align=\"right\">4</td><td align=\"right\">20.0</td><td align=\"right\">149</td><td align=\"right\">26.1</td><td align=\"left\">0.099</td><td align=\"right\">8</td><td align=\"right\">32.0</td><td align=\"right\">267</td><td align=\"right\">29.4</td><td align=\"center\">0.993</td></tr><tr><td align=\"left\">&gt; = 36</td><td align=\"right\">5</td><td align=\"right\">25.0</td><td align=\"right\">98</td><td align=\"right\">17.2</td><td/><td align=\"right\">4</td><td align=\"right\">16.0</td><td align=\"right\">143</td><td align=\"right\">15.7</td><td/></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">20</td><td align=\"right\">100.0</td><td align=\"right\">570</td><td align=\"right\">100.0</td><td/><td align=\"right\">25</td><td align=\"right\">100.0</td><td align=\"right\">908</td><td align=\"right\">100.0</td><td/></tr><tr><td align=\"left\"><bold>Parents status</bold></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Orphan (1 or both parents)</td><td align=\"right\">7</td><td align=\"right\">31.8</td><td align=\"right\">28</td><td align=\"right\">4.2</td><td/><td align=\"right\">10</td><td align=\"right\">34.5</td><td align=\"right\">90</td><td align=\"right\">7.9</td><td/></tr><tr><td align=\"left\">Parents alive</td><td align=\"right\">15</td><td align=\"right\">68.2</td><td align=\"right\">646</td><td align=\"right\">95.8</td><td align=\"left\">&lt; 0.001</td><td align=\"right\">19</td><td align=\"right\">65.5</td><td align=\"right\">1046</td><td align=\"right\">92.1</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">22</td><td align=\"right\">100.0</td><td align=\"right\">674</td><td align=\"right\">100.0</td><td/><td align=\"right\">29</td><td align=\"right\">100.0</td><td align=\"right\">1136</td><td align=\"right\">100.0</td><td/></tr><tr><td align=\"left\"><bold>Presence of proxy indicator</bold></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">&gt; = 1 proxy present</td><td align=\"right\">14</td><td align=\"right\">66.7</td><td align=\"right\">138</td><td align=\"right\">21.5</td><td/><td align=\"right\">15</td><td align=\"right\">53.6</td><td align=\"right\">261</td><td align=\"right\">23.9</td><td/></tr><tr><td align=\"left\">None</td><td align=\"right\">7</td><td align=\"right\">33.3</td><td align=\"right\">505</td><td align=\"right\">78.5</td><td align=\"left\">&lt; 0.001</td><td align=\"right\">13</td><td align=\"right\">46.4</td><td align=\"right\">832</td><td align=\"right\">76.1</td><td align=\"center\">&lt; 0.001</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">21</td><td align=\"right\">100.0</td><td align=\"right\">643</td><td align=\"right\">100.0</td><td/><td align=\"right\">28</td><td align=\"right\">100.0</td><td align=\"right\">1093</td><td align=\"right\">100.0</td><td/></tr><tr><td align=\"left\"><bold>Family history of tuberculosis</bold></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Yes</td><td align=\"right\">4</td><td align=\"right\">19.0</td><td align=\"right\">54</td><td align=\"right\">8.0</td><td/><td align=\"right\">7</td><td align=\"right\">29.2</td><td align=\"right\">120</td><td align=\"right\">18.7</td><td/></tr><tr><td align=\"left\">No</td><td align=\"right\">17</td><td align=\"right\">81.0</td><td align=\"right\">621</td><td align=\"right\">92.0</td><td align=\"left\">0.089</td><td align=\"right\">17</td><td align=\"right\">70.8</td><td align=\"right\">522</td><td align=\"right\">81.3</td><td align=\"center\">0.193</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">21</td><td/><td align=\"right\">675</td><td align=\"right\">100.0</td><td/><td align=\"right\">24</td><td align=\"right\">100.0</td><td align=\"right\">642</td><td align=\"right\">100.0</td><td/></tr><tr><td align=\"left\"><bold>Admission category</bold></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">oedema</td><td align=\"right\">14</td><td align=\"right\">63.6</td><td align=\"right\">583</td><td align=\"right\">84.4</td><td/><td align=\"right\">16</td><td align=\"right\">55.2</td><td align=\"right\">643</td><td align=\"right\">67.8</td><td/></tr><tr><td align=\"left\">No oedema</td><td align=\"right\">8</td><td align=\"right\">36.4</td><td align=\"right\">108</td><td align=\"right\">15.6</td><td align=\"left\">0.017</td><td align=\"right\">13</td><td align=\"right\">44.8</td><td align=\"right\">306</td><td align=\"right\">32.2</td><td align=\"center\">0.017</td></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">22</td><td align=\"right\">100.0</td><td align=\"right\">691</td><td align=\"right\">100.0</td><td/><td align=\"right\">29</td><td align=\"right\">100.0</td><td align=\"right\">949</td><td align=\"right\">100.0</td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Nutritional status at enrolment and the impact of CTC in HIV-positive and HIV-negative children</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"3\"><bold>HIV positive</bold></td><td align=\"center\" colspan=\"3\"><bold>HIV negative</bold></td><td align=\"right\"><bold>P-value</bold></td></tr><tr><td/><td colspan=\"3\"><hr/></td><td colspan=\"3\"><hr/></td><td colspan=\"1\"><hr/></td></tr><tr><td/><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td/><td align=\"right\"><bold>n</bold></td><td align=\"right\"><bold>%</bold></td><td/><td/></tr></thead><tbody><tr><td align=\"left\"><bold>Prospective cohort</bold></td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"><bold>Admission category</bold></td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Oedema</td><td align=\"right\">14</td><td align=\"right\">63.6</td><td/><td align=\"right\">583</td><td align=\"right\">84.4</td><td/><td align=\"right\">0.017†</td></tr><tr><td align=\"left\">Maramus</td><td align=\"right\">1</td><td align=\"right\">4.5</td><td/><td align=\"right\">50</td><td align=\"right\">7.2</td><td/><td/></tr><tr><td align=\"left\">MUAC &lt; 110 mm</td><td align=\"right\">5</td><td align=\"right\">22.7</td><td/><td align=\"right\">41</td><td align=\"right\">5.9</td><td/><td/></tr><tr><td align=\"left\">Others criteria</td><td align=\"right\">2</td><td align=\"right\">9.2</td><td/><td align=\"right\">17</td><td align=\"right\">2.5</td><td/><td/></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">22</td><td align=\"right\">100.0</td><td/><td align=\"right\">691</td><td align=\"right\">100.0</td><td/><td/></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\"><bold>WHM% mean(SD)</bold></td><td/><td/><td align=\"left\">80.5(8.9)</td><td/><td/><td align=\"left\">83.3(12.6)</td><td align=\"right\">0.3</td></tr><tr><td align=\"left\">&gt; = 70%</td><td align=\"right\">20</td><td align=\"right\">90.9</td><td/><td align=\"right\">564</td><td align=\"right\">86.1</td><td/><td align=\"right\">0.755</td></tr><tr><td align=\"left\">&lt; 70%</td><td align=\"right\">2</td><td align=\"right\">9.1</td><td/><td align=\"right\">91</td><td align=\"right\">13.9</td><td/><td/></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">22</td><td align=\"right\">100.0</td><td/><td align=\"right\">655</td><td align=\"right\">100.0</td><td/><td/></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\"><bold>MUAC mean (SD)</bold></td><td/><td/><td align=\"left\">109.2(16.4)</td><td/><td/><td align=\"left\">118.3(16.9)</td><td align=\"right\">0.025</td></tr><tr><td align=\"left\">&gt; = 110 mm</td><td align=\"right\">7</td><td align=\"right\">38.9</td><td/><td align=\"right\">435</td><td align=\"right\">71.3</td><td/><td align=\"right\">0.003</td></tr><tr><td align=\"left\">&lt; 110 mm</td><td align=\"right\">11</td><td align=\"right\">61.1</td><td/><td align=\"right\">175</td><td align=\"right\">28.7</td><td/><td/></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">18</td><td align=\"right\">100.0</td><td/><td align=\"right\">610</td><td align=\"right\">100.0</td><td/><td/></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\"><bold>CTC Outcomes</bold></td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Recovered</td><td align=\"right\">13</td><td align=\"right\">59.1</td><td/><td align=\"right\">523</td><td align=\"right\">83.4</td><td/><td align=\"right\">0.002</td></tr><tr><td align=\"left\">Defaulted</td><td align=\"right\">5</td><td align=\"right\">22.7</td><td/><td align=\"right\">89</td><td align=\"right\">14.2</td><td/><td align=\"right\">&lt; 0.001</td></tr><tr><td align=\"left\">Died</td><td align=\"right\">4</td><td align=\"right\">18.2</td><td/><td align=\"right\">11</td><td align=\"right\">1.8</td><td/><td align=\"right\">&lt; 0.001</td></tr><tr><td align=\"left\">Transfer or still in programme</td><td/><td/><td/><td align=\"right\">4</td><td align=\"right\">0.7</td><td/><td align=\"right\">-</td></tr><tr><td align=\"left\">Median weight gain (IQR) <italic>g/kg/day</italic></td><td align=\"right\">20</td><td/><td align=\"left\">2.8 (1.3–3.9)</td><td align=\"right\">614</td><td/><td align=\"left\">4.7(2.9–6.7)</td><td align=\"right\">0.007</td></tr><tr><td align=\"left\">Median MUAC change (IQR) <italic>mm/day</italic></td><td align=\"right\">9</td><td/><td align=\"left\">0.11(-0.03–0.31)</td><td align=\"right\">361</td><td/><td align=\"left\">0.21(0.05–0.39)</td><td align=\"right\">0.223</td></tr><tr><td align=\"left\">Median LoS (IQR) <italic>days</italic></td><td align=\"right\">20</td><td/><td align=\"left\">56(36–68)</td><td align=\"right\">622</td><td/><td align=\"left\">42(28–63)</td><td align=\"right\">0.25</td></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\"><bold>Retrospective cohort</bold></td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"><bold>Admission category</bold></td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Oedema</td><td align=\"right\">16</td><td align=\"right\">55.2</td><td/><td align=\"right\">643</td><td align=\"right\">67.8</td><td/><td align=\"right\">0.154†</td></tr><tr><td align=\"left\">Marasmus</td><td align=\"right\">6</td><td align=\"right\">20.7</td><td/><td align=\"right\">154</td><td align=\"right\">16.2</td><td/><td/></tr><tr><td align=\"left\">MUAC &lt; 110 mm</td><td align=\"right\">5</td><td align=\"right\">17.2</td><td/><td align=\"right\">70</td><td align=\"right\">7.4</td><td/><td/></tr><tr><td align=\"left\">Others criteria‡</td><td align=\"right\">2</td><td align=\"right\">6.9</td><td/><td align=\"right\">82</td><td align=\"right\">8.6</td><td/><td/></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">29</td><td align=\"right\">100.0</td><td/><td align=\"right\">949</td><td align=\"right\">100.0</td><td/><td/></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\"><bold>WHM% (SD)</bold></td><td/><td/><td align=\"left\">81.0(15.7)</td><td/><td/><td align=\"left\">84.3(14.4)</td><td align=\"right\">0.26</td></tr><tr><td align=\"left\">&gt; = 70</td><td align=\"right\">17</td><td align=\"right\">68.0</td><td/><td align=\"right\">721</td><td align=\"right\">80.5</td><td/><td align=\"right\">0.130</td></tr><tr><td align=\"left\">&lt; 70%</td><td align=\"right\">8</td><td align=\"right\">32.0</td><td/><td align=\"right\">175</td><td align=\"right\">19.5</td><td/><td/></tr><tr><td align=\"left\"><bold>Total</bold></td><td align=\"right\">25</td><td align=\"right\">100.0</td><td/><td align=\"right\">896</td><td align=\"right\">100.0</td><td/><td/></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\"><bold>CTC Outcomes</bold></td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Median Weight gain (IQR) <italic>g/kg/day</italic></td><td align=\"right\">24</td><td/><td align=\"left\">2.2 (1.6–4.0)</td><td align=\"right\">880</td><td/><td align=\"left\">3.1(1.1–5.9)</td><td align=\"right\">0.309</td></tr><tr><td align=\"left\">Median MUAC change (IQR)<italic>mm/day</italic></td><td align=\"right\">11</td><td/><td align=\"left\">0.22(0.01–0.45)</td><td align=\"right\">476</td><td/><td align=\"left\">0.25(0.03–0.48)</td><td align=\"right\">0.891</td></tr><tr><td align=\"left\">Median LOS (IQR) <italic>days</italic></td><td align=\"right\">25</td><td/><td align=\"left\">63(42–128)</td><td align=\"right\">912</td><td/><td align=\"left\">42(28–67)</td><td align=\"right\">0.002</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>The prevalence and diagnostic characteristics of proxy indicators and clinical algorithms for identifying HIV infection in severely malnourished children</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"right\"><bold>Prevalence</bold></td><td align=\"right\"><bold>OR</bold></td><td align=\"right\"><bold>Sensitivity </bold></td><td align=\"right\"><bold>Specificity </bold></td><td align=\"right\"><bold>PPV†</bold></td><td align=\"right\"><bold>NPV‡</bold></td><td align=\"right\"><bold>PLR$</bold></td><td align=\"right\"><bold>NLR&amp;</bold></td></tr><tr><td/><td align=\"right\"><bold>(%)</bold></td><td align=\"right\"><bold>(95%CI)</bold></td><td align=\"right\"><bold>(%)</bold></td><td align=\"right\"><bold>(%)</bold></td><td/><td/><td/><td/></tr></thead><tbody><tr><td align=\"left\"><bold>Individual proxy indicators, symptoms and signs</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">From widow headed household</td><td align=\"right\">1.4</td><td align=\"right\">30.0(7.4–121.6)</td><td align=\"right\">19</td><td align=\"right\">99.2</td><td align=\"right\">44.4</td><td align=\"right\">97.4</td><td align=\"right\">23.8</td><td align=\"right\">0.8</td></tr><tr><td align=\"left\">Looked after by grand-mother</td><td align=\"right\">4.1</td><td align=\"right\">4.1(1.1–14.6)</td><td align=\"right\">13.6</td><td align=\"right\">96.3</td><td align=\"right\">10.7</td><td align=\"right\">97.1</td><td align=\"right\">3.7</td><td align=\"right\">0.9</td></tr><tr><td align=\"left\">Orphan (one or both parents dead)</td><td align=\"right\">5.0</td><td align=\"right\">10.8(4.1–28.5)</td><td align=\"right\">31.8</td><td align=\"right\">95.8</td><td align=\"right\">20</td><td align=\"right\">97.7</td><td align=\"right\">7.6</td><td align=\"right\">0.7</td></tr><tr><td align=\"left\">From female headed household</td><td align=\"right\">7.1</td><td align=\"right\">5.8(2.1–17.2)</td><td align=\"right\">28.6</td><td align=\"right\">93.6</td><td align=\"right\">12.8</td><td align=\"right\">97.6</td><td align=\"right\">4.5</td><td align=\"right\">0.8</td></tr><tr><td align=\"left\"><bold>Symptoms and signs used in algorithms</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Child with tuberculosis</td><td align=\"right\">1.4</td><td align=\"right\">9.2(1.8–47.7)</td><td align=\"right\">10</td><td align=\"right\">98.8</td><td align=\"right\">20</td><td align=\"right\">97.4</td><td align=\"right\">8.3</td><td align=\"right\">0.9</td></tr><tr><td align=\"left\">Minor muco-cutaneous manifestations</td><td align=\"right\">7.2</td><td align=\"right\">4.6(1.6–13.2)</td><td align=\"right\">25</td><td align=\"right\">93.3</td><td align=\"right\">10</td><td align=\"right\">97.6</td><td align=\"right\">3.7</td><td align=\"right\">0.8</td></tr><tr><td align=\"left\"><bold>Variables associated with HIV in the present study</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Death of the father</td><td align=\"right\">3.1</td><td align=\"right\">15.6(4.7–49.9)</td><td align=\"right\">27.3</td><td align=\"right\">97.6</td><td align=\"right\">27.3</td><td align=\"right\">97.6</td><td align=\"right\">11.4</td><td align=\"right\">0.7</td></tr><tr><td align=\"left\">Age &lt; 12 months or age &gt; 59 months</td><td align=\"right\">13.6</td><td align=\"right\">3.7(1.4–9.5)</td><td align=\"right\">35</td><td align=\"right\">85.2</td><td align=\"right\">8.8</td><td align=\"right\">97.5</td><td align=\"right\">2.4</td><td align=\"right\">0.8</td></tr><tr><td align=\"left\">MUAC &lt; 110 mm</td><td align=\"right\">29.6</td><td align=\"right\">3.5(1.9–10.2)</td><td align=\"right\">61.1</td><td align=\"right\">71.3</td><td align=\"right\">5.9</td><td align=\"right\">98.4</td><td align=\"right\">2.1</td><td align=\"right\">0.5</td></tr><tr><td align=\"left\">Absence of oedema</td><td align=\"right\">17.1</td><td align=\"right\">2.9(1.1–7.7)</td><td align=\"right\">36.4</td><td align=\"right\">83.7</td><td align=\"right\">8.3</td><td align=\"right\">93</td><td align=\"right\">2.2</td><td align=\"right\">0.8</td></tr><tr><td align=\"left\">Axillary nodes enlargement</td><td align=\"right\">4.8</td><td align=\"right\">6.7(1.7–26.4)</td><td align=\"right\">23.1</td><td align=\"right\">95.7</td><td align=\"right\">13.6</td><td align=\"right\">97.7</td><td align=\"right\">5.4</td><td align=\"right\">0.8</td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td align=\"left\"><bold>Algorithm and combinations</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">South-African IMCI modified algorithm for paediatric HIV diagnosis</td><td/><td/><td align=\"right\">20.0</td><td align=\"right\">94.5</td><td align=\"right\">8.7</td><td align=\"right\">97.8</td><td align=\"right\">3.6</td><td align=\"right\">0.8</td></tr><tr><td align=\"left\">Original IMCI algorithm</td><td/><td/><td align=\"right\">9.1</td><td align=\"right\">96.7</td><td align=\"right\">8.0</td><td align=\"right\">97.1</td><td align=\"right\">2.8</td><td align=\"right\">0.9</td></tr><tr><td align=\"left\">Action Against Hunger algorithm</td><td/><td/><td align=\"right\">60.0</td><td align=\"right\">62.1</td><td align=\"right\">3.5</td><td align=\"right\">98.3</td><td align=\"right\">1.6</td><td align=\"right\">0.6</td></tr><tr><td align=\"left\" colspan=\"2\"> Presence one or more proxy indicators and MUAC &lt; 110 mm</td><td/><td align=\"right\">95.5</td><td align=\"right\">54.5</td><td align=\"right\">7.1</td><td align=\"right\">99.7</td><td align=\"right\">2.1</td><td align=\"right\">0.1</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>† IMCI = integrated management of childhood illness § Pneumonia was ascertained by asking if the child had breathing abnormalities (fast breathing, chest in-drawing or nasal flaring) and/or severe cough. ¶ All children were assumed to have a history of weight loss/low weight gain.</p></table-wrap-foot>", "<table-wrap-foot><p>† Comparison % oedematous malnutrition; ‡ others criteria = age above 6 months and weight less than 4 kg, child with visible wasting but not meeting marasmus and MUAC criteria and less than 6 months children.</p></table-wrap-foot>", "<table-wrap-foot><p>† PPV = positive predictive value; ‡ NPV = Negative predictive value; § PLR = likelihood ratio for a positive test; &amp; NLR = likelihood ratio for a negative test</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2334-8-106-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>This paper was motivated by a study in which a putative genetic risk marker for disease could not be measured with certainty. The study used a case-control design to assess the association of cervical cancer with a polymorphism in codon 72 of the p53 tumour suppressor gene. DNA specimens from study participants were processed independently and blindly to disease status by three laboratories in different countries. Preliminary analyses showed that inter-laboratory agreement on the genotype was only moderate, which led to considerable ambiguity about its odds ratio (OR) with cervical cancer [##REF##10918193##1##]. The empirical estimates of OR varied widely, depending on how disagreements between laboratory results were treated.</p>", "<p>Statistical latent class models (LCM) have been applied to a wide variety of diagnostic or disease screening data where disease status cannot be established with certainty. Typical scenarios are where a gold standard classification of disease either does not exist or is infeasible to observe [##REF##2324060##2##, ####REF##2765639##3##, ##REF##7582698##4##, ##REF##8817797##5##, ##REF##9871952##6##, ##REF##3877331##7##, ##REF##9351165##8##, ##REF##9519632##9##, ##REF##8805757##10##, ##REF##6359316##11##, ##UREF##0##12##, ##REF##7534756##13##, ##REF##2190288##14##, ##REF##9888282##15##, ##REF##3054000##16##, ##REF##1782636##17####1782636##17##]. The goal of LCM is typically to estimate measurement properties (such as test sensitivities and specificities) of the imperfect methods that are used to assess disease status. These ideas have been applied to meta-analyses as well as to individual studies [##REF##10513757##18##]. In contrast, the motivating case-control study on cervical cancer involved uncertainty about the genetic risk factor, rather than about disease status.</p>", "<p>Our illustrative example was a hospital-based case-control study of cervical cancer and the p53 codon 72 polymorphism, carried out in Brazil [##REF##10918193##1##], where the cases had histologically confirmed invasive squamous cell carcinoma of the cervix. Controls were sampled from women who attended a cervical cancer screening program in the same hospital where the cases were seen. Absence of malignancy in the controls was based on cytological examination of Pap smear samples. p53 codon 72 genotyping was performed blindly by 3 independent laboratories in Montreal, Canada, Sao Paulo, Brazil, and London, UK, randomly labelled here as laboratories A, B and C.</p>", "<p>Misclassification of disease status for the cases was unlikely because histological confirmation of squamous carcinoma was required. Although cervical abnormalities may have existed in previous Pap smears from control women, it is unlikely that any controls would have undetected cervical cancers at the time of study enrolment, because these invasive lesions would have been detected upon examination. To guard against false negatives on cytology, Pap smears from control women were read twice by independent expert cytopathologists [##REF##10918193##1##].</p>", "<p>In order to investigate inter-laboratory variation in test results, a random sample of participants was drawn by an epidemiology team in Montreal and submitted to the Sao Paulo centre, where the DNA specimens were stored. Specimens for selected women were divided into three aliquots, with two being shipped on dry ice to Montreal and London. The laboratories independently reported their classifications of the polymorphism to the epidemiologists in Montreal. Technical details of the laboratory methods varied, as previously described [##REF##10918193##1##].</p>", "<p>The study was not originally designed to assess the association between polymorphism and disease risk, because the index publication on the potential utility of this risk marker appeared several years after it was conducted [##REF##9607760##19##]. However, given the availability of stored specimens for many of the subjects, the authors decided to test the hypothesis on a <italic>post-hoc </italic>basis. The original report of this study included 54 cases and 91 controls. Pairwise comparisons between laboratories indicated crude agreement ranging from 71% to 78%, and chance-corrected kappa statistics of 0.49 to 0.63, implying moderate to substantial inter-laboratory reliability [##REF##843571##20##]. The fact that pairs of results disagree quite frequently (about 25% of the time) underscores the problem of not having a clear-cut definition of how a given woman should be classified if disagreements arise. Table ##TAB##0##1## shows crude and age- and race-adjusted ORs, associated with the homozygous Arg/Arg genotype, vs. a reference category of heterozygous Arg/Pro and homozygous Pro/Pro genotypes combined.</p>", "<p>Faced with the apparent unreliability of the laboratory results, the study investigators adopted alternative definitions of the reference and index categories. For the reference category, the <italic>non-stringent </italic>definition permitted disagreements for the Arg/Pro and the Pro/Pro genotypes, while the <italic>stringent </italic>definition included only genotypes with complete agreement among the laboratories. The index category was defined as: <italic>disagreed </italic>when it included only those subjects with an Arg/Arg genotype result from at least one laboratory but with different results from the other laboratories; <italic>agreed </italic>when it included only Arg/Arg subjects with complete agreement among laboratories; or <italic>all-inclusive </italic>when it allowed any reported Arg/Arg genotype, with or without agreement. Table ##TAB##1##2## shows the OR estimates associated with all 6 combinations of reference and index category definitions, obtained using unconditional logistic regression [##REF##10918193##1##]. The results varied widely, leading the investigators to conclude that \"<italic>When disagreement between laboratories was allowed,...OR was as low as 1.5. In contrast, OR increased to 8.0 after exclusion of discordant genotypes ...Exposure misclassification ... may affect ability to detect the association...\" </italic>[##REF##10918193##1##] The lowest of these OR values (1.5) would represent a relatively weak association between the polymorphism and cervical cancer, while the largest (8.0) would represent a rather strong association, and there is considerable ambiguity about which of any of the empirical OR values is most \"correct\". It should be noted that all these OR estimates, including even the estimates based on excluding the discordant observations, are biased [##REF##6693532##21##], possibly quite seriously. Estimates using the data from only one laboratory are also biased in the presence of measurement error.</p>", "<p>The uncertainty engendered by the wide range of these empirical estimates, and the lack of a preferred estimator motivated us to develop a LCM analysis that could assess the association of the polymorphism with disease, while taking potential inaccuracy of the laboratory results into account. Such an approach should lead to a de-attenuation of the exposure-disease association, giving a more rigorous way to estimate OR. Additionally, investigators can learn about the likely quality of their data, in terms of the accuracy rates of their contributing laboratories.</p>", "<p>In our Methods section, we determine design requirements for the application of LCMs in this situation, according to alternative assumptions about variation in test accuracy. In our Results section, we apply several models to assess the genotype-cervical cancer association, while taking test inaccuracy into account. Use of LCMs for the problem yields maximum likelihood estimates of OR, which have superior statistical properties to the biased empirical estimates mentioned above. Other issues in the application of LCMs to this type of problem are covered in our Discussion.</p>" ]

[ "<title>Methods</title>", "<p>We assume that a true exposure status X exists for the genotype of each study subject, but that it cannot be observed without error – hence X is a latent or unobserved variable. We are interested in the association of disease D (cervical cancer) with the true exposure status X, denoted by DX, but instead we can only observe DE, the association of disease with the observed laboratory results E.</p>", "<p>The accuracy of a laboratory test for a risk factor can be characterised by two measures. First, sensitivity is the probability that an individual whose true exposure X is positive receives a correct positive negative result. Second, specificity is the probability that an individual whose true exposure X is negative receives a correct positive result. The complements (1-sensitivity) and (1-specificity) of these quantities are the false-negative and false-positive rates, these being the probabilities of incorrect results for true positive and true negative individuals, respectively [##UREF##1##22##]. Our proposed LCM estimates the joint probabilities of the set of results for a study participant, conditional on an assumed true state for that individual. The conditional probabilities are then summed over the marginal probability distribution of X, which is also estimated from the data. By suitable specification of alternative models (see below), one can evaluate if accuracy varies significantly between laboratories or by disease status. Additionally we can assess the association of the latent variable with disease, under various assumptions about test accuracy.</p>", "<title>3.1: Required parameters and available degrees of freedom</title>", "<p>In our analytic framework, we are primarily concerned with two types of LCMs. First, we wish to evaluate the measurement accuracy of the exposure data, i.e. the association of the observed genotype test results with respect to the true (but latent) genotype. Here we can either assume test accuracy to be differential or constant between laboratories, between cases and non-cases, or jointly differential by both laboratory and disease status. Second, we wish to estimate the association of disease with the true genotype, and here again we may or may not assume test accuracy to be differential by laboratory and/or disease. Finally, we can compare the LCM results with empirical (non-latent) models which examine the association of disease with the observed genotypes, but which do not admit the possibility of measurement error.</p>", "<p>Table ##TAB##2##3## shows the number of parameters involved in each of these three types of model. This is done for a general specification of the number of laboratories (R), and also for either 1, 2 or 3 laboratories in particular. In the first group of models (models 1–4), the focus is on evaluating test accuracy, and to examine if accuracy is the same or different between laboratories and/or between cases and controls. We examine the association of the set of laboratory results E and X, either conditionally or unconditionally on disease status (D) and laboratory. If the tests are highly accurate, there will be a strong EX association.</p>", "<p>In model 1, we allow the values of test sensitivity and specificity to be different for each laboratory, but accuracy is otherwise assumed to be the same for both cases and controls. Hence if there are R laboratories, there are 2R parameters representing test accuracy. We require two additional parameters, first to fit the marginal distributions of X (the latent exposure variable) and second for D (to constrain the case and control frequencies to agree with their observed values), making 2R + 2 parameters in total. In model 2, accuracy is now additionally permitted to be differential by disease status, which increases the number of model parameters by 2 for each laboratory, giving 4R + 2 parameters in total. In models 3 and 4, accuracy is assumed to be constant (non-differential) across laboratories, and so the number of parameters is independent of the number of laboratories. For model 3, where accuracy is non-differential by disease status, there are two accuracy parameters (sensitivity and specificity, constant across laboratories), and one each for the marginal distributions of X and D as before. For model 4, the two accuracy parameters are potentially different in the case and control groups.</p>", "<p>In the second group of models (5 and 6), we evaluate the relationship between disease and (true) exposure X, or the DX association. In the more general case (model 5), where test accuracy varies by laboratory (but is the same for cases and controls), the parameters are the same as in model 2, except that we now include a term for the conditional probability of D given X, or D|X.</p>", "<p>In the third group (models 7 and 8), we examine the empirical association between D and E, which involves 2R parameters in the more general situation when accuracy is allowed to vary between laboratories. Additionally, we again include a D term to constrain the fitted and observed numbers of cases and controls to agree, making 2R + 1 parameters in total. If accuracy is assumed non-differential between laboratories, there are only 3 parameters – the proportion of study subjects who are cases, and the proportions of cases and controls that are exposed. Empirical models ignore the possibility of measurement error. The empirical approach is often used in practice, but the estimated DE association will in general be biased, unless the exposure assessment is error-free. If the tests are indeed perfect (an unlikely situation in practice), the empirical models suffice and the need for modelling the measurement error process is obviated.</p>", "<p>To estimate the parameters of the various LCMs, we need to verify that there are sufficient degrees of freedom (<italic>df</italic>) available from the observational design. For all the models in Table ##TAB##2##3##, the cross-classification of the R laboratory results by disease status involves 2^R+1data cells, implying that there are 2^R+1- 1 <italic>df </italic>available for parameter estimation after conditioning on the total sample size. For R = 1, 2 and 3 specifically, the available <italic>df </italic>are 3, 7 and 15 respectively. Therefore, among the models assessing the EX association, model 2 (which allows for the most general pattern of test accuracy) requires that there be at least 3 laboratory tests. However, the other models in this group, which assume non-differential test accuracy by disease status and/or by laboratory, can be fitted if R ≥ 2.</p>", "<p>Models 5 and 6 examining the DX association can be fitted if there are at least 2 laboratories. Finally, the empirical evaluation of the {ED association (models 7 and 8) is possible in one or more laboratories.</p>", "<p>Note that having sufficient <italic>df </italic>for parameter estimation does not avoid the issue of parameter identifiability. Because, by definition, the true latent state X is unobservable, there are usually two sets of parameter estimates with the same likelihood and model fit, these being essentially \"mirror images\" of one another [##REF##3054000##16##,##UREF##2##23##]. Thus, for instance the laboratory sensitivity in one solution can be exchanged with a corresponding value of (1-specificity) specificity in the other. In practice, choosing the \"right\" solution is typically straightforward, because it will have inherently far greater plausibility in terms of agreeing with external information on the parameter values. For example, an estimated sensitivity of (say) 90% would almost certainly be more plausible than a 90% false-positive rate.</p>", "<p>Table ##TAB##3##4## summarises the associations that are estimated in each of the models described in Table ##TAB##2##3##, for the specific case of R = 3 laboratories (as we have in our example). For instance, in model 1 the focus is on the test accuracy, through the associations of test results from laboratories A, B and C with the true genotype status X; these associations are represented by the probabilities A|X, B|X, C|X of a positive test result from each laboratory, conditional on the true value of X. We must additionally estimate the prevalence of the latent exposure variable X.</p>", "<p>Model 2 examines test accuracy in more detail, specific to both laboratory and disease status, by fitting the conditional probabilities A|DX, B|DX, C|DX. Models 3 and 4 impose equality constraints on the terms, to force the test accuracy estimates to be the same across laboratories.</p>", "<p>In models 5 and 6, the focus is on the fitted term X|D that defines the association of the genotype with disease, while the models also allow for test accuracy. Finally, models 7 and 8 examine the empirical test positivity rates, conditional on disease state, through terms such as A|D; no allowance is made for the possibility of test errors.</p>", "<p>The LCM models are actually fitted by calculating expected frequencies in the cells of the contingency table formed by a cross-tabulation of the observed variables. These expectations can be represented in a standard log-linear form. [##UREF##3##24##] For instance, for model 5, the log-linear formulation of the expected frequency for the data frequency m_abcdx, corresponding to levels a, b, c, d, and x of the observed laboratory test variables (A, B, C), the disease status D and the latent variable X respectively, is given by</p>", "<p></p>", "<p>where <italic>u </italic>represents the overall mean frequency across all cells, a main effect term such as represents a marginal constraint on the frequencies at each level of A, and the interaction terms such as indicates that the associations such as DX are to be estimated.</p>", "<p>We used the freeware program <italic>lem </italic>[##UREF##4##25##], which provides a flexible framework for latent class analysis. Latent class software, such as <italic>lem</italic>, more easily accommodates the type of data and modelling required for this type of analysis. Programs for the general analysis of log-linear models can also be adopted, if the user is able to specify the requisite latent class models appropriately in a corresponding log-linear format.</p>", "<p>Comparisons between the fits of appropriate pairs of models permits evaluation of the various assumptions, such as those of differential test accuracy between laboratories and disease groups. Statistical significance of the differences in fit between alternative models can be assessed using likelihood ratio statistics.</p>", "<p>The <italic>lem </italic>program allows conditioning on the observed pattern of available data, so that data from women with results available from only one or two laboratories can be used. We assume that data missingness is unrelated to the model parameters of interest, because the chance of an uninformative test result depends primarily on the degree of depletion of the DNA specimen, and not on p53 status. Model fitting is based on the EM algorithm, and iterative proportional fitting, with parameter starting values defined via a random number seed. This method of fitting yields maximum likelihood estimates of the model parameters, which are therefore unbiased in large samples, and have the smallest possible variance. These statistical properties imply strong advantages of the LCM parameter estimates, compared to the <italic>ad hoc </italic>estimates described earlier.</p>" ]

[ "<title>Results</title>", "<p>Our analysis is based on a larger sample of participants obtained subsequent to the original report [##REF##10918193##1##], with 142 cases and 162 controls identified using the same methods as previously. Table ##TAB##4##5## shows the numbers of participants with polymorphism classifications available from the various combinations of laboratories. Laboratory B did more tests because they were able to salvage additional DNA samples from the frozen cervical specimens. Laboratories varied in their diligence in obtaining informative test results, and their potential to do so also varied by the amount of fractionated sample material available to them.</p>", "<title>Assessment of laboratory accuracy</title>", "<p>Table ##TAB##5##6## shows results from the first group of models in Table ##TAB##2##3##, examining the accuracy of the laboratory classifications of the polymorphism. Model 1 estimates the prevalence of the latent genotype X, and the probability of each laboratory result (A, B, or C) conditional on X, while conditioning on the observed number of cases and controls through inclusion of the variable D. Model 2 is similar, but it conditions the probability of laboratory results to depend on D as well as X. A likelihood ratio test between models 1 and 2 gives χ²= 9.2 on 6 <italic>df </italic>(p = 0.16), indicating no strong evidence of differential test accuracy between cancer cases and controls, while still allowing differential accuracy by laboratory. This is reassuring, given that DNA samples from cases tend to be more plentiful than from controls. (Case biopsy samples contain more cells than cervical cell swabs from controls). Specimens with a greater quantity of DNA permit replication of results whenever the interpretation of the first assay was uninformative.</p>", "<p>A similar comparison of models 3 and 4 also addresses the issue of possibly differential accuracy by disease status, but now assuming that the laboratories have equal accuracy; the likelihood ratio test is χ²= 4.8 on 2 <italic>df </italic>(p = 0.09), suggesting that accuracy is not significantly related to disease status. This seems reasonable, because it is unlikely on biological grounds that errors in classifying this polymorphism would be related to disease [##REF##10918193##1##].</p>", "<p>Other comparisons between the models of Table ##TAB##5##6## can address variation in accuracy across laboratories. For instance, a comparison of models 1 and 3 tests for equality between laboratories while assuming independence of accuracy and disease status, while a similar comparison of models 2 and 4 allows for a dependence of accuracy on disease. These tests give χ²= 8.6 on 4 <italic>df </italic>(p = 0.07) and χ²= 13.0 on 8 <italic>df </italic>(p = 0.11), assuming non-differential or differential test accuracy by disease status, respectively, thus giving weak evidence of inter-laboratory differences in accuracy. There is a suggestion that laboratory A has lower specificity, while laboratory C has lower sensitivity. However, these differences were not strongly supported by the likelihood ratio tests, which gave only borderline significance.</p>", "<title>Association of polymorphism with disease</title>", "<p>Table ##TAB##6##7## shows the results of models focussed on the association of the true genotype variable X with disease. The likelihood ratio test comparing models 5 and 6 (χ²= 8.0 on 4 <italic>df</italic>, p = 0.09) again weakly suggests that laboratory accuracy varies, and the pattern of parameter estimates is similar to those in Table ##TAB##5##6##. These models additionally estimate the conditional probabilities of X for given values of D (cases or controls), which in turn lead to their ORs. Model 5 gives estimates P(X = +|case) = 0.340 (SE = 0.066) and P(X = +|control) = 0.220 (SE = 0.048), where + and - indicate presence or absence of the Arg/Arg genotype respectively. This implies an OR of 1.83 (95%CI = 0.97, 3.46). The corresponding conditional probabilities in model 6 (with laboratories constrained to have equal accuracy) are 0.378 (SE = 0.072) and 0.237 (SE = 0.054), and an associated OR of 1.96. (95%CI = 1.02, 3.75). Given that there is no strong evidence of inter-laboratory differences in accuracy, the model 6 estimate of OR would be the preferred value.</p>", "<title>Comparison with empirical results</title>", "<p>Table ##TAB##7##8## shows the empirical associations of laboratory results with disease. A comparison of models 7 and 8 assesses the possibility of different strength of association with disease by laboratory. Their likelihood ratio test (χ²= 4.78 on 4 <italic>df</italic>, p = 0.31) indicates no strong evidence for different associations by laboratory. The empirical estimates of OR are 2.48 (95%CI 1.10 – 5.60), 1.59 (95%CI 0.90 – 2.80), and 1.84 (95%CI 1.10 – 5.60), for laboratories A, B, C respectively.</p>" ]

[ "<title>Discussion</title>", "<p>Variation in measuring p53 expression has been recognized before, in the context of bladder cancer studies [##REF##10815908##26##]. In this paper, we have illustrated the use of LCMs to evaluate the association of a genotype with cancer, while taking measurement error in the genotype into account. This approach is attractive for the rapidly increasing number of studies relating genetic traits to various diseases, but the models are also potentially applicable to a wide variety of other epidemiological investigations. The data discussed here came from several laboratories, but the same approach could be applied to studies where different methods are used to assess exposure or putative susceptibility to a risk factor, for instance questionnaires <italic>vs</italic>. medical records concerning risk determinants, self-report <italic>vs</italic>. proxy reports for dietary consumption, or different methods within the same laboratory.</p>", "<p>We used several models to investigate the possibility of differential test accuracy by laboratory. These models can be fitted whenever the number of tests per subject is at least 2. For data with exactly 2 measurements, one can permit accuracy to be differential by disease status, but one cannot allow for differences between laboratories (or between methods in general). When there are 3 or more measurements per subject, one can examine the possibility of accuracy being differential by <italic>both </italic>disease and laboratory.</p>", "<p>Use of LCMs when there is uncertainty about risk status is somewhat more feasible than when it is the disease status that may be misclassified. For the latter, one requires at least <italic>three </italic>measurements in order to estimate test accuracy and disease prevalence in a single population, or two measurements with data from two or more populations, assuming one can ignore the possibility of population by test interactions [##REF##3877331##7##,##UREF##5##27##]. The particular case of two independent measurements in two populations was discussed in detail by Hui and Walter [##REF##7370371##28##], this scenario being one of very few that admit a closed-form solution for the parameter estimates.</p>", "<p>In analyses concerned with uncertainty about disease status, conditional independence of test errors is often assumed, but this assumption may not always be valid in practice. However, conditionally dependent errors can be included in the model if there are additional measurements available [##REF##8841648##29##, ####REF##9330426##30##, ##REF##3830260##31##, ##REF##9290225##32####9290225##32##], but this presents an additional burden on the investigators, and it may not be feasible to include such additional measurements.</p>", "<p>In contrast, when it is the risk factor that involves measurement error (as in the present example), the conditional independence assumption can be examined more easily, because of the more limited data requirements. In our data, we found no strong evidence of test accuracy being dependent on disease status, a reasonable finding given the underlying biology and the laboratory testing methods. We also tested the conditional independence assumption by adding terms such as AB|X to model 1. None of these terms was statistically significant, so there was no evidence of a departure from the conditional independence assumption. Drews et al. [##REF##8347743##33##] describe an alternative latent class approach to situations with two measurements having non-differential and conditionally dependent errors, but the error correlations must either be known (somewhat unrealistic in practice) or at least taken to have given, fixed values.</p>", "<p>We also found only weak evidence of differential accuracy by laboratory. However, with the given data (having only one result per woman for each laboratory), we were obliged to assume no subject-by-laboratory interaction, or in other words conditionally independent error rates by laboratories. This last interaction could be examined if there were replicated observations in the same laboratories.</p>", "<p>The main objective of genetic studies of the type we have discussed is to obtain the best possible estimate of the OR between a polymorphism and disease. The LCMs we have used include all the available data, and yield maximum likelihood estimates of OR. While the test accuracy of laboratories is not a main focus, the latent class method does give estimates of accuracy as a useful by-product. Also, the evaluation of the fit of alternative LCMs that examine test accuracy provides guidance on the preferred way to allow for test inaccuracy when the polymorphism-disease association is addressed in later models. In our example, we found no convincing evidence of differential test accuracy by laboratory or disease status, which implied that the preferred model for the polymorphism OR should be the one (here, model 6) where accuracy is constrained to be equal in all laboratory-disease groups of data.</p>", "<p>In our example, we exploited the existence of data from women whose samples had been analysed by more than one laboratory. Practicalities limited the number of samples where sufficient material was available for replicated testing, especially given the wide geographical spread of the participating laboratories. If there is primary interest in assessing test accuracy (as opposed to primary interest in the polymorphism OR), then an appropriate study should imply a sample design having more replicated observations with the analytic focus being on test variation between, and possibly within, laboratories.</p>" ]

[ "<title>Conclusion</title>", "<p>Our analysis provided an estimate of OR for the genotype-cancer association. Subject to the validity of the assumed model, this estimate enjoys the general properties of maximum likelihood estimates, including asymptotic unbiasedness and minimum variance. The model-based estimate also avoids the ambiguous and arbitrary choices that must be made between the various empirical estimates available when the genotype classifications disagree for some study subjects, as exemplified by the wide range of empirical ORs in Table ##TAB##1##2##, and as seen in the laboratory-specific estimates from model 7. Also, if the reliability of the data is low, the latent class OR estimate will tend to have a lower standard error and narrower confidence limits than the various empirical estimates. In our example, in which reliability was moderate or substantial, the latent class OR estimate was still somewhat more precise than the estimates for laboratories A and C. It was also statistically significant, whereas the empirical results for laboratories B and C were not.</p>", "<p>An additional benefit of the LCM approach is that it yields estimates of the accuracy of the test method. In the absence of a definitive (i.e. an error-free gold standard) classification of exposure, the accuracy values can be used to calculate the predictive values associated with given test results, an attractive feature for clinical applications. The accuracy results may also help to identify deficiencies in data quality, e.g. from certain laboratories or observational methods.</p>", "<p>The methods used here involved a binary risk factor, but they could easily be extended to cover multinomial exposures. [##REF##8347743##33##,##REF##7742401##34##] Extensions to the basic LCMs of Hui and Walter [##REF##7370371##28##] have been proposed to allow for differential misclassification between cases and controls [##REF##8347743##33##, ####REF##7742401##34##, ##REF##10521867##35####10521867##35##]; these approaches require specification of a covariate that defines two subgroups of cases and controls, across which the error rates of each observational method are assumed constant. Further extensions to allow for additional or continuous covariates can be envisaged. Potential difficulties with such extensions are the number of extra parameters required and the sparser distribution of the observations over a larger number of data cells when suitable covariates exist, or the unavailability of suitable covariates in other cases. The validity of the maximum likelihood parameter estimates and likelihood ratio tests to compare models might then be a concern. Others have commented [##REF##15180668##36##,##REF##11414591##37##] that likelihood methods may not perform well in distinguishing competing models in this context.</p>", "<p>On the basis of the present re-assessment, we believe that previous attempts to compensate for the measurement error in the original study [##REF##10918193##1##] may have led to over-estimates of the OR. A recent meta-analysis of all case-control studies on the association between p53 codon 72 polymorphism and cervical cancer risk indicated an average effect that was consistent with the LCM estimates presented here [##REF##14744727##38##]. Likewise, the ORs we obtained in a recent case-control study specifically designed to verify the association, and which used improved methods to assess the polymorphism (involving less measurement error) [##REF##16122882##39##], were consistent with the present latent class-based estimates.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Researchers wanting to study the association of genetic factors with disease may encounter variability in the laboratory methods used to establish genotypes or other traits. Such variability leads to uncertainty in determining the strength of a genotype as a risk factor. This problem is illustrated using data from a case-control study of cervical cancer in which some subjects were independently assessed by different laboratories for the presence of a genetic polymorphism. Inter-laboratory agreement was only moderate, which led to a very wide range of empirical odds ratios (ORs) with the disease, depending on how disagreements were treated.</p>", "<p>This paper illustrates the use of latent class models (LCMs) and to estimate OR while taking laboratory accuracy into account. Possible LCMs are characterised in terms of the number of laboratory measurements available, and if their error rates are assumed to be differential or non-differential by disease status and/or laboratory.</p>", "<title>Results</title>", "<p>The LCM results give maximum likelihood estimates of laboratory accuracy rates and the OR of the genetic variable and disease, and avoid the ambiguities of the empirical results. Having allowed for possible measurement error in the expure, the LCM estimates of exposure – disease associations are typically stronger than their empirical equivalents. Also the LCM estimates exploit all the available data, and hence have relatively low standard errors.</p>", "<title>Conclusion</title>", "<p>Our approach provides a way to evaluate the association of a polymorphism with disease, while taking laboratory measurement error into account. Ambiguities in the empirical data arising from disagreements between laboratories are avoided, and the estimated polymorphism-disease association is typically enhanced.</p>" ]

[ "<title>Authors' contributions</title>", "<p>SDW developed the methodology presented in this paper, carried out the statistical analysis of the data, and drafted the manuscript. ELF has done research on the impact of exposure misclassification in cancer etiology studies and was responsible for the case-control study that generated the data used to illustrate the methods developed by SDW in the paper. Both authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>The authors thank Drs. Luisa Villa, Greg Matlashewski, and Alan Storey for their valuable contributions as co-investigators in the original case-control study. The work was partly supported by funding from the Natural Sciences and Engineering and Research Council.</p>" ]

[]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Association of invasive cervical cancer with p53 arg/arg genotype¹using individual laboratory results</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"2\"><bold>OR¹(95% CI)</bold></td></tr><tr><td/><td align=\"center\"><bold>Crude</bold></td><td align=\"center\"><bold>Adjusted</bold>²</td></tr></thead><tbody><tr><td align=\"left\">Laboratory A</td><td align=\"center\">2.5 (1.1–5.6)</td><td align=\"center\">3.2 (1.3–7.9)</td></tr><tr><td align=\"left\">Laboratory B</td><td align=\"center\">2.2 (1.0–5.1)</td><td align=\"center\">2.4 (1.0–5.9)</td></tr><tr><td align=\"left\">Laboratory C</td><td align=\"center\">1.8 (0.7–4.8)</td><td align=\"center\">2.8 (0.9–8.4)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Association of invasive cervical cancer with p53 arg/arg genotype¹, with various approaches to inter-laboratory disagreements</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Referent category definition</bold>^{<bold>1</bold>}</td><td align=\"center\"><bold>Index category (Arg/Arg genotype) definition</bold>^{<bold>2</bold>}</td><td align=\"center\" colspan=\"2\"><bold>OR (95% CI)</bold></td></tr><tr><td/><td/><td align=\"center\"><bold>Crude</bold></td><td align=\"center\"><bold>Adjusted</bold>^{<bold>3</bold>}</td></tr></thead><tbody><tr><td align=\"center\">Non-stringent</td><td align=\"center\">Disagreed</td><td align=\"center\">1.6 (0.6–4.2)</td><td align=\"center\">1.5 (0.5–3.9)</td></tr><tr><td/><td align=\"center\">All-inclusive</td><td align=\"center\">2.4 (1.2–5.0)</td><td align=\"center\">2.4 (1.1–5.3)</td></tr><tr><td/><td align=\"center\">Agreed</td><td align=\"center\">2.6 (1.0–6.9)</td><td align=\"center\">3.4 (1.2–9.9)</td></tr><tr><td align=\"center\">Stringent</td><td align=\"center\">Disagreed</td><td align=\"center\">2.7 (0.9–8.1)</td><td align=\"center\">2.8 (0.9–8.9)</td></tr><tr><td/><td align=\"center\">All-inclusive</td><td align=\"center\">4.1 (1.7–10.0)</td><td align=\"center\">5.0 (1.9–13.3)</td></tr><tr><td/><td align=\"center\">Agreed</td><td align=\"center\">4.5 (1.5–13.4)</td><td align=\"center\">8.0 (2.3–28.5)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Number of parameters in latent class model, by number of laboratories (R)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Model number</bold></td><td align=\"center\"><bold>Association of interest</bold></td><td align=\"center\"><bold>Accuracy differential by laboratory</bold></td><td align=\"center\"><bold>General number of labs (R)</bold></td><td align=\"center\"><bold>R = 1</bold></td><td align=\"center\"><bold>R = 2</bold></td><td align=\"center\"><bold>R = 3</bold></td></tr></thead><tbody><tr><td align=\"center\">1</td><td align=\"center\">Observed vs. true exposure status: EX</td><td align=\"center\">Yes</td><td align=\"center\">2R + 2</td><td align=\"center\">4</td><td align=\"center\">6</td><td align=\"center\">8</td></tr><tr><td align=\"center\">2</td><td/><td align=\"center\">Yes*</td><td align=\"center\">4R + 2</td><td align=\"center\">6</td><td align=\"center\">10</td><td align=\"center\">14</td></tr><tr><td align=\"center\">3</td><td/><td align=\"center\">No</td><td align=\"center\">4</td><td align=\"center\">4</td><td align=\"center\">4</td><td align=\"center\">4</td></tr><tr><td align=\"center\">4</td><td/><td align=\"center\">No*</td><td align=\"center\">6</td><td align=\"center\">6</td><td align=\"center\">6</td><td align=\"center\">6</td></tr><tr><td align=\"center\">5</td><td align=\"center\">Disease vs. true exposure status: DX</td><td align=\"center\">Yes</td><td align=\"center\">2R + 3</td><td align=\"center\">5</td><td align=\"center\">7</td><td align=\"center\">9</td></tr><tr><td align=\"center\">6</td><td/><td align=\"center\">No</td><td align=\"center\">5</td><td align=\"center\">5</td><td align=\"center\">5</td><td align=\"center\">5</td></tr><tr><td align=\"center\">7</td><td align=\"center\">Disease with observed exposure status: DE</td><td align=\"center\">Yes</td><td align=\"center\">2R + 1</td><td align=\"center\">3</td><td align=\"center\">5</td><td align=\"center\">7</td></tr><tr><td align=\"center\">8</td><td/><td align=\"center\">No</td><td align=\"center\">3</td><td align=\"center\">3</td><td align=\"center\">3</td><td align=\"center\">3</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Terms involved in latent class models, for R = 3 laboratories</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Model</bold></td><td align=\"center\"><bold>Terms of interest</bold></td><td align=\"center\"><bold>Other terms fitted</bold></td></tr></thead><tbody><tr><td align=\"center\">1</td><td align=\"center\">Test accuracy (A|X, B|X, C|X)</td><td align=\"center\">True exposure prevalence (X)</td></tr><tr><td align=\"center\">2</td><td align=\"center\">Test accuracy, differential by disease (A|DX, B|DX, C|DX)</td><td align=\"center\">True exposure prevalence (X)</td></tr><tr><td align=\"center\">3</td><td align=\"center\">Test accuracy, constant across labs (A|X = B|X = C|X)</td><td align=\"center\">True exposure prevalence (X)</td></tr><tr><td align=\"center\">4</td><td align=\"center\">Test accuracy, constant across labs, differential by disease (A|DX = B|DX = C|DX)</td><td align=\"center\">True exposure prevalence (X)</td></tr><tr><td align=\"center\">5</td><td align=\"center\">True exposure by disease (X|D)</td><td align=\"center\">Test accuracy (A|X, B|X, C|X)</td></tr><tr><td align=\"center\">6</td><td align=\"center\">True exposure by disease (X|D)</td><td align=\"center\">Test accuracy, constant across labs (A|X = B|X = C|X)</td></tr><tr><td align=\"center\">7</td><td align=\"center\">Empirical exposure by disease (A|D, B|D, C|D)</td><td/></tr><tr><td align=\"center\">8</td><td align=\"center\">Empirical exposure by disease constant across labs (A|D = B|D = C|D)</td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Number of cases and controls with p53 classifications available, by laboratory</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td/><td align=\"center\" colspan=\"6\"><bold>Lab C</bold></td></tr><tr><td/><td/><td align=\"center\" colspan=\"3\"><bold>Cases</bold></td><td align=\"center\" colspan=\"3\"><bold>Controls</bold></td></tr><tr><td align=\"center\"><bold>Lab A</bold></td><td align=\"center\"><bold>Lab B</bold></td><td align=\"center\"><bold>Arg/Arg</bold></td><td align=\"center\"><bold>Other</bold></td><td align=\"center\"><bold>NA</bold></td><td align=\"center\"><bold>Arg/Arg</bold></td><td align=\"center\"><bold>Other</bold></td><td align=\"center\"><bold>NA</bold></td></tr></thead><tbody><tr><td align=\"center\">Arg/Arg</td><td align=\"center\">Arg/Arg</td><td align=\"center\">5</td><td align=\"center\">2</td><td align=\"center\">4</td><td align=\"center\">7</td><td align=\"center\">5</td><td align=\"center\">1</td></tr><tr><td/><td align=\"center\">Other</td><td align=\"center\">0</td><td align=\"center\">4</td><td align=\"center\">0</td><td align=\"center\">1</td><td align=\"center\">2</td><td align=\"center\">0</td></tr><tr><td/><td align=\"center\">NA</td><td align=\"center\">1</td><td align=\"center\">0</td><td align=\"center\">3</td><td align=\"center\">3</td><td align=\"center\">1</td><td align=\"center\">1</td></tr><tr><td align=\"center\">Other</td><td align=\"center\">Arg/Arg</td><td align=\"center\">1</td><td align=\"center\">0</td><td align=\"center\">1</td><td align=\"center\">0</td><td align=\"center\">4</td><td align=\"center\">0</td></tr><tr><td/><td align=\"center\">Other</td><td align=\"center\">1</td><td align=\"center\">12</td><td align=\"center\">0</td><td align=\"center\">1</td><td align=\"center\">33</td><td align=\"center\">2</td></tr><tr><td/><td align=\"center\">NA</td><td align=\"center\">0</td><td align=\"center\">1</td><td align=\"center\">3</td><td align=\"center\">1</td><td align=\"center\">7</td><td align=\"center\">4</td></tr><tr><td align=\"center\">NA</td><td align=\"center\">Arg/Arg</td><td align=\"center\">2</td><td align=\"center\">1</td><td align=\"center\">26</td><td align=\"center\">0</td><td align=\"center\">0</td><td align=\"center\">10</td></tr><tr><td/><td align=\"center\">Other</td><td align=\"center\">0</td><td align=\"center\">0</td><td align=\"center\">65</td><td align=\"center\">0</td><td align=\"center\">1</td><td align=\"center\">44</td></tr><tr><td/><td align=\"center\">NA</td><td align=\"center\">0</td><td align=\"center\">1</td><td align=\"center\">9</td><td align=\"center\">1</td><td align=\"center\">1</td><td align=\"center\">32</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T6\"><label>Table 6</label><caption><p>Results for latent class models focussing on laboratory error rates</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Model</bold></td><td align=\"center\"><bold>Log- likelihood</bold></td><td align=\"center\"><bold>Disease Groups</bold></td><td align=\"center\"><bold>Lab</bold></td><td align=\"center\"><bold>Sensitivity (SE)</bold></td><td align=\"center\"><bold>Specificity (SE)</bold></td></tr></thead><tbody><tr><td align=\"center\">1</td><td align=\"center\">-441.64</td><td align=\"center\">All</td><td align=\"center\">A</td><td align=\"center\">0.94 (0.07)</td><td align=\"center\">0.90 (0.04)</td></tr><tr><td/><td/><td/><td align=\"center\">B</td><td align=\"center\">0.94 (0.08)</td><td align=\"center\">0.94 (0.03)</td></tr><tr><td/><td/><td/><td align=\"center\">C</td><td align=\"center\">0.70 (0.10)</td><td align=\"center\">0.95 (0.05)</td></tr><tr><td align=\"center\">2</td><td align=\"center\">-437.06</td><td align=\"center\">Cases</td><td align=\"center\">A</td><td align=\"center\">0.89 (0.10)</td><td align=\"center\">0.76 (0.10)</td></tr><tr><td/><td/><td/><td align=\"center\">B</td><td align=\"center\">1.00 (0.00)</td><td align=\"center\">0.97 (0.07)</td></tr><tr><td/><td/><td/><td align=\"center\">C</td><td align=\"center\">0.77 (0.15)</td><td align=\"center\">0.94 (0.06)</td></tr><tr><td/><td/><td align=\"center\">Controls</td><td align=\"center\">A</td><td align=\"center\">1.00 (0.00)</td><td align=\"center\">1.00 (0.00)</td></tr><tr><td/><td/><td/><td align=\"center\">B</td><td align=\"center\">0.73 (0.11)</td><td align=\"center\">0.93 (0.03)</td></tr><tr><td/><td/><td/><td align=\"center\">C</td><td align=\"center\">0.59 (0.11)</td><td align=\"center\">0.95 (0.03)</td></tr><tr><td align=\"center\">3</td><td align=\"center\">-445.94</td><td align=\"center\">All</td><td align=\"center\">A</td><td align=\"center\">0.83 (0.06)</td><td align=\"center\">0.93 (0.02)</td></tr><tr><td/><td/><td/><td align=\"center\">B</td><td align=\"center\">0.83 (0.06)</td><td align=\"center\">0.93 (0.02)</td></tr><tr><td/><td/><td/><td align=\"center\">C</td><td align=\"center\">0.83 (0.06)</td><td align=\"center\">0.93 (0.02)</td></tr><tr><td align=\"center\">4</td><td align=\"center\">-443.56</td><td align=\"center\">Cases</td><td align=\"center\">A</td><td align=\"center\">0.90 (0.08)</td><td align=\"center\">0.88 (0.05)</td></tr><tr><td/><td/><td/><td align=\"center\">B</td><td align=\"center\">0.90 (0.08)</td><td align=\"center\">0.88 (0.05)</td></tr><tr><td/><td/><td/><td align=\"center\">C</td><td align=\"center\">0.90 (0.08)</td><td align=\"center\">0.88 (0.05)</td></tr><tr><td/><td/><td align=\"center\">Controls</td><td align=\"center\">A</td><td align=\"center\">0.77 (0.10)</td><td align=\"center\">0.96 (0.03)</td></tr><tr><td/><td/><td/><td align=\"center\">B</td><td align=\"center\">0.77 (0.10)</td><td align=\"center\">0.96 (0.03)</td></tr><tr><td/><td/><td/><td align=\"center\">C</td><td align=\"center\">0.77 (0.10)</td><td align=\"center\">0.96 (0.03)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T7\"><label>Table 7</label><caption><p>Results for latent class models focussing on association of latent exposure variable and disease</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Model</bold></td><td align=\"center\"><bold>Log-likelihood</bold></td><td align=\"center\"><bold>Lab</bold></td><td align=\"center\"><bold>Sensitivity (SE)</bold></td><td align=\"center\"><bold>Specificity (SE)</bold></td><td align=\"center\"><bold>Estimated prevalence in cases (SE)</bold></td><td align=\"center\"><bold>Estimated prevalence in controls (SE)</bold></td><td align=\"center\"><bold>Odds ratio (CI)</bold></td></tr></thead><tbody><tr><td align=\"center\">5</td><td align=\"center\">-439.87</td><td align=\"center\">A</td><td align=\"center\">0.94 (0.07)</td><td align=\"center\">0.90 (0.05)</td><td align=\"center\">0.34 (0.07)</td><td align=\"center\">0.22 (0.05)</td><td align=\"center\">1.83 (0.97–3.46)</td></tr><tr><td/><td/><td align=\"center\">B</td><td align=\"center\">0.91 (0.11)</td><td align=\"center\">0.94 (0.06)</td><td/><td/><td/></tr><tr><td/><td/><td align=\"center\">C</td><td align=\"center\">0.69 (0.10)</td><td align=\"center\">0.95 (0.03)</td><td/><td/><td/></tr><tr><td align=\"center\">6</td><td align=\"center\">-443.85</td><td align=\"center\">A</td><td align=\"center\">0.82 (0.07)</td><td align=\"center\">0.94 (0.03)</td><td align=\"center\">0.38 (0.07)</td><td align=\"center\">0.24 (0.05)</td><td align=\"center\">1.96 (1.02–3.75)</td></tr><tr><td/><td/><td align=\"center\">B</td><td align=\"center\">0.82 (0.07)</td><td align=\"center\">0.94 (0.03)</td><td/><td/><td/></tr><tr><td/><td/><td align=\"center\">C</td><td align=\"center\">0.82 (0.07)</td><td align=\"center\">0.94 (0.03)</td><td/><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T8\"><label>Table 8</label><caption><p>Results for empirical models focussing on association of laboratory values and disease</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Model</bold></td><td align=\"center\"><bold>Log-likelihood</bold></td><td align=\"center\"><bold>Lab</bold></td><td align=\"center\"><bold>Observed prevalence in cases (SE)</bold></td><td align=\"center\"><bold>Observed prevalence in controls (SE)</bold></td></tr></thead><tbody><tr><td align=\"center\">7</td><td align=\"center\">-475.23</td><td align=\"center\">A</td><td align=\"center\">0.50 (0.08)</td><td align=\"center\">0.29 (0.05)</td></tr><tr><td/><td/><td align=\"center\">B</td><td align=\"center\">0.34 (0.04)</td><td align=\"center\">0.24 (0.04)</td></tr><tr><td/><td/><td align=\"center\">C</td><td align=\"center\">0.32 (0.08)</td><td align=\"center\">0.21 (0.05)</td></tr><tr><td align=\"center\">8</td><td align=\"center\">-477.62</td><td align=\"center\">A</td><td align=\"center\">0.37 (0.03)</td><td align=\"center\">0.24 (0.03)</td></tr><tr><td/><td/><td align=\"center\">B</td><td align=\"center\">0.37 (0.03)</td><td align=\"center\">0.24 (0.03)</td></tr><tr><td/><td/><td align=\"center\">C</td><td align=\"center\">0.37 (0.03)</td><td align=\"center\">0.24 (0.03)</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2156-9-51-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>ℓ</mml:mi><mml:mi>n</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mrow><mml:mi>a</mml:mi><mml:mi>b</mml:mi><mml:mi>c</mml:mi><mml:mi>d</mml:mi><mml:mi>x</mml:mi></mml:mrow></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mi>u</mml:mi><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mi>a</mml:mi><mml:mi>A</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mi>b</mml:mi><mml:mi>B</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mi>c</mml:mi><mml:mi>C</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mi>d</mml:mi><mml:mi>D</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mi>x</mml:mi><mml:mi>X</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>a</mml:mi><mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mi>A</mml:mi><mml:mi>X</mml:mi></mml:mrow></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>b</mml:mi><mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mi>B</mml:mi><mml:mi>X</mml:mi></mml:mrow></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>c</mml:mi><mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mi>C</mml:mi><mml:mi>X</mml:mi></mml:mrow></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>d</mml:mi><mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mi>D</mml:mi><mml:mi>X</mml:mi></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2156-9-51-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>u</mml:mi><mml:mi>a</mml:mi><mml:mi>A</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2156-9-51-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>d</mml:mi><mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mi>D</mml:mi><mml:mi>X</mml:mi></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>¹Odds ratio (OR) relative to reference category combining Arg/Pro and Pro/Pro genotypes.</p><p>²OR adjusted for age and race.</p></table-wrap-foot>", "<table-wrap-foot><p>¹<underline>Non-stringent</underline> definition allows inter-laboratory disagreement for the Arg/Pro and the Pro/Pro genotypes; <underline>stringent</underline> definition includes only genotypes with complete agreement among the three laboratories (adapted from reference 1).</p><p>²<underline>Disagreed</underline>: includes only subjects with an Arg/Arg genotype determined by at least one laboratory but with different results from the other laboratories; <underline>Agreed</underline>: includes only Arg/Arg subjects with complete agreement among laboratories; <underline>All-inclusive</underline>: includes any reported Arg/Arg genotype with or without agreement among laboratories or in isolation</p><p>³OR adjusted for age and race.</p></table-wrap-foot>", "<table-wrap-foot><p>* error rates are (additionally) differential by disease status</p><p>D- disease; X-true genotype status; E-genotype status measured by laboratory test.</p></table-wrap-foot>", "<table-wrap-foot><p>= : indicates terms constrained to be equal</p></table-wrap-foot>" ]

[]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p><italic>Calluna vulgaris </italic>L. (Hull.), an exclusive species within the genus <italic>Calluna</italic>, has increased its economic weight, and not only within the German horticultural industry over the last few decades: In 2005 almost 100 million plants were produced in Germany, of which about 30% were exported to other European countries [##UREF##0##1##] where the demand is also still increasing. Although merely a handful of breeders are commercially active in breeding <italic>C. vulgaris</italic>, more than 300 varieties now exist, which are or have been protected at the Bundessortenamt, Hannover (BSA) [##UREF##1##2##] and/or the Community Plant Variety Office (CPVO), Angers, France [##UREF##2##3##]. More than 50% of applications for variety protection at the CPVO date from 2003 or later, which supports the argument of the increasing importance of <italic>C. vulgaris</italic>.</p>", "<p>Breeding efforts in <italic>C. vulgaris </italic>primarily aim at a special type of its inflorescence, the so-called bud flowers (Fig. ##FIG##0##1##). Flowers of these plants do not open during the entire reproduction phase from August to December which makes them appear visually attractive for a long period of time when not many other flowering ornamental outdoor plants are available in the northern hemisphere. This phenotype is closely linked with and possibly caused by a lack of anthers. This connection, in turn, has a severe impact on breeding methods because interesting bud-flowering genotypes are only applicable as the female parent in crossings. In addition, there is only sparse information and hypotheses available concerning the inheritance of this trait. Therefore – and since <italic>C. vulgaris </italic>is a vegetatively propagated crop – breeding in <italic>C. vulgaris </italic>over the past few decades was to a large extent performed by selection of spontaneous mutations, rather than by systematic crossings (personal communications with breeders). The actual variety composition in Europe offers a mixture of normal flowering and bud flowering types (state: 01/2008) with main focus on the latter (~85%). Some special forms (e.g. 'Radnor' with filled flowers or 'Peace' as a multi-bracteate type) are present as well. However, due to the problems described above, the actual gene pool used in breeding of <italic>C. vulgaris </italic>is presumably quite narrow.</p>", "<p>Therefore, in this study the genetic diversity within the species <italic>C. vulgaris </italic>was examined with molecular DNA techniques, comprising a selection of 64 economic important and partially still-protected varieties from Germany, including varieties from other European countries and the USA, 5 genotypes resulting from crossings, as well as a selection of 5 wild plants of different origin. Moreover, 3 different genotypes of <italic>Erica </italic>spp. were included in this study as an anticipated outgroup [see Additional file ##SUPPL##0##1##].</p>", "<p>In the case of <italic>C. vulgaris</italic>, variety protection assessments as executed by the CPVO and as described in the Protocol for Distinctness, Uniformity and Stability Tests for <italic>Calluna </italic>L. (Hull.), LING, Scots Heather (CPVO-TP/94/Final of 06/11/2003), comprises 22 phenotypic traits in total but only 18 traits for bud-flowering varieties, which are tested in 2 flowering seasons with 30 plants (replications). Herein, problems arise from continually increasing applications for protection of bud-flowering genotypes, from their partial identicalness in many of these traits and from the subjectiveness that is inherent in the measurement of phenotypic traits. Moreover, breeding of bud flowering types requires backcrossing, which is – in contrast to mutant selection or 'cosmetic breeding' – 'true breeding', but which also contributes to the narrow gene pool. Previously, these drawbacks led to some juridical disputes in the field of variety derivation in <italic>C. vulgaris </italic>in Germany.</p>", "<p>The problem of variety derivation and the need for an appropriate protection system was already identified decades ago and is especially pressing in the context of global marketing. In Europe, the Act of Convention from 1991 followed on from a Convention on the Protection of Plant Varieties [##UREF##3##4##] and first introduced the term 'Plant Breeder's Rights'. Today it is acknowledged by 65 member states (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.upov.int\"/>, state: 01/2008). Variety protection in these member states is based on DUS-tests (distinctness, stability, uniformity: see above). Despite increasing testing efforts, these tests remain sketchy since the investigated traits may be influenced by several factors, e.g. environmental changes, and are evaluated by subjective ratings so that molecular markers have become a desirable tool [##UREF##4##5##]. But also the so-called 'fingerprinting techniques' – although widely recommended as a supplement to phenotypic tests e.g. by [##UREF##5##6##] – entail a number of problems, since only a random sample i.e. a subset of the genome, can be examined. Therefore, any statistical method applied to this problem has to be able to maintain a delicate balance in order to avoid excessive identification of false positives on the one hand, as well as false negatives on the other [##UREF##5##6##, ####UREF##6##7##, ##UREF##7##8####7##8##].</p>", "<p>Several case studies have been recently published in the context of ED-conflicts, however, for the most part these do not concern vegetatively propagated species, since it is generally assumed that variation in such varieties does not occur, which would allow clear-cut molecular genotyping. [##UREF##8##9##] tested essential derivation in various vegetatively propagated ornamentals (<italic>Rhododendron</italic>, <italic>Rosa</italic>, <italic>Phalaenopsis</italic>) by AFLP-genotyping and constructing UPGMA-dendrograms. However, these investigations relied on the assumption of total genetic stability within vegetatively propagated varieties and therefore dispensed with any statistical analysis. From our point of view, this is not appropriate for all vegetatively propagated species, because – for example with <italic>Calluna </italic>– phenotypic variations (sports) are well-known and are based on genotypic variation. From these experiences we support EDV-identifying systems with respect to statistical validation as with the one introduced by [##UREF##9##10##] for lettuce and barley [##UREF##10##11##], which is based on the definition of a minimum distance (threshold) for distinctness. Such procedures are necessary since proving identity is more difficult than proving distinction with molecular markers [##UREF##11##12##]. Lettuce is a self-fertilizer and consequently genetic variation within varieties can be expected to be very low. Moreover total variation between today's cultivars should be somewhat reduced due to an intensive breeding history. For this reason, [##UREF##10##11##] suggested that ED-conflicts should not be analyzed through the construction of a dendrogram visualizing hypothetical kinship relations, but instead by the examination of all pairwise genetic distances within an appropriate reference population, and then comparing these results to the distance between actual varieties in question.</p>", "<p>As a result, another aim of our study, drawing on the publication by [##UREF##10##11##], was to implement a comparable concept of identification of EDVs in <italic>C. vulgaris </italic>based on molecular data resulting from RAPD and iSSR techniques. Our system proposal is critically evaluated with regard to essential premises e.g. variation and stability [##UREF##12##13##], its success in <italic>C. vulgaris</italic>, and its practicability in the future.</p>", "<p>The results presented here were obtained during a BMWi-(Federal Ministry of Economics and Technology) funded cooperation between the IGZ and a German breeding company (Heidepflanzen Peter de Winkel, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.dewinkel.de\"/>). Thus, variety denotation is ciphered in cases where the breeder's interests may be affected.</p>" ]

[ "<title>Methods</title>", "<title>DNA techniques: isolation of genomic DNA (gDNA)</title>", "<p>gDNA of <italic>C. vulgaris </italic>genotypes was isolated according to [##UREF##20##27##]. About 200 mg young leaf tissue (stored over night and frozen in liquid nitrogen) was homogenized in 2 ml tubes in a mixer mill (MM301, Retsch) using 2 stainless steal balls (Ø = 5 mm). The tissue was resuspended in buffer A (50 mM Tris-HCl pH 8.0, 5 mM EDTA, 350 mM sorbitol, 1% β-mercaptoethanole, 10% PEG-6000) and centrifuged for 1 min at 4°C and 8,000 rpm (Sigma 3K30, rotor-no. 12148). The resulting pellet was again resuspended in buffer B (50 mM Tris-HCl pH 8.0, 5 mM EDTA, 350 mM sorbitol, 1% β-mercaptoethanole, 1% sodiumsarcosyle, 0.1% CTAB, 710 mM NaCl) and incubated for 30 min at 60°C. After adding 0.8 volumes chloroform-isoamyl alcohol 24:1, the samples were centrifuged for 15 min at 4°C at 15,300 rpm. The supernatant was transferred to a new 2 ml reaction tube and incubated at -20°C for 30 min after adding 0.75 volumes isopropanole. After centrifugation (5 min at 4°C at 5,000 rpm) the pellet was washed with 70% ethanol, air-dried and resuspended in 500 μl TE buffer. To each sample, 1 ng RNAse (Carl Roth GmbH) was added, followed by incubation for 15 min at 37°C. Subsequently, phenol-chloroform extraction step was performed twice and the resulting supernatant containing purified gDNA was pelleted at -20°C for 60 min after addition of 0.1 volumes 3 M sodium acetate and 0.75 volumes isopropanole. The precipitated gDNA was washed twice with 70% ethanol, air-dried and resuspended in 100 μl TE. Long-time storage was achieved at -20°C.</p>", "<title>DNA techniques: PCR amplification and electrophoresis</title>", "<p>Amplifications of RAPD-fragments generated from random decamer primers (Carl Roth GmbH) were performed in a Primus 96 advanced thermocycler (peqlab GmbH) using the following protocol: 5 min at 95°C, [1 min at 95°C, 1 min at 35°C, 1 min at 72°C]_35×, 10 min at 72°C. The reaction mixture for a total volume of 25 μl contained 1× reaction buffer, 2.5 mM MgCl₂, 1 U Taq.-DNA Polymerase (recombinant, Invitrogen), 0.2 mM of each dNTP (Invitrogen), 0.5 μM primer (Carl Roth GmbH, MWG Biotech AG), 10 ng gDNA and the adequate amount of sterile deionized H₂O.</p>", "<p>Amplification of iSSR-Fragments was performed following the same protocol as described for RAPDs, with the altered annealing temperatures according to primer length. Table ##TAB##2##3## provides an overview of primers used in this study; these were chosen after screening 60 decamer primers for reproducibility. Decamer primers were obtained as random primer kits from Carl Roth GmbH; iSSR primers were synthesized by MWG Biotech AG. iSSR primers given in Table ##TAB##2##3## were chosen by referring to the common di- and trinucloetide motifs in plants (AC/TG)_nand (AAG/TTC)_n(e.g. [##REF##9734070##28##]).</p>", "<p>Electrophoretic separation of the amplification products was performed in 23 × 25 cm 1.5% agarose gels by applying 7 V/cm for 2.5 hours. The gel contained ethidium bromide for visualization of fragments at 254 nm. Documentation was carried out with a digital imaging system (Biostep GmbH).</p>", "<p>Reactions were repeated at least twice before fragments were used for distance calculations.</p>", "<title>Statistics: gel analysis and phylogenetic calculations</title>", "<p>Gel analysis (band detection, noise reduction, size calibration, fragment matching) was performed with the Phoretix 1D Advanced software (Nonlinear Dynamics). The selection of bands derived from each primer was performed by objective criteria (e.g. thresholds for routine band detection and matching and recommendations to ensure reproducibility, e.g. the exclusion of fragments of very high and very low size). The banding data were transformed to a computable 0/1 matrix in the common [OTU × band] layout.</p>", "<p>Phylogenetic as well as dendrogram calculations were conducted with the NTSYSpc 2.20 L software (^©1986–2006, Applied Biostatistics Inc.). Qualitative banding values were computed using the <italic>SimQual </italic>module with the similarity coefficient of Dice [##UREF##21##29##] and Nei and Li [##REF##291943##30##], respectively. Subsequent UPGMA clustering was conducted within the <italic>Sahn </italic>module, while the module <italic>Treeview </italic>was used to visualize the data set as a dendrogram.</p>", "<p>For Bootstrapping using the Dice coefficient, Winboot [##UREF##22##31##] was used (replications given in the text) which finally constructs a majority-rule consensus tree based on the <italic>Consense </italic>module of the PHYLIP software.</p>", "<title>EDV-testing by application of the tail principle</title>", "<p>A system first published by [##UREF##10##11##] for EDV identification in lettuce and barley was adapted for <italic>C. vulgaris </italic>as follows. Since a priori pedigree information is unavailable for <italic>C. vulgaris </italic>and the application of the <italic>pedigree principle </italic>– a threshold selection based upon inclusion of 'identity by descent' probabilities – was not possible, we selected the <italic>tail principle </italic>for the identification of a threshold from a distribution of pair-wise similarities from a reference-set. The configuration of the Reference Set (25 varieties) for each Test Set (2 varieties) was adapted by using the information gained from our phylogenetic results which matches the integration of the <italic>calibration principle</italic>. In the context of [##UREF##10##11##], we decided not to include known EDVs in this set, e.g. mutation-derived varieties or sports, since these would only represent extremely high values within the reference-set and would complicate data interpretation. Both polymorphic and monomorphic markers were analyzed. Detached primer-wise computation of similarity values becomes applicable by assuming a genetic independence between primers and a uniform distribution of primer binding sequences throughout the genome. Arithmetic means, standard deviations, standard errors as well as its medians were calculated according to [##UREF##9##10##] and [##UREF##10##11##] in analyzing the inner- and inter-set-similarities and also to define a threshold for identifying EDVs in <italic>Calluna</italic>. This threshold was positioned in such a way that varieties known to be ED exceed the threshold and non-EDVs do not.</p>", "<p>As carried out before for RAPD and iSSR values, these data were computed for their similarity values with the coefficient of Dice (Nei &amp; Li) by NTSYSpc 2.20 L. Only in case of Dice similarity values above the threshold and non-overlapping error bars, a pair of genotypes is categorized as being essentially derived. In a case where similarity values are below the threshold or where they are above but with overlapping error bars, a pair of genotypes is categorized as not essentially derived.</p>" ]

[ "<title>Results</title>", "<title>Estimation of genetic diversity and kinship relations within <italic>C. vulgaris</italic></title>", "<p>Using RAPD- and iSSR-techniques, we achieved a total of 129 (RAPD) and 39 (iSSR) distinguishable and reproducible bands. This corresponds to 9.9 bands/RAPD primer and 7.8 bands/iSSR primer. The combined results of RAPD and iSSR studies are shown in the dendrogram in Fig. ##FIG##1##2##. While the three <italic>Erica </italic>genera do cluster as an outgroup, all tested genotypes from the <italic>Calluna </italic>species cluster to the right of one node. Interestingly, the wild-types from Thuringia (Ruhla) and from the Italian Alps (San Remo) cluster as an additional outgroup within the <italic>Calluna </italic>species while the other wild-types available (Löhnstein, Niederohe, Tiefenthal, all from the Lüneburger Heide in Germany) are grouped within the rest of the <italic>Calluna </italic>genotypes.</p>", "<p>The statistical significance of our data was investigated with the resampling method of bootstrapping as initially described by [##UREF##13##14##] using the software Winboot and n = 10,000 replications. Those few nodes with moderate support (50% &lt; p &lt; 85%), as well as strong support (p ≥ 85%), which appeared both in the NTSYSpc-constructed tree as well as in the majority-rule consensus tree of Winboot, are marked with * and **, respectively, in Fig. ##FIG##1##2##. The linked genotypes to the right of these nodes may be considered to be linked in real kinship. Despite the high number of analyzed bands, all other linkages are statistically unconfirmable within the present data set.</p>", "<title>Identifying EDVs in <italic>C. vulgaris</italic></title>", "<p>Due to former juridical conflicts concerning property rights of varieties in the genus <italic>Calluna </italic>we endeavored to develop a reliable statistical system for identifying EDVs in this species based on the results from the first part of our study. Since the dendrogram analysis did not support statistically significant decisions on kinship relations and probably would not do so even after analysis with a clearly expanded data set, we decided to implement a method based on a procedure published by [##UREF##10##11##] for similar analyses in lettuce. We therefore created appropriate Reference Sets of 25 varieties (Table ##TAB##0##1##) for each pair of tested genotypes (Test Set) in question and then computed primer-wise and pair-wise similarity values within each set. The Test Sets were chosen to represent non-ambiguous EDV or clear non-EDV cases for proof of concept, as well as several cases of interest in <italic>Calluna </italic>(Table ##TAB##1##2##). This non-ambiguousness was derived from personal communications with the involved breeding company in case of the EDV-pair. The test of a BC₁against the parents as a clear non-EDV case was performed with our own crossings.</p>", "<p>After extensive testing we selected a threshold provided by the highest Dice value of the 98% lowest values of all pairwise comparisons within the reference set (Fig. ##FIG##2##3##). This threshold was chosen in order to prevent the BC₁individual from being categorized as essentially derived from the backcross parent which constitutes an essential prerequisite for validation of our test since backcrossing is the normal breeding system in bud-flowering <italic>Calluna</italic>. The 98% thresholds in both Reference Sets differ due to the necessary adjustment of the reference set according to the test in question (exchange of wild-type genotypes against varieties from the upper cluster of the dendrogram): 98%-Set A: 0.865 Dice similarity value, 98%-Set B: 0.893 Dice similarity value.</p>", "<p>For proof of concept we tested, on the one hand, one pair of individuals ('Maria' and <italic>Maria Hell</italic>), from which it was known that the latter was derived from the first one. On the other hand, an individual from a backcross progeny was tested against both parents, which should result in the categorization of being non-derived. As expected, the first result was positive and the second one negative, using the threshold as given above (Fig. ##FIG##2##3##). Moreover, similarity between the BC₁individual and the backcross parent was clearly higher than between the BC₁individual and the second parent. The Dice value of the comparison with the backcross parent was actually slightly above the threshold; however, overlapping error bars indicated that the similarity was nevertheless not sufficiently high for these two genotypes to be categorized as essentially derived.</p>", "<p>Regarding the 'true tests', the results were negative for several pairs of morphologically similar cultivars from different breeders (Fig. ##FIG##2##3##), as well as for wild genotypes of different origin (Fig. ##FIG##3##4##). In contrast, when testing the cultivars 'Melanie' and 'Anette', their genetic similarity was found to exceed the threshold, thus confirming the public data supplied by the BSA according to which 'Anette' is a sport of 'Melanie'. This was also confirmed for 'Melanie' vs. 'Sandy' and 'Annegret' vs. 'Anneliese' (Fig. ##FIG##2##3##).</p>", "<p>The last test concerned a pair of cultivars ('Fritz Kircher' vs. CV7) which have in a former, non-public study been characterized as being essentially derived from one another using dendrogram analysis. In our investigation, however, their genetic similarity is lower than the threshold, thus clearly indicating an absence of essential derivation.</p>" ]

[ "<title>Discussion</title>", "<p>Until now, molecular data on genetic diversity within the species <italic>C. vulgaris </italic>was only available for regionally restricted wild-type populations [##REF##10383687##15##, ####REF##10632855##16##, ##REF##11903905##17####11903905##17##], not for varieties used in commercial breeding. An actual and urgent necessity for a comprehensive study in <italic>C. vulgaris </italic>can be deduced from several points: the number of applications for variety protection is currently increasing considerably, whereas the information given in the registration schedules is at least occasionally unreliable or equivocal (e.g. a bud flowering variety is said to be the result of selfing of another bud-flowering genotype, which is biologically impossible due to the total loss of anthers in bud-flowering genotypes). This leads to an ambiguous situation with regard to variety derivation. Additionally, molecular data are needed for concerted breeding works and the elimination of coincidence in this process.</p>", "<p>Since it is technically simple to accomplish and requires no a priori sequence information, iSSR- and RAPD-PCR [##UREF##14##18##, ####UREF##15##19##, ##REF##8020964##20####8020964##20##] are widely used techniques in different species; but RAPDs in particular may be 'considered the practice of PCR without a clue' [##UREF##16##21##]. All the same, both techniques provide a uniformly distributed amplification of DNA fragments throughout the genome of eukaryotic organisms due to the nature of their origin, and were shown to be an adequate molecular tool for studying DNA polymorphisms (e.g. [##REF##11501435##22##,##UREF##17##23##]). The same was true for our investigations as we observed very robust inner-laboratory reproducibility: here, a value as low as 0.46% of missing data within the 77 × 168 similarity matrix was achieved.</p>", "<p>The dendrogram resulting from the combined computation of both RAPD and iSSR banding patterns showed a low genetic variability within the species <italic>C. vulgaris</italic>: almost all tested varieties and genotypes are grouped at a Dice/Nei &amp; Li similarity value of 0.80, or even higher. This confirmed our hypothesis of a narrow gene pool, which was expected by the breeding experiences and methods applied of the participating company (personal communications) and its competitors. Moreover, one has to bear in mind that <italic>C. vulgaris </italic>is the only species within the genus <italic>Calluna </italic>and that crossing with other genera of the <italic>Ericaceae </italic>is thus impossible, thereby assisting in the conservation in nature, too, of a slender genetic diversity. We esteem the clear discrimination between <italic>Erica </italic>and <italic>Calluna </italic>as one argument of reassurance for our methodological approach and consider the dendrogram to be unbiased in the sense of an essential prerequisite for picturing genetic data [##UREF##10##11##]. The fact that wild type genotypes from the Lüneburger Heide are grouped this near to economically important varieties is another piece of evidence for our line of argument in respect of a significantly narrow gene pool in <italic>C. vulgaris</italic>. In addition, to our knowledge, breeding in <italic>C. vulgaris </italic>began in exactly this area of Germany by collecting incidentally originated bud-flowering genotypes. Our results might thus confirm this hypothesis, especially since the wild types from Thuringia and the Italian Alps do not cluster within this group.</p>", "<p>Another interesting feature of the resulting dendrogram is that the data were insufficient to support more than the few marked nodes (marked with * or ** in Fig. ##FIG##1##2##) as statistically significant. However, we do not consider the amount of bands i.e. mono-/polymorphisms from our data as generally too sparse, since [##UREF##18##24##] showed that an estimation of diversity within one population using approx. 200 dominant (i.e. AFLP) markers is as efficient as using 50 codominant (i.e. microsatellite) markers. Therefore, it is our suspicion, that the dendrogram method is not suitable for EDV identification in species with narrow gene pools.</p>", "<p>ED issues arise for varieties that successfully passed DUS testing. An EDV is (i) predominantly derived from an initial variety, (ii) clearly distinguishable from it and except for these differences (iii) conforms to the initial variety in the expression of essential characteristics [##UREF##4##5##]. We consider it to be of paramount importance to apply a well adjusted system for identification of these EDVs for each species, and in our case for <italic>C. vulgaris</italic>, since the range of similarities presented in Fig. ##FIG##1##2## proved the hypothesis of some breeders that the economically important varieties (and the genus <italic>Calluna </italic>in general) are closely related and thereby may readily lead to ED disputes, as has already been the case in the past.</p>", "<p>As explained above, construction of a dendrogram proved to be no satisfactory tool for EDV identification in <italic>C. vulgaris </italic>– contrary to the results obtained for other vegetatively propagated species presented by [##UREF##8##9##] for <italic>Phalaenopsis, Rosa </italic>and <italic>Rhododendron</italic>. Another example is given by [##REF##15490105##25##]. Using AFLPs, they proved that <italic>Rosa × hybrida </italic>original varieties are not more closely linked than 0.80 Jaccard's index. In contrast, the genetic similarities in so-called mutant groups were always higher than 0.96 (but not 1.0). Their dendrogram assay is therefore correctly rated as a suitable method to unambiguously distinguish rose EDVs from their initial variety. In addition, the detection of polymorphisms between sports and the original variety may be considered somewhat coincidental since molecular markers only cover a small portion of the target organism's genome. [##UREF##19##26##] demonstrated, that in cut roses RAPD-polymorphisms between a variety and its sports did occur in two varieties, but were not reproducible. Using AFLPs the authors were even able to amplify stable polymorphisms in sports of another variety. However, they were still able to distinguish vegetatively and sexually propagated progenies, since amplification in seedlings constantly resulted in a higher number of polymorphisms.</p>", "<p>We ascribe our differing results to the coincidence of two phenomena in <italic>C. vulgaris</italic>. First, stable genetic conditions – which could be reasonably anticipated for vegetatively propagated species – are worthy of discussion in the context of <italic>Calluna</italic>, since the phenomenon of sport/reversion (a type of somatic mutation) is well-known by breeders.</p>", "<p>Moreover, the very narrow gene pool in <italic>C. vulgaris </italic>gives rise to high genetic similarities, even if a new variety was obtained through crossing, due to the fact that even quite different individual plants, e.g. a wild type from the Lüneburger Heide and a bud flowering variety, show a considerable proportion of monomorphic bands in RAPD and iSSR analyses. Such lack of genetic diversity is our main reason for focusing on a system for EDV-identification involving a reference-set, as this is the important difference to e.g. the rose cases mentioned above: even in <italic>Rosa × hybrida </italic>more than 10,000 varieties exist, resulting from some 150 years of breeding efforts [##REF##15490105##25##], and they are still clearly distinguishable. The opposite situation is, in fact, the result of the differing breeding methods applied in <italic>C. vulgaris</italic>: breeding for a common phenotype (bud-flowering) and repeated back-crossing are generally accepted reasons that promote the development of narrow gene pools [##REF##15490105##25##].</p>", "<p>By working with a system similar to that described for lettuce and barley by [##UREF##10##11##], we were successful in both, identifying well-known essentially derived genotypes as well as discriminating between a genotype resulting from backcrossing and its parents (Fig. ##FIG##2##3##). We considered these results as a proof of concept for our method and additionally analyzed other test-sets whose information of origin we regarded to be unreliable, questionable or simply of interest. Here, information on variety derivation was primarily confirmed by our method as outlined in table ##TAB##1##2##. Moreover, the system discriminated phenotypically similar varieties from different breeders as well as wild genotypes of different origin, thus also confirming the hypotheses.</p>" ]

[ "<title>Conclusion</title>", "<p>As a result of these findings, we would like to suggest the outlined method as an appropriate system for EDV-testing in <italic>C. vulgaris</italic>. Applicability to other vegetatively propagated crops should be tested, as well as the combined use of 'fixed' and 'random/unmapped markers' as suggested by [##UREF##10##11##]. Moreover, we recommend the inclusion of at least three independent gDNA isolations of different individuals per genotype, since inner-varietal identity cannot be presumed and is hard to verify, even in vegetatively propagated crops.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Variety protection is of high relevance for the horticultural community and juridical cases have become more frequent in a globalized economy due to essential derivation of varieties. This applies equally to <italic>Calluna vulgaris</italic>, a vegetatively propagated species from the <italic>Ericaceae </italic>family that belongs to the top-selling pot plants in Europe. We therefore analyzed the genetic diversity of 74 selected varieties and genotypes of <italic>C. vulgaris </italic>and 3 of <italic>Erica </italic>spp. by means of RAPD and iSSR fingerprinting using 168 mono- and polymorphisms. The same data set was utilized to generate a system to reliably identify Essentially Derived Varieties (EDVs) in <italic>C. vulgaris</italic>, which was adapted from a method suggested for lettuce and barley. This system was developed, validated and used for selected tests of interest in <italic>C. vulgaris</italic>.</p>", "<title>Results</title>", "<p>As expected following personal communications with breeders, a very small genetic diversity became evident within <italic>C. vulgaris </italic>when investigated using our molecular methods. Thus, a dendrogram-based assay to detect Essentially Derived Varieties in this species is not suitable, although varieties are propagated vegetatively. In contrast, the system applied in lettuce, which itself applies pairwise comparisons using appropriate reference sets, proved functional with this species.</p>", "<title>Conclusion</title>", "<p>The narrow gene pool detected in <italic>C. vulgaris </italic>may be the genetic basis for juridical conflicts between breeders. We successfully tested a methodology for identification of Essentially Derived Varieties in highly identical <italic>C. vulgaris </italic>genotypes and recommend this for future proof of essential derivation in <italic>C. vulgaris </italic>and other vegetatively propagated crops.</p>" ]

[ "<title>Abbreviations</title>", "<p>AFLP: Amplified Fragment Length Polymorphism; ASSINSEL: 'International Association of Plant Breeders for the Protection of Plant Varieties'; EDV: Essentially Derived Variety/Varieties; BSA: Bundessortenamt; CPVO: Community Plant Variety Office; gDNA: genomic DNA; ISF: International Seed Federation; iSSR: inter Simple Sequence Repeats; RAPD: Randomly Amplified Polymorphic DNA; UPOV: International Union for the Protection of new Varieties of Plants; UPGMA: Unweighted Pair Group Method with Arithmetic Mean</p>", "<title>Authors' contributions</title>", "<p>After methodological setup, TB carried out the complete RAPD section from laboratory work to analysis and drafted the manuscript. JK performed the complete iSSR part from laboratory work to analysis. AH designed the study and participated in drafting the manuscript.</p>", "<p>All authors read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>This study was carried out as part of the BMWi-funded project (KP0172401BN5A) between the IGZ and Heidepflanzen Peter de Winkel, Goch <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.dewinkel.de\"/>. The authors wish specifically to thank the company owner for his support and especially Katja Krüger for her sustained and motivating assistance throughout the daily laboratory work. Additionally, we would like to thank Prof. Dr. T. Debener (University of Hannover) for his support concerning the experimental design and the manuscript review.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Shoots of two <italic>C. vulgaris </italic>genotypes representing the main inflorescence types</bold>. left: normal ('White Mite'), right: bud ('Anneliese').</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Dendrogram consisting of 74 <italic>C. vulgaris </italic>and 3 <italic>Erica </italic>spp. genotypes</bold>. Constructed from 168 mono- and polymorphisms amplified from 13 RAPD and 5 iSSR-primers and based on the Dice/Nei and Li coefficient with subsequent UPGMA-clustering. Nodes with strong support (&gt; 85%) by bootstrapping (n = 10.000, PHYLIP) are marked with **, moderately supported groups (50% – 85%) are marked with *, varieties of interest for the involved company are ciphered by CV# where # is replaced by increasing numbers. Variety encryption is known to the authors and the company, respectively. For purposes of clarity and according to their regional provenance, genotypes have been classified by symbols as indicated.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Identifying essential derivation in <italic>C. vulgaris </italic>I</bold>. Validation of the method using three sets for proof-of-concept: One set of a known essentially derived variety pair and two sets of genotypes involved in backcrossing, marked by black symbols. Additionally six pairs of varieties of interest have been tested against the chosen threshold of 0.865 Dice similarity value, which was derived from Reference Set A.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Identifying essential derivation in <italic>C. vulgaris </italic>II</bold>. Test of three pairs of wild types of different origin using Reference Set B (0.893 Dice similarity value).</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Identifying essential derivation in <italic>C. vulgaris</italic>. </p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"2\"><bold>Reference Sets</bold></td></tr></thead><tbody><tr><td/><td align=\"center\"><bold>Set A</bold></td><td align=\"center\"><bold>Set B</bold></td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td align=\"center\"><bold>1</bold></td><td align=\"center\">Niederohe</td><td align=\"center\">'Sandy'</td></tr><tr><td align=\"center\"><bold>2</bold></td><td align=\"center\">San Remo</td><td align=\"center\">'Annegret'</td></tr><tr><td align=\"center\"><bold>3</bold></td><td align=\"center\" colspan=\"2\">'Adrie'</td></tr><tr><td align=\"center\"><bold>4</bold></td><td align=\"center\" colspan=\"2\">'Allegro'</td></tr><tr><td align=\"center\"><bold>5</bold></td><td align=\"center\" colspan=\"2\">'Boskoop'</td></tr><tr><td align=\"center\"><bold>6</bold></td><td align=\"center\" colspan=\"2\">'Carmen'</td></tr><tr><td align=\"center\"><bold>7</bold></td><td align=\"center\" colspan=\"2\">'C. W. Nix'</td></tr><tr><td align=\"center\"><bold>8</bold></td><td align=\"center\" colspan=\"2\">'Dark Beauty'</td></tr><tr><td align=\"center\"><bold>9</bold></td><td align=\"center\" colspan=\"2\">'Findling'</td></tr><tr><td align=\"center\"><bold>10</bold></td><td align=\"center\" colspan=\"2\">'Glenmorangie'</td></tr><tr><td align=\"center\"><bold>11</bold></td><td align=\"center\" colspan=\"2\">'Johnson's Variety'</td></tr><tr><td align=\"center\"><bold>12</bold></td><td align=\"center\" colspan=\"2\">'Long White'</td></tr><tr><td align=\"center\"><bold>13</bold></td><td align=\"center\" colspan=\"2\">'Mariella'</td></tr><tr><td align=\"center\"><bold>14</bold></td><td align=\"center\" colspan=\"2\">'Marlies'</td></tr><tr><td align=\"center\"><bold>15</bold></td><td align=\"center\" colspan=\"2\">'McDonalds of Glencoe'</td></tr><tr><td align=\"center\"><bold>16</bold></td><td align=\"center\" colspan=\"2\">'Minima Smith's Variety'</td></tr><tr><td align=\"center\"><bold>17</bold></td><td align=\"center\" colspan=\"2\">'Mrs. Pinxteren'</td></tr><tr><td align=\"center\"><bold>18</bold></td><td align=\"center\" colspan=\"2\">'Multicolor'</td></tr><tr><td align=\"center\"><bold>19</bold></td><td align=\"center\" colspan=\"2\">'Orange Queen'</td></tr><tr><td align=\"center\"><bold>20</bold></td><td align=\"center\" colspan=\"2\">'Peace'</td></tr><tr><td align=\"center\"><bold>21</bold></td><td align=\"center\" colspan=\"2\">'Radnor'</td></tr><tr><td align=\"center\"><bold>22</bold></td><td align=\"center\" colspan=\"2\">'Sandhammeren'</td></tr><tr><td align=\"center\"><bold>23</bold></td><td align=\"center\" colspan=\"2\">'Silver Knight'</td></tr><tr><td align=\"center\"><bold>24</bold></td><td align=\"center\" colspan=\"2\">'Underwoodii'</td></tr><tr><td align=\"center\"><bold>25</bold></td><td align=\"center\" colspan=\"2\">'Wickwar Flame'</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Identifying essential derivation in <italic>C. vulgaris</italic>. </p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>#</bold></td><td align=\"center\" colspan=\"2\"><bold>Test-Sets</bold></td><td align=\"center\"><bold>selection criteria</bold></td><td align=\"center\"><bold>Reference Set</bold></td><td align=\"center\"><bold>Hypothesis</bold></td><td align=\"center\"><bold>Result</bold></td></tr></thead><tbody><tr><td align=\"center\">1</td><td align=\"center\">'Maria'</td><td align=\"center\"><italic>Maria Hell</italic></td><td align=\"center\">Maria Hell = known sport of 'Maria' according to information from a breeder</td><td align=\"center\">A</td><td align=\"center\">yes</td><td align=\"center\">yes</td></tr><tr><td align=\"center\">2</td><td align=\"center\">'Maria'</td><td align=\"center\">BC₁-individual</td><td align=\"center\">progeny testing</td><td align=\"center\">A</td><td align=\"center\">no</td><td align=\"center\">no</td></tr><tr><td align=\"center\">3</td><td align=\"center\">'Roter Oktober'</td><td align=\"center\">BC₁-individual</td><td align=\"center\">progeny testing</td><td align=\"center\">A</td><td align=\"center\">no</td><td align=\"center\">no</td></tr><tr><td align=\"center\">4</td><td align=\"center\">'Melanie</td><td align=\"center\">'Anette'</td><td align=\"center\">'Anette' = sport of 'Melanie' according to information given by BSA doc</td><td align=\"center\">A</td><td align=\"center\">yes</td><td align=\"center\">yes</td></tr><tr><td align=\"center\">5</td><td align=\"center\">'Melanie'</td><td align=\"center\">'Sandy'</td><td align=\"center\">'Sandy' = sport of 'Melanie' according to information given by BSA doc</td><td align=\"center\">A</td><td align=\"center\">yes</td><td align=\"center\">yes</td></tr><tr><td align=\"center\">6</td><td align=\"center\">'Annegret'</td><td align=\"center\">'Anneliese'</td><td align=\"center\">'Anneliese' = sport of 'Annegret' according to information given by BSA doc</td><td align=\"center\">A</td><td align=\"center\">Yes</td><td align=\"center\">yes</td></tr><tr><td align=\"center\">7</td><td align=\"center\">'Fritz Kircher'</td><td align=\"center\">CV7</td><td align=\"center\">re-testing results from former investigations</td><td align=\"center\">A</td><td align=\"center\">yes</td><td align=\"center\">no</td></tr><tr><td align=\"center\">8</td><td align=\"center\">'Karla'</td><td align=\"center\">'Venetia'</td><td align=\"center\">similar cultivars from different breeders</td><td align=\"center\">A</td><td align=\"center\">no</td><td align=\"center\">no</td></tr><tr><td align=\"center\">9</td><td align=\"center\">'Minka'</td><td align=\"center\">'Miranda'</td><td align=\"center\">similar cultivars from different breeders</td><td align=\"center\">A</td><td align=\"center\">no</td><td align=\"center\">no</td></tr><tr><td align=\"center\">10</td><td align=\"center\">SanRemo</td><td align=\"center\">Ruhla</td><td align=\"center\">wild-type testing</td><td align=\"center\">B</td><td align=\"center\">no</td><td align=\"center\">no</td></tr><tr><td align=\"center\">11</td><td align=\"center\">Niederohe</td><td align=\"center\">Löhnstein</td><td align=\"center\">wild-type testing</td><td align=\"center\">B</td><td align=\"center\">no</td><td align=\"center\">no</td></tr><tr><td align=\"center\">12</td><td align=\"center\">Niederohe</td><td align=\"center\">SanRemo</td><td align=\"center\">wild-type testing</td><td align=\"center\">B</td><td align=\"center\">no</td><td align=\"center\">no</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>List of iSSR- and RAPD-primers. </p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>type</bold></td><td align=\"center\"><bold>denomination</bold></td><td align=\"center\"><bold>sequence (5' → 3')</bold></td><td align=\"center\"><bold>source</bold></td></tr></thead><tbody><tr><td align=\"center\">iSSR</td><td align=\"center\">17898B</td><td align=\"center\">(CA)₆-gT</td><td align=\"center\">according to [##UREF##16##21##]</td></tr><tr><td/><td align=\"center\">17898C</td><td align=\"center\">(CA)₆-AC</td><td/></tr><tr><td/><td align=\"center\">17899</td><td align=\"center\">(CA)₆-gg</td><td/></tr><tr><td/><td align=\"center\">17901B</td><td align=\"center\">(gT)₆-TT</td><td/></tr><tr><td/><td align=\"center\">P02</td><td align=\"center\">(AAg)₆-Cg</td><td align=\"center\">according to [##REF##9734070##28##]</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\">RAPD</td><td align=\"center\">RX13</td><td align=\"center\">ACgggAgCAA</td><td align=\"center\">random primer kits</td></tr><tr><td/><td align=\"center\">RX14</td><td align=\"center\">ACAggTgCTg</td><td/></tr><tr><td/><td align=\"center\">RY01</td><td align=\"center\">gTggCATCTC</td><td/></tr><tr><td/><td align=\"center\">RY13</td><td align=\"center\">gggTCTCggT</td><td/></tr><tr><td/><td align=\"center\">RY15</td><td align=\"center\">AgTCgCCCTT</td><td/></tr><tr><td/><td align=\"center\">RY16</td><td align=\"center\">gggCCAATgT</td><td/></tr><tr><td/><td align=\"center\">RY17</td><td align=\"center\">gACgTggTgA</td><td/></tr><tr><td/><td align=\"center\">RY18</td><td align=\"center\">gTggAgTCAg</td><td/></tr><tr><td/><td align=\"center\">RZ04</td><td align=\"center\">AggCTgTgCT</td><td/></tr><tr><td/><td align=\"center\">RZ05</td><td align=\"center\">TCCCATgCTg</td><td/></tr><tr><td/><td align=\"center\">RZ07</td><td align=\"center\">CCAggAggAC</td><td/></tr><tr><td/><td align=\"center\">RZ12</td><td align=\"center\">TCAACgggAC</td><td/></tr><tr><td/><td align=\"center\">RZ17</td><td align=\"center\">CCTTCCCACT</td><td/></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p><bold>Complete list of included varieties and genotypes, their country of origin and pedigree information where known (source is given in the last column)</bold>. Column 3 defines the flower type either as normal, bud, multi-bracteate (multi) or filled. Sources of information are either the Bundessortenamt (BSA:doc), the appropriate website (web) of the 'The International Register of Heather Names' <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.heathersociety.org.uk/handy_guide.html\"/> or personal communications (personal contact).</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>Identifying essential derivation in <italic>C. vulgaris</italic>: Reference-Sets A and B chosen for the Test-Sets. Set B differs from A only in the exchange of two wild-type genotypes (Niederohe, San Remo) versus 2 varieties ('Sandy', 'Annegret') as indicated by summarizing both columns of Set A and B for the residual 23 varieties.</p></table-wrap-foot>", "<table-wrap-foot><p>Test Sets and their criteria for selection, the applied Reference Set and our initial hypothesis with regard to whether or not ED was to be expected. The first three tests were used as proof of concept, meaning consistency of hypothesis and validation of the eligibility of the method; tests 4–12 are true testings.</p></table-wrap-foot>", "<table-wrap-foot><p>List of primers used for PCR-amplification of mono- and polymorphic fragments within gDNA of <italic>C. vulgaris</italic>, their sequence, and the source according to which the sequences were selected.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2156-9-56-1\"/>", "<graphic xlink:href=\"1471-2156-9-56-2\"/>", "<graphic xlink:href=\"1471-2156-9-56-3\"/>", "<graphic xlink:href=\"1471-2156-9-56-4\"/>" ]

[ "<media xlink:href=\"1471-2156-9-56-S1.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The bovine cluster of differentiation (CD) 14 is an important player in host innate immunity in that it mediates host defense against Gram-negative bacterial infections and also confers immunity against viral infections [##REF##10074110##1##,##REF##8612135##2##]. It is abundant (about 99,500 to 134,600) on the cell membrane of monocytes and to a lesser extent (about 1,900 to 4,400) on neutrophils (polymorphonuclear neutrophil leukocytes) [##REF##9201263##3##,##REF##8712510##4##]. Two forms exist, a membrane bound form (mCD14) and a soluble (sCD14) form [##REF##7542010##5##]. sCD14 is known to confer lipopolysaccharide (LPS) sensitivity to cells lacking mCD14, including epithelial cells and endothelial cells [##REF##7687581##6##,##REF##7681988##7##]. Also, recombinant bovine sCD14 can sensitize mammary epithelial cells to low concentrations of LPS <italic>in vivo </italic>and <italic>in vitro </italic>thus indicating an important role of sCD14 in initiating host responses to Gram-negative bacterial infections [##REF##12819092##8##,##REF##11943334##9##]. During the periparturient period, individual variations have been noticed in the response of cows to Gram-negative and Gram-positive bacteria infections [##REF##14556694##10##]. Therefore, sequence variations of the CD14 gene may play important roles in the presentation of CD14 molecules and thus, LPS sensitivity.</p>", "<p>The CD14 gene of cattle was initially cloned and sequenced by Ikeda et al. [##REF##9300371##11##] and recently by the bovine genome project. It is mapped to BTA 7. A SNP of human CD14 promoter has been shown to influence the activities of the gene [##REF##11698458##12##]. Baldini et al. [##REF##10226067##13##] reported a SNP in the human CD14 gene promoter involving a C to T transition at position -159 (or -260 by [##REF##10385492##14##]), and associated TT homozygotes with significantly higher levels of sCD14 and lower levels of IgE in children as compared to carriers of TC or CC. Reports of association of the -159 CD14 polymorphism with several human disease conditions have emerged [##REF##15842262##15##, ####REF##14517520##16##, ##REF##18008256##17####18008256##17##]. Other authors however did not find any association between this polymorphism and several diseases [##REF##16085746##18##,##REF##15741437##19##]. Despite the importance of this protein and the effect of the promoter polymorphism in humans, there is no report of sequence variations of this gene in cattle and their possible effects on circulating CD14 levels and disease susceptibility.</p>", "<p>The objectives of this study were therefore to: (1) investigate the CD14 gene of cattle for sequence variations; (2) determine the effect of variations on CD14 expression on the surfaces of monocytes and neutrophils; and (3) use bioinformatics tools to computationally characterize the promoter region. In this study we present information on sequence variations within the CD14 gene of Canadian Holstein and Jersey cows and a possible role of one SNP in influencing the surface expression of the antigen on the surfaces of neutrophils. Furthermore, identified conserved regions of regulatory element motifs may have important regulatory effects on the gene. The results may provide baseline information that may be used in candidate gene studies aimed at defining the role of CD14 in mediating bacterial infections.</p>" ]

[ "<title>Methods</title>", "<title>Animals and genomic DNA extraction</title>", "<p>Genomic DNA was extracted from the blood of 106 Canadian Holstein cows kept at the Howard Webster Centre-Macdonald Teaching Farm, McGill University and the milks of 46 Jersey cows enrolled in the Quebec Dairy Production Centre of Expertise program <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.valacta.com\"/> using Nucleospin Blood Mini Kit (Macherey-Nagel Inc. Easton, PA) as described by the manufacturer. In the case of milk DNA isolation, the manufacturer's protocol was slightly modified. Milk samples were initially centrifuged at 13000 rpm for 15 min at 4°C to remove excess fat before proceeding with the manufacturer's protocol.</p>", "<title>PCR amplification and sequencing</title>", "<p>Four primer pairs were designed with Invitrogen's OligoPerfect™ software (Invitrogen, Canada Inc., Burlington, ON, Canada) based on GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"NW_001495367\">NW_001495367</ext-link> and used to amplify overlapping regions covering the whole <italic>Bos taurus </italic>CD14 gene (Table ##TAB##2##3##). Invitrogen synthesized the primers.</p>", "<p>PCR reactions with all primer pairs were each carried out in a total volume of 45 μL containing 50 ng DNA, 0.25 mM dNTPs, 2.0 to 2.5 mM MgCl₂(Table ##TAB##2##3##), 10 μM of each primer, 2 units Tag DNA polymerase (Fermentas Life Sciences, Burlington, ON, Canada) and 1× <italic>Taq </italic>buffer. The cycling conditions, with PTC-100™ thermal cycler (MJ Research, Inc., Watertown, MA, USA) included an initial denaturation for 2 min at 94°C followed by 30 cycles comprising 30 sec at 94°C, 30 sec at 60°C, 50 sec at 72°C, and a final elongation for 5 min at 72°C. Both directions of amplified PCR products were sequenced by McGill University/Genome Quebec Innovation Centre using the big dye termination technique and an ABI 3700 sequencer.</p>", "<title>Sequence analysis</title>", "<p>Sequences were processed with Chromas, version 1.45 <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.technelysium.com.au/chromas14x.html\"/>) and comparison with other published sequences was done with the multiple sequence alignment program with hierarchical clustering, Multalign <ext-link ext-link-type=\"uri\" xlink:href=\"http://bioinfo.genopole-toulouse.prd.fr/multalin/multalin.html\"/>. CD14 protein sequences of different species (cattle, buffalo, goat, sheep, pig, man, mouse and rat) were aligned or processed with MEGA3.1 software [##REF##15260895##31##] and phylogenetic relationships also constructed with the same software.</p>", "<title>Computational characterization of the bovine CD14 promoter</title>", "<p>The promoter region was analyzed for the presence of putative transcription factor-binding sites using the combined search query against the TRANSFAC database with a maximum allowable string mismatch of 10% [[##UREF##1##32##]; <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.cbil.upenn.edu/cgi-bin/tess/tess\"/>]. The combined search query option was used to take advantage of the full power of combined string and weight matrix searching, pre-filtering of factors, significance p-values, and new information in new databases. Particular attention was paid to binding sites already proven to be of significance in regulating the CD14 gene of other species and other common binding sites in mouse and human.</p>", "<title>Identification of conserved motifs of regulatory elements in the non-coding regions of the CD14 gene of cattle and other species</title>", "<p>Since regions of conserved non-coding sequences between closely related or divergent species are likely to have common functional roles, we searched the region, about 500 bps of the core promoters of cattle (this study or GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>), human (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"U00699\">U00699</ext-link>), mouse (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"X13987\">X13987</ext-link>), rat (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"AF087944\">AF087944</ext-link>) and pig (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"DQ079063\">DQ079063</ext-link>) for described conserved regulatory element (RE) motifs against the NSITEM data base <ext-link ext-link-type=\"uri\" xlink:href=\"http://linux1.softberry.com/berry.phtml?topic=nsitem&amp;group=programs&amp;subgroup=promoter\"/>. These species were chosen because of the availability of complete or partial promoter sequence information.</p>", "<title>Flow cytometry</title>", "<p>Flow cytometry was used to study the effects of identified CD14 SNPs on the expression of CD14 on the surfaces of neutrophils and monocytes in healthy cows. Blood was collected from the caudal vein of 64 Holstein cows with known CD14 genotypes by venipuncture into vacutainer tubes coated with heparin anticoagulant (BD Biosciences, Franklin Lakes, NJ, USA). After collection, samples were stored on ice and analyzed within four hours. One hundred microlitre of heparinized whole blood was placed in a 12 × 75 mm flow cytometric (FCM) tube and incubated with 10 μL of fluorescein isothiocyanate (FITC)-labeled mouse anti human CD14 antibody (ABD Serotec Inc., Raleigh, NC, USA). This was mixed (Barnstead Thermolyne, Dubuque, IOWA, USA) thoroughly and incubated at room temperature on an orbitron rotator (Boekel Ind. Inc., PA, USA) for 30 minutes. Lysis and fixation of erythrocytes was done by adding 2 mL of lysing solution (PHAGOTEST^®Kit, Orpegen Pharma, Heidelberg, Germany) to the mixture. This was mixed gently, incubated for 20 minutes on an orbitron rotator at room temperature and centrifuged at 250 g for 5 minutes at 4°C. The supernatant was aspirated leaving approximately 400 μL of cells in the FCM tube. This was washed with 3 mL of Dulbecco's phosphate buffered saline (DPBS) pH 7.2 (Life Technologies) by centrifuging at 250 g for 5 minutes at 4°C. The supernatant was aspirated as described above and the cells resuspended in 1 mL of DPBS and analyzed by flow cytometry (Becton Dickinson Immunocytochemistry Systems, San José, CA, USA) within 30 minutes. Excitation of samples was at 488 nm; with FITC fluorescence measured at 525 nm ± 10 nm. Acquisition was stopped when 20,000 gated events were collected in the fluorescence cell count histogram. Gating of monocytes and polymorphonuclear leukocytes was based on forward scatter and side scatter dot plots by encircling the populations with amorphous regions. All parameters were recorded with logarithmic amplifications. List mode flow cytometric data from 20,000 events were stored and processed with the Windows Multiple Document Interface for flow cytometry (WinMDI) software version 2.8 (Joseph Trotter, The Sripps Research Institute, <ext-link ext-link-type=\"uri\" xlink:href=\"http://facs.scripps.edu/software.html\"/>) on a personal computer.</p>", "<p>The viability of neutrophils and monocytes in whole blood was determined by propidium iodide (PI) exclusion (50 μg/mL, final concentration) using flow cytometry after cells were incubated for 10 minutes in the dark at room temperature. The cells showed 99% viability.</p>", "<title>Statistical analysis</title>", "<p>Allele frequencies were estimated with GENEPOP program [##UREF##2##33##] while haplotypes and their frequencies were determined with the program PHASE V2.1.1 [##REF##14574645##34##,##REF##11254454##35##]. PHASE implements a Bayesian method of haplotype reconstruction based on genealogies reconstructed from coalescent theory under a Markov Chain Monte Carlo framework and has been shown to outperform other strategies such as the maximum likelihood expectation maximization algorithm in most cases [##REF##11254454##35##].</p>", "<p>Flow cytometric data were analyzed as a one-way ANOVA using the MIXED procedure SAS [##UREF##3##36##]. Treatment means were separated using the least square means option of SAS. Differences between treatment means were tested using Scheffe's Multiple Comparison test and statistical significance was declared at <italic>P </italic>&lt; 0.05.</p>", "<p>Statistical model used: Y_ij= μ + genotype_i+ e_ij</p>" ]

[ "<title>Results</title>", "<title>SNPs in the CD14 gene of Canadian Holsteins and Jersey cows</title>", "<p>Comparison of the CD14 sequences of 106 Canadian Holsteins and 46 Jersey cows with published sequences (GenBank Nos. <ext-link ext-link-type=\"gen\" xlink:href=\"NW_001495367\">NW_001495367</ext-link> and <ext-link ext-link-type=\"gen\" xlink:href=\"D84509\">D84509</ext-link>) revealed a total of five SNPs including one in the 5' untranslated region (UTR) (g.C1291T, numbering is according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>), two in the coding regions (g.A1908G and g.A2318G) and two in the 3' UTR (g.A2601G and g.G2621T) (Table ##TAB##0##1##). Four of the SNPs are transitional mutations while SNP 2621 involves the transversion of guanine to thymine. SNP 1908 is responsible for a non-synonymous codon change in amino acid 175 of the protein, from Asn (<bold>a</bold>ac) to Asp (<bold>g</bold>ac), while SNP 2318 results in a synonymous codon change without a change in amino acid 311 (Pro, cc<bold>a </bold>vs cc<bold>g</bold>) of the protein. The coding region SNPs characterizes three gene alleles <italic>A </italic>(A₁₉₀₈A₂₃₁₈) (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>), <italic>A</italic>₁(A₁₉₀₈G₂₃₁₈) (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148610\">EU148610</ext-link>) and <italic>B </italic>(G₁₉₀₈G₂₃₁₈) (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148611\">EU148611</ext-link>) and two deduced protein variants A (Asn175) (A₁₉₀₈A₂₃₁₈, GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link> or A₁₉₀₈G₂₃₁₈, GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68570\">ABV68570</ext-link>) and B (Asp175) (G₁₉₀₈G₂₃₁₈, GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68571\">ABV68571</ext-link>). Protein variant A is more common in the breeds analyzed with a frequency of 88.6% in Holsteins and fixed in Jerseys (Table ##TAB##0##1##). Within protein variant A, gene allele <italic>A</italic>₁occurred at a very high frequency (80.2%) in Jerseys. The non-coding SNPs occurred at about equal magnitudes in Holsteins while the frequencies of T₁₂₉₁, G₂₆₀₁and T₂₆₂₁were above 80% in Jerseys.</p>", "<p>Comparison of deduced protein sequences with reported CD14 protein sequences for cattle (UniProt AAD32215 and UniProt NP_776433) revealed a further amino acid difference, 209Ser→Thr. Further comparisons revealed that amino acid 175Asn is conserved in cattle, buffalo (UniProtKB ABE68724), goat (UniProtKB ABE68725) and sheep (UniProtKB NP_001070677) while Asp is present at this position in cattle variant B.</p>", "<title>Haplotye structure of the breeds</title>", "<p>Considering the five SNPs identified and the genotype information of all individuals sequenced, the program PHASE V2.2.1 determined a total of four potential haplotype combinations (CAAAG-C₁₂₉₁A₁₉₀₈A₂₃₁₈A₂₆₀₁G₂₆₂₁, CGAAG, TGGGT, TAGGT) in the analyzed breeds. TGGGT was absent in Jersey while CGAAG was absent in both populations. The gene alleles <italic>A1 </italic>and <italic>B </italic>were observed from the sequencing data to be associated with SNPs T₁₂₉₁, G₂₆₀₁and T₂₆₂₁while allele <italic>A </italic>was associated with C₁₂₉₁, A₂₆₀₁and G₂₆₂₁therefore giving rise to three actual haplotypes (TAGGT, TGGGT and CAAAG) in the analyzed populations (Table ##TAB##0##1##). The frequency of the haplotype associated with allele <italic>A </italic>(CAAAG) was highest (65.8%) in Holsteins while the haplotype associated with allele <italic>A</italic>₁(TAGGT) was highest in Jerseys (80.2%) (Table ##TAB##0##1##).</p>", "<title>Effects of CD14 genotypes on the expression of CD14 on the surfaces of monocytes and neutrophils</title>", "<p>Whole blood from healthy Holstein cows (animals showing no outward symptoms of infection and farm record indicating milk somatic cell counts below 200,000 cells/ml) with different CD14 genotypes were incubated with fluorescein isothiocyanate (FITC)-labeled mouse anti-human CD14 antibody to determine the effects of genotypes on the expression of CD14 antigens on the surfaces of monocytes and neutrophils. The results are presented in Figure ##FIG##0##1## and Table ##TAB##1##2##. In Figure ##FIG##0##1##, a higher percentage of gated monocyte cells were found in the LogFITC region labeled M2 (log10²and above) which indicates a higher fluorescence intensity coming from cells with the most CD14 antigens on their surfaces and termed the high expression region. On the other hand, more neutrophils were in the M1 or low expression region (Log10¹to 10²). The mean channel fluorescence (MCF) intensities (for all cows) observed for the gated regions were M1 = 32.04 (range 16.67 – 65.18), M2 = 288.97 (149.85 – 566.47) and M3 = 201.44 (73.57 – 437.56) for monocytes and M1 = 22.76 (16.20 – 35.52), M2 = 267.33 (148.26 – 536.87) and M3 = 34.76 (19.49–60.83) for neutrophils. As presented, the M3 (M1 + M2) MCF intensity for monocyte was higher than for neutrophils. SNP A1908G that changes amino acid 175Asn to Asp, and thus protein A to B, is significantly (P &lt; 0.01) associated with a higher number of neutrophils in the M2 or higher expression zone. 7.26% of gated neutrophils from cows of genotype 1908AG were found in M2 as compared to 4.36% from cows of genotype 1908AA (Table ##TAB##1##2##). In the M1 zone, the percentage of neutrophils from cows of both genotypes was the same. For monocytes, a higher number of total cells stained was observed for 1291CT (P &lt; 0.05) and the other genotypes (2318AG, 2601AG and 2621GT) in perfect linkage disequilibrium with C1291T but this difference disappeared when Scheffe's adjustments was applied to means. A similar result was recorded for haplotypes.</p>", "<title>Characterization of CD14 promoter and comparative analysis ofCD14 proteins</title>", "<p>The CD14 sequence analyzed in this study is made up of 2630 bps with a 1213 bp promoter (Figure ##FIG##1##2##), two exons and one intron (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>). With the use of bioinformatics tools and published information in the literature, we were able to identify putative transcription factor binding sites (TFBSs) (with 100% match against searched data bases) on the bovine CD14 promoter. The putative TFBSs described here are, in particular, those already demonstrated to control the expression of the gene in human and rat. The putative motifs are shown in Figure ##FIG##1##2## (boxed) and include amongst others 9 PU.1, 7AP-1, 5 SP1, 4 C/EBP, 3 c-Myb and 2 AP-2 sites.</p>", "<p>We also searched for conserved regulatory motifs in the core promoters of the CD14 genes of different species (cattle, human, mouse, rat and pig) through a query in the NSITEM data base <ext-link ext-link-type=\"uri\" xlink:href=\"http://linux1.softberry.com/berry.phtml?topic=nsitem&amp;group=programs&amp;subgroup=promoter\"/>. The results indicated a total of 63 motifs of 50 regulatory elements (REs) that were conserved in the analyzed breeds with zero to 3 bp mismatches (Figure ##FIG##2##3##). Nine of the motifs representing 6 REs including C/EBP and Spi-1/PU.1 were conserved in all the species while 4 motifs of 3 REs including AP1 and C/EBP-alpha were conserved in 4 species including cattle (Figure ##FIG##2##3##). Furthermore, 60.30% (38) of the motifs were conserved between cattle and at least two other species. Sequence alignment of the same region revealed a perfect conservation, both in nucleotide number and orientation of the TATA box in cattle, human, mouse and rat (data not shown). The TATA box of pig differed from the others by only one bp mismatch.</p>", "<p>Further, we checked the degree of conservation between the CD14 proteins of cattle with those of different species. Protein sequences compared were those deduced in this work, variants A (<ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link>) and B (<ext-link ext-link-type=\"gen\" xlink:href=\"ABV68571\">ABV68571</ext-link>), other published bovine sequences UniProt: BAA21517, AAD32215 and NP_776433, protein sequences of buffalo (<ext-link ext-link-type=\"gen\" xlink:href=\"ABE68724\">ABE68724</ext-link>), goat (<ext-link ext-link-type=\"gen\" xlink:href=\"ABE68725\">ABE68725</ext-link>), sheep (<ext-link ext-link-type=\"gen\" xlink:href=\"NP_001070677\">NP_001070677</ext-link>), pig (<ext-link ext-link-type=\"gen\" xlink:href=\"AAY98033\">AAY98033</ext-link>), mouse (<ext-link ext-link-type=\"gen\" xlink:href=\"CAA32166\">CAA32166</ext-link>), rat (<ext-link ext-link-type=\"gen\" xlink:href=\"NP_068512\">NP_068512</ext-link>) and human variant 1 (<ext-link ext-link-type=\"gen\" xlink:href=\"NP_001035110\">NP_001035110</ext-link>) and 2 (<ext-link ext-link-type=\"gen\" xlink:href=\"NP_000582\">NP_000582</ext-link>). The analysis revealed extensive conservations in the amino acid composition and structure. For the bovine proteins, the sequence of <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link> is the same as <ext-link ext-link-type=\"gen\" xlink:href=\"BAA21517\">BAA21517</ext-link> while <ext-link ext-link-type=\"gen\" xlink:href=\"AAD32215\">AAD32215</ext-link> and <ext-link ext-link-type=\"gen\" xlink:href=\"NP_776433\">NP_776433</ext-link> differed from <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link> by having amino acid 209 changed from Ser to Thr. This indicates the presence of a further CD14 protein variant in cattle here named C. Furthermore, the amino acids of the CD14 proteins of buffalo, sheep and goat shared high conservation rates of 97.05%, 95.17%, and 87.40% respectively, with the bovine <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link>, followed by pig (76.94%) (Figure ##FIG##3##4##). The rate of amino acid conservation of bovine A variant was less with the human (72.39%), mouse (61.66%) and rat (60.59%) proteins. While bovine, buffalo, goat, sheep and pig proteins are made up of 373 amino acids, the human protein has two more, and the rat and mouse have one more and seven less animo acids, respectively. The signal peptide, made up of the first 20 amino acids, was highly conserved (only one amino acid difference, 14Ser to Pro) between the bovine and buffalo/goat/sheep proteins. The difference between cattle and the pig was 4 amino acids while being highly diverged with human, mouse and rat (7 to 12 amino acid differences). These relationships were further represented phylogenetically (Figure ##FIG##3##4##). As depicted in Figure ##FIG##3##4##, three main groups were evident; the rat and mouse proteins in group one (bootstrap value 100), the human proteins in another group (bootstrap value 100) while members of the Artiodactyla order (bovine, buffalo, goat, sheep) and pig formed a group of their own (bootstrap value 99). In the third group, the pig formed an outcrop of its own, while a closer relationship was visible between cattle and buffalo on the one hand (bootstrap value 72) and, goat and sheep on the other hand (bootstrap value 78).</p>" ]

[ "<title>Discussion</title>", "<p>We report here sequence variations of the bovine CD14 gene of Canadian Holstein and Jersey cows, through genomic DNA sequencing and computational analysis of the promoter. A gene is made up of both coding and non-coding regions which are all important in its expression and functionality. The complete description of a gene must therefore contain necessary information about the protein coding regions [##REF##10404615##20##] and non-coding regions. The CD14 gene is an important component in host immunity [##REF##10074110##1##,##REF##12819092##8##] and detailed information on its structure and sequence variations as shown in this study may provide further insight into its mode of action.</p>", "<p>The coding region SNPs in our study and comparative analysis of our sequences with published sequences show that the CD14 gene of cattle codes for three putative CD14 proteins-A (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link> and <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68570\">ABV68570</ext-link>), B (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68571\">ABV68571</ext-link>) and C (UniProt AAD32215 and UniProt NP_776433), with A and B described herein. The A variant, fixed in Jerseys and with a high frequency of 88.6% in Holsteins, may be the original wild type allele for the gene. Furthermore, the sequence of UniProt BAA21517 is similar to variant A and the haplotyes that contain the A variant SNPs are at the highest frequencies in the studied breeds. The other variants may therefore be the result of recent mutational events. Further three SNPs described in the 5' and 3'UTRs and one synonymous SNP in the coding region of the gene indicates a higher sequence variation for the gene in Canadian Holsteins than Jerseys.</p>", "<p>The variations, both in the coding and non-coding regions of the gene may affect the surface expression of CD14 molecules on monocytes and neutrophils. Interestingly, the coding region SNP that gave rise to the B variant of the protein (g.A1908G or p.Asn175Asp) had the greatest effect by being associated with the highest number of neutrophils expressing more CD14 molecules on their surfaces. It is well known that, monocytes express more CD14 receptors on their surfaces, about 99,500 to 134,600 as compared to 1,900 to 4,400 for neutrophils (3). This difference was clearly shown by the pattern of expression depicted in Figure ##FIG##0##1## whereby, more monocytes were recorded in the M2 gated zone (region of higher expression) and more neutrophils in the M1 zone (lower expression). These results suggest that, the characteristic B protein SNP or g.A1908G may play a role in cell surface expression of CD14 on neutrophils. Also, the SNP in the 5'UTR region could be important in influencing the expression of this receptor on monocytes. Our data was however based on a small sample size necessitating further verifications on a larger scale. Neutrophils form a major line of defense against bacterial infections and their effectiveness depends on their availability at the site of infections. For Gram-negative bacterial infections, the CD14 molecule confers LPS sensitivity to neutrophils [##REF##8712510##4##], which is necessary to initiate host immune responses. The complex of TLR4, CD14 and myeloid differentiation protein 2, enhanced by the presence of LPS binding protein is crucial in LPS signaling; leading to the release of cytokines [##REF##8675286##21##, ####UREF##0##22##, ##REF##17846506##23####17846506##23##].</p>", "<p>Even though no promoter polymorphism was detected in this study, a promoter polymorphism of the gene in human is a risk factor in several diseases [##REF##15842262##15##, ####REF##14517520##16##, ##REF##18008256##17####18008256##17##]. The promoter region in bovine may probably be under strong purifying selection which may explain the lack of SNPs in this region. This is a positive factor considering the important role of the gene in mediating Gram-negative bacteria attack and the possible effect of the 1908 SNP on the abundance of the molecule on the surfaces of neutrophils. Determination of the roles of the individual SNPs and haplotypes on the activities of the gene under disease conditions will further shed more light on their biological significance.</p>", "<p>Our analysis on promoter characterization indicates that part of the exon 1 reported by Ikeda et al. [##REF##9300371##11##] constitutes the promoter. Since evolutionary pressures lead to the conservation of important non-protein-coding regulatory regions, including transcription factor binding sites (TFBs) across closely related species, identification of TFBs described in the CD14 genes of human [##REF##7512565##24##] and rat [##REF##10854787##25##] in the present study was expected. In particular, the perfect conservation of the TATA box, 9 motifs of 6 REs motifs across cattle, human, rat, mouse and pig and 4 other motifs across at least 4 of these species including cattle shows common regions in the core promoter that act together in the same biological context to control the expression of gene products and functions. In our study, up to 5 SP1 and 8 AP1 sites (with no bp mismatch) were identified which may indicate possible roles in controlling the expression of the gene as demonstrated in the rat and humans [##REF##7512565##24##,##REF##10854787##25##]. In the rat, Lui et al. [##REF##10854787##25##] through mobility shift assays demonstrated that the SP1 and AP1 elements located respectively at positions -836 and -270 were required for basal promoter activity in liver cells. Also, Zhang et al. [##REF##7512565##24##] showed that the SP1 transcription factor bound to three different regions of the human CD14 promoter and that a mutation of the major SP1 binding site decreased tissue specific promoter activity. One of the AP1 sites in our study, also shared by c-Fos and c-Jun (position 960–967, Figure ##FIG##1##2##), is similar to an AP1 site in rat promoter were JunD and Fra-2 proteins have been shown to bind [##REF##10854787##25##]. This site is also thought to transactivate the basal expression of the gene [##REF##10854787##25##]. This site in the mouse also plays a major role in the expression of the CD14 gene in macrophages [##REF##1383228##26##]. Furthermore, three motifs of PEA3 interestingly were conserved in the promoters of all species studied. PEA3 belongs to the ETS transcription factor super family and is known to appear on the promoters of many cellular genes, including HER-2/neu [##REF##10655108##27##] and CD226 antigen [##REF##16887814##28##]. Conservation of C/EBP or CCAAT/enhancer binding proteins motifs in the studied species may be explained by their involvement in many aspects of cell growth. The high conservation of the amino acids of the proteins of bovine, buffalo, sheep and goat proteins was reflected in the tree of their phylogenetic relationships and is in line with other studies that found a high rate of conservation of genes and protein coding nucleotide positions between bovine, sheep and goat genes [##REF##15589326##29##,##REF##16573533##30##]. This further supports the fact that information from the sequencing of the bovine genome will greatly enhance studies in other very closely related species.</p>" ]

[ "<title>Conclusion</title>", "<p>Overall, this study provides information on sequence variations of the CD14 gene of Canadian Holstein and Jersey cows. The identified variations and association data have provided information that may shade more light on cell surface expression of CD14 by neutrophils, which are needed to control bacterial infections. Further data on the biological significance of the mutations is however necessary. Our computational analysis highlighted on the regulatory element motifs present in the promoter region of the gene. The comparative analysis with other species revealed conserved regions of regulatory element motifs that may have important regulatory effects on the gene.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>CD14 is an important player in host innate immunity in that it confers lipopolysaccharide sensitivity to cell types like neutrophils, monocytes and macrophages. The study was aimed at characterizing the CD14 gene of cattle for sequence variations and to determine the effect of variations on the expression of the protein on the surfaces of monocytes and neutrophils in healthy dairy cows.</p>", "<title>Results</title>", "<p>Five SNPs were identified: two within the coding regions (g.A1908G and g.A2318G, numbering is according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>), one in the 5' (g.C1291T) and two in the 3' (g.A2601G and g.G2621T) untranslated regions. SNP 1908 changes amino acid 175 of the protein (p.Asn175Asp, numbering is according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link>), while SNP 2318 involves a synonymous codon change. Coding region SNPs characterized three gene alleles <italic>A </italic>(GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>), <italic>A</italic>₁(GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148610\">EU148610</ext-link>) and <italic>B </italic>(GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148611\">EU148611</ext-link>) and two deduced protein variants A (<ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link> and <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68570\">ABV68570</ext-link>) and B (<ext-link ext-link-type=\"gen\" xlink:href=\"ABV68571\">ABV68571</ext-link>). Protein variant A is more common in the breeds analyzed. All SNPs gave rise to 3 haplotypes for the breeds. SNP genotype 1908AG was significantly (P &lt; 0.01) associated with a higher percentage of neutrophils expressing more CD14 molecules on their surfaces. The promoter region contains several transcription factor binding sites, including multiple AP-1 and SP1 sites and there is a high conservation of amino acid residues between the proteins of closely related species.</p>", "<title>Conclusion</title>", "<p>The study has provided information on sequence variations within the CD14 gene and proteins of cattle. The SNP responsible for an amino acid exchange may play an important role in the expression of CD14 on the surfaces of neutrophils. Further observations involving a larger sample size are required to validate our findings. Our SNP and association analyses have provided baseline information that may be used at defining the role of CD14 in mediating bacterial infections. The computational analysis on the promoter and comparative analysis with other species has revealed regions of regulatory element motifs that may indicate important regulatory effects on the gene.</p>" ]

[ "<title>Authors' contributions</title>", "<p>EMI-B carried out the molecular genetic studies, sequence and protein comparisons, data analysis and drafted the manuscript. J-WL and XZ conceived the study and participated in its design. XZ attracted funding for the project and coordinated the work. AEI carried out the flow cytometric analysis of cells and analyzed the resulting data. All authors read and approved the draft.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This research was financed by NSERC, Alberta Milk, Dairy Farmers of New Brunswick, Nova Scotia, Ontario and Prince Edward Island, Novalait Inc., Dairy Farmers of Canada, Canadian Dairy Network, AAFC, PHAC, Technology PEI Inc., Université de Montréal and University of Prince Edward Island through the Canadian Bovine Mastitis Research Network. We thank Jaime Sanchez-Dardon for technical support in flow cytometry.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Flow cytometyric analyses of relationship of surface expression of CD14 on monocytes and neutrophils with CD14 genotypes. Cells were stained with fluorescein isothiocyanate (FITC)-labelled mouse anti human CD14 antibody and 20,000 events were gated and analyzed with Windows Multiple Document Interface for flow cytometry (WinMDI) software version 2.8. (a) Histogram showing a control sample that was not stained with antibody and occupies the Log10⁰to 10¹region and known as the control zone. M1 (Log10¹to 10²) is the zone of low expression or lower fluorescence zone indicating lower number of CD14 antigens on cells, M2 (Log10²and higher) is the zone of higher fluorescence emitted by a higher rate of absorption by more CD14 antigens on cells and M3 is the total area of expression. (b) Histogram showing stained monocytes with a higher percentage of cells in M2 and a higher overall MCF intensity of 201.44 as compared to 34.76 for polymorphonuclear neutrophils; (c) Histogram showing stained polymorphonuclear neutrophils with a higher percentage of cells in M1.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Putative transcription factor binding sites (with 100% match) on the bovine CD14 promoter (1213 bps). Recognition sequences are shaded. The rectangle is used where recognition sequences overlap. The arrow points to the first nucleotide of exon 1 at position 1214. Numbering is according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Conserved motifs of regulatory elements in the core promoter regions (about 500 bps) of cattle (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link>), rat (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"AF087944\">AF087944</ext-link>), mouse (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"X13987\">X13987</ext-link>), human (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"U00699\">U00699</ext-link>) and pig (GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"DQ079063\">DQ079063</ext-link>). + indicates the presence of a motif and • its absence. ♣ indicates motifs that are conserved in all five species and * in four species including cattle. A maximum of three base pair mismatches was allowed.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>A Neighbor-joining dendrogram of the phylogenetic relations among the CD14 proteins of cattle, buffalo, goat, sheep, pig, human, mouse and rat. Species common names are preceded by their GenBank numbers. The degree of amino acid conservation between the bovine proteins and other species is represented in percentages. On the nodes are percent bootstrap values.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>SNPs, gene alleles, protein variants, haplotypes and their frequencies in the analyzed breeds</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\">Parameter</td><td align=\"center\">Holstein (n = 106)</td><td align=\"center\">Jersey (n = 43)</td></tr></thead><tbody><tr><td align=\"center\">C1291T*</td><td/><td/></tr><tr><td align=\"center\"><italic>C</italic></td><td align=\"center\">0.658</td><td align=\"center\">0.198</td></tr><tr><td align=\"center\"><italic>T</italic></td><td align=\"center\">0.342</td><td align=\"center\">0.802</td></tr><tr><td align=\"center\">A1908G</td><td/><td/></tr><tr><td align=\"center\"><italic>A</italic></td><td align=\"center\">0.886</td><td align=\"center\">1.000</td></tr><tr><td align=\"center\"><italic>G</italic></td><td align=\"center\">0.114</td><td align=\"center\">-</td></tr><tr><td align=\"center\">A2318G</td><td/><td/></tr><tr><td align=\"center\"><italic>A</italic></td><td align=\"center\">0.658</td><td align=\"center\">0.198</td></tr><tr><td align=\"center\"><italic>G</italic></td><td align=\"center\">0.342</td><td align=\"center\">0.802</td></tr><tr><td align=\"center\">A2601G</td><td/><td/></tr><tr><td align=\"center\"><italic>A</italic></td><td align=\"center\">0.658</td><td align=\"center\">0.198</td></tr><tr><td align=\"center\"><italic>G</italic></td><td align=\"center\">0.342</td><td align=\"center\">0.802</td></tr><tr><td align=\"center\">G2621T</td><td/><td/></tr><tr><td align=\"center\"><italic>G</italic></td><td align=\"center\">0.658</td><td align=\"center\">0.198</td></tr><tr><td align=\"center\"><italic>T</italic></td><td align=\"center\">0.342</td><td align=\"center\">0.802</td></tr><tr><td align=\"center\">Gene alleles (haplotypes)</td><td/><td/></tr><tr><td align=\"center\"><italic>A </italic>(<italic>CAAAG</italic>)</td><td align=\"center\">0.658</td><td align=\"center\">0.198</td></tr><tr><td align=\"center\"><italic>A</italic>₁(<italic>TAGGT</italic>)</td><td align=\"center\">0.228</td><td align=\"center\">0.802</td></tr><tr><td align=\"center\"><italic>B </italic>(<italic>TGGGT</italic>)</td><td align=\"center\">0.114</td><td align=\"center\">-</td></tr><tr><td align=\"center\">Protein variants</td><td/><td/></tr><tr><td align=\"center\">A or **175Asn</td><td align=\"center\">0.886</td><td align=\"center\">1.000</td></tr><tr><td align=\"center\">B or 175Asp</td><td align=\"center\">0.114</td><td align=\"center\">-</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Effects of CD14 genotypes on its expression (in %) on the surfaces of monocytes and neutrophils in Holstein cows (n = 64)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Parameter</td><td align=\"left\">Genotypes</td><td align=\"left\">No.</td><td align=\"center\" colspan=\"3\">Monocytes</td><td align=\"center\" colspan=\"3\">Neutrophils</td></tr><tr><td colspan=\"3\"/><td colspan=\"3\"><hr/></td><td colspan=\"3\"><hr/></td></tr><tr><td/><td/><td/><td align=\"left\">*Low</td><td align=\"left\">High</td><td align=\"left\">Total</td><td align=\"left\">Low</td><td align=\"left\">High</td><td align=\"left\">Total</td></tr></thead><tbody><tr><td align=\"left\">**C1291T</td><td align=\"left\">CC</td><td align=\"left\">28</td><td align=\"left\">29.57 ± 1.54</td><td align=\"left\">57.00 ± 1.81</td><td align=\"left\">86.57^a± 0.96</td><td align=\"left\">73.25 ± 2.52</td><td align=\"left\">4.48 ± 0.62</td><td align=\"left\">77.74 ± 2.73</td></tr><tr><td/><td align=\"left\">CT</td><td align=\"left\">30</td><td align=\"left\">28.25 ± 1.48</td><td align=\"left\">61.26 ± 1.75</td><td align=\"left\">89.51^b± 0.93</td><td align=\"left\">76.21 ± 2.43</td><td align=\"left\">5.54 ± 0.60</td><td align=\"left\">81.75 ± 2.64</td></tr><tr><td/><td align=\"left\">TT</td><td align=\"left\">6</td><td align=\"left\">32.02 ± 3.32</td><td align=\"left\">58.94 ± 3.91</td><td align=\"left\">90.97^b± 2.07</td><td align=\"left\">81.57 ± 5.43</td><td align=\"left\">4.11 ± 1.34</td><td align=\"left\">85.68 ± 5.91</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">A1908G (Asn175Asp)</td><td align=\"left\">AA</td><td align=\"left\">51</td><td align=\"left\">29.10 ± 1.14</td><td align=\"left\">59.12 ± 1.36</td><td align=\"left\">88.22 ± 0.74</td><td align=\"left\">75.25 ± 1.88</td><td align=\"left\">4.36^A± 0.43</td><td align=\"left\">79.60 ± 2.03</td></tr><tr><td/><td align=\"left\">AG</td><td align=\"left\">13</td><td align=\"left\">29.49 ± 2.26</td><td align=\"left\">59.39 ± 2.70</td><td align=\"left\">88.88 ± 1.47</td><td align=\"left\">76.09 ± 3.72</td><td align=\"left\">7.26^B± 0.86</td><td align=\"left\">83.35 ± 4.03</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Haplotypes</td><td align=\"left\">CAAAG, CAAAG</td><td align=\"left\">28</td><td align=\"left\">29.57 ± 1.56</td><td align=\"left\">57.00 ± 1.82</td><td align=\"left\">86.57^a± 0.97</td><td align=\"left\">73.25 ± 2.55</td><td align=\"left\">4.48^a± 0.59</td><td align=\"left\">77.74 ± 2.78</td></tr><tr><td/><td align=\"left\">CAAAG, TAGGT</td><td align=\"left\">20</td><td align=\"left\">27.75 ± 1.84</td><td align=\"left\">62.46 ± 2.16</td><td align=\"left\">90.21^b± 1.14</td><td align=\"left\">77.36 ± 3.01</td><td align=\"left\">4.45^a± 0.70</td><td align=\"left\">81.81 ± 3.28</td></tr><tr><td/><td align=\"left\">CAAAG, TGGGT</td><td align=\"left\">10</td><td align=\"left\">29.24 ± 2.60</td><td align=\"left\">58.85 ± 3.05</td><td align=\"left\">88.09^a± 1.61</td><td align=\"left\">73.90 ± 4.26</td><td align=\"left\">7.73^bc± 0.98</td><td align=\"left\">81.63 ± 4.65</td></tr><tr><td/><td align=\"left\">TAGGT, TGGGT</td><td align=\"left\">3</td><td align=\"left\">30.33 ± 4.75</td><td align=\"left\">61.18 ± 5.56</td><td align=\"left\">91.51^a± 2.95</td><td align=\"left\">83.41 ± 7.78</td><td align=\"left\">5.68^ab± 1.80</td><td align=\"left\">89.08 ± 8.48</td></tr><tr><td/><td align=\"left\">TAGGT, TAGGT</td><td align=\"left\">3</td><td align=\"left\">33.72 ± 4.75</td><td align=\"left\">56.72 ± 5.56</td><td align=\"left\">90.42^a± 2.95</td><td align=\"left\">79.73 ± 7.78</td><td align=\"left\">2.55^a± 1.80</td><td align=\"left\">82.28 ± 8.48</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Primers used in the amplification of the whole bovine CD14 gene and other PCR conditions</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Primer</td><td align=\"left\">*Primer sequence</td><td align=\"left\">MgCl₂concentration</td><td align=\"left\">Amplicon size (bp)</td></tr></thead><tbody><tr><td align=\"left\">BoCD14.83330F</td><td align=\"left\">5'ATT ACC TTC TTC TGC ACC TCC A 3'</td><td align=\"left\">2.5 mM</td><td align=\"left\">1578</td></tr><tr><td align=\"left\">BoCD14.84907R</td><td align=\"left\">5' GGC AGC CTC TGA GAG TTT ATG T 3'</td><td/><td/></tr><tr><td/><td/><td/><td/></tr><tr><td align=\"left\">BoCD14.84746F</td><td align=\"left\">5' CTT CCT GTT ATA GCC CCT TTC C 3'</td><td align=\"left\">2.5 mM</td><td align=\"left\">832</td></tr><tr><td align=\"left\">BoCD14.85577R</td><td align=\"left\">5' CAC GAT ACG TTA CGG AGA CTG A 3'</td><td/><td/></tr><tr><td/><td/><td/><td/></tr><tr><td align=\"left\">BoCD14.85456F</td><td align=\"left\">5' GGG TAC TCT CGT CTC AAG GAA C 3'</td><td align=\"left\">2.0 mM</td><td align=\"left\">825</td></tr><tr><td align=\"left\">BoCD14.86280R</td><td align=\"left\">5' CTG AGC CAA TTC ATT CCT CTT C 3'</td><td/><td/></tr><tr><td/><td/><td/><td/></tr><tr><td align=\"left\">BoCD14.86081F</td><td align=\"left\">5' ACC TGA CTC TGG ACG GAA ATC 3'</td><td align=\"left\">2.0 mM</td><td align=\"left\">747</td></tr><tr><td align=\"left\">BoCD14.86827R</td><td align=\"left\">5' TAC AGG AGA GCA ACC CTG AAA 3'</td><td/><td/></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>*SNP Numbers are according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link></p><p>** Amino acid residue numbering is according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link></p></table-wrap-foot>", "<table-wrap-foot><p>^{a, b, c or A, B}Means for each parameter and within the same column differ significantly. a, b, c indicates significance (P &lt; 0.05) with non adjusted means while A, B indicate significance with both non-adjusted means and when Scheffe's adjustments were applied to means (P &lt; 0.01). SNPs 1291, 2318 and 2601 are in 100% linkage disequilibrium and possess the same information, like wise are gene alleles and haplotypes. Only values for SNP1291 and haplotypes have been represented in the table.</p><p>*Low indicates the percentage of cells in the region of Log10¹to 10², high &gt; log10²and total, all cells stained.</p><p>**SNP numbering is according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"EU148609\">EU148609</ext-link> and amino acid residue numbering is according to GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"ABV68569\">ABV68569</ext-link>.</p></table-wrap-foot>", "<table-wrap-foot><p>*Primers were designed based on GenBank No. <ext-link ext-link-type=\"gen\" xlink:href=\"NW_001495367\">NW_001495367</ext-link>. Numbers preceding F or R in primer names represent their positions on the reference sequence.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2156-9-50-1\"/>", "<graphic xlink:href=\"1471-2156-9-50-2\"/>", "<graphic xlink:href=\"1471-2156-9-50-3\"/>", "<graphic xlink:href=\"1471-2156-9-50-4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>In recent years, the discovery and development of genetic markers such as microsatellites has led to a rapid growth in the number of molecular studies, with the isolation and development of microsatellite loci becoming relatively quick and straight-forward in many taxa [reviewed in [##REF##11903900##1##]]. The method of Armour <italic>et al</italic>. 1994 has been successfully used for birds [##REF##11091342##2##,##UREF##0##3##], mammals [##UREF##1##4##,##UREF##2##5##] and fish [##UREF##3##6##] but this success has not been ubiquitous across all taxa and the identification of microsatellite markers in many invertebrates has proved to be difficult, e.g. mosquitoes [##REF##17420178##7##] and butterflies [##REF##15140111##8##,##REF##16701315##9##]. Even after successful microsatellite isolation, primers developed for certain species suffer from multiple banding patterns and random primer binding, leading to difficulties in obtaining reliable genotype data. The presence of repeat regions in the genome, known as SINEs (short interspersed nuclear elements), repeat microsatellite regions, and a general failure to amplify a specific product, have often been cited as potential causes for inaccuracies in data sets, for example in nematodes [##REF##16765463##10##], lepidopterans [##REF##15140111##8##,##REF##16701315##9##,##REF##17298557##11##] and marine molluscs [##UREF##4##12##, ####UREF##5##13##, ##REF##9734082##14####9734082##14##].</p>", "<p>In many marine invertebrates the situation is further complicated by a deficiency in the number of heterozygotes observed (relative to Hardy-Weinberg expectation) (see Additional file ##SUPPL##0##1##), with both allozyme and microsatellite studies documenting the phenomenon in many marine bivalve and gastropod populations [##REF##9734082##14##, ####UREF##6##15##, ##UREF##7##16##, ##REF##8587499##17##, ##UREF##8##18##, ##UREF##9##19##, ##UREF##10##20##, ##UREF##11##21####11##21##]. While some of the species presented in Additional file ##SUPPL##0##1## are hermaphroditic, e.g. <italic>Physa acuta</italic>, and thus may be expected to show heterozygote deficiency (due to high potential for selfing which would increase homozygosity in the population), there are many more examples (see Additional file ##SUPPL##0##1##) where species display separate sexes and would not necessarily be expected to display heterozygote deficiency.</p>", "<p>With this in mind, one should consider the numerous factors which can cause heterozygote deficiencies. These include poor primer design and optimisation, null alleles [##REF##7735527##22##], genotyping errors (stuttering or large allele dropout [##REF##10790319##23##]), mutation, inbreeding effects, SINEs, non-random mating and population admixture. However, despite extensive examination in some cases [##REF##8587499##17##,##UREF##12##24##] the causes behind these heterozygote deficiencies still remain unknown and, as a consequence, the ability to utilise traditional techniques for primer and population assessments may be significantly compromised.</p>", "<p>In \"model\" populations, which are expected to conform to HWE, testing for heterozygote deficiency in loci is commonly used as a means to assess designed primers and genotype reliability. In such cases identification of an unreliable primer pair, or a locus affected by null alleles, is simply a matter of identifying heterozygote deficiency at that locus. This task is becoming increasingly simple due to the wide array of tools available for population and locus assessment (e.g. G<sc>ENEPOP</sc>[##UREF##13##25##]; C<sc>ERVUS</sc>[##REF##9633105##26##]; M<sc>ICROCHECKER</sc>, [##UREF##14##27##]). Once identified, the primer pair can then be removed from further study. Difficulties can arise, however, when attempting to identify unreliable primers and null alleles in populations that naturally display heterozygote deficiency (see Additional file ##SUPPL##0##1##). In these cases, the use of traditional analysis and software are no longer effective, as they often assume that populations conform to HWE and display expected heterozygosity. Thus, identifying unreliable primers by analysing departures from HWE is ineffective due to the fact that the population itself is naturally deficient. In these cases primer reliability is normally assumed without additional investigation beyond these standard tests (Additional file ##SUPPL##0##1##). As a result, unreliable primers may often remain undetected.</p>", "<p>In this study microsatellite loci were isolated and examined for heterozygote deficiency in the prosobranch mollusc <italic>Hydrobia ulvae</italic>. A previous study by Haase [##UREF##7##16##] using allozyme markers suggests that <italic>H. ulvae </italic>populations are naturally heterozygote deficient; therefore further testing was conducted here to see if the same conclusions would be supported by microsatellite genotyping data. A novel technique was then used to assess microsatellite marker reliability by designing two microsatellite marker primer sets to amplify the same loci. By altering the binding sites, designing additional primers (B-primers) to amplify the same loci as the original primer set (A-primers), and comparing genotypes between A and B, a method is presented to assess the reliability of Primer set_A. Additionally, the method can also be used to identify inaccuracies due to PCR amplification failure and erroneous genotype scoring.</p>", "<p><italic>Hydrobia ulvae </italic>is a widespread and abundant member of the benthic fauna of estuarine habitats. It is dioecious with sexes being easily identified through dissection. On the west coast of Wales this species has peaks of spawning activity in spring and autumn and produces planktotrophic larvae that remain in the plankton for up to four weeks before settlement [##UREF##15##28##]. This period of development affords the potential for dispersal to new habitats and mixing with geographically separate populations. The species provides an interesting case for molecular analysis as the pelagic dispersal phase raises fascinating questions on gene flow, differentiation, recruitment, and inbreeding, but there remains the potential for self-recruitment of estuarine populations [##UREF##16##29##].</p>", "<p>The objectives of this study were: (1) to isolate polymorphic microsatellite loci for <italic>Hydrobia ulvae</italic>, (2) to conduct standard tests for HWE, heterozygosity, linkage equilibrium and sex linkage and (3) to assess the use of alternative primer sets for the same loci as a tool to assess primer reliability.</p>" ]

[ "<title>Methods</title>", "<title>DNA isolation</title>", "<p>Genomic DNA was extracted for microsatellite library development and genotyping using a modified CTAB procedure with proteinase K digestion and a chloroform-isoamylalcohol protocol [##REF##8122306##53##].</p>", "<title>Microsatellite library preparation</title>", "<p>An enriched <italic>Hydrobia ulvae </italic>microsatellite library was prepared using a method based on Armour <italic>et al. </italic>[##REF##8069306##54##], with modifications described by Gibbs <italic>et al. </italic>[##REF##9589582##55##]. Six <italic>Hydrobia ulvae </italic>individuals of unknown sex were pooled to obtain a sufficient amount of genomic DNA from which to prepare the library. The pooled genomic DNA (2 μg) was digested using <italic>Mbo</italic>I (Qbiogene, Cambridge, UK), and ligated to double-stranded Sau-L linkers [##REF##1348122##56##]. Size-selected (200–800 bp), digested genomic DNA fragments were then denatured and hybridised against double-stranded denatured dinucleotide and tetranucleotide sequences that had been bound to nylon Hybond membrane (Amersham Pharmaceuticals Ltd, Buckinghamshire, UK). The Armour [##REF##8069306##54##]<italic>et al. </italic> pre-enrichment PCR was not performed [as in [##REF##9589582##55##]]. The dinucleotide sequences (AC.GT)_nand (AG.CT)_nwere obtained as DNA Alternating Copolymers (Amersham Pharmaceuticals Ltd, Buckinghamshire, UK) and the tetranucleotide sequences (TTTC.GAAA)_n, (GTAA.TTAC)_n, (GATA.TATC)_n, (CTAA.TTAG)_nand (TAAA.TTTA)_nwere prepared using two rounds of PCR amplification as in Armour <italic>et al. </italic>[##REF##8069306##54##]. Once recovered, the microsatellite-enriched <italic>Hydrobia ulvae </italic>fragments were separated from the Sau-L linkers by digestion with <italic>Mbo</italic>I. Fragments were then ligated into <italic>Bam</italic>HI/BAP-dephosphorylated pUC19 vector (Qbiogene) and transformed into XL1-Blue competent cells (Stratagene). Transformant colonies were grown overnight at 37°C on agar plates containing Luria Broth, Ampicillin, X-gal and IPTG, bound to a Hybond nylon filter and screened with [α³²P]-dCTP and/or [α³²P]-dATP radiolabelled dinucleotide and tetranucleotide sequences as described above. In total, 902 transformants were screened and 265 produced a positive autoradiograph signal (4 of 18 from the dinucleotide library and 261 of 884 from the tetranucleotide enriched library). Sequence data for 153 clones was obtained using an ABI 3730 capillary sequencer with BigDye chemistry (Applied Biosystems). All sequences were checked for duplication using stand-alone BLAST software (protocol available at [##UREF##26##57##] and MEGA v.3.1 sequence alignment software [##REF##15260895##58##]. Of the 153 sequences examined, 118 were found to be unique.</p>", "<title>PCR and genotyping</title>", "<p>When genotyping individual <italic>Hydrobia ulvae</italic>, each PCR reaction contained approximately 5–10 ng of genomic DNA, 0.5 μM of each primer, 1.5–2 mM MgCl₂(Additional file ##SUPPL##1##2##), 0.2 mM of each dNTP and 0.05 units of <italic>Taq </italic>DNA polymerase (Bioline, London, UK) in the manufacturer's buffer [final concentration: 16 mM (NH₄)₂SO₄, 67 mM Tris-HCl (pH 8.8 at 25°C), 0.01% Tween-20]. PCR amplification was performed using a thermal cycler (MJ Research model PTC DNA engine) with the following program: one cycle of 3 min at 94°C, followed by: 35 cycles of 94°C for 30 s, annealing temperature (Additional file ##SUPPL##1##2##) for 30 s, 72°C for 45 s and a final extension cycle of 10 min at 72°C. PCR products were visualised on 2% agarose gel, pre-stained with ethidium bromide. Amplified product sizes were compared to the size expected based on the cloned allele sequence and checked for the presence of large alleles (&gt; 500 bp), a potential by-product of SINE insertions [##REF##15140111##8##,##REF##16701315##9##].</p>", "<p>A fraction of the fluorescently-labelled PCR product was diluted to one part per thousand and loaded on an ABI 3730 DNA Analyzer, along with ROX500 size marker (Applied Biosystems). Allele sizes were assigned using GENEMAPPER 3.7 software (Applied Biosystems).</p>" ]

[ "<title>Results</title>", "<title>Microsatellite library preparation, Primer set_A design and testing</title>", "<p>The first step was to isolate polymorphic microsatellite loci for <italic>H. ulvae </italic>and test to see if the population examined displayed heterozygote deficiency. Samples were collected from a presumed single population of <italic>Hydrobia ulvae </italic>from an area of approximately 1.0 m²in the Dyfi estuary, Wales (52° 31' 31.9\" N; 4° 2' 40.2\" W) using a 0.5 mm mesh sieve to remove individuals from the sediment. Specimens were stored alive in sea water of 28 ‰ at 14°C until required for DNA extraction.</p>", "<p>Prior to DNA extraction, <italic>Hydrobia ulvae </italic>individuals were dissected and checked for parasitic infection. Those that showed signs of infection were removed from further study to avoid the possibility of isolating microsatellites from the parasites. Extraction of <italic>Hydrobia </italic>DNA and preparation of an enriched library yielded 118 unique microsatellite sequences that were submitted to the EMBL Nucleotide Sequence Database [##UREF##17##30##] [EMBL <ext-link ext-link-type=\"gen\" xlink:href=\"AM409397\">AM409397</ext-link>–<ext-link ext-link-type=\"gen\" xlink:href=\"AM409514\">AM409514</ext-link>]. For details of DNA extraction and enriched library development, see Methods. All the obtained sequences were confirmed to be unique using BLASTN version 2.2.4 [##REF##9254694##31##]. Forty-seven loci possessed sufficient flanking sequence for primer design (in terms of repeat length, sequence length and base-pair composition of the flanking sequence). Primer pairs were designed for 17 loci using P<sc>RIMER</sc>3 [##UREF##18##32##], labelled with a fluorescent dye, and optimised with respect to temperature (gradient range 50–70°C) and MgCl₂concentration (1.5 mM, 2.0 mM and 2.5 mM). A short chain of nucleotides (GTTTCTT), known as a pigtail, was also added to the 5' end of each reverse primer to reduce stutter bands during PCR amplification [##REF##8780871##33##].</p>", "<p>The subsequent PCR and genotyping of 16 snails from the Dyfi estuary (including the six individuals from which the library was made) using 17 designed primers, resulted in nine primers failing to amplify a specific product, and one locus, <italic>Hulv-10 </italic>(<ext-link ext-link-type=\"gen\" xlink:href=\"AM409406\">AM409406</ext-link>; Forward-TCGTACCAGGAAAGGCCTCAG, Reverse-CCACGTCACTTTCGGTGCTC) being monomorphic in all 16 individuals tested. For each of the seven remaining polymorphic loci a further 48 <italic>Hydrobia ulvae </italic>individuals were genotyped. The primers designed for these seven loci are hereafter described as Primer set_A (Additional file ##SUPPL##1##2##). The expected and observed heterozygosity of each locus (as amplified with Primer set_A) was calculated and null allele frequency estimates obtained using CERVUS version 2.0 [##REF##9633105##26##], while deviations from HWE and linkage equilibrium were examined using exact tests in GENEPOP (v3.4) [##UREF##13##25##]. Results showed that the seven polymorphic loci (when amplified using Primer set_A) were highly variable, with the number of alleles ranging from 16 to 27 (Additional file ##SUPPL##1##2##). Six of the loci were found to display a heterozygote deficiency when compared with the expectation under HWE. The exception was locus <italic>Hulv-04_ A</italic>, which was in Hardy-Weinberg equilibrium (Additional file ##SUPPL##1##2##). Two of the loci (<italic>Hulv-06 </italic>and <italic>Hulv-07</italic>) showed very low levels of heterozygosity, suggesting possible unreliability in the primers (Additional file ##SUPPL##1##2##). However, to justify the removal of these loci and designed primers, further examination was required, as low heterozygosity is not necessarily a precursor for unreliable markers. Another notable result was the amplification success shown by <italic>Hulv-01_A</italic>, <italic>Hulv-06_A </italic>and <italic>Hulv-07_A </italic>primers. At these loci, amplification success was slightly lower than expected (83.3%, 83.3% and 70.8% of individuals amplified, respectively). Given that high levels of amplification failure can be used as an indicator of unreliable primers, the opportunity was taken to explore the connection between amplification failure and primer reliability by examining these loci using the primer redesign technique described.</p>", "<p>As well as heterozygote analysis, tests were conducted using the software M<sc>ICROCHECKER</sc>[##UREF##14##27##] to examine genotyping error, presence of null alleles, allele frequency estimates, allele stutter and allele dropout (in which smaller alleles are preferentially amplified over larger alleles [##REF##10790319##23##]). Interestingly, all results were negative except for null alleles, which the software highlighted could be a potential factor in causing the heterozygote deficiency at all but one of the loci, <italic>Hulv-04</italic>. For this reason, and to ensure that alleles were not failing to amplify due to the PCR conditions, each locus was PCR-amplified twice using each primer set (A &amp; B). On both occasions the same genotypes were observed. Alleles were resolved on an agarose gel and allele sizes were checked to see if they were consistent with the size expected based on the sequence of the cloned allele. Examination of the gel showed no indication of large alleles (&gt; 500 bp), suggesting that neither SINE insertions, nor duplicate microsatellite regions, were present in the products. Pedigrees were not available for checking the inheritance of null alleles due to difficulties in cultivating the species in the laboratory.</p>", "<p>Loci were also examined for sex linkage by visually comparing the genotypes of individuals of known sex (sexes assigned after dissection), and the results suggested no association between apparent homozygosity and sex in the genotyped individuals. Similarly, all loci were checked for linkage disequilibrium using G<sc>ENEPOP</sc> v3.4 without Bonferroni correction and no linkage disequilibrium was observed.</p>", "<title>Comparing genotype data from two sets of primers for each locus</title>", "<p>In order to test genotype outputs and Primer set_A reliability, microsatellite primers were redesigned (Primer set_B) (Additional file ##SUPPL##2##3##) for six loci (<italic>Hulv-01</italic>, <italic>Hulv-03</italic>, <italic>Hulv-04</italic>, <italic>Hulv-05</italic>, <italic>Hulv-06 </italic>&amp;<italic> Hulv-07</italic>) to amplify the same loci as Primer set_A. \"B\" primers were not designed for <italic>Hulv-02 </italic>due to difficulty in identifying new primer sites (due to the length and base-pair composition of the flanking sequence), so this locus could not be verified. In all other cases, primers were designed at sites as distinct as possible from the original primer sites. In some cases, short flanking regions hindered primer design and thus only one of the primers in a pair could be significantly shifted. However, where possible the 3' end of the primer was redesigned so that primer reliability could still be assessed. Single mismatches at or near the 3' end of the primer are known to affect the efficiency of polymerase extension and oligonucleotide stability [##REF##2506529##34##] and will have a greater effect upon amplified products than mismatches elsewhere [##REF##3413095##35##,##REF##2179874##36##], thus where possible an alteration in the 3' end of the primer sequence was attempted. The relevance of this restriction in the current study and to the technique as a whole is discussed later.</p>", "<p>Redesigned Primer set _B was optimised, amplified and visualised in exactly the same way as Primer set_A, using only the DNA templates from individuals which yielded successful amplification in Primer set_A. Samples which failed to amplify a product in Primer set_A were not examined using Primer set_B, as regardless of the output from Primer set_B in these cases, no comparison could ever be made. This was also done so that direct comparison of successful amplifications could be made between the primer sets. Alleles were scored independently on separate ABI3730 gel loading runs for each primer set (A and B) to prevent any bias in the data.</p>", "<p>To assess primer data, the presence and size of alleles from Primer set_B were compared with those originally amplified by Primer set_A. Allele sizes were expected to show a consistent and predictable difference in size between the primer sets within each locus, due to the difference in product size caused by the redesign and movement of primers. To be classified as a successful match between primers, both alleles in a heterozygote needed to be the same for both primer sets. Likewise, in homozygotes the single alleles needed to display the same product after the predictable size difference was taken into account.</p>", "<p>Results indicate that when both primer sets (A and B) were used in conjunction to amplify the six target loci, four of these (<italic>Hulv-01</italic>, <italic>Hulv-03</italic>, <italic>Hulv-04</italic>, &amp;<italic> Hulv-05</italic>) showed more than 95% agreement (allele-allele) between the alleles amplified by the two primer sets A and B (Additional file ##SUPPL##3##4##), suggesting that these primers from Primer set_A had amplified the target loci correctly, and thus are reliable. However, when comparison was made between Primer set_A and Primer set_B for the remaining two loci (<italic>Hulv-06 </italic>&amp;<italic> Hulv-07</italic>), a high degree of discrepancy was observed (0% &amp; only 25% alleles matching, respectively) (Additional file ##SUPPL##3##4##). These discrepancies included a change from heterozygotes to homozygotes or vice versa, while some samples also displayed new unexpected allele sizes. Samples which failed to amplify the same target loci in each primer set were examined to ensure that poor DNA was not the underlying cause. No individuals consistently failed to amplify and the individuals failing to amplify differed between loci and thus poor DNA samples were not to blame for amplification failures.</p>", "<p>When these findings are coupled with the previously identified decreased amplification success at these loci, it is clear that <italic>Hulv-06_A </italic>and <italic>Hulv-07_A </italic>primers are not suitable for further use in population studies due to their unreliability. Interestingly, despite the fact that <italic>Hulv-01_A </italic>was previously shown to display similar levels of amplification failure as <italic>Hulv-06_A </italic>and <italic>Hulv-07_A</italic>, analysis using redesigned markers has highlighted that <italic>Hulv-01_A </italic>primers were in fact producing reliable genotype data. Therefore the redesign technique presented here demonstrates that decreased amplification success does not necessary mean a marker is unreliable.</p>" ]

[ "<title>Discussion</title>", "<p>This study has shown that in order to identify and confidently remove unreliable markers in heterozygote deficient populations, additional techniques are required beyond those currently applied to microsatellite data. The currently used tests such as examination of HWE, heterozygosity, amplification failure, allele frequency distributions and linkage equilibrium are not sufficient to identify reliability in primer sets. This may particularly be the case for invertebrate species, many of which commonly display departures from HWE and heterozygote deficiencies [##REF##9734082##14##,##UREF##9##19##]. The results show that the majority of the loci examined (6/7) displayed heterozygote deficiency (Additional file ##SUPPL##1##2##) and <italic>H. ulvae </italic>therefore exhibits similar characteristics to many of the marine and freshwater invertebrates (as detailed in Additional file ##SUPPL##0##1##). The technique presented here has shown that after primer redesign, <italic>Hulv-06 </italic>and <italic>Hulv-07 </italic>cannot be reliably genotyped due to the discrepancies between each primer set. Therefore it is concluded that these loci, along with designed primers, should not be used in future studies. Similarly, the technique has also provided increased confidence in the remaining loci (<italic>Hulv-01</italic>, <italic>Hulv-03</italic>, <italic>Hulv-04 </italic>and <italic>Hulv-05</italic>), which all showed similar genotype outputs regardless of the primer set implemented. For this reason, the technique described here presents a useful method for helping to assess the reliability of designed primers in heterozygote-deficient populations.</p>", "<p>Previous published studies have suggested many causes for heterozygote deficiency in invertebrates, including selection, inbreeding, mutation and null alleles, SINEs, poor primer design and amplification error [##REF##9734082##14##,##UREF##9##19##]. However, attempts to ascertain the exact source of the heterozygote deficiency have often had limited success [##UREF##11##21##,##UREF##12##24##], and while it is not the purpose of this study to identify the source of the heterozygote deficiency, the technique and results obtained can be used to shed light into the cause of the heterozygosity in the species.</p>", "<p>The first factor to consider is selection. While some microsatellite loci have been shown to be affected by selection [##REF##11903909##37##,##UREF##19##38##], the majority are still considered to be selectively neutral [##REF##16643306##39##]. This study took care not to incorporate tri-nucleotide repeats, which have been shown to occur in coding regions of the genome and thus are more likely to be subject to selection [##UREF##20##40##]. Therefore, while it cannot be categorically ruled out, selection is not likely to be the cause of the observed deficiencies. Similarly, given what is known about the life stages of the target organism, inbreeding seems unlikely. In species with limited dispersal or direct development, it is simple to imagine inbreeding as a dominant force [##REF##14871357##41##,##UREF##21##42##], which can be detrimental to species fitness [##UREF##22##43##,##UREF##23##44##]. However, given that <italic>Hydrobia ulvae </italic>has a dispersal phase of up to four weeks [##UREF##15##28##], a good potential for mixing of progeny by tides and currents, and the likelihood of widespread settlement of larvae within the estuary, it seems highly improbable that inbreeding is occurring. Another alternative potential cause is the Wahlund effect, which is a heterozygote deficiency due to the accidental pooling of discrete sub-populations. Indeed, by disregarding sub-structuring within populations one would expect to see a common deficiency across the majority of the loci as observed in this study [##REF##9734082##14##,##UREF##21##42##]. However, it is unlikely that samples taken from a homogeneous area of less than 1 m², as done here, could contain many different sub-populations [##UREF##7##16##]. Nonetheless, in order to be certain, additional genotypic analysis would be required on individuals from identified cohorts and known local spatial localities [##REF##14871357##41##].</p>", "<p>Perhaps the most interesting factor to consider with regard to the technique presented here is the presence of null alleles and their potential effect upon levels of heterozygosity. Null alleles represent base-pair mutations in the primer regions which cause primer binding to weaken and/or fail, resulting in a failure to amplify certain alleles [##REF##8574449##45##]. As a result, the presence of null alleles in data sets has been commonly suggested as a contributor to heterozygote deficiency [##REF##8488841##46##,##UREF##24##47##] (Additional file ##SUPPL##0##1##). While microsatellite regions are often highly polymorphic due to a high rate of mutation through replication slippage and proof-reading events [##UREF##25##48##], the flanking regions surrounding microsatellite repeat regions are generally considered to be more conserved. However, given the very high levels of polymorphism shown in microsatellite loci examined for <italic>H. ulvae </italic>(Additional file ##SUPPL##1##2##), it is possible that the sequence flanking the repeat regions may also exhibit increased levels of mutation which would certainly reduce the effectiveness of primer binding and result in an abundance of null alleles. Microsatellite loci in humans have been estimated to have mutation rates of about 10^-4[##REF##8401493##49##]. However, microsatellite mutation is known to vary between different taxa [##REF##11102705##50##,##REF##9724780##51##], and while little is known specifically about the mutation rate in marine molluscs, several studies, including this current study, have shown high polymorphism in microsatellite loci in marine invertebrates [##REF##9734082##14##,##UREF##10##20##,##REF##10652081##52##], suggesting that mutation rates may be high. For this reason further investigation in marine invertebrates is required (i.e. genetic sequencing) to determine whether mutations in the flanking regions introduce errors into genotype data and consequently influence levels of heterozygosity.</p>", "<p>Despite this, there are several reasons to suggest that mutation and null alleles are not the explanation for the overall heterozygote deficiency observed in <italic>Hydrobia ulvae</italic>. First, given their nature one would typically expect null alleles to occur at a minority of loci and not across the majority of loci as seen in this study [##REF##9734082##14##,##UREF##9##19##,##REF##16643306##39##]. Secondly, results from the double-primer technique show similar heterozygote deficiencies and null allele frequencies in both primer sets A and B in all loci with the exception of <italic>Hulv-06 </italic>and <italic>Hulv-07 </italic>(Additional file ##SUPPL##3##4##). If nulls were the explanation for the heterozygote deficiency then we would not expect both primer sets to be equally affected.</p>", "<p>Mutation and null alleles do however, present one possible explanation for the poor match between primer sets for <italic>Hulv-06 </italic>and <italic>Hulv-07</italic>, particularly as the predicted null allele frequencies at these loci were shown by the software M<sc>ICROCHECKER</sc> to be much higher than for all the other loci examined (Additional file ##SUPPL##3##4##). Indeed, when coupled with the low amplification rate of individuals shown by <italic>Hulv-06_A </italic>and <italic>Hulv-07_A </italic>primers (83.3 and 70.8%, respectively), there is evidence to suggest mutation in the flanking regions and null alleles. Alternatively, the mismatch between primer sets A and B for <italic>Hulv-06 </italic>and <italic>Hulv-07 </italic>could be due to similarity in the flanking regions of different loci. In the study by Meglecz <italic>et al. </italic>[##REF##15140111##8##] on lepidopterans (butterflies and moths), high flanking sequence similarity was observed at a number of loci, which led to difficulties in designing effective primer sets. Likewise, in a study on the marine gastropod <italic>Littorina saxatilis</italic>, anomalous large alleles were identified that may signify the presence of flanking similarity in marine invertebrates [##UREF##10##20##]. While no evidence has been found in the present study for large alleles above 500 bp as described by Sokolov <italic>et al. </italic>[##UREF##10##20##], it is entirely possible that flanking region similarity could introduce errors into genotyped data and thus disrupt the levels of heterozygosity observed. However, while further investigation into the matter is required, the fact that very little sequence similarity was identified in this study suggests that it is not a dominant factor.</p>", "<p>While the precise nature of the overall heterozygote deficiency still remains unclear, it is concluded that mutation and null alleles are the likely explanation for the unreliability observed at two of the loci examined (<italic>Hulv-06 </italic>and <italic>Hulv-07</italic>). Drawing definitive conclusions on the presence of null alleles, the cause of heterozygote deficiency and primer reliability, has been limited in this study due to constraints on primer redesign imposed by the restricted length of flanking regions. However, when redesigning primers, attempts were made to make B-primers as distinct from A-primers as possible, particularly at the 3' end of the primer, as this region is known to affect product amplification and polymerase extension [##REF##2506529##34##,##REF##2179874##36##,##REF##3413095##35##]. Given that the occurrence of restricted flanking regions in microsatellite studies has not been commonly reported, and that information regarding the length of flanking regions in different species is scarce, it is difficult to estimate the likelihood of primer redesign complications in other species; nevertheless, it is believed that the technique presented here provides an important contribution to the field, as it offers a practical alternative to merely assuming primer reliability in heterozygote-deficient populations. Most importantly, this study has highlighted the need for increased caution in assessing primer reliability using only standard assessments, especially in populations commonly exhibiting heterozygote deficiency and deviations from HWE. Moreover, this study has illustrated that additional testing of designed primers in heterozygote deficient populations can identify potentially unreliable markers, amplification errors and genotyping failures, which are commonly associated with microsatellite studies [##UREF##4##12##,##REF##16643306##39##]. For these reasons, the methodology presented will be of potential interest/application to future studies on invertebrate and non-invertebrate species alike. Indeed, the technique will be of value in all microsatellite studies that seek increased confidence in genotype assignments. While doubling expenditure on primers may be seen as a disadvantage, the benefits make the technique a sound investment when one considers the high cost of running unreliable primers in large-scale population analysis. The technique will also be of particular use in studies where small primer sets are used due to difficulties in microsatellite isolation. In these cases, it is imperative that all primers consistently amplify a reliable product, given the low number being implemented. Similarly, the genotype data validation provided by this technique will also enable increased confidence in species where multiple banding patterns and anomalous large alleles are often noted [##UREF##10##20##].</p>", "<p>In addition to the technique proposed, five novel <italic>Hydrobia ulvae </italic>polymorphic loci (<italic>Hulv-01</italic>, <italic>Hulv-02</italic>, <italic>Hulv-03</italic>, <italic>Hulv-04 </italic>&amp;<italic> Hulv-05</italic>) (Additional file ##SUPPL##1##2##) are presented for further use in population studies. Given their level of polymorphism, these microsatellites, and the primers described, will provide valuable tools in the study of genetic mixing and population differentiation in <italic>Hydrobia ulvae</italic>.</p>" ]

[ "<title>Conclusion</title>", "<p>While it is clear that <italic>Hydrobia ulvae </italic>is characteristic of its taxon with regard to heterozygosity levels, the methods described here provide a useful tool for assessing genotype data and primer reliability in studies where increased confidence in microsatellites is desired. While traditional examination of HWE and heterozygote deficiency has been shown to be insufficient to identify unreliable primers in naturally deficient populations, the use of dual primers provides a simple alternative to merely assuming reliability. In studies that have previously only used traditional methods, or assumed primer reliability, the technique may serve to identify unreliable primer pairs early in the study before the cost of genotyping multiple individuals is incurred.</p>", "<p>Therefore, while it is clear that standard examinations and software can serve to highlight many primer reliability issues in model populations, heterozygote deficient and other non-model populations require the additional validation provided by designing a second primer set. In order for microsatellite primers to be used effectively in population studies and to justify the lengthy developmental process, they need to be reliable and of sufficient power, a situation which is more likely when validation steps are taken.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>In studies where microsatellite markers are employed, it is essential that the primers designed will reliably and consistently amplify target loci. In populations conforming to Hardy-Weinberg equilibrium (HWE), screening for unreliable markers often relies on the identification of heterozygote deficiencies and subsequent departures from HWE. However, since many populations naturally deviate from HWE, such as many marine invertebrates, it can be difficult to distinguish heterozygote deficiencies resulting from unreliable markers from natural processes. Thus, studies of populations that are suspected to deviate from HWE naturally would benefit from a method to validate genotype data-sets and test the reliability of the designed primers. Levels of heterozygosity are reported for the prosobranch mollusc <italic>Hydrobia ulvae </italic>(Pennant) together with a method of genotype validation and primer assessment that utilises two primer sets for each locus. Microsatellite loci presented are the first described for the species <italic>Hydrobia ulvae</italic>; the five loci presented will be of value in further study of populations of <italic>H. ulvae</italic>.</p>", "<title>Results</title>", "<p>We have developed a novel method of testing primer reliability in naturally heterozygote deficient populations. After the design of an initial primer set, genotyping in 48 <italic>Hydrobia ulvae </italic>specimens using a single primer set (Primer set_A) revealed heterozygote deficiency in six of the seven loci examined. Redesign of six of the primer pairs (Primer set_B), re-genotyping of the successful individuals from Primer set_A using Primer set_B, and comparison of genotypes between the two primer sets, enabled the identification of two loci (<italic>Hulv-06 </italic>&amp;<italic> Hulv-07</italic>) that showed a high degree of discrepancy between primer sets A and B (0% &amp; only 25% alleles matching, respectively), suggesting unreliability in these primers. The discrepancies included changes from heterozygotes to homozygotes or vice versa, and some individuals who also displayed new alleles of unexpected sizes. Of the other four loci examined (<italic>Hulv-01</italic>, <italic>Hulv-03</italic>, <italic>Hulv-04</italic>, &amp;<italic> Hulv-05</italic>), all showed more than 95% agreement between primer sets. <italic>Hulv-01</italic>, <italic>Hulv-03</italic>, &amp;<italic> Hulv-</italic>05 displayed similar levels of heterozygosity with both primer sets suggesting that these loci are indeed heterozygote deficient, while <italic>Hulv-08 </italic>showed no deficiency in either primer set.</p>", "<title>Conclusion</title>", "<p>The simple method described to identify unreliable markers will prove a useful technique for many population studies, and also emphasises the dangers in using a single primer set and assuming marker reliability in populations shown to naturally deviate from HWE.</p>" ]

[ "<title>Authors' contributions</title>", "<p>RJB collected and processed specimens for examination, carried out the molecular genetics studies and drafted the manuscript. DAD supervised the molecular work, assisted with interpretation of data and assisted in drafting of the manuscript. GJH prepared the microsatellite-enriched genomic library. JJB contributed to the initial design of the project. JDF conceived the study, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>The authors acknowledge the support of the Natural Environment Research Council (NERC) at the NERC Molecular Genetics Facility (Sheffield) and Aberystwyth University for funding this study. The authors thank Professor Terry Burke for guidance throughout and for constructive comments on the manuscript, Dr Joanne Porter for advice on molecular analyses and Rory Geoghegan and Gareth Owen for assistance in sampling and sample storage.</p>" ]

[]

[]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p>Review of heterozygosity in microsatellite studies of marine and freshwater gastropods.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p>Characterisation of seven polymorphic microsatellite loci (Primer set_A) for <italic>Hydrobia ulvae</italic>.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p>Characterisation of redesigned Primer set_B for <italic>Hydrobia ulvae</italic>.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S4\"><caption><title>Additional file 4</title><p>Comparison of microsatellite genotyping data using two primer sets (A and B) for <italic>Hydrobia ulvae</italic>.</p></caption></supplementary-material>" ]

[]

[]

[ "<media xlink:href=\"1471-2156-9-55-S1.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2156-9-55-S2.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2156-9-55-S3.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2156-9-55-S4.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Gliomas, the most common brain tumor, are currently classified as astrocytic, ependymal, oligodendroglial and choroid plexus tumors. Among astrocytic tumors, glioblastoma (World Health Organization grade IV [##REF##17618441##1##]) is the most lethal primary malignant brain tumor. Although considerable progress has been made in its treatment, the clinical prognosis associated with this tumor remains poor.</p>", "<p>Histone deacetylases (HDACs) have recently become recognized as a promising target for cancer therapy, including for the treatment of glioblastomas [##REF##16450343##2##]. Together with histone acetyltransferases (HATs), HDACs are responsible for chromatin packaging, which influences the transcription process. In general, increased levels of acetylation (high HAT levels) are associated with increased transcriptional activity, whereas decreased acetylation levels (high HDAC levels) are associated with repression of transcription (reviewed in [##REF##11257101##3##]). HDACs are classified into 4 major categories based on their homology to yeast HDACs, including structure and cellular localization (Figure ##FIG##0##1##). Class I and class II HDAC proteins share a common enzymatic mechanism that is the Zn-catalyzed hydrolysis of the acetyl-lysine amide bond. Human class I HDACs includes HDAC1, -2, -3, and -8, which are enzymes similar to the yeast transcriptional regulator Rpd3, generally localized to the nucleus [##REF##11139331##4##,##REF##17325692##5##]. These enzymes are ubiquitously expressed (with the exception of <italic>HDAC8</italic>, which has higher expression levels in the liver) and seems to function as a complex with other proteins [##REF##12429021##6##]. HDAC1 and -2 only show activity within a protein complex, which consists of proteins necessary for modulating their deacetylase activity and DNA binding, and the recruitment of HDACs to gene promoters [##REF##10444591##7##]. Wilson AJ et al. [##REF##16533812##8##] have suggested that multiple class I HDAC members are also involved in repressing p21 and that the growth inhibitory and apoptotic effects induced by HDAC inhibitors are probably mediated through the inhibition of multiple HDACs.</p>", "<p>Class II HDACs includes HDAC4, -5, -6, -7, -9a, -9b, and -10, which are homologous to yeast Hda1. These class II enzymes can be found in the nucleus and cytoplasm, suggesting potential extranuclear functions by regulating the acetylation status of nonhistone substrates [##REF##10220385##9##,##REF##12711221##10##]. HDAC members of class II are abundantly expressed in skeletal muscle, heart, brain, tissues with low levels of mitotic activity [##REF##10640276##11##,##REF##9891014##12##]. Functionally, Class II HDACs is thought to act as transcriptional corepressors by deacetylating nucleosomal histones. These enzymes do not bind directly to DNA; they are thought to be recruited to distinct regions of the genome by sequence specific DNA binding proteins [##REF##10748098##13##, ####REF##10487761##14##, ##REF##10523670##15####10523670##15##].</p>", "<p>Class III HDACs is composed of the Sirtuins (SIRT) proteins 1–7, which are homologous to the yeast Sir2 protein and require NAD⁺for deacetylase activity in contrast to the zinc-catalyzed mechanism used by class I and II HDACs [##REF##15189148##16##, ####REF##10693811##17##, ##REF##15506920##18####15506920##18##].</p>", "<p>An additional HDAC expressed by higher eukaryotes is a Zn-dependent HDAC (HDAC11 in mammals). This enzyme is phylogenetically different from both class I and class II enzymes and is therefore classified separately as class IV [##REF##11948178##19##] reviewed in [##REF##17325692##5##].</p>", "<p>The use of HDAC inhibitors (HDACis) for the treatment of cancer is an area of active investigation. In gliomas, HDACis have been used for the treatment of glioblastoma in combination with radiation therapy and chemotherapy. Some authors have demonstrated that HDACis have a radiosensitizing effect on glioblastoma cells <italic>in vitro </italic>and <italic>in </italic>vivo [##REF##17707275##20##, ####UREF##0##21##, ##REF##17347490##22##, ##REF##15234053##23####15234053##23##] and also seem to be associated with inhibition of glioma cell growth by both cell-cycle arrest and apoptosis [##REF##15024582##24##, ####REF##17310267##25##, ##REF##16383255##26####16383255##26##]. Despite the widespread use of HDACis, the mechanistic implications remain to be elucidated.</p>", "<p>To this date, there are no studies that demonstrate the status of global HDAC gene expression and protein levels in astrocytomas. The purpose of this study was to evaluate and compare mRNA and protein levels of class I, II and IV of HDACs in low grade and high grade astrocytomas and normal brain tissue and to correlate the findings with the malignancy in astrocytomas.</p>" ]

[ "<title>Methods</title>", "<title>Patients Samples</title>", "<p>For this study, tumor samples of 43 patients (19 men and 24 women) ranging in age from 1.3 to 79 years (mean age 24.6 years, with a median and a standard deviation of 12.8 ± 22.6 years) were evaluated. The histopathologic diagnoses were 20 low-grade gliomas (13 grade I and 7 grade II) and 23 high-grade gliomas (5 grade III and 18 glioblastomas). In addition, 11 samples of normal cerebral tissue were analyzed. Frozen tumor and normal specimens were microdissected. Diagnoses were based on 2007 World Health Organization criteria [##REF##17618441##1##].</p>", "<p>For tumor microdissection, tumor samples were placed on a cooled platform and immediately positioned on the cutting base of the cryostat under Tissue Tek (Fisher Scientific, Pittsburg, PA). After rapid freezing in liquid nitrogen, the sample was cut and immediately captured on a coverslip, stained with hematoxylin and eosin, and evaluated by image apposition. The area of interest in the original cryopreserved tumor block was then trimmed, and the microdissected sample was transferred to a previously identified tube, which was immediately placed under dry ice.</p>", "<p>Prior to initiation, the research here presented was approved by the Research Ethics Committee of the University Hospital of the Faculty of Medicine of University of Sao Paulo, processes number 9375/2003 and 7645/99. The mentioned Committee is in agreement with the Helsinki Declaration requirements for research carried out on humans. Informed consent was also taken from each patient (or their legal representative) involved in this project, also in accordance to the Helsinki Declaration.</p>", "<title>RNA extraction and cDNA synthesis</title>", "<p>Total cellular RNA was extracted using Trizol^®Reagent (Invitrogen, Carlsbad, CA, USA) and RNA was reverse transcribed to single-stranded cDNA using a High Capacity Kit (Applied Biossystems, Foster City, CA, USA) according to the manufacturer's protocol.</p>", "<title>Quantitative real-time polymerase chain reaction (qRT-PCR)</title>", "<p>Messenger RNA expression level for each <italic>HDAC </italic>was evaluated using an ABI 7500 machine (Applied Biosystems, Foster City, CA, USA). Amplifications were obtained using on demand TaqMan^®probes (Applied Biosystems, Foster City, CA, USA) for each <italic>HDAC</italic>. For relative quantification of gene expression, standard curves were constructed for each gene by considering at least 3 points in triplicate of 10-fold serial dilution of cDNA in water, starting from 1:10 of a volume of undiluted cDNA transcribed from 1.0 μg of total RNA. The slopes of standard curves ranged from -3.17 to -3.87. Blank and standard controls (calibrators) were run in parallel to verify amplification efficiency within each experiment. To normalize differences in the amount of total cDNA added to each reaction, <italic>β-glucuronidase </italic>(<italic>GUS β</italic>) gene expression was used as an endogenous control. As a calibrator sample (reference sample for relative quantification), the U343 cell line was used. To obtain the Ct (cycle) values, we established a threshold of 0.1. All reactions were made in duplicate, and all procedures were carried out at 4°C.</p>", "<title>Western Blotting</title>", "<p>For protein analysis, 30 μg of each sample was loaded and separated by sodium dodecyl sulphate-polyacrylamide gel electrophoresis [##REF##5432063##27##]. Proteins were transferred to nitrocellulose membranes, and the membranes were then incubated in 1% Tris-buffered saline Tween-20 (TBST) containing 5% (w/v) dried non-fat milk for 1 hour at room temperature. The primary antibodies were diluted at 1:3000 in TBST containing 5% (w/v) milk: HDAC9 (Abcam, Cambridge, MA), Acetyl-Lys H3 (Abcam, Cambridge, MA), and Acetyl-Lys H4 (Upstate Biotechnology Lake Placid, NY), and the membranes were incubated for 1 hour at room temperature, The membranes were then washed 3 times with TBST, incubated with the secondary antibody (1:5000 in TBST) labeled by horseradish peroxidase (Abcam, Cambridge, MA) for 1 hour at room temperature, and washed 3 times with TBST. The secondary antibody was visualized using electron chemiluminescent reagent (Pierce, Rockford, IL). Films were exposed from 10 to 60 seconds and developed.</p>", "<title>Statistical analysis</title>", "<p>Comparison of gene expression between groups of tumor was performed by nonparametric testes Mann-Whitney and Kruskall-Wallis. The level of significance was set at <italic>p </italic>&lt; 0.05 in all analyses.</p>" ]

[ "<title>Results</title>", "<title>Global expression of <italic>HDAC </italic>genes in gliomas and normal brain tissue</title>", "<p>The expression of 12 <italic>HDAC </italic>genes was analyzed using relative quantification of mRNA levels in normal brain, astrocytomas grades I, II and III, and glioblastomas (Figure ##FIG##1##2##). Class I <italic>HDAC </italic>genes (<italic>HDAC</italic>1, -2, -3, and -8) showed lower levels of expression (relative expression of 0.5 to 6.0 approximately) compared to the other classes studied. The highest values of expression for class I were seen for <italic>HDAC8 </italic>(2.0 to 6.0). Expression of <italic>HDAC </italic>class II and class IV were higher, with values of approximately 1.0 to 150.0. The highest level of mRNA was observed for <italic>HDAC9a </italic>and <italic>HDAC9b</italic>, with relative expression reaching values above 100 for normal brain tissue and grade I astrocytomas; however, <italic>HDAC6 </italic>and <italic>HDAC7 </italic>showed levels of expression comparable with those of class I (approximately 1.0 and 3.0).</p>", "<p>Table ##TAB##0##1## shows the median values for each <italic>HDAC </italic>gene in all groups analyzed. For class I, the highest median value observed was 2.87 (<italic>HDAC8 </italic>in low-grade astrocytoma) and the lowest median value was 0.62 (<italic>HDAC3 </italic>in astrocytoma grade III). For class II, the highest median was 61.51 (<italic>HDAC9b </italic>in normal brain) and the lowest was 0.56 (<italic>HDAC7 </italic>in glioblastoma). <italic>HDAC11</italic>, the only member of class IV, had the lowest median value for normal brain tissue (3.67) and the highest for glioblastoma (1.02).</p>", "<title>Comparison of the <italic>HDAC </italic>mRNA levels in low- and high-grade gliomas and normal brain tissue</title>", "<p>We compared mRNA levels of <italic>HDAC </italic>genes (the medians of relative expression) in low- and high-grade gliomas and also in normal brain. Among class I <italic>HDACs</italic>, significant differences in gene expression between tumor groups were not observed (Table ##TAB##1##2##).</p>", "<p>Seven of 8 class II <italic>HDAC </italic>genes (exception for <italic>HDAC4</italic>) were expressed at lower levels in high-grade astrocytomas compared to low-grade astrocytomas (<italic>p </italic>&lt; 0.05; Table ##TAB##1##2##). The same significant difference for low-grade and high-grade gliomas was observed for <italic>HDAC11 </italic>(class IV).</p>", "<p>We compared the most malignant form of astrocytoma (glioblastoma) with the other 3 tumors (astrocytomas grades I, II and III) and normal brain in order to establish a correlation between <italic>HDAC </italic>expression and tumor grade (Table ##TAB##2##3##). As mentioned above, no significant difference in <italic>HDAC </italic>expression was observed for class I; however, for classes II and IV, there was a decrease in expression of these genes in glioblastoma compared to that in other groups. Moreover, this downregulation appeared to follow a pattern in which lower-grade tumors had a larger number of <italic>HDAC </italic>genes at lower expression levels. Comparison between glioblastoma and grade III astrocytoma showed that 4 of 8 genes were expressed at lower levels in glioblastoma samples (<italic>HDAC4</italic>, <italic>-6</italic>, <italic>-7</italic>, <italic>and -11</italic>). Comparison with low-grade astrocytoma (grades I and II) showed that the expression of 6 of 8 genes was lower in glioblastomas (except for <italic>HDAC5 </italic>and -<italic>7</italic>). Finally, when we compared glioblastoma with normal brain, 7 of 8 genes studied, with the exception of <italic>HDAC7</italic>, were expressed at lower levels in glioblastoma.</p>", "<title>Protein analysis: Acetyl H3 but not Acetyl H4 correlates with mRNA levels</title>", "<p>In order to validate the data obtained from qRT-PCR, western blot analysis was performed for HDAC9b protein. This protein was chosen because we found the highest levels of mRNA expression for HDAC9b. The results obtained for HDAC9b western blot analysis confirmed the data obtained in quantitative mRNA analysis. The protein levels of HDAC9a were higher in normal brain tissue and low-grade astrocytoma than in the grade III astrocytoma and glioblastoma (Figure ##FIG##2##3A##).</p>", "<p>Anti-acetyl histone H3 and anti-acetyl histone H4 antibodies were also used to verify the level of acetylated histones H3 and H4 and to correlate the findings with histone deacetylase activity in the groups studied (Figure ##FIG##2##3b##).</p>", "<p>Considering the large number of <italic>HDAC </italic>genes with low levels of expression in glioblastomas, we expected that the levels of acetylated histones were higher in those tumors. Interestingly, when we analyzed the acetylation levels of the H3 and H4 histones, H3 histone acetylation, but not H4 histone acetylation, correlated with the data obtained by qRT-PCR (Figure ##FIG##2##3B##). Glioblastoma samples showed higher levels of acetylated histone H3 than normal brain and low-grade gliomas.</p>" ]

[ "<title>Discussion</title>", "<p>In this study we evaluated and compared mRNA levels of 12 <italic>HDAC </italic>genes in astrocytomas and normal brain tissue. As mentioned before, the acetylation levels of histones is a process regulated by two groups of important enzymes: HATs and HDACs, and the balance of the activities of these two enzymes is tightly related to the gene expression status in the cell. The regulation of HDACs has been studied, but is not yet well established. The reason for that is maybe because HDACs seem to be regulated at multiple levels including cellular compartmentalization, association with other factors, abundance, activity states and genome-wide distribution.</p>", "<p>Considering that altered gene expression is frequently observed in cancer, a relationship between HDACs and cancer progression has been postulated. Wade P. [##REF##11257101##3##] suggested that loss of targeting of class I HDACs through disruption of a transcriptional corepressor and the inappropriate redistribution of class II resulted in the misregulated gene expression in cancer.</p>", "<p>Although a crucial role for HDACs in gene transcription and their possible involvement in cancer has been proposed, no studies have demonstrated the expression profile of 3 classes of HDACs simultaneously in brain tumor.</p>", "<p>Our study did not find differential gene expression of class I <italic>HDACs </italic>in high-grade and low-grade gliomas which may indicate that class of deacetylases seems to not be directly involved with malignancy of gliomas. Only a few studies have evaluated the level of class I <italic>HDAC </italic>expression in cancer: Huang BH et al. [##REF##15665816##28##] demonstrated that <italic>HDAC1 </italic>and <italic>HDAC2 </italic>seem to be upregulated in colon cancer; Choi JH et al. [##REF##11749695##29##] demonstrated an overexpression of <italic>HDAC1 </italic>mRNA 68% of gastric cancer tissues studied by them (17 of 25) and elevated expression of HDAC1 protein was also detected in 61% of the gastric cancer samples (11 of 18). Expression of class I <italic>HDAC3 </italic>was also shown elevated in astrocytic glial tumors compared to nonmalignant gliosis [##REF##17007107##30##]. On the other hand, Ozdag H. et al. [##REF##16638127##31##] showed that <italic>HDAC1 </italic>is significantly lower in colorectal cancer samples in comparison to normal colorectal tissues.</p>", "<p>For class II <italic>HDACs</italic>, downregulation of its expression in glioblastoma compared to low-grade gliomas and normal brain tissue was demonstrated and statistically confirmed in our study, indicating a negative correlation between <italic>HDAC </italic>expression and malignancy in gliomas. In agreement with our study, some authors have demonstrated downregulation of these deacetylases and their relationship with prognosis in cancer as well. Ozdag H. et al. [##REF##16638127##31##] showed that <italic>HDAC5 </italic>and <italic>HDAC7 </italic>are significantly lower in colorectal cancer samples in comparison to normal colorectal tissues. These authors also showed downregulation of <italic>HDAC5 </italic>in renal tumors compared to normal renal tissue. Downregulation of <italic>HDAC </italic>gene expression has also been observed in lung cancer: reduced expression of class II <italic>HDAC </italic>genes (mostly <italic>HDAC10</italic>) seems to be significantly associated with poor prognosis, suggesting that class II <italic>HDACs </italic>may repress critical genes that may have important roles in lung cancer progression [##UREF##1##32##]. Like in our study, these authors demonstrated that class I <italic>HDACs </italic>seem to have no correlation with malignancy of those tumors.</p>", "<p>The downregulation of <italic>HDAC </italic>genes in cancer may be difficult to explain, since multiple factors are involved in the regulation of these enzymes. Unlike class I HDACs, which are predominantly localized in the nucleus, class II HDACs actively shuttle between the cytoplasm and nucleus being under control of classic cellular signaling pathways, and cellular localizations represents a fundamental mechanism for them [##REF##9891014##12##,##REF##10487761##14##]. Additionally, class II <italic>HDACs </italic>seem to have additional levels of regulation, which makes elucidation of the mechanism of gene transcription regulation more complicated.</p>", "<p>Our study seems to reveal the involvement of class II HDAC<italic>s </italic>in glioma malignancy. If we consider that, in general, increased levels of acetylation (downregulation of HDACs) is related to higher transcriptional activity, we could predict that gene transcription in malignant gliomas may be upregulated. In that case we could infer that some proto-oncogenes might be overexpressed and somehow leading to the malignancy. The overexpression of proto-oncogenes in gliomas has been well documented. Besides <italic>EGFR</italic>, which is found overexpressed in 40% of gliomas, the genes <italic>N-MYC</italic>, <italic>C-MYC</italic>, <italic>PDGFR-α, MYB</italic>, <italic>K-RAS</italic>, <italic>CDK</italic>-4 and <italic>MDM</italic>2 are the most commonly amplified oncogenes in gliomas [##REF##10671694##33##, ####REF##16619307##34##, ##REF##11550301##35##, ##REF##15742604##36##, ##REF##9643506##37####9643506##37##]. However, due to the fact that these deacetylases are regulated at several levels, additional information about the functionality of HDACs in gliomas is required.</p>", "<p>In order to correlate mRNA and protein levels of HDACs, we analyzed HDAC9 protein levels in all groups of tumors studied. Western blot analysis showed that HDAC9 protein is expressed at a higher concentration in normal brain and low-grade gliomas than in high-grade gliomas, validating the data obtained by real-time PCR.</p>", "<p>We also evaluated HDAC activity by analyzing histone acetylation levels in tumor samples and normal brain. When we analyzed the levels of H3 and H4 acetylated histones, we observed an increased level in acetylation of H3 but not of H4 histone in glioblastomas compared to low-grade astrocytomas and normal brain tissue. Considering the low levels of <italic>HDAC </italic>expression in glioblastomas, it was expected that the levels of acetylated histones should be more elevated in those tumors, as demonstrated here for histone H3. On the other hand, the lack of correlation between low <italic>HDAC </italic>expression and high histone H4 deacetylation levels in glioblastomas could be explained by the existence of cofactors or unidentified regulators. Some few authors have already demonstrated that the specificity of HDACs depends on cofactors which makes HDAC specificity a complicated process [##REF##10346897##38##, ####REF##17428445##39##, ##REF##14511685##40##, ##REF##16079916##41####16079916##41##]. It is tempting to speculate that class II of HDACs could be responsible for deacetylation of histone H3 more than histone H4. However, the data here presented are not enough to infer about the specificity of HDACs in astrocytomas. Moreover, it has been observed that differences between HDAC subtype specificity do not coincide with the division into class I and class II enzymes. HDAC1, HDAC3 (class I), and HDAC6 (class II) seem to be very similar in substrate specificity and mainly differ in the degree of specificity [##REF##17428445##39##]. Until now, a lot of data has been generated about HDAC research, but the natural substrates of different HDACs and their substrate specificities is still not well understood.</p>", "<p>The class IV HDAC11 enzyme seems to be an unusual member of the HDAC family. Its sequence is not homologous to any other HDAC, and it may have distinct physiological roles. This enzyme, like members of the class II HDACs is expressed more in brain, heart, skeletal muscle, and kidney [##REF##11948178##19##]. In our study, we also found high levels of <italic>HDAC11 </italic>in normal brain tissue and significant differential expression was also observed for this enzyme in low-grade and high-grade gliomas. Little is known about this unique HDAC, and its relation to malignancy of gliomas is yet to be elucidated.</p>", "<title>HDACis</title>", "<p>Even considering that to evaluate the effect of HDAC inhibitors in gliomas was not the aim of the present study, the finding that most of the <italic>HDAC </italic>genes are downregulated in glioblastomas could lead us to consider that HDACis may not be effective in the treatment of high-grade gliomas. Some studies have demonstrated radiosensitization of gliomas after HDACi treatment [##UREF##0##21##,##REF##17347490##22##]. Therefore, despite the low levels of <italic>HDAC </italic>gene expression in glioblastomas, HDACis seem to be potential therapeutical targets for glioma treatment. The explanation to this may point to the existence of nonhistone substrates for HDACs. Although histones are the most thoroughly studied as HDAC substrates, several reports have shown that HDACs are also responsible for the deacetylation of diverse types of nonhistone proteins, including transcriptional factors, signal transduction mediators, and molecular chaperones (a summarized table of nonhistone substrates for HDACs can be found in [##REF##16450343##2##]). Additionally, recent evidences suggest that modulation of gene expression through histone remodeling might not be the only process responsible for the antiproliferative action of HDACis [##REF##11304533##42##,##REF##12892709##43##]. Although the present study has demonstrated that most of <italic>HDAC</italic>s genes are downregulated in glioblastomas, no experiment was performed in order to analyze the effect of HDAC inhibitors on glioma treatment, therefore the effect of these inhibitors on glioma treatment should be addressed in a separate manuscript.</p>" ]

[ "<title>Conclusion</title>", "<p>Our study has established a negative correlation between <italic>HDAC </italic>gene expression and glioma grade, suggesting that class II and class IV <italic>HDACs </italic>might play an important role in glioma malignancy. This differential <italic>HDAC </italic>expression may provide insight into development of novel treatment approaches for this devastating disease. However, a more complete understanding of the biological function and specificity of the diverse HDAC isoforms and their involvement in the cancer process is necessary.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Glioblastoma is the most lethal primary malignant brain tumor. Although considerable progress has been made in the treatment of this aggressive tumor, the clinical outcome for patients remains poor. Histone deacetylases (HDACs) are recognized as promising targets for cancer treatment. In the past several years, HDAC inhibitors (HDACis) have been used as radiosensitizers in glioblastoma treatment. However, no study has demonstrated the status of global <italic>HDAC </italic>expression in gliomas and its possible correlation to the use of HDACis. The purpose of this study was to evaluate and compare mRNA and protein levels of class I, II and IV of HDACs in low grade and high grade astrocytomas and normal brain tissue and to correlate the findings with the malignancy in astrocytomas.</p>", "<title>Methods</title>", "<p>Forty-three microdissected patient tumor samples were evaluated. The histopathologic diagnoses were 20 low-grade gliomas (13 grade I and 7 grade II) and 23 high-grade gliomas (5 grade III and 18 glioblastomas). Eleven normal cerebral tissue samples were also analyzed (54 total samples analyzed). mRNA expression of class I, II, and IV <italic>HDACs </italic>was studied by quantitative real-time polymerase chain reaction and normalized to the housekeeping gene <italic>β-glucuronidase</italic>. Protein levels were evaluated by western blotting.</p>", "<title>Results</title>", "<p>We found that mRNA levels of class II and IV <italic>HDACs </italic>were downregulated in glioblastomas compared to low-grade astrocytomas and normal brain tissue (7 in 8 genes, <italic>p </italic>&lt; 0.05). The protein levels of class II HDAC9 were also lower in high-grade astrocytomas than in low-grade astrocytomas and normal brain tissue. Additionally, we found that histone H3 (but not histone H4) was more acetylated in glioblastomas than normal brain tissue.</p>", "<title>Conclusion</title>", "<p>Our study establishes a negative correlation between <italic>HDAC </italic>gene expression and the glioma grade suggesting that class II and IV <italic>HDACs </italic>might play an important role in glioma malignancy. Evaluation of histone acetylation levels showed that histone H3 is more acetylated in glioblastomas than normal brain tissue confirming the downregulation of <italic>HDAC </italic>mRNA in glioblastomas.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interest.</p>", "<title>Authors' contributions</title>", "<p>Experiments and collection of data were performed by AKBL-E, MAAC, ETV, FJNM and RGPQ. Collection of tumor samples was performed by HRM and CGCJr. Microdissection of the tumor samples was performed by LN. AKBL-E was responsible for data analysis and interpretation and also wrote the manuscript. Manuscript reviewing was made by AKBL-E, MAAC, CAS and LGT.</p>", "<title>Pre-publication history</title>", "<p>The pre-publication history for this paper can be accessed here:</p>", "<p><ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2407/8/243/prepub\"/></p>" ]

[ "<title>Acknowledgements</title>", "<p>We thank Dr. John F. de Groot and Dr. Vinay Puduvalli from MD Anderson Cancer Center (Houston, Tx, USA) for providing Actin and Acetyl-H4 antibodies, respectively. We also thank Priscila C. Leite for the technical assistance in this work.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Classification of classes I, II, and IV HDACs by structure and cellular localization.</bold>[##REF##16450343##2##,##REF##12429021##6##,##REF##15955865##44##,##UREF##2##45##].</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Relative expression of <italic>HDAC </italic>genes in astrocytomas and normal brain tissue</bold>. Class I in purple, class II in orange, and class IV in green. Data obtained by qRT-PCR analysis using Taqman^®probes for each gene using the <italic>GUS β </italic>gene as housekeeping. Standard curves were constructed for each gene by considering at least 3 points in triplicate of 10-fold serial dilution of cDNA in water, starting from 1:10 of a volume of undiluted cDNA transcribed from 1.0 μg of total RNA. The slopes of standard curves ranged from -3.17 to -3.87. Blank and standard controls (calibrators) were run in parallel to verify amplification efficiency within each experiment. To obtain the Ct (cycle) values, we established a threshold of 0.1. *AI, AII and AIII mean astrocytoma grade I, II and III respectively. GBM means glioblastoma.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Protein analysis</bold>. (A) HDAC9 and (B) acetyl H3 and acetyl H4 in glioblastoma (GBM), astrocytoma grade III (AIII), astrocytoma grades I and II (AI and II), and normal brain. Actin was used as endogenous control.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Relative expression (medians) of <italic>HDAC </italic>genes in different groups of tumor and normal brain tissue.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Normal Brain</bold></td><td align=\"center\"><bold>Astrocytoma grades I and II</bold></td><td align=\"center\"><bold>Astrocytoma grade III</bold></td><td align=\"center\"><bold>Glioblastoma</bold></td></tr></thead><tbody><tr><td align=\"left\"><bold><italic>Class I</italic></bold></td><td/><td/><td/><td/></tr><tr><td align=\"left\"><italic>HDAC1</italic></td><td align=\"center\">0.98</td><td align=\"center\">1.19</td><td align=\"center\">1.93</td><td align=\"center\">1.03</td></tr><tr><td align=\"left\"><italic>HDAC2</italic></td><td align=\"center\">0.68</td><td align=\"center\">0.72</td><td align=\"center\">1.30</td><td align=\"center\">0.95</td></tr><tr><td align=\"left\"><italic>HDAC3</italic></td><td align=\"center\">0.85</td><td align=\"center\">0.73</td><td align=\"center\">0.62</td><td align=\"center\">0.69</td></tr><tr><td align=\"left\"><italic>HDAC8</italic></td><td align=\"center\">1.30</td><td align=\"center\">2.87</td><td align=\"center\">2.08</td><td align=\"center\">1.64</td></tr><tr><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"><bold><italic>Class II</italic></bold></td><td/><td/><td/><td/></tr><tr><td align=\"left\"><italic>HDAC4</italic></td><td align=\"center\">3.27</td><td align=\"center\">3.33</td><td align=\"center\">8.43</td><td align=\"center\">1.42</td></tr><tr><td align=\"left\"><italic>HDAC5</italic></td><td align=\"center\">1.71</td><td align=\"center\">2.72</td><td align=\"center\">2.27</td><td align=\"center\">0.91</td></tr><tr><td align=\"left\"><italic>HDAC6</italic></td><td align=\"center\">1.72</td><td align=\"center\">3.05</td><td align=\"center\">2.57</td><td align=\"center\">1.33</td></tr><tr><td align=\"left\"><italic>HDAC7</italic></td><td align=\"center\">0.61</td><td align=\"center\">0.80</td><td align=\"center\">0.81</td><td align=\"center\">0.56</td></tr><tr><td align=\"left\"><italic>HDAC9a</italic></td><td align=\"center\">30.74</td><td align=\"center\">24.46</td><td align=\"center\">30.51</td><td align=\"center\">11.49</td></tr><tr><td align=\"left\"><italic>HDAC9b</italic></td><td align=\"center\">61.51</td><td align=\"center\">17.10</td><td align=\"center\">18.32</td><td align=\"center\">7.43</td></tr><tr><td align=\"left\"><italic>HDAC10</italic></td><td align=\"center\">6.54</td><td align=\"center\">7.29</td><td align=\"center\">11.04</td><td align=\"center\">4.26</td></tr><tr><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"><bold><italic>Class IV</italic></bold></td><td/><td/><td/><td/></tr><tr><td align=\"left\"><italic>HDAC11</italic></td><td align=\"center\">3.67</td><td align=\"center\">3.30</td><td align=\"center\">3.05</td><td align=\"center\">1.02</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Statistical comparison (Kruskall-Wallis analysis) of levels of <italic>HDAC </italic>expression (medians) between low-grade (astrocytomas grade I and II) and high-grade (astrocytomas grade III and glioblastomas) astrocytomas.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Low-grade astrocytomas × high-grade astrocytomas </bold><break/><bold>(<italic>p </italic>value)</bold></td></tr></thead><tbody><tr><td align=\"left\"><bold><italic>Class I</italic></bold></td><td/></tr><tr><td align=\"left\"><italic>HDAC1</italic></td><td align=\"center\">ns</td></tr><tr><td align=\"left\"><italic>HDAC2</italic></td><td align=\"center\">ns</td></tr><tr><td align=\"left\"><italic>HDAC3</italic></td><td align=\"center\">ns</td></tr><tr><td align=\"left\"><italic>HDAC8</italic></td><td/></tr><tr><td/><td/></tr><tr><td align=\"left\"><bold><italic>Class II</italic></bold></td><td/></tr><tr><td align=\"left\"><italic>HDAC4</italic></td><td align=\"center\">ns</td></tr><tr><td align=\"left\"><italic>HDAC5</italic></td><td align=\"center\">&lt; 0.05</td></tr><tr><td align=\"left\"><italic>HDAC6</italic></td><td align=\"center\">&lt; 0.01</td></tr><tr><td align=\"left\"><italic>HDAC7</italic></td><td align=\"center\">&lt; 0.05</td></tr><tr><td align=\"left\"><italic>HDAC9a</italic></td><td align=\"center\">&lt; 0.01</td></tr><tr><td align=\"left\"><italic>HDAC9b</italic></td><td align=\"center\">&lt; 0.0001</td></tr><tr><td align=\"left\"><italic>HDAC10</italic></td><td align=\"center\">&lt; 0.001</td></tr><tr><td/><td/></tr><tr><td align=\"left\"><bold><italic>Class IV</italic></bold></td><td/></tr><tr><td align=\"left\"><italic>HDAC11</italic></td><td align=\"center\">&lt; 0.001</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Statistical comparison (Mann-Whitney analysis) of levels of <italic>HDAC </italic>expression (medians) between tumor groups.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"4\"><italic>p </italic>value</td></tr><tr><td/><td colspan=\"4\"><hr/></td></tr><tr><td/><td align=\"center\"><italic>HDAC1</italic></td><td align=\"center\"><italic>HDAC2</italic></td><td align=\"center\"><italic>HDAC3</italic></td><td align=\"center\"><italic>HDAC8</italic></td></tr></thead><tbody><tr><td align=\"left\">Glioblastoma x</td><td/><td/><td/><td/></tr><tr><td align=\"left\">Normal brain</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">ns</td></tr><tr><td align=\"left\">AI and II</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">ns</td></tr><tr><td align=\"left\">AIII</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">ns</td></tr><tr><td/><td colspan=\"4\"><hr/></td></tr><tr><td/><td align=\"center\"><italic>HDAC4</italic></td><td align=\"center\"><italic>HDAC5</italic></td><td align=\"center\"><italic>HDAC6</italic></td><td align=\"center\"><italic>HDAC7</italic></td></tr><tr><td colspan=\"5\"><hr/></td></tr><tr><td align=\"left\">Glioblastoma x</td><td/><td/><td/><td/></tr><tr><td align=\"left\">Normal brain</td><td align=\"center\">&lt;0.0001</td><td align=\"center\">0.018</td><td align=\"center\">0.011</td><td align=\"center\">ns</td></tr><tr><td align=\"left\">AI and II</td><td align=\"center\">0.012</td><td align=\"center\">ns</td><td align=\"center\">0.001</td><td align=\"center\">ns</td></tr><tr><td align=\"left\">AIII</td><td align=\"center\">0.010</td><td align=\"center\">ns</td><td align=\"center\">0.010</td><td align=\"center\">0.008</td></tr><tr><td/><td colspan=\"4\"><hr/></td></tr><tr><td/><td align=\"center\"><italic>HDAC9a</italic></td><td align=\"center\"><italic>HDAC9b</italic></td><td align=\"center\"><italic>HDAC10</italic></td><td align=\"center\"><italic>HDAC11</italic></td></tr><tr><td colspan=\"5\"><hr/></td></tr><tr><td align=\"left\">Glioblastoma x</td><td/><td/><td/><td/></tr><tr><td align=\"left\">Normal brain</td><td align=\"center\">0.0096</td><td align=\"center\">&lt;0.0001</td><td align=\"center\">0.015</td><td align=\"center\">0.0001</td></tr><tr><td align=\"left\">AI and II</td><td align=\"center\">0.01</td><td align=\"center\">0.013</td><td align=\"center\">0.001</td><td align=\"center\">0.0003</td></tr><tr><td align=\"left\">AIII</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">ns</td><td align=\"center\">0.023</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>ns: no significant</p></table-wrap-foot>", "<table-wrap-foot><p>AI, AII, and AIII mean astrocytoma grades I, II and III, respectively.</p><p>ns: no significant</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2407-8-243-1\"/>", "<graphic xlink:href=\"1471-2407-8-243-2\"/>", "<graphic xlink:href=\"1471-2407-8-243-3\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The relationship between genetic diversity and population size offers a number of tantalizing insights into demographic influences on evolution [##UREF##0##1##, ####REF##14631042##2##, ##UREF##1##3####1##3##]. While life history characteristics of species tend to make the effective population size (N_e) of a species much lower than the actual census size [##UREF##2##4##, ####UREF##3##5##, ##REF##12454077##6####12454077##6##], neutral theory [##REF##5713805##7##] predicts a proportional relationship between genetic diversity and N_e[##UREF##1##3##,##UREF##4##8##]. Research has shown many cases in which N_eas estimated from genetic markers is several orders of magnitude lower than would be predicted based on census size (N) and a species' reproductive traits [##UREF##5##9##], and it has been suggested that extremely high variance in reproductive success (the \"sweepstakes\" models of [##UREF##0##1##,##REF##12454077##6##]) or genome-wide selective sweeps [##REF##11794777##10##,##REF##16645093##11##] may be causal mechanisms.</p>", "<p>Here, using the results of Pringle and colleagues [##UREF##6##12##,##UREF##7##13##] and a simple numerical model, we quantify N_efor populations whose dispersal is subject to persistent directional flow and find a complementary mechanism for the reduction of N_e. We do this in a linear domain, such as a benthic population in a stream or a coastline, though the results can be easily generalized to different geometries. We find that <italic>physical </italic>drift in the ocean or in a stream supplements <italic>genetic </italic>drift as a mechanism for losing genetic diversity, and thus asymmetric dispersal – where larvae are more likely to settle to one side of their parent then another – will reduce N_ein a given species domain. This mechanism for the reduction of N_ewill be shown to be distinct from sweepstakes models. In the sweepstakes models, N_eis reduced by variability in reproductive success between individuals in the same region. In contrast, physical drift will be shown to reduce N_eby creating differential reproductive success between individuals in different regions.</p>", "<p>Previous work [e.g. [##REF##17248820##14##,##REF##18707443##15##]] has shown that in a linked series of populations that exchange differing numbers of migrants, the \"downstream\" sink population (i.e., the one that received more immigration from the \"upstream\" source population than the \"upstream\" population received from it) lost endemic alleles and eventually acquired the allele structure of the upstream population. These results suggest that the N_eof the entire group of populations would tend to that of only the upstream population, given sufficiently large and asymmetric migration. While their work treated this effect in discrete demes, we examine a continuously distributed species along a coast or river in which there is no <italic>a priori </italic>partition into separate populations – a pattern that closely resembles many marine and freshwater systems. In addition, we assume that local density dependent effects limit the population and that there is no significant immigration from outside of the population being considered. We first find the region of the species domain in which the species is maintained by propagules released in that region, and not by migration from elsewhere. Then we show that it is the population of this region that acts as a source of allelic diversity and defines N_efor the species over its entire domain. By identifying allelic retention as a spatially defined component of coastal diversity, this work has implications for the design of marine reserves. While genetic diversity has typically not been considered in the placement or size of marine reserves [##UREF##8##16##], there are clear associations between local genetic diversity and a population's resilience to stress and environmental change [##UREF##9##17##].</p>", "<title>An estimate of N_ein an advective environment</title>", "<p>In order to determine N_e, it is necessary to divide the populations into source and sink regions. To do so, we define a \"retentive population\" as a demographically stable group of individuals that can persist without immigration from outside its domain. This definition is equivalent to that of Booke [##UREF##10##18##] as discussed in [##REF##16629801##19##]. In coastal oceans, and other nearly one-dimensional systems such as rivers, Pringle and Wares [##UREF##6##12##] showed that a population and any alleles it contains could persist within a region if</p>", "<p></p>", "<p>where N_alleleis the mean number of a given allele (or class of alleles, <italic>sensu </italic>[##REF##11701630##20##]) in the offspring that recruit and successfully reach reproductive age per gene copy per adult per lifetime, <italic>considering the effects of density dependence on reproductive success</italic>. If an allele is neutral, N_alleleis equivalent to mean lifetime reproductive success per adult [##UREF##6##12##]. L_advis the average distance a successfully recruiting larva is moved downstream from its parent before recruiting, and L_diffis the standard deviation of that distance for all successful recruiting larvae an adult releases. These criteria assume that kurtosis of the dispersal kernel is close to that of a Gaussian; for other kernels, a correction has been developed (Pringle et al., in review).</p>", "<p>Allelic diversity persists within this retentive population, as the population is not supported by migration from elsewhere. We define the concept of allelic \"persistence\" relative to the expectation for the rate at which neutral alleles are lost or go to fixation in a finite randomly mating population [##REF##17248440##21##]; with time, all allelic diversity may be transient. Alleles are considered \"persistent\" in a given population if they are expected to be lost or go to fixation at the rate predicted for a neutral gene in a population of that size [##REF##17248440##21##,##REF##5713805##22##]. As will be seen below, alleles in an advective domain that originate outside of a \"retentive population\" will be lost more rapidly than the neutral prediction, and will go to fixation far less often.</p>", "<p>With advection an entire species domain cannot be a retentive population, for the criteria in equation (1) above cannot be met throughout the species range if L_advis not zero. At demographic equilibrium, each adult will (on average) generate one surviving offspring and thus one copy each for each copy of an allele it carries – thus the average of N_alleleover the species range is 1 for neutral alleles, which does not satisfy equation (1). Retention of some allelic diversity occurs because the reproductive success per adult is not evenly distributed spatially, and so in some places is great enough to satisfy eq. 1 [##UREF##6##12##]. One location where enhanced reproductive success must occur is the upstream edge of the model domain, for there can be no subsidy of this region by immigration from farther upstream. Byers and Pringle [##UREF##7##13##] note that if the successful reproduction at the upstream edge of the domain, and therefore N_allelefor a neutral allele, is greater than that needed to satisfy eq. (1), the population will increase at that point. This suggests that the population at the upstream edge will increase until, due to density dependent effects, the average of N_alleleover the upstream retention zone is reduced until it just satisfies eq. (1). At the upstream edge of the domain there will be a region where eq. (1) is satisfied, and novel allelic diversity can be retained. Since most larvae are transported downstream a mean distance L_adv, this upstream region also supplies migrants to downstream regions. Thus, the upstream edge is a retentive population where alleles will only change in frequency due to stochastic drift in allele frequency and the accompanying probability of fixation.</p>", "<p>The size and census population of the region of enhanced reproductive success, and thus N_e, will depend on the nature of the spatial variation in habitat quality, L_advand L_diff. Here we examine the case in which the habitat is spatially uniform downstream of the upstream edge of the habitat. The mean transport will move an average propagule nL_advdownstream of its parents after n generations, while the stochastic component of transport will move the propagule a standard deviation of n^0.5L_diffaround that point [##UREF##7##13##]. These two distances are equal after n = L²_diff/L²_advgenerations. Substituting this expression into either of the distances defined above gives the distance L_reten= L²_diff/L_adv, suggesting L_retenis the fundamental length scale of this system, and is the distance over which the effects of mean and stochastic propagule transport are balanced. This suggestion is confirmed with dimensional analysis [##UREF##11##23##] by noting that \"generation\" is a discrete time-like dimension, and that the relations nL_advand n^0.5L_diffsuggest underlying parameters with units of velocity (time/distance) and diffusivity (distance²/time). From these, only a single dimensionally consistent length scale can be formed, and it is L_reten. Multiplying this distance by the carrying capacity per unit length of the environment (H_dens) provides a scaling for N_ecomparable to that of [##REF##16153045##24##]:</p>", "<p></p>", "<p>and the numerical modeling described below confirms the appropriateness of this scale. (We assume that eq. (1) can be satisfied even when the population is close to its carrying capacity. When this is not true, the population is marginal at this location [##UREF##7##13##], and the estimate of N_ewill be further reduced). We expect this estimate of N_eto be reduced relative to standard drift expectations by increasing mean propagule transport (L_adv), and that this effect is diminished by increased stochastic transport (L_diff). Since L_advand L_diffare significantly smaller than species ranges for most coastal species [##UREF##6##12##,##UREF##7##13##,##UREF##12##25##], this value of N_eshould be much less than the census population size of the entire domain or metapopulation [as in [##UREF##13##26##]].</p>", "<p>Downstream of the retentive population that defines N_e, N_allelewill not satisfy (1) – and alleles are not retained – if L_advis non-zero. Allelic diversity in downstream regions will be set by the allelic composition of migrants from upstream. So, in an advective environment the evolution of allelic diversity in the entire population will be governed by the allelic diversity in the retentive population, and the N_efor the entire population should approach the census population of the retentive population given by eq. (2). However, heterogeneity in abiotic (<italic>i.e. </italic>oceanography, temperature, salinity) as well as biotic (<italic>i.e. </italic>physiological responses) factors that result in an interruption or severe reduction of larval transport from upstream may lead to the formation of additional retention zones within a species' geographic range that harbor additional diversity [##UREF##6##12##,##UREF##7##13##]. As descendants of adults in the retentive populations drift downstream in large species domains, they may acquire additional allelic diversity through mutation. It will be argued below that for most realistic population and dispersal parameters this latter effect is small.</p>" ]

[ "<title>Methods &amp; results</title>", "<p>To test these ideas, we use a simple numerical model, similar to those used by Pringle and colleagues [##UREF##6##12##,##UREF##7##13##]. In our model, a haploid semelparous individual produces a fixed number of larvae that disperse on average a distance L_advdownstream, with standard deviation of L_diffwith a Gaussian dispersal kernel [##UREF##14##27##] or with Laplace's distribution. Density dependence exists because if more than one propagule recruits to the same location, one is randomly chosen to survive. The model domain is finite, and any propagules that leave the domain die. The results shown below are computed in a model with a low population carrying capacity per unit length (of order 1 individual/km), due to limited computational resources. However, the validity of these results is not sensitive to the magnitude of this parameter.</p>", "<p>To illustrate how advection reduces genetic diversity in a population, two domains are initialized with five different alleles each in different parts of the domain in the numerical model (figure ##FIG##0##1##). In one domain, mean transport of larvae L_advis zero; in the other it is 4 km/generation to the right. In both, the stochastic component of larval transport L_diffis 10 km/generation. In the case with no mean larval transport, all genetic diversity is retained. However, when there is mean larval transport, only the upstream allele persists and the other alleles are lost downstream, for only the upstream allele begins in the retentive region that lies within L_reten= 25 km of the upstream edge of the domain.</p>", "<p>A second numerical experiment illustrates how the presence of directional larval dispersal changes the spatial structure of the population system. In these model runs, L_diffis fixed to 100 km and L_advis varied from 0 to 116 km. There is a mutation rate μ = 10^-3such that larvae randomly carry a new allele with this frequency (a smaller, more realistic μ does not change the results, but dramatically increases computation time). In these model runs, N_alleleis uniform in the interior and small near the edges when L_advis zero, but as L_advincreases, N_allelebecomes largest near the upstream edge of the species domain, and is one in the interior of the model domain (figure ##FIG##1##2A##). Examining N_alleledivided by the value that just satisfies eq. (1) (figure ##FIG##1##2B##), we find that N_allelejust satisfies eq. (1) in the retention zone that lies within L_retenfrom the upstream edge of the domain, and does not elsewhere. In the model, time to fixation or extinction of all novel alleles is tracked as a function of their origin. As discussed above, enhanced reproductive success within a distance L_retenfrom the upstream edge allows novel alleles to persist longer in the upstream retentive population, for a time appropriate to N_eas given by Eq. (1), while those in downstream regions are lost much more quickly (figure ##FIG##1##2C##). The region in which novel alleles are retained, and the density of these upstream regions, decreases in size as L_advincreases (figure ##FIG##1##2D##), as predicted in the expression for L_reten.</p>", "<p>To determine N_eas a function of L_advand L_diff, we calculate the inbreeding effective population size N_e[##REF##15489538##28##] given the mean lifetime of a novel allele in the system. We initialize the model with two alleles, each randomly distributed and each comprising 50% of the population. The neutral time to fixation in such a model will be 2.7 N_e[##REF##17248440##21##], and so we estimate N_efrom the average fixation time of 100 model runs. In Figure ##FIG##2##3##, we run the model in three domains of sizes L_domain= 10³, 4 × 10³, and 1.6 × 10⁴km. In each domain, we fix L_diffto 200 km, and vary L_advfrom 0 to 110km, and compare the estimated N_efrom (2) to the estimation from fixation time in an upstream region of the model L_retenin size. Once there is fixation in this upstream region, the allele fixes rapidly in the rest of the species domain in approximately L_domain/L_advgenerations. When the size of the domain is less than L_retenin extent, N_eis limited to the population census size (figure ##FIG##2##3##). Thus when L_advis small, N_eis nearly equal to the census population of the entire population, though somewhat smaller due to loss of larvae from the edges of the domain caused by stochastic larval transport. When the domain size is greater than L_reten, eq. (2) captures the variability of N_ewith L_advvery well, capturing the several order of magnitude decline in N_ewith increasing L_adv. As mentioned above, the estimate of N_efrom (2) is, for most values of L_adv, very much smaller than – and not dependent upon – the census population size. When the model is re-run with Laplace's dispersal kernel, the results shown in Figure ##FIG##2##3## remain unchanged (not shown), suggesting that these results are not very sensitive to the kurtosis of the dispersal kernel.</p>" ]

[ "<title>Methods &amp; results</title>", "<p>To test these ideas, we use a simple numerical model, similar to those used by Pringle and colleagues [##UREF##6##12##,##UREF##7##13##]. In our model, a haploid semelparous individual produces a fixed number of larvae that disperse on average a distance L_advdownstream, with standard deviation of L_diffwith a Gaussian dispersal kernel [##UREF##14##27##] or with Laplace's distribution. Density dependence exists because if more than one propagule recruits to the same location, one is randomly chosen to survive. The model domain is finite, and any propagules that leave the domain die. The results shown below are computed in a model with a low population carrying capacity per unit length (of order 1 individual/km), due to limited computational resources. However, the validity of these results is not sensitive to the magnitude of this parameter.</p>", "<p>To illustrate how advection reduces genetic diversity in a population, two domains are initialized with five different alleles each in different parts of the domain in the numerical model (figure ##FIG##0##1##). In one domain, mean transport of larvae L_advis zero; in the other it is 4 km/generation to the right. In both, the stochastic component of larval transport L_diffis 10 km/generation. In the case with no mean larval transport, all genetic diversity is retained. However, when there is mean larval transport, only the upstream allele persists and the other alleles are lost downstream, for only the upstream allele begins in the retentive region that lies within L_reten= 25 km of the upstream edge of the domain.</p>", "<p>A second numerical experiment illustrates how the presence of directional larval dispersal changes the spatial structure of the population system. In these model runs, L_diffis fixed to 100 km and L_advis varied from 0 to 116 km. There is a mutation rate μ = 10^-3such that larvae randomly carry a new allele with this frequency (a smaller, more realistic μ does not change the results, but dramatically increases computation time). In these model runs, N_alleleis uniform in the interior and small near the edges when L_advis zero, but as L_advincreases, N_allelebecomes largest near the upstream edge of the species domain, and is one in the interior of the model domain (figure ##FIG##1##2A##). Examining N_alleledivided by the value that just satisfies eq. (1) (figure ##FIG##1##2B##), we find that N_allelejust satisfies eq. (1) in the retention zone that lies within L_retenfrom the upstream edge of the domain, and does not elsewhere. In the model, time to fixation or extinction of all novel alleles is tracked as a function of their origin. As discussed above, enhanced reproductive success within a distance L_retenfrom the upstream edge allows novel alleles to persist longer in the upstream retentive population, for a time appropriate to N_eas given by Eq. (1), while those in downstream regions are lost much more quickly (figure ##FIG##1##2C##). The region in which novel alleles are retained, and the density of these upstream regions, decreases in size as L_advincreases (figure ##FIG##1##2D##), as predicted in the expression for L_reten.</p>", "<p>To determine N_eas a function of L_advand L_diff, we calculate the inbreeding effective population size N_e[##REF##15489538##28##] given the mean lifetime of a novel allele in the system. We initialize the model with two alleles, each randomly distributed and each comprising 50% of the population. The neutral time to fixation in such a model will be 2.7 N_e[##REF##17248440##21##], and so we estimate N_efrom the average fixation time of 100 model runs. In Figure ##FIG##2##3##, we run the model in three domains of sizes L_domain= 10³, 4 × 10³, and 1.6 × 10⁴km. In each domain, we fix L_diffto 200 km, and vary L_advfrom 0 to 110km, and compare the estimated N_efrom (2) to the estimation from fixation time in an upstream region of the model L_retenin size. Once there is fixation in this upstream region, the allele fixes rapidly in the rest of the species domain in approximately L_domain/L_advgenerations. When the size of the domain is less than L_retenin extent, N_eis limited to the population census size (figure ##FIG##2##3##). Thus when L_advis small, N_eis nearly equal to the census population of the entire population, though somewhat smaller due to loss of larvae from the edges of the domain caused by stochastic larval transport. When the domain size is greater than L_reten, eq. (2) captures the variability of N_ewith L_advvery well, capturing the several order of magnitude decline in N_ewith increasing L_adv. As mentioned above, the estimate of N_efrom (2) is, for most values of L_adv, very much smaller than – and not dependent upon – the census population size. When the model is re-run with Laplace's dispersal kernel, the results shown in Figure ##FIG##2##3## remain unchanged (not shown), suggesting that these results are not very sensitive to the kurtosis of the dispersal kernel.</p>" ]

[ "<title>Discussion</title>", "<p>The concept of \"effective population size\" is typically used as a numerical trait of a population more than as a descriptor of biological reality [##UREF##13##26##]. But N_eis intended to reflect the number of individuals that contribute to the evolutionary potential of a species [##REF##17247074##29##]; in advective environments, we have shown that this contribution is driven mostly by a small upstream portion of a species geographic range. Given selective neutrality of the genes being studied in a standard population genetics survey, it is generally assumed that the number of successful offspring an individual will have is independent of its genotypic state or geographic location [##UREF##1##3##]. Other results have suggested that for some species, habitat may be structured by varying quality that will determine the distribution of offspring numbers [##REF##12454077##6##], <italic>i.e. </italic>a \"nest-site\" model [##UREF##1##3##]. In these cases, however, it is typically assumed that location itself (and thus the potential for lower variance in reproductive success) is not heritable. In the case of populations under mean advection, however, the upstream retention zone will represent a heritable component of reproductive success for any larvae spawned there because a small fraction of these larvae retain their parents' geographic reproductive advantage.</p>", "<p>The attention given to estimating N_ein natural populations has recently been focused on a number of demographic causes for reduced N_e/N ratios [##REF##12454077##6##,##UREF##13##26##,##REF##12206244##30##]. Here we show a significant environmental interaction that can strongly affect diversity in continuously distributed species. While overall allelic diversity in the species' domain will likely include a large number of potentially transient alleles that form in downstream regions at a rate 2 N_eμ (or N_eμ for haploid markers), in an advective environment most of these are more quickly lost due to advection (a time in generations of about L_domain/L_adv) than to stochastic genetic drift (Figure ##FIG##1##2##). The retention of diversity is thus going to be related to the size of demographically stable and retentive populations such as the upstream edge of the domain, a size that is specific to a species' reproductive and larval dispersal traits. As L_advincreases, this source region becomes smaller, and with uniform density of individuals reduces N_econcomitantly.</p>", "<p>Thus, different species with distinct larval dispersal traits can have distinct N_e/N ratios in the same region, all else being equal. This mechanism does not hinge on the reproductive \"sweepstakes\" between individuals at the same location – instead, it is an effect of the differential reproductive success of individuals from different regions, and the effects of mean larval transport. Mean larval transport, L_adv, can change from generation to generation [##UREF##7##13##], and therefore the size of the retention region can vary from generation to generation. This will produce a fluctuating N_efrom generation to generation, and years of especially strong mean advection could reduce net diversity. Thus N_ecan be reduced not only by year-to-year variation in reproductive success, but also by inter-annual changes in the physical environment that affect larval dispersal.</p>", "<p>A growing body of literature attempts to link patterns of genetic diversity with patterns of biodiversity, for the purposes of elucidating the mechanisms underlying broad-scale biogeographic structure and for conservation-focused predictions [##REF##15212375##31##,##REF##16032574##32##]. However, in strongly advective environments the link between population genetic structure and community structure may be tenuous because what appears to be panmixia – extensive, range-wide gene flow – may instead represent an extended source-sink metacommunity [##UREF##13##26##]. Our predictions suggest that in populations whose dispersal is subject to strong advection (<italic>e.g.</italic>, high dispersal from the upstream populations to downstream) N_ewill be more disassociated from actual census size than for populations less affected by advection.</p>", "<p>To test this prediction, one might imagine comparing species with very different dispersal strategies, or comparing the same or similar species in two locations with different dispersal conditions. However, while classical population genetics predicts elevated diversity <italic>across </italic>populations of low-dispersal species, recent isotropic descriptions of metapopulation structure [##UREF##13##26##] show that due to unequal contribution of some populations to subsequent generations, limited dispersal can actually reduce N_e. Metapopulation structure in general may bias the measurement of N_eand gene flow measures [##REF##12586728##33##], and Foltz [##REF##14745529##34##] suggests a strong role of purifying selection and/or local adaptation in limiting diversity in dispersal-limited species. Altogether, these factors may confound comparisons of Ne/N ratios among distantly related taxa; thus it may be more appropriate to make intraspecific or sister-taxon comparisons, in which closely-related equilibrium (e.g. not introduced or range-expanded) populations exist on a \"strong-advection\" coast and a \"weak-advection\" coast. Such a comparison can be made for species with pelagic larvae that are distributed on both the Pacific coast of the Baja California peninsula (strong advection) and in the semi-enclosed Gulf of California (weak advection). Our work would predict higher diversity in the weak advection Gulf of California region; given six appropriate studies, four support this hypothesis [##REF##16262860##35##, ####REF##15723663##36##, ##REF##11681740##37##, ##REF##10937240##38####10937240##38##], one is equivocal but at least some population samples in the Gulf have higher diversity [##REF##11246411##39##], and in one case the Gulf sister species is apparently less diverse [##REF##11161747##40##].</p>" ]

[ "<title>Conclusion</title>", "<p>Overall, some of the lowest N_e/N ratios observed are for species with broad dispersal potential in regions where ocean currents would be expected to generate a large L_adv[##REF##12454077##6##,##UREF##15##41##]. While a species' fecundity may be associated with dispersal potential in marine organisms [##UREF##16##42##], fecundity alone is not predictive of Ne/N [##UREF##17##43##] – suggesting that other factors, including advection, may be involved in the relationship between N_eand actual census size. In the end, any mechanism that increases the variance in reproductive success among individuals, whether due to stochastic, biological, or spatial processes, will reduce genetic variation in a species [##REF##15489538##28##]. Here we argue that one of many processes that must be considered when analyzing the genetic diversity of coastal species is the interaction between a species and its dispersal. Dispersal must be considered in the persistent discussion of N_e/N ratios (<italic>e.g.</italic>, [##REF##16645093##11##]) if there is to be a better empirical understanding of how variation is maintained in natural populations, and for management and conservation questions [##UREF##18##44##]. Our work explicitly defines the size of domain that will be evolutionarily important and is relevant to marine reserve design, in that if a reserve is larger than this region of size L_retenit will also be protecting 'sink' regions; if it is smaller than this length, reserve size will be a limiting factor on total genetic diversity.</p>" ]

[ "<title>Background</title>", "<p>Genetic estimates of effective population size often generate surprising results, including dramatically low ratios of effective population size to census size. This is particularly true for many marine species, and this effect has been associated with hypotheses of \"sweepstakes\" reproduction and selective hitchhiking.</p>", "<title>Results</title>", "<p>Here we show that in advective environments such as oceans and rivers, the mean asymmetric transport of passively dispersed reproductive propagules will act to limit the effective population size in species with a drifting developmental stage. As advection increases, effective population size becomes decoupled from census size as the persistence of novel genetic lineages is restricted to those that arise in a small upstream portion of the species domain.</p>", "<title>Conclusion</title>", "<p>This result leads to predictions about the maintenance of diversity in advective systems, and complements the \"sweepstakes\" hypothesis and other hypotheses proposed to explain cases of low allelic diversity in species with high fecundity. We describe the spatial extent of the species domain in which novel allelic diversity will be retained, thus determining how large an appropriately placed marine reserve must be to allow the persistence of endemic allelic diversity.</p>" ]

[ "<title>Authors' contributions</title>", "<p>This work was developed as equal-authorship collaboration between JPW, who conceived of the study and drafted the manuscript, and JMP who developed the numerical simulations and scalings and coordinated the results, and helped to draft the manuscript. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>Many thanks to John Wakeley, Scott Small, Mike Hickerson, Robin Waples, Bob Holt, and two anonymous reviewers for comments and discussions during the writing of this manuscript. This work was supported by University of Georgia Research Foundation funds to JPW and NSF grant OCE-0453792 to JMP.</p>" ]

[ "<fig id=\"F1\" position=\"float\"><label>Figure 1</label><caption><p><bold>In both (A) and (B), the domain is initialized with haploid adults containing 5 different alleles, each geographically isolated to 1/5 of the domain, and each adult colored according to its allelic composition.</bold> The model is run for 400 generations. In (A), L_adv= 0 km and L_diff= 10 km. The allelic composition diffuses isotropically away from initial positions, and no allele is favored over others. In (B), L_adv= 4, so larvae preferentially disperse towards positive x (to the right) and the upstream allele quickly dominates the entire domain. L_retenin (B) is 25 km.</p></caption></fig>", "<fig id=\"F2\" position=\"float\"><label>Figure 2</label><caption><p><bold>(A) N_allelefor a 1-dimensional ocean with a mean current from left to right as a function of the alongshore distance and the mean larval transport distance, L_adv.</bold> The heavy black line represents the width of the retention zone L_reten= L²_diff/L_advfrom the upstream edge of the domain. (B) N_allelenormalized by the critical value of N_alleleneeded to allow retention, as given by eq. (1). (C) The logarithm of average persistence time of a novel allele in generations as a function of the location where the allele first appeared. (D) The population density per length of the domain, normalized by the carrying capacity. The stochastic component of larval transport, L_diff, is 100 km, the carrying capacity of the domain is 1 individual/km, and the domain is 2048 km in size.</p></caption></fig>", "<fig id=\"F3\" position=\"float\"><label>Figure 3</label><caption><p><bold>(Thick Black Line) Estimate of N_efrom equation (2).</bold> (Dashed Thin Lines) Census population in domain. The squares (□) are for a domain 1024 km in size, the circles (○) are for a domain 4096 km in size, and the diamonds (◇) for a domain 16384 km in size. (Solid Lines) Estimates of N_efrom a numerical model with varying L_advand a constant L_diffof 200 km. For the purposes of simulation, the carrying capacity of the domain is about 0.5 individuals per kilometer.</p></caption></fig>" ]

[]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label>log(N_allele) &gt; L²_adv/(2L²_diff)</disp-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label>N_e= H_densL²_diff/L_adv,</disp-formula>" ]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2148-8-235-1\"/>", "<graphic xlink:href=\"1471-2148-8-235-2\"/>", "<graphic xlink:href=\"1471-2148-8-235-3\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The number of solved RNA secondary structures has increased dramatically in the past decade, and several databases are available to search and download specific classes of RNA secondary structures [##REF##11869452##1##, ####REF##15608164##2##, ##REF##9847214##3##, ##REF##16381838##4##, ##REF##15608160##5####15608160##5##]. However, for purposes such as improving RNA energy models [##REF##17646296##6##,##REF##16873527##7##], evaluating RNA secondary structure prediction software, obtaining distributions of naturally occuring structural features, or searching RNA molecules with specific motifs, researchers need to easily access a much larger set of known RNA secondary structures, ideally all known RNA secondary structures. RNA STRAND aims to provide this capability, in addition to easy search, analysis and download features. Figure ##FIG##0##1## shows an example of an RNA secondary structure and highlights some of its structural features.</p>", "<p>Previous RNA databases provide secondary structure information, but are specialised in a different direction or follow different goals. The Rfam Database [##REF##15608160##5##] contains a large collection of non-coding RNA families; however, many of the corresponding secondary structures are computationally predicted. The Comparative RNA Web Site [##REF##11869452##1##] specialises in ribosomal RNA and intron RNA molecules. The Sprinzl tRNA database [##REF##15608164##2##] specialises in tRNA molecules, the RNase P database [##REF##9847214##3##] specialises in RNase P RNA molecules, and the SRP and tmRNA databases [##REF##16381838##4##] specialise in SRP RNA and tmRNA molecules, respectively. Pseudobase [##REF##11125088##8##] contains short RNA fragments that have pseudoknots. The RAG (RNA-As-Graphs) Database [##REF##14962931##9##] classifies and analyses RNA secondary structures according to their topological characteristics based on the description of RNAs as graphs, but its collection of structures is very limited.</p>", "<p>A number of previous databases contain three-dimensional (3D) RNA structures; however, as opposed to proteins, the number of solved RNA 3D structures is much smaller than the number of solved RNA secondary structures. (Only 18% of all RNA molecules we collected have known 3D structures.) As such, all these databases do not include molecules whose secondary structures are known but 3D structures are unknown; examples include: the RCSB Protein Data Bank [##REF##12520059##10##], the Nucleic Acids Database [##REF##1384741##11##], the RNA Structure Database [##REF##12520063##12##] and the Structural Classification of RNA (SCOR) database [##REF##14681389##13##]. NCIR [##REF##11752347##14##] contains non-canonical base pairs in 3D RNA molecules. FR3D [##REF##17694311##15##] provides a collection of 3D RNA structural motifs found in the RCSB Protein Data Bank. Finally, there are other RNA databases that provide RNA sequences, but no experimental structural information, such as the SubViral RNA Database [##REF##16519798##16##], which contains a collection of over 2600 sequences of viroids, the hepatitis delta virus and satellite RNAs, but only mfold-predicted secondary structures.</p>", "<p>RNA STRAND spans a more comprehensive range of RNA secondary structures than do previous databases. It currently provides highly accurate secondary structures for 4666 RNA molecules. Since some users of RNA STRAND will likely develop new thermodynamic models, prediction tools or statistical analyses, our data is exclusively determined by carefully conducted comparative sequence analysis [##REF##11869452##1##], or by experimental methods such as NMR or X-ray crystallography [##REF##12520059##10##]. All information has been obtained from publicly available RNA databases. Our goal in creating this database is to provide comprehensive information on structural features – such as types and sizes for stems and loops, pseudoknot complexity and base pair types – that can be interactively analysed or downloaded within and across functional classes of molecules. Such information could be used, for example, to understand what type of structural motifs are common in a specific set of RNA molecules; to estimate the accuracy of RNA secondary structure computational prediction methods; or to improve current thermodynamic models for RNA secondary structure prediction.</p>" ]

[]

[]

[ "<title>Utility and discussion</title>", "<p>RNA STRAND v2.0 contains 4666 RNA molecules or interacting complexes of various types, and an abundance of RNA structural motifs (see also Table ##TAB##0##1##). This represents a considerable amount of data from which to draw significant statistics and trends about RNA secondary structures. from which to draw significant statistics and trends about RNA secondary structures. In what follows we illustrate how the information in RNA STRAND can be used for various purposes.</p>" ]

[ "<title>Conclusion</title>", "<p>In this paper, we presented RNA STRAND, a new database for RNA secondary structure data that provides access to detailed information on known secondary structures as well as statistical analyses of structural aspects of various types of RNAs. We believe that such information will be useful in the context of understanding RNA structure and function; in particular, we expect it to further facilitate the development and evaluation of energy models for secondary structure prediction. Our database is flexible and extensible; it provides a convenient web interface to its major functions and supports searches according to many criteria, including properties of secondary structure elements. The database is publicly accessible and supports the submission of new RNA structures by the research community. We are committed to keeping RNA STRAND up-to-date with new structures that are added to the eight databases of provenance, and we invite submissions of all types of RNA secondary structures, which will help to further expand the database and increase its usefulness.</p>", "<p>In the future, we intend to add RNA secondary structures obtained from the SHAPE technique [##REF##15783204##40##,##REF##17406453##41##], and also to provide further search options such as searches by specific structural motifs.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>The ability to access, search and analyse secondary structures of a large set of known RNA molecules is very important for deriving improved RNA energy models, for evaluating computational predictions of RNA secondary structures and for a better understanding of RNA folding. Currently there is no database that can easily provide these capabilities for almost all RNA molecules with known secondary structures.</p>", "<title>Results</title>", "<p>In this paper we describe RNA STRAND – the RNA secondary STRucture and statistical ANalysis Database, a curated database containing known secondary structures of any type and organism. Our new database provides a wide collection of known RNA secondary structures drawn from public databases, searchable and downloadable in a common format. Comprehensive statistical information on the secondary structures in our database is provided using the RNA Secondary Structure Analyser, a new tool we have developed to analyse RNA secondary structures. The information thus obtained is valuable for understanding to which extent and with which probability certain structural motifs can appear. We outline several ways in which the data provided in RNA STRAND can facilitate research on RNA structure, including the improvement of RNA energy models and evaluation of secondary structure prediction programs. In order to keep up-to-date with new RNA secondary structure experiments, we offer the necessary tools to add solved RNA secondary structures to our database and invite researchers to contribute to RNA STRAND.</p>", "<title>Conclusion</title>", "<p>RNA STRAND is a carefully assembled database of trusted RNA secondary structures, with easy on-line tools for searching, analyzing and downloading user selected entries, and is publicly available at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.rnasoft.ca/strand\"/>.</p>" ]

[ "<title>Construction and Content</title>", "<p>Figure ##FIG##1##2## describes the four main modules that comprise RNA STRAND. To create the database, we first collected the data from various external sources, then we processed the data and prepared it for a MySQL relational database. Next, we installed and populated the database, and finally we prepared dynamic web pages that interact with the database. In what follows we describe in detail the construction and content of each module.</p>", "<title>External sources</title>", "<p>The current release v2.0 of RNA STRAND contains a total of 4666 entries (RNA sequences and secondary structures) of the following provenance:</p>", "<p>• RCSB Protein Data Bank (PDB) [##REF##12520059##10##]: 1059 entries, obtained from three dimensional NMR and X-ray atomic structures containing RNA molecules only, or RNA molecules and proteins (only the RNAs were included in RNA STRAND), in PDB format. These include ribozymes, ribosomal RNAs, transfer RNAs, synthetic structures, and complexes containing more than one RNA molecule. Out of the 1059 entries, 575 contain at least two RNA molecules; these are easily searchable from the RNA STRAND web site. The RNA secondary structures were generated from the tertiary structures using RNAView [##REF##12824344##17##], which is also used for secondary structure visualisation in the Nucleic Acid Database [##REF##1384741##11##].</p>", "<p>• Comparative RNA Web Site, version 2 [##REF##11869452##1##]: 1056 entries of ribosomal and intronic RNA molecules obtained by covariance-based comparative sequence analysis.</p>", "<p>• tmRNA database [##REF##16381838##4##]: 726 entries of transfer messenger RNA sequences and secondary structures determined by comparative sequence analysis.</p>", "<p>• Sprinzl tRNA Database (September 2007 edition) [##REF##15608164##2##]: 622 transfer RNA sequences and secondary structures obtained by comparative sequence analysis from the tRNA sequences data set. The genomic tRNA and tRNA gene sets from the Sprinzl tRNA database contain genomic sequences, and thus we think they are not as relevant for understanding function and folding of functional RNA molecules.</p>", "<p>• RNase P Database [##REF##9847214##3##]: 454 Ribonuclease P RNA sequences and secondary structures obtained by comparative sequence analysis.</p>", "<p>• SRP Database [##REF##16381838##4##]: 383 entries of Signal Recognition Particle RNA sequences and secondary structures determined by comparative sequence analysis.</p>", "<p>• Rfam Database, version 8.1 [##REF##15608160##5##]: 313 entries from 19 Rfam families, including hammerhead ribozymes, telomerase RNAs, RNase MRP RNAs and RNase E 5' UTR elements (only the seeds have been used). Of the 607 Rfam families in version 8.1, 172 have the secondary structure flag \"published\", while the remaining 435 families have been predicted using Pfold [##REF##15608160##5##]. For several reasons, we decided to include only 19 of the 172 \"published\" families: (1) some of these families come from other databases that we have included directly, such as structures from the RNase P Database or SRP Database; (2) most of the secondary structures are actually predicted computationally and then published in the papers cited by Rfam, such as families RF00013, RF00035, RF00161 or RF00625. Since the Rfam database provides only very limited information about the reliability of the structures it contains, we have studied all 172 families and decided which families to include based on the cited papers. The details regarding the decision for each family are described in Supplementary Material 1, accessible from the main page of the RNA STRAND web site.</p>", "<p>• Nucleic Acid Database (NDB) [##REF##1384741##11##]: 53 entries which occur in NDB and not in PDB (note that NDB and PDB have a large overlap of RNA structures); these include transfer RNAs and synthetic RNAs obtained by X-ray crystallography.</p>", "<p>Table ##TAB##0##1## provides some additional information on these RNAs; information and statistics on the current database contents are also available from the main page of the RNA STRAND site.</p>", "<p>In the future, we intend to regularly check the aforementioned databases for new entries. With our current tools, keeping the database up-to-date will be relatively easy.</p>", "<title>Data processing</title>", "<title>Unique IDs</title>", "<p>We created a unique and stable identifier for each entry in the RNA STRAND database. Future releases will keep all previous IDs unchanged.</p>", "<title>Conversion scripts</title>", "<p>One of the challenging tasks in collecting the RNA STRAND data arose from the fact that the external sources offer data in various formats. We have built tools to convert from all these formats to the CT format, which we use to store all structures internally, and to RNAML, BPSEQ, dot-parentheses and FASTA formats when requested by a user. The format descriptions are accessible on the \"Help\" page online.</p>", "<title>Validation</title>", "<p>All external databases we have used in the current version of RNA STRAND, except Rfam, contain highly curated RNA secondary or tertiary structures, therefore we trust the curation methods of these sources. For Rfam we selected a set of reliable structures based on the cited papers, as described in the previous section. Once we converted all the secondary structure external files into the CT format, we checked all files in order to make sure the secondary structures are valid (i.e., one base is paired with at most one other base, and if base at position <italic>i </italic>is paired with base at position <italic>j</italic>, then base at position <italic>j </italic>is also paired with base at position <italic>i</italic>.). When performed on our present data, this validation step revealed several inconsistencies in some of the external files, which we brought to the attention of the respective database owners.</p>", "<title>RNA Secondary Structure Analyser</title>", "<p>The structural statistics that form the core part of RNA STRAND were generated using the RNA Secondary Structure Analyser, which takes as input an RNA secondary structure description, for example in CT format, and outputs a wide range of secondary structure information. While many of these features, such as the number and composition of stems, are rather straightforward to determine, in some cases, more advanced algorithmic techniques have to be applied – as is the case, for example, for the minimal number of base pairs that need to be removed to render a structure pseudoknot free. For this specific task, we implemented a dynamic programming algorithm that removes the minimum number of base pairs [##UREF##0##18##]; however, more sophisticated approaches could be used, such as those recently described by Smit et al. [##REF##18230758##19##]. The complete output of the analyser run for each individual database entry can be accessed easily from the RNA STRAND web interface, and a description of the output can be found in the online Supplementary Material 2.</p>", "<title>MySQL database</title>", "<p>All the data obtained from the RNA Secondary Structure Analyser were inserted into a relational database implemented in MySQL (version 5.0.26). The main table is MOLECULE, with one row per RNA entry in the database. This table contains as primary key the unique RNA STRAND ID of the entry and further comprises various descriptive fields, including: organism, reference, length, RNA type, external source, external ID, sequence, three levels of abstract shapes using the RNAshapes representation [##REF##16357029##20##], the method of secondary structure determination, and a link to the respective CT file. (Since RNAshapes version 2.1.5 cannot obtain the abstract shape of pseudoknotted secondary structures, we first removed a minimum number of base pairs to render the structure pseudoknot-free.) Furthermore, there is one table per secondary structure feature, where the table MOLECULE is connected to each of these tables in a one-to-many relationship. For example, the table STEM contains information such as the number of base pairs and the estimated free energy change for that stem, using parameters by Xia et al. [##REF##9778347##21##]. Accurately estimating the free energy change of entire structures is currently challenging, due to structural motifs for which current energy models are incomplete, such as pseudoknots, non-canonical base pairs, and modified nucleotides. Other similar tables include HAIRPIN_LOOP, MULTI_LOOP and PSEUDOKNOT.</p>", "<p>An additional table TMP_MOLECULE is used to temporarily hold new submissions received via the web interface; for these, we manually check the submission information by checking the cited paper, after which, if the submission is accepted, all further steps required to permanently add the respective RNA(s) to the database are performed automatically.</p>", "<title>Web interface</title>", "<p>The web interface to RNA STRAND has been created using a set of PHP scripts (version 5.1.2). The main functions of the web interface are searching, browsing, analysis, downloading and uploading.</p>", "<title>Searching and browsing</title>", "<p>The user specifies one or more search criteria in a web-based form. The general criteria include RNA type (e.g., 16S Ribosomal RNA), organism of origin (e.g., <italic>E. coli</italic>), external source (e.g., RCSB Protein Data Bank), length (in bases), the number of molecules in the complex, whether it is a fragment, a sequence pattern using the standard IUPAC nucleic acid codes, an abstract structure or fragment using the RNAshapes representation [##REF##16357029##20##] and whether or not to include non-redundant sequences.</p>", "<p>We define a set of entries to be non-redundant if their sequences are pairwise distinct. On a search page the user can request a non-redundant set that satisfies some search criteria. In this case, if two entries have identical RNA sequences, one of them will be selected arbitrarily. In the remainder of this paper, when we refer to a number of non-redundant entries matching some criteria, we mean a largest non-redundant set of entries satisfying the specified criteria. Currently there are 4104 non-redundant entries out of the 4666 entries in RNA STRAND v2.0.</p>", "<p>Advanced searches are supported based on 21 additional search criteria on secondary structure elements, such as selection of RNA molecules having at least one pseudoknot, or hairpin loops with a specific sequence – for example GNRA hairpin loops. The set of database entries that match all of the specified criteria simultaneously is returned in the form of a table.</p>", "<p>Using advanced search criteria, users can search for entries with various structural motifs. For example, when looking for a Y shape with an additional hairpin, one would search for entries that have exactly one multi-loop, three multi-loop branches, three hairpins, one molecule in the complex, and no pseudoknots. This search returns 31 entries, most of which are <italic>ciliate telomerase RNAs </italic>from Rfam. If pseudoknots are allowed, then <italic>vertebrate telomerase RNAs </italic>from Rfam are also included, yielding 36 search results. An equivalent pseudoknot-free search can be obtained by typing in the abstract shape [ [] [] ] [] (where matching brackets represent one interrupted or uninterrupted stem). Pseudoknots are currently not permitted in the abstract shape representation [##REF##16357029##20##].</p>", "<p>Support for inspecting large fractions of the database contents is provided via searches with no or very general criteria. For example, it is easy to obtain a list of all RNase P RNA structures contained in the database.</p>", "<p>Details on individual entries from the result list of any search can be displayed by clicking on an RNA STRAND ID link of the results table. This single entry display comprises general information about the entry, links to the original database entry for this molecule, a secondary structure diagram, details of its secondary structure elements and features, links to other RNA STRAND entries with the same sequence (i.e., redundant entries), links to the sequence and secondary structure specification in five formats (CT, RNAML, BPSEQ, dot-parentheses and FASTA), and a link to the complete output of the RNA Secondary Structure Analyser.</p>", "<title>Analysis</title>", "<p>In addition to the aforementioned analysis information for individual entries, RNA STRAND also provides histograms or cumulative distribution functions of various molecule characteristics (such as number of pseudoknots per molecule) or structural features (such as number of branches per multi-loop) for all structures in the database or for user-selected subsets, as obtained from the search page. In addition, correlations between various molecule characteristics and molecule length can be obtained. For an unbiased analysis, the user has the option of normalising the data by RNA type (such as tRNA), in which case for each particular RNA type, one data point is obtained by averaging over all the data for molecules of that type. Finally, the user can choose to remove the outliers of the distributions. We use a common definition, according to which a data point is an outlier if, and only if, it is smaller than <italic>Q</italic>₁- 1.3·(<italic>Q</italic>₃- <italic>Q</italic>₁) or greater than <italic>Q</italic>₁+ 1.3·(<italic>Q</italic>₃- <italic>Q</italic>₁), where <italic>Q</italic>₁and <italic>Q</italic>₃are the first and third quartiles, respectively. Such analyses may guide research pertaining to understanding structural features in naturally occuring RNA molecules, as we outline in the \"Utility and discussion\" section.</p>", "<title>Downloading</title>", "<p>The set of molecules selected via the search page can be downloaded in one of five supported formats: CT, RNAML, BPSEQ, dot-parentheses and FASTA. Thus, researchers can use specifically selected structures locally.</p>", "<title>Uploading</title>", "<p>RNA STRAND supports public submission of RNA secondary structures to the database via its web interface. The structure file can be in any of the four supported secondary structure formats (CT, RNAML, BPSEQ and dot-parentheses) or in the PDB tertiary structure format. Since RNA STRAND is a curated database, newly submitted structures are checked for accuracy and completeness by one of the database administrators before they are added to the database. New additions to the public databases that constitute our external sources will be added to RNA STRAND regularly. This is complemented by the public submission option, which is intended for submission of structures that do not yet belong to any of these databases.</p>", "<title>Obtaining statistics of naturally occuring RNA structural features</title>", "<p>We performed statistical analyses using the RNA STRAND web interface. Our first observation concerns the number and complexity of pseudoknots. According to the current data from RNA STRAND v2.0, pseudoknots occur rather commonly, especially in longer molecules: 74% of all (non-redundant) entries with 100 or more nucleotides contain pseudoknots. We compared the stem length (i.e., the number of base pairs in uninterrupted stems) with the minimal number of base pairs that need to be removed per pseudoknot to render the structure pseudoknot free (we denote this number by # PKBP; note that for over 95% of the pseudoknots, the bases counted in determining # PKBP form one uninterrupted stem; also, there is no overlap between the base pairs counted in determining the stem length and the base pairs counted in determining # PKBP). Table ##TAB##1##2## shows that when considering all RNA types in the database, the median, mean and standard deviation of the two measures, stem length and # PKBP, are very similar, even when we normalise by RNA type. (For normalised analysis, instead of using one data point per molecule or per structural feature, we use one data point for each RNA type, where this point is determined by averaging all data points for the respective class of RNAs. This way, the user can avoid biasing the analysis when there are substantially more structures for some RNA types than for others.) However, for 16S and 23S ribosomal RNA molecules the stem length tends to be significantly larger than # PKBP, whereas for transfer messenger RNA molecules in particular and ribonuclease P RNA molecules to some extent, # PKBP is larger than the stem length. This observation is interesting in the context of computational approaches for RNA secondary structure prediction which ignore pseudoknots [##REF##10329189##22##], add pseudoknots hierarchically in a second stage [##UREF##1##23##], or simultaneously add stems in pseudoknotted and non-pseudoknotted regions [##REF##16199760##24##,##REF##9925784##25##].</p>", "<p>Our second observation concerns the abundance of non-canonical base pairs and the pairing type of their immediate neighbours. (We consider all C-G, A-U and G-U pairs to be <italic>canonical base pairs</italic>, and all other base pairs to be <italic>non-canonical</italic>.) Figure ##FIG##2##3## shows a histogram for the 729 non-redundant entries whose structures were determined by all-atom methods (these include structures from the Protein Data Bank and the Nucleic Acid Database). For this data set, non-canonical A-G base pairs are the most abundant, representing 55% of all non-canonical base pairs, and G-G pairs are the least abundant, representing only 4% of all non-canonical base pairs. The plot also shows that a relatively small fraction of non-canonical base pairs have as immediate neighbours canonical base pairs. Interestingly, for all seven types of non-canonical base pairs, more pairs are adjacent to at least one other non-canonical base pair than surrounded by two canonical base pairs. For example, 55% of all A-A pairs are adjacent to at least one other non-canonical base pair. This may suggest that non-canonical base pairs are sufficiently stable energetically to form several consecutive base pairs.</p>", "<p>Finally, we found rather strong linear correlations between the number of nucleotides of the RNAs in our database and the number of stems, hairpin loops, bulges, internal loops and multi-loops; the Pearson correlation coefficients are <italic>r </italic>= 0.95, 0.95, 0.92, 0.91 and 0.92, respectively. This is consistent with the idea that the local formation of these secondary structure elements is relatively independent of the overall size of the molecule and in agreement with the current thermodynamic energy models of RNA secondary structure, which assume additive and independent energy contributions for these structural elements. Interestingly, the correlation between the RNA length and the number of pseudoknots is significantly weaker (<italic>r </italic>= 0.64), suggesting that pseudoknots may not follow the same linearity principle.</p>", "<title>Evaluating energy-based secondary structure prediction programs</title>", "<p>The RNA STRAND database can be used to evaluate the prediction accuracy of energy-based RNA secondary structure prediction software. RNA STRAND v2.0 contains 3704 non-redundant entries containing one molecule that can be used to evaluate software such as CONTRAfold [##REF##16873527##7##] or mfold [##REF##12824337##26##], 403 non-redundant entries containing complexes of two or more molecules that can be used to evaluate sofware for interacting molecules [##REF##15644199##27##,##UREF##2##28##], and 1957 non-redundant single-molecules with pseudoknots that can be used to evaluate secondary structure prediction programs with pseudoknots [##UREF##1##23##, ####REF##16199760##24##, ##REF##9925784##25####9925784##25##,##REF##12926009##29##].</p>", "<p>We have selected 2518 structures out of the 3704 non-redundant entries containing one molecule, after we eliminated the entries with unknown nucleotides and overly large loops. (Specifically, entries having hairpin loops, bulges, internal loops or multi-loops with more than 50, 50, 50 and 100 unpaired bases, respectively, were removed.) In addition, we have removed all non-canonical base pairs and the minimum number of base pairs needed to render the structures pseudoknot-free. The resulting structures are used as ground-truth reference structures. We evaluated the sensitivity and positive predictive value (PPV) of CONTRAfold [##REF##16873527##7##] and SimFold [##UREF##3##30##] with various free energy parameter sets, see Figure ##FIG##3##4##. Sensitivity is the number of correctly predicted base pairs divided by the number of base pairs in the reference structure, and PPV is the number of correctly predicted base pairs divided by the number of predicted base pairs. \"SimFold Turner99\" in Figure ##FIG##3##4## refers to SimFold using the free energy parameters described by Mathews et al. [##REF##10329189##22##], and is essentially equivalent to mfold 3.1 [##REF##12824337##26##]. On this large set, the average sensitivity of prediction is 0.63, while the average PPV is 0.57.</p>", "<p>\"CONTRAfold 1.1 151Rfam\" is the CONTRAfold software version 1.1, as reported by Do et al. [##REF##16873527##7##]. The CONTRAfold prediction program uses a trade-off parameter <italic>γ </italic>between sensitivity and PPV, and thus we report predictions for <italic>γ </italic>ranging from 2 to 20. When the target of one measure is fixed to the value obtained with \"SimFold Turner99\", the other is similar as well, showing that on this data set, CONTRAfold 1.1 gives similar average prediction accuracy as \"SimFold Turner99\". The remaining points of Figure ##FIG##3##4## are described in the following section.</p>", "<title>Improving RNA energy models</title>", "<p>More importantly, RNA STRAND can facilitate approaches for improving the free energy models underlying energy-based RNA secondary structure prediction software [##REF##17646296##6##,##REF##16873527##7##]. In this context, it can be very useful to exploit training data consisting of RNA sequences with known secondary structures, and the size and variety of such data are key for obtaining good results.</p>", "<p>Figure ##FIG##3##4## shows the average sensitivity and PPV of various programs measured on the 2518 structures mentioned in the previous section, and trained on various training sets.</p>", "<p>\"CONTRAfold 1.1 151Rfam\" was trained on a small set of 151 structures from various Rfam families [##REF##16873527##7##], while \"CONTRAfold 2.0 STRAND1.3\" was trained on 3427 pre-processed structures (i.e., split and restricted) of average length 178 nucleotides from version 1.3 of the RNA STRAND database, as used by Andronescu et al. [##REF##17646296##6##]. The figure shows that using the much larger set in the latter case gives an improvement of roughly 7% in prediction accuracy.</p>", "<p>To demonstrate even further the importance of using a large and comprehensive set of known RNA secondary structures for obtaining high-quality free energy parameters, we have used the current version of RNA STRAND v2.0 to obtain a new training set of 2246 structures of average length 246 nucleotides. Using the Maximum Likelihood parameter estimation method described by Andronescu et al. [##REF##17646296##6##], which is similar to CONTRAfold [##REF##16873527##7##], we have improved the average accuracy of prediction even further, as shown by the data point labelled \"SimFold STRAND2.0\" in Figure ##FIG##3##4##. This gives an improvement of 8% in average sensitivity and 10% in average PPV compared to the Turner99 parameters, when measured on our test set of 2518 structures. (Note that, since CONTRAfold and SimFold use different energy models and prediction algorithms, it is more appropriate to make comparisons between different versions of each, than it is to compare CONTRAfold versus SimFold).</p>", "<p>These results provide clear evidence for the key role of large and carefully assembled sets of RNA secondary structures, such as provided by RNA STRAND, in the context of determining RNA free energy models. In the future, we are planning to use the RNA STRAND data to train free energy parameters for pseudoknotted structures. Existing energy models for RNA secondary structure prediction methods with pseudoknots are often ad-hoc [##REF##9925784##25##,##REF##12926009##29##], and we believe that by using data-driven methods in conjunction with the 1957 non-redundant RNA STRAND entries representing RNAs with pseudoknots, it will be possible to obtain improved energy models for pseudoknotted structure prediction.</p>", "<title>Other uses of RNA STRAND</title>", "<p>The numerous search criteria supported by the RNA STRAND web interface allow users to select and study molecules with specific structural features. For example, Tyagi and Mathews [##REF##17507661##31##] studied the computational prediction accuracy of helical coaxial stacking in multi-loops. RNA STRAND v2.0 conveniently allows the selection and download of 189 non-redundant entries with all-atom structures that have at least one multi-loop. Other examples include the use of naturally occuring pseudoknotted structures that can be used to evaluate computational methods to render a pseudoknotted RNA secondary structure pseudoknot free [##REF##18230758##19##], or to evaluate RNA secondary structure visualisation tools [##REF##16845039##32##].</p>", "<p>In recent work on the role of RNA structure in splicing, Rogic et al. [##REF##18664289##33##] needed to identify thermodynamically stable stems that maximally shorten the distance between mRNA donor sites and branchpoint sequences. Since the optimal free energy of such stems is unknown, Rogic et al. wished to determine the most probable ranges of possible free energies for uninterrupted stems. By selecting all molecules on the RNA STRAND web site, they obtained distributions of estimated stem free energies (obtained with parameters by Xia et al. [##REF##9778347##21##]), which were used to support a new model for the role of RNA secondary stucture in mRNA splicing.</p>", "<p>In addition, RNA STRAND can facilitate the design of optical melting experiments [##REF##9778347##21##], whose goal is to better understand the thermodynamics of RNA structure formation, and to improve RNA secondary structure prediction accuracy. When designing optical melting experiments, usually a set of known RNA secondary structures is first assembled to determine what types of structural motifs that have not been previously studied appear frequently in naturally occuring RNAs [##REF##18020450##34##,##REF##17958380##35##]. The RNA STRAND web interface, as well as the abundance of reliable RNA structures in the RNA STRAND database, can be very useful in this context. For example, a significant number of multi-loops (16% in all non-redundant RNA STRAND entries) have five or more branches, but, to the best of our knowledge, optical melting experiments only exist for multi-loops with up to four branches [##REF##11389613##36##,##REF##11790109##37##]. Moreover, 30% of the internal loops in all non-redundant RNA STRAND entries have seven or more unpaired bases, and 13% have an absolute asymmetry (i.e., absolute difference between the number of unpaired bases on each side) of at least three, while only limited optical melting experiments exist to cover these cases [##REF##1711369##38##,##REF##16548530##39##].</p>", "<title>Availability and requirements</title>", "<p>RNA STRAND is publicly available at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.rnasoft.ca/strand\"/>. The RNA Secondary Structure Analyser, as well as the database tables, are available upon request from the authors.</p>", "<title>Authors' contributions</title>", "<p>MA collected the data, implemented conversion and validation scripts, implemented the MySQL database and part of the PHP scripts, performed the statistical analyses and helped to draft the manuscript. VB implemented the vast majority of the PHP scripts and most of the RNA Secondary Structure Analyser. HHH and AC conceived the project, specified the design of the RNA Secondary Structure Analyser, supervised MA's and VB's work, and helped to write the manuscript. All authors checked the accuracy of the database and web interface, read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We thank Baharak Rastegari, Yinglei S. Zhao, Mohammad Safari and Jack Jia, who provided an efficient computer program for pseudoknotted structure parsing; Robin Gutell, Christian Zwieb, James Brown and Mathias Sprinzl for providing data and help with the CRW, SRPDB &amp; tmRDB, RNase P DB and Sprinzl tRNA DB, respectively; Simon Moxon and Jennifer Daub for help using the Rfam database; Robert Giegerich and David Mathews for useful discussions; Alex Brown for help with the web interface; and Farheen Rawji for her work on an early version of the RNA Secondary Structure Analyser.</p>", "<p>Funding for this work was provided by: Mathematics of Information Technology and Complex Systems Network of Centres of Excellence (to AC and HH); Natural Sciences and Engineering Research Council of Canada Discovery Grant Program (238788 to HH and 217192 to AC).</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>RNA secondary structure example</bold>. Schematic representation of the secondary structure for the RNase P RNA molecule of <italic>Methanococcus marapaludis </italic>from the RNase P Database; the RNA STRAND ID for this molecule is ASE_00199. Solid grey lines represent the ribose-phosphate backbone. Dotted grey lines represent missing nucleotides. Solid circles mark base pairs. Dashed boxes mark structural features. We define an RNA secondary structure as a set of <italic>base pairs </italic>[##REF##10329189##22##]. In this work, we consider all C-G, A-U and G-U base pairs as <italic>canonical</italic>, and all other base pairs as <italic>non-canonical</italic>. However, we note that from the point of view of the planar edge-to-edge hydrogen bonding interaction [##REF##11345429##42##], there are C-G, A-U and G-U base pairs that do not interact via Watson-Crick edges, and vice-versa [##REF##11752347##14##,##REF##11345429##42##]. Comparative sequence analysis tools do not currently describe bond types. A number of structural motifs can be identified in a secondary structure: A <italic>stem </italic>is composed of one or more consecutive base pairs. A <italic>hairpin loop </italic>contains one closing base pair, and all the bases between the paired bases are unpaired. An <italic>internal loop </italic>is a loop with two closing base pairs, and all bases between them are unpaired. A <italic>bulge loop </italic>can be seen as a variant of an internal loop in which there are no unpaired bases on one side. A <italic>multi-loop </italic>is a loop which has at least three closing base pairs; stems emanating from these base pairs are called <italic>multi-loop branches</italic>. A <italic>pseudoknot </italic>is a structural motif that involves non-nested, crossing base pairs.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Database schema</bold>. Construction of RNA STRAND, from data collection to data presentation via dynamic web pages.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Histogram of the occurence of non-canonical base pairs</bold>. Histogram of non-canonical base pairs in the 729 non-redundant entries whose structures were determined by NMR or X-ray crystallography.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Prediction accuracy achieved by various energy models</bold>. Sensitivity vs. positive predictive value (PPV) of various secondary structure prediction methods. Sensitivity is the number of correctly predicted base pairs divided by the number of base pairs in the reference structure, PPV is the number of correctly predicted base pairs, divided by the number of predicted base pairs. Higher prediction accuracy is achieved when the free energy parameters are obtained by training on a larger set of structures. The CONTRAfold prediction program uses a trade-off parameter <italic>γ </italic>between sensitivity and PPV, and thus we report predictions for <italic>γ </italic>ranging from 2 to 20.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>The main RNA types included in RNA STRAND v2.0.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">RNA type</td><td align=\"left\">Main source(s)</td><td align=\"center\">#</td><td align=\"center\" colspan=\"2\">Length</td><td align=\"center\" colspan=\"2\">% PKBP</td></tr><tr><td/><td/><td align=\"center\">entries</td><td align=\"right\">mean</td><td align=\"left\">std</td><td align=\"right\">mean</td><td align=\"left\">std</td></tr></thead><tbody><tr><td align=\"left\">Transfer messenger RNA</td><td align=\"left\">tmRDB [##REF##16381838##4##]</td><td align=\"center\">726</td><td align=\"right\">368</td><td align=\"left\">86</td><td align=\"right\">21.0</td><td align=\"left\">6.1</td></tr><tr><td align=\"left\">16S ribosomal RNA</td><td align=\"left\">CRW [##REF##11869452##1##], PDB [##REF##12520059##10##]</td><td align=\"center\">723</td><td align=\"right\">1529</td><td align=\"left\">286</td><td align=\"right\">1.8</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\">Transfer RNA</td><td align=\"left\">Sprinzl DB [##REF##15608164##2##], PDB [##REF##12520059##10##]</td><td align=\"center\">707</td><td align=\"right\">76</td><td align=\"left\">21</td><td align=\"right\">0.1</td><td align=\"left\">2.3</td></tr><tr><td align=\"left\">Ribonuclease P RNA</td><td align=\"left\">RNase P DB [##REF##9847214##3##]</td><td align=\"center\">470</td><td align=\"right\">323</td><td align=\"left\">71</td><td align=\"right\">5.7</td><td align=\"left\">3.2</td></tr><tr><td align=\"left\">Signal rec. particle RNA</td><td align=\"left\">SRPDB [##REF##16381838##4##], PDB [##REF##12520059##10##]</td><td align=\"center\">394</td><td align=\"right\">220</td><td align=\"left\">111</td><td align=\"right\">0.0</td><td align=\"left\">0.0</td></tr><tr><td align=\"left\">23S ribosomal RNA</td><td align=\"left\">CRW [##REF##11869452##1##], PDB [##REF##12520059##10##]</td><td align=\"center\">205</td><td align=\"right\">2699</td><td align=\"left\">716</td><td align=\"right\">2.4</td><td align=\"left\">1.1</td></tr><tr><td align=\"left\">5S ribosomal RNA</td><td align=\"left\">CRW [##REF##11869452##1##], PDB [##REF##12520059##10##]</td><td align=\"center\">161</td><td align=\"right\">115</td><td align=\"left\">21</td><td align=\"right\">0.0</td><td align=\"left\">0.0</td></tr><tr><td align=\"left\">Group I intron</td><td align=\"left\">CRW [##REF##11869452##1##], PDB [##REF##12520059##10##]</td><td align=\"center\">152</td><td align=\"right\">563</td><td align=\"left\">412</td><td align=\"right\">5.8</td><td align=\"left\">2.2</td></tr><tr><td align=\"left\">Hammerhead ribozyme</td><td align=\"left\">Rfam [##REF##15608160##5##], PDB [##REF##12520059##10##]</td><td align=\"center\">146</td><td align=\"right\">61</td><td align=\"left\">24</td><td align=\"right\">0.0</td><td align=\"left\">0.0</td></tr><tr><td align=\"left\">Group II intron</td><td align=\"left\">CRW [##REF##11869452##1##], PDB [##REF##12520059##10##]</td><td align=\"center\">42</td><td align=\"right\">1298</td><td align=\"left\">829</td><td align=\"right\">1.4</td><td align=\"left\">3.5</td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"left\">All molecules</td><td align=\"left\">All of the above</td><td align=\"center\">4666</td><td align=\"right\">527</td><td align=\"left\">722</td><td align=\"right\">5.3</td><td align=\"left\">9.1</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Statistics on the complexity of pseudoknots in RNA STRAND v2.0.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">RNA type</td><td align=\"center\">#</td><td align=\"center\" colspan=\"3\">Stem length</td><td align=\"center\" colspan=\"3\"># PKBP</td></tr><tr><td colspan=\"2\"/><td colspan=\"3\"><hr/></td><td colspan=\"3\"><hr/></td></tr><tr><td/><td align=\"center\">entries</td><td align=\"center\">median</td><td align=\"center\">mean</td><td align=\"right\">std</td><td align=\"center\">median</td><td align=\"center\">mean</td><td align=\"right\">std</td></tr></thead><tbody><tr><td align=\"left\">16S ribosomal RNA</td><td align=\"center\">644</td><td align=\"center\">4.00</td><td align=\"center\">4.30</td><td align=\"right\">2.50</td><td align=\"center\">3.00</td><td align=\"center\">2.50</td><td align=\"right\">0.68</td></tr><tr><td align=\"left\">23S ribosomal RNA</td><td align=\"center\">93</td><td align=\"center\">4.00</td><td align=\"center\">4.14</td><td align=\"right\">2.39</td><td align=\"center\">2.00</td><td align=\"center\">3.75</td><td align=\"right\">3.12</td></tr><tr><td align=\"left\">Transfer messenger RNA</td><td align=\"center\">657</td><td align=\"center\">4.00</td><td align=\"center\">4.11</td><td align=\"right\">2.24</td><td align=\"center\">5.00</td><td align=\"center\">5.51</td><td align=\"right\">1.00</td></tr><tr><td align=\"left\">Ribonuclease P RNA</td><td align=\"center\">433</td><td align=\"center\">4.00</td><td align=\"center\">4.45</td><td align=\"right\">2.51</td><td align=\"center\">4.00</td><td align=\"center\">5.18</td><td align=\"right\">1.36</td></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"left\">All, non-redundant</td><td align=\"center\">4104</td><td align=\"center\">4.00</td><td align=\"center\">4.35</td><td align=\"right\">2.44</td><td align=\"center\">4.00</td><td align=\"center\">4.14</td><td align=\"right\">1.86</td></tr><tr><td align=\"left\">All, non-redundant &amp; normalised</td><td align=\"center\">4104</td><td align=\"center\">4.96</td><td align=\"center\">5.05</td><td align=\"right\">0.58</td><td align=\"center\">4.65</td><td align=\"center\">4.95</td><td align=\"right\">1.78</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Overview of the main RNA types in version 2.0 of the RNA STRAND database, their provenance, the number of RNAs, the mean length and standard deviation for each type. % PKBP denotes the percentage of the base pairs that need to be removed in order to render the structure pseudoknot-free. Most of the major RNA types are represented by a large number of molecules.</p></table-wrap-foot>", "<table-wrap-foot><p>The columns represent the RNA type, the number of entries for each type, the median, mean and standard deviation of the stem length (i.e., number of adjacent base pairs) and the minimum number of base pairs to break in order to open pseudoknots (# PKBP). For each row, a non-redundant set was selected, and outliers were removed.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-340-1\"/>", "<graphic xlink:href=\"1471-2105-9-340-2\"/>", "<graphic xlink:href=\"1471-2105-9-340-3\"/>", "<graphic xlink:href=\"1471-2105-9-340-4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>1 Background</title>", "<p>High density oligonucleotide tiling arrays allow the investigation of transcriptional activity, protein-DNA interactions and chromatin structure across a whole genome. Tiling arrays have been used in a wide range of studies, including investigation of transcription factor activity [##REF##14980218##1##] and of histone modifications in animals [##REF##15680324##2##] and plants [##REF##17439305##3##], as well as DNA methylation [##REF##16949657##4##]. Analyses of these data are usually based either on a sliding window [##REF##14980218##1##,##REF##15539566##5##], or on hidden Markov models (HMMs) [##REF##15961467##6##, ####REF##16046496##7##, ##REF##16672042##8####16672042##8##]. Other approaches have been suggested, e.g., by Huber <italic>et al</italic>. [##REF##16787969##9##] and Reiss <italic>et al</italic>. [##REF##18056063##10##], but are less common.</p>", "<p>Parameter estimates for sliding window approaches as well as hidden Markov models are typically <italic>ad hoc</italic>. Although there are some notable exceptions in gene expression studies [##REF##16672042##8##,##REF##16098113##11##], no established procedures exist to obtain good parameter estimates from tiling array data, especially in the context of chromatin immunoprecipitation (ChIP-chip) experiments. Attempts have been made to obtain parameter estimates by integrating genome annotations into the analysis [##REF##17038339##12##]. While this may provide good results when investigating transcriptional activity in well studied organisms, it is limited by the quality of available annotations. For ChIP-chip studies the required annotation data is unavailable. A method for the localisation of transcription factors from ChIP-chip experiments by Keleş [##REF##17447925##13##] does obtain the required parameter estimates from the data and allows for variations in length of enriched regions.</p>", "<p>Methods designed for the analysis of ChIP-chip data focus almost exclusively on the study of transcription factors [[##REF##15961467##6##,##REF##16046496##7##,##REF##18056063##10##], and [##REF##17447925##13##]]. While this is an important class of experiments, ChIP-chip studies are not limited to transcription factors, and the analysis of other ChIP-chip experiments may require new methods. One other area of active research that utilises ChIP-chip experiments is the study of histone modifications and chromatin structure [##REF##17439305##3##]. Although both types of experiment employ the same technology, there are several important differences between them. Most importantly, the 147 bp of DNA bound by a histone complex are considerably longer than the typical transcription factor binding site, and the histone modifications of interest are expected to affect several neighbouring histones. Consequently the ChIP fragments derived from a transcription factor binding site all originate from a small region containing the given binding site while regions affected by histone modifications can be much longer than the ChIP fragments used. As a result of this, the data from histone modification experiments usually contain long regions of interest encompassing several non-overlapping ChIP fragments, rather than the short and relatively isolated peaks produced by transcription factor studies.</p>", "<p>Here we consider the analysis of data from a histone modification study in <italic>Arabidopsis </italic>[##REF##17439305##3##]. These data consist of four ChIP samples for histone H3 with lysine 27 trimethylation (H3K27me3) and four histone H3 ChIP samples that act as a control. The aim of this analysis is to identify and characterise regions throughout the genome that exhibit enrichment for H3K27me3. It is desirable to use a method which is specifically designed for the analysis of histone modifications or flexible enough to accomodate the varying length of enriched regions. Furthermore, the method should obtain all parameter estimates from the data without the use of genome annotations and be robust towards outliers. Amongst the methods discussed above TileMap [##REF##16046496##7##] comes closest to these requirements. Although it was developed with transcription factor analysis in mind it is general enough that it should provide useful results for other ChIP-chip experiments. This is emphasised by its application to histone modification [##REF##17439305##3##] and DNA methylation [##REF##16949657##4##] data as well as transtription factor analysis [##REF##17090591##14##,##REF##17322403##15##]. TileMap obtains some, but not all, of the required parameter estimates from the data. To provide a method which meets the requirements oulined above we develop a two state HMM with <italic>t </italic>emission distributions. All parameter estimates for the model are obtained by maximum likelihood estimation using the Baum-Welch algorithm [##UREF##0##16##] and Viterbi training [##UREF##1##17##]. These methods have the advantage that no prior knowledge about parameter values is required and one need not rely on frequently unavailable genome annotations. To assess the performance of our model, we apply it to simulated and real data. Results are compared to those produced by TileMap. The remainder of this article is structured as follows. In Section 2 the hidden Markov model is developed and MLEs for all parameters are derived in Section 4. The performance of the resulting model is assessed in terms of sensitivity and specificity on simulated data in Sections 2.3.3–2.3.6. In Section 2.3.7 the model is used to analyse a public ChIP-chip data set [##REF##17439305##3##] and results are compared to the original analysis of these data.</p>" ]

[ "<title>4 Methods</title>", "<title>4.1 Baum-Welch Algorithm</title>", "<p>The Baum-Welch algorithm [##UREF##0##16##] used to estimate parameters for our model is outlined in Section 2.2.2; further details are given below. Computing the likelihood of the long observation sequences produced by tiling arrays involves products of many small contributions. This typically results in likelihoods below machine precision. To avoid this effect computations are carried out in log-space, using the identity</p>", "<p></p>", "<p>In the following we use ^ln∑ to denote summations which should be computed via Equation (4). The sequence of probe statistics <italic>Y </italic>is split into <italic>D </italic>observation sequences <italic>Y </italic>^{(<italic>d</italic>)}such that the distance between probes within each observation sequence is at most max_gap and the distance between the end points of different observation sequences is greater than max_gap.</p>", "<p>The emission distribution of state <italic>S</italic>_{<italic>i </italic>}is given as</p>", "<p></p>", "<p>For a given parameter set <italic>θ </italic>we can obtain new parameter estimates for transition probabilities by calculating</p>", "<p></p>", "<p></p>", "<p>Here <italic>α</italic>_{<italic>k </italic>}and <italic>β</italic>_{<italic>k </italic>}are known as forward and backward variables. For observation sequence <italic>d</italic>, <italic>d </italic>= 1, ..., <italic>D</italic>, they are defined as</p>", "<p></p>", "<p></p>", "<p>where 1 ≤ <italic>i </italic>≤ <italic>N</italic>, 1 ≤ <italic>j </italic>≤ <italic>N</italic>, 1 ≤ <italic>k </italic>&lt;<italic>K</italic>_{<italic>d </italic>}and</p>", "<p></p>", "<p></p>", "<p>where 1 ≤ <italic>i </italic>≤ <italic>N</italic>, 1 ≤ <italic>i </italic>≤ <italic>N</italic>, <italic>k </italic>= <italic>K</italic>_{<italic>d </italic>}- 1, ..., 1. Note that ln [<italic>P</italic>(<italic>Y </italic>^{(<italic>d</italic>)}; <italic>θ</italic>)] is given by We then calculate</p>", "<p></p>", "<p></p>", "<p></p>", "<p>Combining the estimates from all observation sequences we obtain new parameter estimates for the transition probabilities:</p>", "<p></p>", "<p></p>", "<p>Calculations for the re-estimation of <italic>θ</italic>₂may involve negative values and cannot be carried out in log-space.</p>", "<p>To obtain the required parameter estimates we first define and then compute</p>", "<p></p>", "<p></p>", "<p></p>", "<p>There is no closed form estimate for <italic>ν</italic>_{<italic>i</italic>}. To obtain one has to find a solution to the equation</p>", "<p></p>", "<p>where <italic>ψ </italic>is the digamma function. Standard root-finding techniques are employed to find a solution to (20).</p>", "<title>4.2 Viterbi Training</title>", "<p>Viterbi training provides a faster alternative to the Baum-Welch algorithm. See Section 2.2.3 for a high level description of the algorithm. Details of the parameter estimation procedure are given below. Instead of calculating the conditional expectation of the complete data log likelihood, this algorithm first computes the most likely state sequence <italic>Q </italic>given the observation sequence <italic>Y </italic>and the current model <italic>θ</italic>. The sequence <italic>Y </italic>is partitioned according to <italic>Q</italic>, assigning each observation to the state that it most likely originated from. New estimates for <italic>θ</italic>₁are then obtained by calculating</p>", "<p></p>", "<p></p>", "<p>Updates for <italic>μ </italic>and <italic>σ </italic>are obtained as in Section 2.2.1. The degrees of freedom <italic>ν </italic>can be either fixed in advance or estimated from the data using Equation (20) by setting if and otherwise.</p>", "<title>4.3 Simulated Data</title>", "<p>In a first step following the original analysis by [##REF##17439305##3##], TileMap [##REF##16046496##7##] is used with the HMM option to define enriched and non-enriched probes. Note that, although this classification of probes is not perfect, it can be assumed that most probes are assigned to the correct group. The length distribution of enriched and non-enriched regions detected by TileMap is used to determine the length distributions for the simulated data after removing all regions that contain less than 10 probes (Figure ##FIG##9##10##). Data are generated by first determining the length of enriched and non-enriched regions from the empirical length distributions and then sampling data points from the respective TileMap generated clusters. Following this procedure, 600 sequences with one to ten enriched regions in each sequence are generated. A second dataset is generated by applying the model described in Section 2. Note that, although this procedure relies on the classifications produced by the respective models, the resampling procedure will place individual probe values in a new context of surrounding probes, which may lead to different probe calls in the analysis of the simulated data. Prior to analysis all data are quantile normalised.</p>" ]

[ "<title>2 Results and discussion</title>", "<p>Tiling array data consists of a series of measurements taken along the genome. Typically, microarray probes are designed to interrogate the genome at regular intervals. Design constraints such as probe affinity and uniqueness cause differences in probe density along the genome and can lead to large gaps between probes. Here we assume that the probe density is homogeneous except for a number of large gaps where the distance between adjacent probes is larger than max_gap. In the following analyses we use max_gap = 200 bp. This is identical to the value used by Zhang <italic>et al</italic>. [##REF##17439305##3##], allowing for a direct comparison of results. Consider a ChIP-chip tiling array experiment with two conditions, a ChIP sample <bold>X</bold>₁targeting the protein of interest and a control sample <bold>X</bold>₂. Each sample <bold>X</bold>_{<italic>l </italic>}has <italic>m</italic>_{<italic>l </italic>}replicates (<italic>l </italic>= 1, 2) providing measurements for <italic>K </italic>genomic locations. The measurements for each probe are summarised by the \"shrinkage <italic>t</italic>\" statistic [##REF##17402924##18##]:</p>", "<p></p>", "<p>where is a James-Stein shrinkage estimate of the probe variance obtained by calculating</p>", "<p></p>", "<p>and are the usual unbiased empirical variances and is the estimated optimal pooling parameter</p>", "<p></p>", "<p>Other moderated <italic>t </italic>statistics have been suggested and could be used instead, most notably the empirical Bayes <italic>t </italic>statistic used by Ji and Wong [##REF##16046496##7##] and the moderated <italic>t </italic>of Smyth [##REF##16646809##19##]. All of these approaches are designed to increase performance compared to the ordinary <italic>t </italic>statistic by incorporating information from all probes on the microarray into individual probe statistics. Here we choose the \"shrinkage <italic>t</italic>\" because it does not require any knowledge about the underlying distribution of probe values while providing similar performance compared to more complex models [##REF##17402924##18##].</p>", "<title>2.1 Hidden Markov Model</title>", "<p>To detect enriched regions we use a two state discrete time hidden Markov model with continuous emission distributions and homogeneous transition probabilities (Figure ##FIG##0##1##), i.e., the transition probabilities depend only on the current state of the model. The use of homogeneous transition probabilities assumes equally-spaced probes within each observation sequence as well as a geometric distribution of the length of enriched regions. As discussed above there will be some variation in probe distances. Using a relatively small value for max_gap ensures that the assumption of homogeneity holds at least approximately. The two states of the model correspond to enrichment or no enrichment in the ChIP sample. The model is characterised by the set of states <bold>S </bold>= {<italic>S</italic>₁, <italic>S</italic>₂}, the initial state distribution <italic>p</italic>, the matrix of transition probabilities <italic>A </italic>and the state specific emission density functions <italic>f</italic>_{<italic>i</italic>}, <italic>i </italic>= 1, 2. The emission distribution of state <italic>S</italic>_{<italic>i </italic>}is modelled as a <italic>t </italic>distribution with location parameter <italic>μ</italic>_{<italic>i</italic>}, scale parameter <italic>σ</italic>_{<italic>i</italic>}, and <italic>ν</italic>_{<italic>i </italic>}degrees of freedom.</p>", "<p>The use of <italic>t </italic>distributions has the advantage that their sensitivity to outliers can be adjusted via the degrees of freedom parameter, making them more robust and versatile than normal distributions. This is particularly useful when <italic>ν </italic>is estimated from the data [##UREF##2##20##]. It should be noted that the <italic>y</italic>_{<italic>k </italic>}modelled here are from a <italic>t</italic>-like distribution (Equation (1)). While this in itself might suggest the use of <italic>t </italic>distributions for the <italic>f</italic>_{<italic>i</italic>}s, they are primarily chosen for their robustness. In the following we will refer to this model by its parameter vector <italic>θ </italic>= (<italic>θ</italic>₁, <italic>θ</italic>₂), where <italic>θ</italic>₁is the ordered pair (<italic>p</italic>, <italic>A</italic>) and <italic>θ</italic>₂the ordered triple (<italic>μ</italic>, <italic>σ</italic>, <italic>ν</italic>).</p>", "<p>Given a hidden Markov model <italic>θ </italic>and an observation sequence <italic>Y</italic>, it is possible to compute the sequence of states <italic>Q </italic>= <italic>q</italic>₁<italic>q</italic>₂...<italic>q</italic>_{<italic>K </italic>}that is most likely to produce <italic>Y</italic>. There are several approaches to obtaining <italic>Q </italic>[##UREF##3##21##]. Usually <italic>Q </italic>is computed either by maximising the posterior probabilities <italic>P</italic>(<italic>q</italic>_{<italic>k </italic>}= <italic>S</italic>_{<italic>i</italic>}|<italic>Y</italic>; <italic>θ</italic>), <italic>k </italic>= 1, ..., <italic>K </italic>or by calculating the sequence that maximises <italic>P</italic>(<italic>Q|Y</italic>; <italic>θ</italic>). The latter provides the single most likely sequence of states and can be computed efficiently by the Viterbi algorithm [##UREF##4##22##]. For the particular model used here both approaches are equivalent.</p>", "<title>2.2 Parameter Estimation</title>", "<p>In this section we will discuss two different approaches to estimate <italic>θ </italic>for the model described in Section 2.1. The methods under consideration are the EM algorithm, which is usually known as the Baum-Welch algorithm in the context of HMMs, and Viterbi training. While the Baum-Welch algorithm is guaranteed to converge to a local maximum of the likelihood function, it is computationally intensive. Viterbi training provides a faster alternative but may not converge to a local maximum.</p>", "<title>2.2.1 Initial Estimates</title>", "<p>Both optimisation algorithms discussed here require initial parameter estimates. These are obtained from the data by first partitioning the vector of observations <italic>Y </italic>into two clusters using <italic>k</italic>-means clustering [##UREF##5##23##]. From these clusters the location and scale parameters of the corresponding states are obtained as the mean and variance of the observations in the cluster. In the following, <italic>ν</italic>₁= <italic>ν</italic>₂= 6 is used as initial estimate for the degrees of freedom parameters.</p>", "<title>2.2.2 Baum-Welch Algorithm</title>", "<p>The Baum-Welch algorithm [##UREF##0##16##] is a well established iterative method for estimating parameters of HMMs. It represents the EM algorithm [##UREF##6##24##] for the specific case of HMMs. This algorithm can be used to optimise the transition parameters <italic>θ</italic>₁as well as the emission parameters <italic>θ</italic>₂. Each iteration of the algorithm consists of two phases. During the first phase, the current parameter estimates are used to determine for each probe statistic in the observation sequence how likely it is to be produced by the different states of the model. In the second phase, parameters for the emission distributions of each state are estimated using contributions from all observations, according to the probability that they were produced by the respective state of the model. The state transition parameters are updated in a similar fashion, accounting for the probability of transitions between states based on the observation sequence and the current model. After each iteration this procedure results in a model which explains the observed data better than the previous one, approaching a locally optimal solution. Using this method parameter estimates are updated until convergence is achieved. The details of the resulting algorithm are outlined in Section 4.1.</p>", "<p>This method of parameter estimation is computationally expensive and time-consuming for a typical tiling array data set. The computing time can be reduced by fixing the degrees of freedom for the emission distributions in advance, thus avoiding the root-finding required for the estimation of these parameters. While this does not provide the same flexibility as estimating the required degree of robustness from the data it reduces the complexity of the optimisation problem. It is noted by Liu and Rubin [##UREF##7##25##] that attempts to estimate the degrees of freedom are more likely to produce results which are of little practical interest. The impact on classification performance of this choice is investigated in Section 2.3.</p>", "<p>The formulation of the Baum-Welch algorithm used in this article is based on the description given by Rabiner [##UREF##3##21##] and on the EM algorithm derived by Peel and McLachlan [##UREF##8##26##] for fitting mixtures of <italic>t </italic>distributions.</p>", "<title>2.2.3 Viterbi Training</title>", "<p>While the Baum-Welch algorithm described in Section 2.2.2 is expected to provide good parameter estimates, it is computationally expensive. A faster model-fitting procedure can be devised by replacing the first phase of the Baum-Welch algorithm with a maximisation step. This method was introduced in [##UREF##1##17##] as segmental <italic>k</italic>-means and is now commonly referred to as Viterbi training. Unlike the Baum-Welch algorithm which allows each probe statistic to contribute to the parameter estimates for all states, Viterbi training assigns each observation to the state that is most likely to produce the given probe statistic. Thus each observation contributes to exactly one state of the model. While each iteration of this method is faster than one iteration of the Baum-Welch algorithm some iterations may decrease the likelihood of the model, thus failing to advance it towards a useful solution. See Section 4.2 for further details on the implementation of Viterbi training used here.</p>", "<title>2.3 Testing</title>", "<title>2.3.1 Simulated Data</title>", "<p>To assess the ability to distinguish between enriched and non-enriched probes of the models obtained by the different parameter estimation methods discussed in Section 2.2, we simulate data with known enriched regions. To ensure that the simulation study is providing meaningful results, it is based as closely on real data as possible. To this end, two independent analyses of the H3K27me3 data published by Zhang <italic>et al</italic>. [##REF##17439305##3##] are carried out, one using TileMap [##REF##16046496##7##], the other based on our model. The result of each analysis is used to generate a new dataset with known enriched regions. See Section 4 for further details. In the following these data are referred to as datasets I and II respectively. Since the simulation procedure is likely to bias results towards the model that was used in the process, we concentrate on the analysis of dataset I, with some results for dataset II presented for comparison. The use of data based on both models allows us to consider their performance under advantageous and disadvantageous conditions.</p>", "<title>2.3.2 Performance Measure</title>", "<p>The performance of different models on these data is determined in terms of false positive and false negative rates at probe level. While the relative importance of false positives and false negatives depends on the experiment under consideration, they are often equally problematic in the context of ChIP-chip experiments, especially when considering experiments which investigate differences between different cell lines or developmental stages, where all incorrect classifications are of equal concern. In this context, we define false positives as probes that are classified as non-enriched by the analysis of the real data but are called enriched in the subsequent analysis of simulated data, and vice versa for false negatives.</p>", "<p>The output of each model is the estimated posterior probability of enrichment for each probe. In practice, probe calls (\"enriched\" or \"non-enriched\") are generated from this posterior probability based on a 0.5 cut-off. For any given model, classification performance will change with the chosen threshold. Thus we assess model performance across a range of cut-offs, reporting the relative number of false positives and false negatives as well as the error rate. The latter is also used to determine the cut-off that minimises incorrect classification results, and model performance is judged on the numbers of incorrect classifications at this optimal cut-off and at the usual 0.5 cut-off, and on the distance between the optimal cut-off and 0.5. The trade-off between sensitivity and specificity provided by the different models is characterised with ROC curves and the associated AUC values.</p>", "<p>Another measure of interest is the ability to characterise the length distribution of enriched regions correctly. When studying chromatin structure the extent of structural changes is of interest; this is the case for the data studied in Section 2.3.7. This property of the different models is investigated in Section 2.3.6.</p>", "<title>2.3.3 Estimating Degrees of Freedom</title>", "<p>We now consider the performance of both the Baum-Welch procedure and Viterbi training when all model parameters, including the degrees of freedom <italic>ν</italic>, are estimated from the data. Both parameter estimation methods are used to fit an HMM to datasets I and II, and the performance of resulting models is assessed in terms of the achieved error rate (Figure ##FIG##1##2##), ROC curves (Figure ##FIG##2##3##) and their associated AUC (Table ##TAB##0##1##) for both datasets. To assess how well these methods perform in comparison to an established algorithm, we also fit a TileMap model to the two simulated datasets. The three models are compared to each other, as well as an <italic>ad hoc </italic>model which simply uses, without optimisation, the initial parameter estimates used by the two parameter optimisation methods. When comparing the performance of these models on both simulated datasets, it is important to consider that the simulation procedure introduces a bias towards the underlying model.</p>", "<p>Estimating all parameters from the data with either the Baum-Welch algorithm or Viterbi training leads to models with high sensitivity, producing fewer false negatives than TileMap for any given cut-off [see Additional file ##SUPPL##0##1##]. At the same time they lead to an increased number of false positives [see Additional file ##SUPPL##1##2##] compared to TileMap, indicating a slight reduction of specificity. When considering the error rate it becomes apparent that both Baum-Welch and Viterbi training provide a favourable trade-off between sensitivity and specificity. These models reduce the number of incorrect classifications compared to TileMap both at the usual 0.5 cut-off and at the optimal cut-off. Moreover, while the Baum-Welch algorithm and Viterbi training both lead to models with an optimal cut-off close to 0.5 (0.51 and 0.42 respectively), TileMap provides an optimal cut-off of 0.19, indicating that it underestimates the posterior probability of enrichment. This becomes even more apparent when considering the result for dataset II where the optimal cut-off for TileMap is at 0.002 compared to 0.5 for Baum-Welch and 0.41 for Viterbi training. This result suggests that TileMap is more tuned towards avoiding false positives than false negatives. From the above results we estimate that the weight given to false positives by TileMap is approximately 3.2 and 26 times larger than the weight for false negatives on datasets I and II respectively. The ROC curves (Figure ##FIG##2##3##) provide further evidence that the models with MLEs outperform TileMap. Although all three models perform well on dataset I, both parameter optimisation methods lead to better results than TileMap. The benefits of optimising parameter estimates are further highlighted by the performance of the model with <italic>ad hoc </italic>estimates that is used as starting point for the optimisation procedures. On both datasets, optimised parameters provide a notable increase in performance, with TileMap performing only slightly better than the <italic>ad hoc </italic>model on dataset II.</p>", "<title>2.3.4 Fixed Degrees of Freedom</title>", "<p>Estimating <italic>ν</italic>, the degrees of freedom, for <italic>t </italic>distributions from the data is time-consuming and may not be very accurate, especially for relatively large values of <italic>ν</italic>. In this section we investigate the effect of fixing <italic>ν </italic>a priori for both states of the model. Only the case <italic>ν</italic>₁= <italic>ν</italic>₂is considered here. The remaining parameters are estimated from the training data using the Baum-Welch algorithm and Viterbi training with <italic>ν </italic>= 3, 4, ..., 50. For each value of <italic>ν</italic>, we report the error rate (Figure ##FIG##3##4##) as well as the AUC (Figure ##FIG##4##5##) on the simulated data.</p>", "<p>For the best combination of <italic>ν </italic>and cut-off, both parameter estimation methods result in models with a classification performance comparable to the case of variable degrees of freedom (Figure ##FIG##1##2##). While the Baum-Welch algorithm tends to produce models with an optimal cut-off close to 0.5, Viterbi training only achieves this for large values of <italic>ν</italic>. Notably, the best classification performance of the Viterbi trained model is achieved with 14 degrees of freedom and a 0.37 cut-off compared to 7 degrees of freedom and a 0.49 cut-off from Baum-Welch. This results in a decreased performance of the Viterbi model relative to the Baum-Welch model at the 0.5 cut-off.</p>", "<title>2.3.5 Convergence</title>", "<p>To reduce the time required for parameter estimation it is useful to limit the number of iterations. While each iteration of the Baum-Welch algorithm is guaranteed to improve the likelihood of the model, small changes to the parameter values do not necessarily lead to significant changes in the classification result. Furthermore, Viterbi training is not guaranteed to converge to a local maximum of the likelihood function and a likelihood based convergence criterion may not be appropriate for this method. Here we investigate the convergence of both algorithms based on the error rate and AUC to gauge the number of iterations required to achieve good classification results. Parameter estimation is performed with 60 iterations for both algorithms. Current estimates are used to classify the test data at every 5^thiteration and AUC (Figure ##FIG##4##5##) and error rate (Figure ##FIG##5##6##) are determined.</p>", "<p>The most striking difference in the convergence behaviour of the two methods is that Viterbi training appears to obtain good parameter estimates within a small number of iterations. Further iterations of the algorithm do not improve results substantially, whereas the Baum-Welch procedure provides parameter estimates that are better than the ones obtained by Viterbi training, both in terms of likelihood and classification performance, but takes substantially longer to obtain these estimates. The Baum-Welch algorithm not only requires more iterations than Viterbi training, but the time required for each iteration is also longer.</p>", "<title>2.3.6 Length Distribution of Enriched Regions</title>", "<p>When studying histone modifications one possible characteristic of interest is the length of enriched regions. To assess how accurately the different methods reflect the length distribution of enriched regions, we compare the length of regions predicted by TileMap and by the model (using Baum-Welch parameter estimates) to the length distribution of enriched regions in the simulated data (the \"true length distribution\"). Note that this length distribution may vary from the one found in real data. Nevertheless this comparison highlights some of the differences between the two models. Quantile-quantile plots of the respective length distributions show that TileMap systematically underestimates the length of enriched regions (Figure ##FIG##6##7## (bottom left) and Figure ##FIG##7##8## (bottom left)). While this effect is relatively small on dataset I there is some indication that it increases with region length and long regions may not be characterised appropriately by TileMap (Figure ##FIG##6##7## (top left)). This observation is further supported by the length distribution of enriched regions produced by TileMap on dataset II (Figure ##FIG##7##8## (left)). Enriched regions in dataset II are generally longer than regions in dataset I. This difference is not captured by TileMap. Both TileMap and the Baum-Welch trained model produce several regions that are shorter than the shortest enriched region in the simulated data (Figure ##FIG##6##7## (bottom)). There are two possible explanations for these short regions. They may be caused by underestimating the length of enriched regions, possibly splitting one enriched region into several predicted regions, or they may represent spurious enriched results produced by the model. In each case there is the possibility that the occurrence of extremely short regions is caused either by an intrinsic shortcoming of the model or by artifacts introduced during the simulation process. Since the simulation relies on TileMap to identify enriched and non-enriched probes it is inevitable that some probes will be misclassified. Subsequently these probes may be included in the simulated data, causing short disruptions of enriched and non-enriched regions. A sufficiently sensitive model could detect these unintended changes between enriched and non-enriched states.</p>", "<p>To investigate further which of these is the case, we first examine the number of enriched probes contained in the short regions found by the Baum-Welch model and by TileMap respectively. The model with Baum-Welch parameter estimates found 126 regions with less than 10 probes. These regions contain a total of 866 probes of which 717 are in enriched regions. While this indicates that the majority of short regions is due to underestimating the length of enriched regions, several spurious probe calls remain. TileMap produced 249 regions with less than 10 probes, containing a total of 1781 probes, of which 1753 are in enriched regions. This is strong evidence that almost all of these short regions are caused by underestimating the length of enriched regions, and is consistent with the above observation that TileMap systematically underestimates the length of enriched regions.</p>", "<p>To investigate whether the spurious short regions produced by the Baum-Welch model are due to an intrinsic shortcoming of the model or are artifacts introduced by the simulation procedure, we turn to real data. Here we focus on enriched regions containing only a single probe, which are most likely to be false positives. On dataset I the Baum-Welch model produced six of these extremely short regions. One of these probes is a true positive from an enriched region containing ten probes, i.e., the length of this region is underestimated by the Baum-Welch model. Of the remaining five probes three are identical, leaving three unique probes to be investigated further. For each of these three probes, we determine its position in the real data and its distance from enriched regions identified by TileMap and by our model (Section 2.3.7). Two of the probes are found to be located close to enriched regions identified by TileMap (142 and 391 bp) and all three probes are contained within enriched regions identified by our model [see Additional file ##SUPPL##2##3##]. This suggests that these probes may have been misclassified by TileMap during the original analysis, leading to an overestimation of the number of false positives produced by the Baum-Welch model on dataset I.</p>", "<title>2.3.7 Application to ChIP-Chip Data</title>", "<p>To investigate the performance of our model further, we apply it to the data of [##REF##17439305##3##] and compare the result to the original analysis. Based on the results of the simulation study (Sections 2.3.3–2.3.6) we use the following procedure:</p>", "<p>1. Quantile normalise and log transform data;</p>", "<p>2. Calculate probe statistics (Equation (2));</p>", "<p>3. Obtain initial estimates (Section 2.2.1);</p>", "<p>4. Use 5 iterations of Viterbi training to improve initial estimates;</p>", "<p>5. Use 15 iterations of Baum-Welch algorithm to obtain maximum likelihood estimates;</p>", "<p>6. Apply resulting model to data to identify enriched regions.</p>", "<p>This results in the detection of 5285 H3K27me3 regions covering 12.9 Mb of genomic sequence. Of these enriched regions, 3962 (~75%) are overlapping at least one annotated transcript. A total of 4982 or about 18.9% of all annotated genes are found to be enriched for H3K27me3. While most of the enriched regions cover a single gene, some regions are found to contain up to seven genes (Figure ##FIG##8##9(b)##). Enriched regions are predominantly longer than 1 kb with some extending over more than 20 kb (Figure ##FIG##8##9(c)##).</p>", "<p>To assess whether there is a difference between regions of the genome that show H3K27me3 enrichment and the rest of the genome, we investigate the density of genes in the neighbourhood of genes that appear to be regulated by H3K27me3, and compare this to the gene density in other regions of the genome. For this purpose we obtain the gene density for the 50 kb upstream and downstream of each gene as (bp annotated as genes)/100 kb. The resulting gene densities for genes with and without enriched regions are summarised in Figure ##FIG##8##9(a)##. There are visible differences between the two distributions which we test for significance with a two sided Kolmogorov-Smirnov test; this results in an approximate <italic>p</italic>-value of 2 × 10^-15. The significance of this result is further confirmed by a resampling experiment: the smallest <italic>p</italic>-value obtained from a series of 10000 resampled datasets is 1 × 10^-6.</p>" ]

[ "<title>2 Results and discussion</title>", "<p>Tiling array data consists of a series of measurements taken along the genome. Typically, microarray probes are designed to interrogate the genome at regular intervals. Design constraints such as probe affinity and uniqueness cause differences in probe density along the genome and can lead to large gaps between probes. Here we assume that the probe density is homogeneous except for a number of large gaps where the distance between adjacent probes is larger than max_gap. In the following analyses we use max_gap = 200 bp. This is identical to the value used by Zhang <italic>et al</italic>. [##REF##17439305##3##], allowing for a direct comparison of results. Consider a ChIP-chip tiling array experiment with two conditions, a ChIP sample <bold>X</bold>₁targeting the protein of interest and a control sample <bold>X</bold>₂. Each sample <bold>X</bold>_{<italic>l </italic>}has <italic>m</italic>_{<italic>l </italic>}replicates (<italic>l </italic>= 1, 2) providing measurements for <italic>K </italic>genomic locations. The measurements for each probe are summarised by the \"shrinkage <italic>t</italic>\" statistic [##REF##17402924##18##]:</p>", "<p></p>", "<p>where is a James-Stein shrinkage estimate of the probe variance obtained by calculating</p>", "<p></p>", "<p>and are the usual unbiased empirical variances and is the estimated optimal pooling parameter</p>", "<p></p>", "<p>Other moderated <italic>t </italic>statistics have been suggested and could be used instead, most notably the empirical Bayes <italic>t </italic>statistic used by Ji and Wong [##REF##16046496##7##] and the moderated <italic>t </italic>of Smyth [##REF##16646809##19##]. All of these approaches are designed to increase performance compared to the ordinary <italic>t </italic>statistic by incorporating information from all probes on the microarray into individual probe statistics. Here we choose the \"shrinkage <italic>t</italic>\" because it does not require any knowledge about the underlying distribution of probe values while providing similar performance compared to more complex models [##REF##17402924##18##].</p>", "<title>2.1 Hidden Markov Model</title>", "<p>To detect enriched regions we use a two state discrete time hidden Markov model with continuous emission distributions and homogeneous transition probabilities (Figure ##FIG##0##1##), i.e., the transition probabilities depend only on the current state of the model. The use of homogeneous transition probabilities assumes equally-spaced probes within each observation sequence as well as a geometric distribution of the length of enriched regions. As discussed above there will be some variation in probe distances. Using a relatively small value for max_gap ensures that the assumption of homogeneity holds at least approximately. The two states of the model correspond to enrichment or no enrichment in the ChIP sample. The model is characterised by the set of states <bold>S </bold>= {<italic>S</italic>₁, <italic>S</italic>₂}, the initial state distribution <italic>p</italic>, the matrix of transition probabilities <italic>A </italic>and the state specific emission density functions <italic>f</italic>_{<italic>i</italic>}, <italic>i </italic>= 1, 2. The emission distribution of state <italic>S</italic>_{<italic>i </italic>}is modelled as a <italic>t </italic>distribution with location parameter <italic>μ</italic>_{<italic>i</italic>}, scale parameter <italic>σ</italic>_{<italic>i</italic>}, and <italic>ν</italic>_{<italic>i </italic>}degrees of freedom.</p>", "<p>The use of <italic>t </italic>distributions has the advantage that their sensitivity to outliers can be adjusted via the degrees of freedom parameter, making them more robust and versatile than normal distributions. This is particularly useful when <italic>ν </italic>is estimated from the data [##UREF##2##20##]. It should be noted that the <italic>y</italic>_{<italic>k </italic>}modelled here are from a <italic>t</italic>-like distribution (Equation (1)). While this in itself might suggest the use of <italic>t </italic>distributions for the <italic>f</italic>_{<italic>i</italic>}s, they are primarily chosen for their robustness. In the following we will refer to this model by its parameter vector <italic>θ </italic>= (<italic>θ</italic>₁, <italic>θ</italic>₂), where <italic>θ</italic>₁is the ordered pair (<italic>p</italic>, <italic>A</italic>) and <italic>θ</italic>₂the ordered triple (<italic>μ</italic>, <italic>σ</italic>, <italic>ν</italic>).</p>", "<p>Given a hidden Markov model <italic>θ </italic>and an observation sequence <italic>Y</italic>, it is possible to compute the sequence of states <italic>Q </italic>= <italic>q</italic>₁<italic>q</italic>₂...<italic>q</italic>_{<italic>K </italic>}that is most likely to produce <italic>Y</italic>. There are several approaches to obtaining <italic>Q </italic>[##UREF##3##21##]. Usually <italic>Q </italic>is computed either by maximising the posterior probabilities <italic>P</italic>(<italic>q</italic>_{<italic>k </italic>}= <italic>S</italic>_{<italic>i</italic>}|<italic>Y</italic>; <italic>θ</italic>), <italic>k </italic>= 1, ..., <italic>K </italic>or by calculating the sequence that maximises <italic>P</italic>(<italic>Q|Y</italic>; <italic>θ</italic>). The latter provides the single most likely sequence of states and can be computed efficiently by the Viterbi algorithm [##UREF##4##22##]. For the particular model used here both approaches are equivalent.</p>", "<title>2.2 Parameter Estimation</title>", "<p>In this section we will discuss two different approaches to estimate <italic>θ </italic>for the model described in Section 2.1. The methods under consideration are the EM algorithm, which is usually known as the Baum-Welch algorithm in the context of HMMs, and Viterbi training. While the Baum-Welch algorithm is guaranteed to converge to a local maximum of the likelihood function, it is computationally intensive. Viterbi training provides a faster alternative but may not converge to a local maximum.</p>", "<title>2.2.1 Initial Estimates</title>", "<p>Both optimisation algorithms discussed here require initial parameter estimates. These are obtained from the data by first partitioning the vector of observations <italic>Y </italic>into two clusters using <italic>k</italic>-means clustering [##UREF##5##23##]. From these clusters the location and scale parameters of the corresponding states are obtained as the mean and variance of the observations in the cluster. In the following, <italic>ν</italic>₁= <italic>ν</italic>₂= 6 is used as initial estimate for the degrees of freedom parameters.</p>", "<title>2.2.2 Baum-Welch Algorithm</title>", "<p>The Baum-Welch algorithm [##UREF##0##16##] is a well established iterative method for estimating parameters of HMMs. It represents the EM algorithm [##UREF##6##24##] for the specific case of HMMs. This algorithm can be used to optimise the transition parameters <italic>θ</italic>₁as well as the emission parameters <italic>θ</italic>₂. Each iteration of the algorithm consists of two phases. During the first phase, the current parameter estimates are used to determine for each probe statistic in the observation sequence how likely it is to be produced by the different states of the model. In the second phase, parameters for the emission distributions of each state are estimated using contributions from all observations, according to the probability that they were produced by the respective state of the model. The state transition parameters are updated in a similar fashion, accounting for the probability of transitions between states based on the observation sequence and the current model. After each iteration this procedure results in a model which explains the observed data better than the previous one, approaching a locally optimal solution. Using this method parameter estimates are updated until convergence is achieved. The details of the resulting algorithm are outlined in Section 4.1.</p>", "<p>This method of parameter estimation is computationally expensive and time-consuming for a typical tiling array data set. The computing time can be reduced by fixing the degrees of freedom for the emission distributions in advance, thus avoiding the root-finding required for the estimation of these parameters. While this does not provide the same flexibility as estimating the required degree of robustness from the data it reduces the complexity of the optimisation problem. It is noted by Liu and Rubin [##UREF##7##25##] that attempts to estimate the degrees of freedom are more likely to produce results which are of little practical interest. The impact on classification performance of this choice is investigated in Section 2.3.</p>", "<p>The formulation of the Baum-Welch algorithm used in this article is based on the description given by Rabiner [##UREF##3##21##] and on the EM algorithm derived by Peel and McLachlan [##UREF##8##26##] for fitting mixtures of <italic>t </italic>distributions.</p>", "<title>2.2.3 Viterbi Training</title>", "<p>While the Baum-Welch algorithm described in Section 2.2.2 is expected to provide good parameter estimates, it is computationally expensive. A faster model-fitting procedure can be devised by replacing the first phase of the Baum-Welch algorithm with a maximisation step. This method was introduced in [##UREF##1##17##] as segmental <italic>k</italic>-means and is now commonly referred to as Viterbi training. Unlike the Baum-Welch algorithm which allows each probe statistic to contribute to the parameter estimates for all states, Viterbi training assigns each observation to the state that is most likely to produce the given probe statistic. Thus each observation contributes to exactly one state of the model. While each iteration of this method is faster than one iteration of the Baum-Welch algorithm some iterations may decrease the likelihood of the model, thus failing to advance it towards a useful solution. See Section 4.2 for further details on the implementation of Viterbi training used here.</p>", "<title>2.3 Testing</title>", "<title>2.3.1 Simulated Data</title>", "<p>To assess the ability to distinguish between enriched and non-enriched probes of the models obtained by the different parameter estimation methods discussed in Section 2.2, we simulate data with known enriched regions. To ensure that the simulation study is providing meaningful results, it is based as closely on real data as possible. To this end, two independent analyses of the H3K27me3 data published by Zhang <italic>et al</italic>. [##REF##17439305##3##] are carried out, one using TileMap [##REF##16046496##7##], the other based on our model. The result of each analysis is used to generate a new dataset with known enriched regions. See Section 4 for further details. In the following these data are referred to as datasets I and II respectively. Since the simulation procedure is likely to bias results towards the model that was used in the process, we concentrate on the analysis of dataset I, with some results for dataset II presented for comparison. The use of data based on both models allows us to consider their performance under advantageous and disadvantageous conditions.</p>", "<title>2.3.2 Performance Measure</title>", "<p>The performance of different models on these data is determined in terms of false positive and false negative rates at probe level. While the relative importance of false positives and false negatives depends on the experiment under consideration, they are often equally problematic in the context of ChIP-chip experiments, especially when considering experiments which investigate differences between different cell lines or developmental stages, where all incorrect classifications are of equal concern. In this context, we define false positives as probes that are classified as non-enriched by the analysis of the real data but are called enriched in the subsequent analysis of simulated data, and vice versa for false negatives.</p>", "<p>The output of each model is the estimated posterior probability of enrichment for each probe. In practice, probe calls (\"enriched\" or \"non-enriched\") are generated from this posterior probability based on a 0.5 cut-off. For any given model, classification performance will change with the chosen threshold. Thus we assess model performance across a range of cut-offs, reporting the relative number of false positives and false negatives as well as the error rate. The latter is also used to determine the cut-off that minimises incorrect classification results, and model performance is judged on the numbers of incorrect classifications at this optimal cut-off and at the usual 0.5 cut-off, and on the distance between the optimal cut-off and 0.5. The trade-off between sensitivity and specificity provided by the different models is characterised with ROC curves and the associated AUC values.</p>", "<p>Another measure of interest is the ability to characterise the length distribution of enriched regions correctly. When studying chromatin structure the extent of structural changes is of interest; this is the case for the data studied in Section 2.3.7. This property of the different models is investigated in Section 2.3.6.</p>", "<title>2.3.3 Estimating Degrees of Freedom</title>", "<p>We now consider the performance of both the Baum-Welch procedure and Viterbi training when all model parameters, including the degrees of freedom <italic>ν</italic>, are estimated from the data. Both parameter estimation methods are used to fit an HMM to datasets I and II, and the performance of resulting models is assessed in terms of the achieved error rate (Figure ##FIG##1##2##), ROC curves (Figure ##FIG##2##3##) and their associated AUC (Table ##TAB##0##1##) for both datasets. To assess how well these methods perform in comparison to an established algorithm, we also fit a TileMap model to the two simulated datasets. The three models are compared to each other, as well as an <italic>ad hoc </italic>model which simply uses, without optimisation, the initial parameter estimates used by the two parameter optimisation methods. When comparing the performance of these models on both simulated datasets, it is important to consider that the simulation procedure introduces a bias towards the underlying model.</p>", "<p>Estimating all parameters from the data with either the Baum-Welch algorithm or Viterbi training leads to models with high sensitivity, producing fewer false negatives than TileMap for any given cut-off [see Additional file ##SUPPL##0##1##]. At the same time they lead to an increased number of false positives [see Additional file ##SUPPL##1##2##] compared to TileMap, indicating a slight reduction of specificity. When considering the error rate it becomes apparent that both Baum-Welch and Viterbi training provide a favourable trade-off between sensitivity and specificity. These models reduce the number of incorrect classifications compared to TileMap both at the usual 0.5 cut-off and at the optimal cut-off. Moreover, while the Baum-Welch algorithm and Viterbi training both lead to models with an optimal cut-off close to 0.5 (0.51 and 0.42 respectively), TileMap provides an optimal cut-off of 0.19, indicating that it underestimates the posterior probability of enrichment. This becomes even more apparent when considering the result for dataset II where the optimal cut-off for TileMap is at 0.002 compared to 0.5 for Baum-Welch and 0.41 for Viterbi training. This result suggests that TileMap is more tuned towards avoiding false positives than false negatives. From the above results we estimate that the weight given to false positives by TileMap is approximately 3.2 and 26 times larger than the weight for false negatives on datasets I and II respectively. The ROC curves (Figure ##FIG##2##3##) provide further evidence that the models with MLEs outperform TileMap. Although all three models perform well on dataset I, both parameter optimisation methods lead to better results than TileMap. The benefits of optimising parameter estimates are further highlighted by the performance of the model with <italic>ad hoc </italic>estimates that is used as starting point for the optimisation procedures. On both datasets, optimised parameters provide a notable increase in performance, with TileMap performing only slightly better than the <italic>ad hoc </italic>model on dataset II.</p>", "<title>2.3.4 Fixed Degrees of Freedom</title>", "<p>Estimating <italic>ν</italic>, the degrees of freedom, for <italic>t </italic>distributions from the data is time-consuming and may not be very accurate, especially for relatively large values of <italic>ν</italic>. In this section we investigate the effect of fixing <italic>ν </italic>a priori for both states of the model. Only the case <italic>ν</italic>₁= <italic>ν</italic>₂is considered here. The remaining parameters are estimated from the training data using the Baum-Welch algorithm and Viterbi training with <italic>ν </italic>= 3, 4, ..., 50. For each value of <italic>ν</italic>, we report the error rate (Figure ##FIG##3##4##) as well as the AUC (Figure ##FIG##4##5##) on the simulated data.</p>", "<p>For the best combination of <italic>ν </italic>and cut-off, both parameter estimation methods result in models with a classification performance comparable to the case of variable degrees of freedom (Figure ##FIG##1##2##). While the Baum-Welch algorithm tends to produce models with an optimal cut-off close to 0.5, Viterbi training only achieves this for large values of <italic>ν</italic>. Notably, the best classification performance of the Viterbi trained model is achieved with 14 degrees of freedom and a 0.37 cut-off compared to 7 degrees of freedom and a 0.49 cut-off from Baum-Welch. This results in a decreased performance of the Viterbi model relative to the Baum-Welch model at the 0.5 cut-off.</p>", "<title>2.3.5 Convergence</title>", "<p>To reduce the time required for parameter estimation it is useful to limit the number of iterations. While each iteration of the Baum-Welch algorithm is guaranteed to improve the likelihood of the model, small changes to the parameter values do not necessarily lead to significant changes in the classification result. Furthermore, Viterbi training is not guaranteed to converge to a local maximum of the likelihood function and a likelihood based convergence criterion may not be appropriate for this method. Here we investigate the convergence of both algorithms based on the error rate and AUC to gauge the number of iterations required to achieve good classification results. Parameter estimation is performed with 60 iterations for both algorithms. Current estimates are used to classify the test data at every 5^thiteration and AUC (Figure ##FIG##4##5##) and error rate (Figure ##FIG##5##6##) are determined.</p>", "<p>The most striking difference in the convergence behaviour of the two methods is that Viterbi training appears to obtain good parameter estimates within a small number of iterations. Further iterations of the algorithm do not improve results substantially, whereas the Baum-Welch procedure provides parameter estimates that are better than the ones obtained by Viterbi training, both in terms of likelihood and classification performance, but takes substantially longer to obtain these estimates. The Baum-Welch algorithm not only requires more iterations than Viterbi training, but the time required for each iteration is also longer.</p>", "<title>2.3.6 Length Distribution of Enriched Regions</title>", "<p>When studying histone modifications one possible characteristic of interest is the length of enriched regions. To assess how accurately the different methods reflect the length distribution of enriched regions, we compare the length of regions predicted by TileMap and by the model (using Baum-Welch parameter estimates) to the length distribution of enriched regions in the simulated data (the \"true length distribution\"). Note that this length distribution may vary from the one found in real data. Nevertheless this comparison highlights some of the differences between the two models. Quantile-quantile plots of the respective length distributions show that TileMap systematically underestimates the length of enriched regions (Figure ##FIG##6##7## (bottom left) and Figure ##FIG##7##8## (bottom left)). While this effect is relatively small on dataset I there is some indication that it increases with region length and long regions may not be characterised appropriately by TileMap (Figure ##FIG##6##7## (top left)). This observation is further supported by the length distribution of enriched regions produced by TileMap on dataset II (Figure ##FIG##7##8## (left)). Enriched regions in dataset II are generally longer than regions in dataset I. This difference is not captured by TileMap. Both TileMap and the Baum-Welch trained model produce several regions that are shorter than the shortest enriched region in the simulated data (Figure ##FIG##6##7## (bottom)). There are two possible explanations for these short regions. They may be caused by underestimating the length of enriched regions, possibly splitting one enriched region into several predicted regions, or they may represent spurious enriched results produced by the model. In each case there is the possibility that the occurrence of extremely short regions is caused either by an intrinsic shortcoming of the model or by artifacts introduced during the simulation process. Since the simulation relies on TileMap to identify enriched and non-enriched probes it is inevitable that some probes will be misclassified. Subsequently these probes may be included in the simulated data, causing short disruptions of enriched and non-enriched regions. A sufficiently sensitive model could detect these unintended changes between enriched and non-enriched states.</p>", "<p>To investigate further which of these is the case, we first examine the number of enriched probes contained in the short regions found by the Baum-Welch model and by TileMap respectively. The model with Baum-Welch parameter estimates found 126 regions with less than 10 probes. These regions contain a total of 866 probes of which 717 are in enriched regions. While this indicates that the majority of short regions is due to underestimating the length of enriched regions, several spurious probe calls remain. TileMap produced 249 regions with less than 10 probes, containing a total of 1781 probes, of which 1753 are in enriched regions. This is strong evidence that almost all of these short regions are caused by underestimating the length of enriched regions, and is consistent with the above observation that TileMap systematically underestimates the length of enriched regions.</p>", "<p>To investigate whether the spurious short regions produced by the Baum-Welch model are due to an intrinsic shortcoming of the model or are artifacts introduced by the simulation procedure, we turn to real data. Here we focus on enriched regions containing only a single probe, which are most likely to be false positives. On dataset I the Baum-Welch model produced six of these extremely short regions. One of these probes is a true positive from an enriched region containing ten probes, i.e., the length of this region is underestimated by the Baum-Welch model. Of the remaining five probes three are identical, leaving three unique probes to be investigated further. For each of these three probes, we determine its position in the real data and its distance from enriched regions identified by TileMap and by our model (Section 2.3.7). Two of the probes are found to be located close to enriched regions identified by TileMap (142 and 391 bp) and all three probes are contained within enriched regions identified by our model [see Additional file ##SUPPL##2##3##]. This suggests that these probes may have been misclassified by TileMap during the original analysis, leading to an overestimation of the number of false positives produced by the Baum-Welch model on dataset I.</p>", "<title>2.3.7 Application to ChIP-Chip Data</title>", "<p>To investigate the performance of our model further, we apply it to the data of [##REF##17439305##3##] and compare the result to the original analysis. Based on the results of the simulation study (Sections 2.3.3–2.3.6) we use the following procedure:</p>", "<p>1. Quantile normalise and log transform data;</p>", "<p>2. Calculate probe statistics (Equation (2));</p>", "<p>3. Obtain initial estimates (Section 2.2.1);</p>", "<p>4. Use 5 iterations of Viterbi training to improve initial estimates;</p>", "<p>5. Use 15 iterations of Baum-Welch algorithm to obtain maximum likelihood estimates;</p>", "<p>6. Apply resulting model to data to identify enriched regions.</p>", "<p>This results in the detection of 5285 H3K27me3 regions covering 12.9 Mb of genomic sequence. Of these enriched regions, 3962 (~75%) are overlapping at least one annotated transcript. A total of 4982 or about 18.9% of all annotated genes are found to be enriched for H3K27me3. While most of the enriched regions cover a single gene, some regions are found to contain up to seven genes (Figure ##FIG##8##9(b)##). Enriched regions are predominantly longer than 1 kb with some extending over more than 20 kb (Figure ##FIG##8##9(c)##).</p>", "<p>To assess whether there is a difference between regions of the genome that show H3K27me3 enrichment and the rest of the genome, we investigate the density of genes in the neighbourhood of genes that appear to be regulated by H3K27me3, and compare this to the gene density in other regions of the genome. For this purpose we obtain the gene density for the 50 kb upstream and downstream of each gene as (bp annotated as genes)/100 kb. The resulting gene densities for genes with and without enriched regions are summarised in Figure ##FIG##8##9(a)##. There are visible differences between the two distributions which we test for significance with a two sided Kolmogorov-Smirnov test; this results in an approximate <italic>p</italic>-value of 2 × 10^-15. The significance of this result is further confirmed by a resampling experiment: the smallest <italic>p</italic>-value obtained from a series of 10000 resampled datasets is 1 × 10^-6.</p>" ]

[ "<title>3 Conclusion</title>", "<p>With the use of MLEs for all model parameters, our model clearly improves classification performance on simulated data compared to <italic>ad hoc </italic>estimates, and outperforms TileMap. While our model produced some short regions that appear to be false positives, they are readily explained as a result of the simulation process. Comparison of results on simulated and real data suggests that TileMap produced a large number of false negatives in the original analysis used as the basis for the simulation. Inevitably, these false negatives were selected as part of non-enriched regions during the simulation process. The fact that the model with Baum-Welch parameter estimates was able to identify these isolated enriched probes despite the non-enriched contexts where they appeared emphasises the high sensitivity of the model.</p>", "<p>TileMap's apparent tendency to penalise false positives more than false negatives clearly contributes to its relatively low performance in our comparisons which are based on the assumption that both types of error are equally problematic. While this is the case for the application considered here, one may argue that false positives are indeed of greater concern in some cases. When this is the case, TileMap's trade-off between sensitivity and specificity may lead to better results. However, it should be noted that the relative weights given to false positives and false negatives by TileMap can vary substantially between datasets. The parameter estimation procedure used for our model on the other hand provides consistent performance at the chosen cut-off.</p>", "<p>The model-fitting procedure derived from the results of the simulation study (Sections 2.3.3–2.3.6) provides a fast and reliable approach to parameter estimation. This method retains all the favourable properties of the Baum-Welch algorithm while utilising the reduced computing time provided by Viterbi training. The use of MLEs ensures that model parameters are appropriate for the data. Results from the simulation study show that estimating model parameters from the data improves the model's ability to recognise enriched regions of varying length and generally improves classification performance.</p>", "<title>3.1 Future Work</title>", "<p>The analysis of the H3K27me3 data (Section 2.3.7) largely confirms the analysis of [##REF##17439305##3##] although there are some notable differences. Most importantly, the H3K27me3 regions detected by our analysis are longer than the ones determined by TileMap (Figure ##FIG##9##10##). While Zhang <italic>et al</italic>. [##REF##17439305##3##] found few regions longer than 1 kb, our analysis indicates that over 70% of enriched regions have a length of at least 1 kb, with the longest region spanning over 20 kb. Accordingly we find more regions that extend over several genes (Figure ##FIG##8##9(b)##). This may have implications for conclusions about the spreading of H3K27me3 regions in <italic>Arabidopsis</italic>.</p>", "<p>At this stage, the biological significance of the observed difference in gene density in the neighbourhood of enriched and non-enriched genes is unclear. However, it indicates that the two groups of genes differ in a significant way. This suggests that the partition into enriched and non-enriched genes produced by our analysis is indeed meaningful.</p>", "<p>The hidden Markov model presented in this article uses homogeneous transition probabilities, assuming that all probes are spaced out equally along the genome. To satisfy this assumption at least approximately, we use a fixed cut-off of 200 bp to partition the sequence of probe statistics such that there are no large gaps between probes. This arbitrary cut-off could be avoided by using a continuous time hidden Markov model.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Tiling arrays are an important tool for the study of transcriptional activity, protein-DNA interactions and chromatin structure on a genome-wide scale at high resolution. Although hidden Markov models have been used successfully to analyse tiling array data, parameter estimation for these models is typically <italic>ad hoc</italic>. Especially in the context of ChIP-chip experiments, no standard procedures exist to obtain parameter estimates from the data. Common methods for the calculation of maximum likelihood estimates such as the Baum-Welch algorithm or Viterbi training are rarely applied in the context of tiling array analysis.</p>", "<title>Results</title>", "<p>Here we develop a hidden Markov model for the analysis of chromatin structure ChIP-chip tiling array data, using <italic>t </italic>emission distributions to increase robustness towards outliers. Maximum likelihood estimates are used for all model parameters. Two different approaches to parameter estimation are investigated and combined into an efficient procedure.</p>", "<title>Conclusion</title>", "<p>We illustrate an efficient parameter estimation procedure that can be used for HMM based methods in general and leads to a clear increase in performance when compared to the use of <italic>ad hoc </italic>estimates. The resulting hidden Markov model outperforms established methods like TileMap in the context of histone modification studies.</p>" ]

[ "<title>5 Availability</title>", "<p>The parameter estimation methods used in this article are available as part of the R package tileHMM from the authors' webpage <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.bioinformatics.csiro.au/TileHMM/\"/> and from CRAN. The simulated data used in this study is available from the authors' web page.</p>", "<title>6 Authors' contributions</title>", "<p>PH conducted the research and wrote the manuscript. DB critically revised the manuscript. GS conceived the project. DB and GS provided supervision to PH. All authors have read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>PH is supported by an MQRES scholarship from Macquarie University and a top-up scholarship from CSIRO. The authors would like to thank Michael Buckley for his helpful suggestions.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Hidden Markov model for the analysis of ChIP-chip tiling array data.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Error rate for different models on datasets I and II</bold>. Error rate resulting from the different models on dataset I (left) and II (right). When the total number of incorrect probe calls is considered, both parameter estimation procedures outperform TileMap on dataset I for cut-offs larger than 0.2. Both Baum-Welch and Viterbi training provide models with an optimal cut-off close to 0.5, while TileMap significantly underestimates the posterior probability resulting in an optimal cut-off of 0.19. The models with optimised parameters show similar performance on both datasets. On dataset II TileMap's performance is reduced in comparison to the results on dataset I. The main differences between the models considered here occur at error rates of 0–0.08. The relevant area of the figures in the top row is magnified in the plots below.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>ROC curves for different models on datasets I and II</bold>. TileMap and the models with Baum-Welch and Viterbi training parameter estimates show similar performance on dataset I (left) with a small advantage for the models with optimised parameters. Comparison with a model using <italic>ad hoc </italic>parameter estimates highlights the performance increase achieved by optimising model parameters. On dataset II (right) TileMap performs similarly to the model with <italic>ad hoc </italic>parameter estimates. Figures on the bottom provide a close-up view of the plots above.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Model performance for different choices of <italic>ν</italic></bold>. The Baum-Welch model (red) performs better for relatively small values of <italic>ν </italic>while Viterbi training (blue) favours larger <italic>ν</italic>. For the optimal choice of <italic>ν </italic>the Baum-Welch parameter estimates lead to an optimal cut-off close to 0.5.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>AUC for different choices of <italic>ν </italic>and increasing numberof iterations</bold>. Change in AUC for different choices of <italic>ν </italic>(left). The Baum-Welch model performs better for relatively small values of <italic>ν </italic>while Viterbi training favours larger <italic>ν</italic>. Improvements in AUC with increasing number of iterations (right). The performance of the Viterbi trained model improves substantially during the first five iterations. Further iterations only produce small changes in the AUC. The Baum-Welch method requires more iterations to obtain the same AUC as as the Viterbi model. After 20 iterations the Baum-Welch model starts to outperform the Viterbi model.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Error rate at optimal and 0.5 cutoff for increasing number of iterations</bold>. Parameter estimates obtained by the Baum-Welch algorithm (filled symbols) and Viterbi training (open symbols) improve model performance with increasing nuber of iterations. Viterbi training quickly approaches its optimal solution and initially outperforms Baum-Welch. The final model produced by the Baum-Welch algorithm provides a lower error rate than Viterbi training.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p><bold>Length distribution of enriched regions from dataset I</bold>. Quantile-quantile plots comparing length distributions of enriched regions found with TileMap (left) and with the model based on maximum likelihood estimates (right) to the true length distribution of enriched regions in dataset I. Figures on the bottom provide a close-up view of the plots above. Each dot represents a percentile of the length distributions.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p><bold>Length distribution of enriched regions from dataset II</bold>. Quantile-quantile plots comparing length distributions of enriched regions found with TileMap (left) and with the model based on maximum likelihood estimates (right) to the true length distribution of enriched regions in dataset II. Figures on the bottom provide a close-up view of the plots above. Each dot represents a percentile of the length distributions.</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p><bold>Analysis of ChIP-chip data</bold>. (a) Gene density in areas surrounding genes that contain H3K27me3 enriched regions and genes that do not contain enriched regions. (b) Number of genes found in H3K27me3 regions. While most enriched regions cover a single gene, there is a substantial number of H3K27me3 regions that cover several genes and enriched regions are found to contain up to seven genes. (c) Length distribution of H3K27me3 regions.</p></caption></fig>", "<fig position=\"float\" id=\"F10\"><label>Figure 10</label><caption><p><bold>Length distribution of enriched regions from real data</bold>. Length distribution of enriched regions as determined by TileMap (blue) and Baum-Welch (red). Region length is determined in terms of probes per region. Both distributions were truncated at 10 for the simulation, ensuring that all regions in the simulated data contain at least ten probes.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>AUC for different models on both simulated datasets.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\">TileMap</td><td align=\"center\">Baum-Welch</td><td align=\"center\">Viterbi-Training</td><td align=\"center\">Viterbi-EM</td><td align=\"center\">ad hoc</td></tr></thead><tbody><tr><td align=\"left\">dataset I</td><td align=\"center\">0.9986</td><td align=\"center\">0.9998</td><td align=\"center\">0.9997</td><td align=\"center\">0.9998</td><td align=\"center\">0.9869</td></tr><tr><td align=\"left\">dataset II</td><td align=\"center\">0.9749</td><td align=\"center\">0.9995</td><td align=\"center\">0.9994</td><td align=\"center\">0.9995</td><td align=\"center\">0.9728</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-343-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>y</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>x</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:mn>1</mml:mn><mml:mi>k</mml:mi></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>x</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mrow><mml:mn>2</mml:mn><mml:mi>k</mml:mi></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:msqrt><mml:mrow><mml:mfrac><mml:mrow><mml:msubsup><mml:mi>v</mml:mi><mml:mrow><mml:mn>1</mml:mn><mml:mi>k</mml:mi></mml:mrow><mml:mo>∗</mml:mo></mml:msubsup></mml:mrow><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mn>1</mml:mn></mml:msub></mml:mrow></mml:mfrac><mml:mo>+</mml:mo><mml:mfrac><mml:mrow><mml:msubsup><mml:mi>v</mml:mi><mml:mrow><mml:mn>2</mml:mn><mml:mi>k</mml:mi></mml:mrow><mml:mo>∗</mml:mo></mml:msubsup></mml:mrow><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:mrow></mml:mfrac></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-343-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>v</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mi>k</mml:mi></mml:mrow><mml:mo>∗</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-343-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>v</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mi>k</mml:mi></mml:mrow><mml:mo>∗</mml:mo></mml:msubsup><mml:mo>=</mml:mo><mml:msubsup><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>l</mml:mi><mml:mo>∗</mml:mo></mml:msubsup><mml:msubsup><mml:mi>s</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mtext>median</mml:mtext></mml:mrow><mml:mn>2</mml:mn></mml:msubsup><mml:mo>+</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:msubsup><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>l</mml:mi><mml:mo>∗</mml:mo></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo><mml:msubsup><mml:mi>s</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mi>k</mml:mi></mml:mrow><mml:mn>2</mml:mn></mml:msubsup><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-343-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>s</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mi>k</mml:mi></mml:mrow><mml:mn>2</mml:mn></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-343-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>l</mml:mi><mml:mo>∗</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM3\"><label>(3)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1471-2105-9-343-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>l</mml:mi><mml:mo>*</mml:mo></mml:msubsup><mml:mo>=</mml:mo><mml:mi>min</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>,</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>K</mml:mi></mml:msubsup><mml:mrow><mml:mover accent=\"true\"><mml:mrow><mml:mtext>Var</mml:mtext></mml:mrow><mml:mo stretchy=\"true\">︿</mml:mo></mml:mover><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>s</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mi>k</mml:mi></mml:mrow><mml:mn>2</mml:mn></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>K</mml:mi></mml:msubsup><mml:mrow><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>s</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mi>k</mml:mi></mml:mrow><mml:mn>2</mml:mn></mml:msubsup><mml:mo>−</mml:mo><mml:msubsup><mml:mi>s</mml:mi><mml:mrow><mml:mi>l</mml:mi><mml:mtext>median</mml:mtext></mml:mrow><mml:mn>2</mml:mn></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>.</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM4\"><label>(4)</label>ln(<italic>x </italic>+ <italic>y</italic>) = ln(<italic>x</italic>) + ln (1 + <italic>e</italic>^{ln(<italic>y</italic>)-ln(<italic>x</italic>)}).</disp-formula>", "<disp-formula id=\"bmcM5\"><label>(5)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1471-2105-9-343-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>f</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>y</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo>;</mml:mo><mml:msub><mml:mi>μ</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>σ</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mi>Γ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mn>2</mml:mn></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>Γ</mml:mi><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mn>2</mml:mn></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msup></mml:mrow><mml:mrow><mml:msub><mml:mi>σ</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:msqrt><mml:mrow><mml:mi>π</mml:mi><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mfrac><mml:mrow><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>y</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>μ</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:msub><mml:mi>σ</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msup></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac><mml:mo>.</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM6\"><label>(6)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1471-2105-9-343-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>ξ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>P</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>q</mml:mi><mml:mi>k</mml:mi><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msubsup><mml:mi>q</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>|</mml:mo><mml:msup><mml:mi>Y</mml:mi><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>d</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:msup><mml:mo>;</mml:mo><mml:mi>θ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM7\"><label>(7)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1471-2105-9-343-i9\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable columnalign=\"left\"><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mo>=</mml:mo></mml:mtd><mml:mtd columnalign=\"left\"><mml:mrow><mml:msubsup><mml:mi>α</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>+</mml:mo><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>f</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>y</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>;</mml:mo><mml:msub><mml:mi>θ</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>+</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow/></mml:mtd><mml:mtd columnalign=\"left\"><mml:mrow><mml:mo>+</mml:mo><mml:msubsup><mml:mi>β</mml:mi><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo><mml:mi>j</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>−</mml:mo><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>P</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msup><mml:mi>Y</mml:mi><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>d</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:msup><mml:mo>;</mml:mo><mml:mi>θ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>.</mml:mo></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM8\"><label>(8)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M10\" name=\"1471-2105-9-343-i10\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>α</mml:mi><mml:mrow><mml:mn>1</mml:mn><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>p</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>+</mml:mo><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>f</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>y</mml:mi><mml:mn>1</mml:mn><mml:mi>d</mml:mi></mml:msubsup><mml:mo>;</mml:mo><mml:msub><mml:mi>θ</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM9\"><label>(9)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M11\" name=\"1471-2105-9-343-i11\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable columnalign=\"left\"><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:msubsup><mml:mi>α</mml:mi><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo><mml:mi>j</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mtd><mml:mtd columnalign=\"left\"><mml:mo>=</mml:mo></mml:mtd><mml:mtd columnalign=\"left\"><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>ln</mml:mi><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>α</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:mo stretchy=\"false\">]</mml:mo><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow/></mml:mtd><mml:mtd columnalign=\"left\"><mml:mrow/></mml:mtd><mml:mtd columnalign=\"left\"><mml:mrow><mml:mo>+</mml:mo><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>f</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>y</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>;</mml:mo><mml:msub><mml:mi>θ</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>,</mml:mo></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM10\"><label>(10)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M12\" name=\"1471-2105-9-343-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>β</mml:mi><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mn>0</mml:mn><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM11\"><label>(11)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M13\" name=\"1471-2105-9-343-i13\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>β</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>+</mml:mo><mml:mi>ln</mml:mi><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mi>f</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>y</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>;</mml:mo><mml:msub><mml:mi>θ</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>+</mml:mo><mml:msubsup><mml:mi>β</mml:mi><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo><mml:mi>j</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M14\" name=\"1471-2105-9-343-i14\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mrow><mml:msubsup><mml:mi>α</mml:mi><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM12\"><label>(12)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M15\" name=\"1471-2105-9-343-i15\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>γ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>P</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>q</mml:mi><mml:mi>k</mml:mi><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>|</mml:mo><mml:msup><mml:mi>Y</mml:mi><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>d</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:msup><mml:mo>;</mml:mo><mml:mi>θ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM13\"><label>(13)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M16\" name=\"1471-2105-9-343-i16\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mo>=</mml:mo><mml:msubsup><mml:mi>α</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>+</mml:mo><mml:msubsup><mml:mi>β</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>−</mml:mo><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>P</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msup><mml:mi>Y</mml:mi><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>d</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:msup><mml:mo>;</mml:mo><mml:mi>θ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM14\"><label>(14)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M17\" name=\"1471-2105-9-343-i17\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mo>=</mml:mo><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:msubsup><mml:mi>ξ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle><mml:mo>.</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM15\"><label>(15)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M18\" name=\"1471-2105-9-343-i18\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>a</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>=</mml:mo><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:munderover><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:munderover><mml:mrow><mml:msubsup><mml:mi>ξ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle><mml:mo>−</mml:mo><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:munderover><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:munderover><mml:mrow><mml:msubsup><mml:mi>γ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM16\"><label>(16)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M19\" name=\"1471-2105-9-343-i19\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>p</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>=</mml:mo><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow/></mml:mtd></mml:mtr></mml:mtable><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:munderover><mml:mrow><mml:msubsup><mml:mi>γ</mml:mi><mml:mrow><mml:mn>1</mml:mn><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle><mml:mo>−</mml:mo><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>D</mml:mi><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>.</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M20\" name=\"1471-2105-9-343-i20\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>=</mml:mo><mml:msubsup><mml:mi>γ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM17\"><label>(17)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M21\" name=\"1471-2105-9-343-i21\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>+</mml:mo><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>y</mml:mi><mml:mi>k</mml:mi><mml:mi>d</mml:mi></mml:msubsup><mml:mo>−</mml:mo><mml:msub><mml:mi>μ</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM18\"><label>(18)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M22\" name=\"1471-2105-9-343-i22\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>μ</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:msubsup><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub></mml:mrow></mml:msubsup><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:msubsup><mml:mi>y</mml:mi><mml:mi>k</mml:mi><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:msubsup><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub></mml:mrow></mml:msubsup><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM19\"><label>(19)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M23\" name=\"1471-2105-9-343-i23\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>σ</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:msubsup><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub></mml:mrow></mml:msubsup><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msubsup><mml:mi>y</mml:mi><mml:mi>k</mml:mi><mml:mi>d</mml:mi></mml:msubsup><mml:mo>−</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>μ</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:msubsup><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub></mml:mrow></mml:msubsup><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac><mml:mo>.</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M24\" name=\"1471-2105-9-343-i24\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>ν</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM20\"><label>(20)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M25\" name=\"1471-2105-9-343-i25\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable columnalign=\"left\"><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mo>−</mml:mo><mml:mi>ψ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mn>2</mml:mn></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mn>2</mml:mn></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>+</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo></mml:mrow></mml:mrow></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:mo>+</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub></mml:mrow></mml:msubsup><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub></mml:mrow></mml:munderover><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow></mml:mstyle><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>−</mml:mo><mml:msubsup><mml:mi>u</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>+</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:mrow><mml:mrow><mml:mo>+</mml:mo><mml:mi>ψ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mn>2</mml:mn></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>−</mml:mo><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>ν</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mn>2</mml:mn></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM21\"><label>(21)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M26\" name=\"1471-2105-9-343-i26\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>p</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mi>D</mml:mi></mml:mfrac><mml:mo>|</mml:mo><mml:mo>{</mml:mo><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>,</mml:mo><mml:mn>...</mml:mn><mml:mo>,</mml:mo><mml:mi>D</mml:mi><mml:mo>:</mml:mo><mml:msubsup><mml:mi>q</mml:mi><mml:mn>1</mml:mn><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>}</mml:mo><mml:mo>|</mml:mo><mml:mo>,</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM22\"><label>(22)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M27\" name=\"1471-2105-9-343-i27\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>a</mml:mi><mml:mo>ˆ</mml:mo></mml:mover><mml:mrow><mml:mi>i</mml:mi><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mo>|</mml:mo><mml:mo>{</mml:mo><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>,</mml:mo><mml:mn>...</mml:mn><mml:mo>,</mml:mo><mml:mi>D</mml:mi><mml:mo>:</mml:mo><mml:msubsup><mml:mi>q</mml:mi><mml:mi>k</mml:mi><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mtext> and </mml:mtext><mml:msubsup><mml:mi>q</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>}</mml:mo><mml:mo>|</mml:mo></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>d</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>D</mml:mi></mml:msubsup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>K</mml:mi><mml:mi>d</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac><mml:mo>.</mml:mo></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M28\" name=\"1471-2105-9-343-i28\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M29\" name=\"1471-2105-9-343-i29\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mi>q</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>,</mml:mo><mml:msubsup><mml:mi>q</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M30\" name=\"1471-2105-9-343-i30\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>τ</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mi>i</mml:mi></mml:mrow><mml:mi>d</mml:mi></mml:msubsup><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow></mml:semantics></mml:math></inline-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p><bold>False negative probe calls resulting from different models</bold>. For any given cut-off TileMap produces more false negatives than the Baum-Welch and Viterbi trained models.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p><bold>False positive probe calls resulting from different models</bold>. For any given cut-off TileMap produces fewer false positives than the Baum-Welch and Viterbi trained models.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p><bold>Origin of isolated enriched probes in dataset I</bold>. The isolated enriched probes identified in dataset I by the Baum-Welch model originate from enriched regions identified by the Baum-Welch model in the real data. Two out of three probes are located close to enriched regions identified by TileMap.</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>All models with optimised parameters outperform TileMap on both simulated datasets. While TileMap performs well on dataset I it is only slightly better than the model with <italic>ad hoc </italic>parameter estimates.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-343-1\"/>", "<graphic xlink:href=\"1471-2105-9-343-2\"/>", "<graphic xlink:href=\"1471-2105-9-343-3\"/>", "<graphic xlink:href=\"1471-2105-9-343-4\"/>", "<graphic xlink:href=\"1471-2105-9-343-5\"/>", "<graphic xlink:href=\"1471-2105-9-343-6\"/>", "<graphic xlink:href=\"1471-2105-9-343-7\"/>", "<graphic xlink:href=\"1471-2105-9-343-8\"/>", "<graphic xlink:href=\"1471-2105-9-343-9\"/>", "<graphic xlink:href=\"1471-2105-9-343-10\"/>" ]

[ "<media xlink:href=\"1471-2105-9-343-S1.png\" mimetype=\"image\" mime-subtype=\"png\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-343-S2.png\" mimetype=\"image\" mime-subtype=\"png\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-343-S3.png\" mimetype=\"image\" mime-subtype=\"png\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Skin constantly renews throughout adult life. The proliferative compartment of epidermis is confined to the basal layer, where it harbors stem cells, and their progeny, transient amplifying cells [##REF##3719664##1##, ####REF##5117212##2##, ##REF##3981041##3####3981041##3##]. Stem cells are predominantly quiescent <italic>in situ</italic>. Transient amplifying cells are more rapidly cycling, and after dividing for a limited period of time cease to proliferate and undergo terminal differentiation while moving towards the skin surface [##REF##11978538##4##]. Slow cycling stem cells of the murine epidermis were identified by the retention of BrdU or [³H]thymidine after prolonged chase [##REF##6943171##5##, ####REF##2364430##6##, ##REF##10201531##7##, ##REF##10966107##8##, ##REF##14671312##9####14671312##9##]. Research aimed at isolating stem cells directly from human tissue has to be based on different methodological approaches. Putative human interfollicular stem cells have been enriched based on the expression of β1 integrin [##REF##8500165##10##], transferin receptor [##REF##9520465##11##], connexin 43 [##REF##11851883##12##], an isoform of <italic>CD133 </italic>[##REF##12042327##13##] and desmosomal proteins [##REF##12953062##14##]. However, it has not been determined whether these cells represent distinct populations, or belong to overlapping cell subsets. Databases generated from gene expression profiles of stem cells provide useful resources in evaluating putative stem cell populations. The lack or low levels of MHCI molecules have been reported in stem cells of several tissues [##REF##12114532##15##, ####REF##12832694##16##, ##REF##12077603##17##, ##REF##12738887##18##, ##REF##15277692##19##, ##REF##9282991##20####9282991##20##]. Downregulation of MHCI transcripts has been observed in mouse hair follicle stem cells [##REF##15024388##21##]. We have previously isolated a subpopulation of human basal keratinocytes with low/negative MHCI expression (α6⁺/MHCI^-) [##REF##16360835##22##]. Cells with α6⁺/MHCI^-phenotype constitute a small fraction of the basal layer (0.5–2%) as determined by flow cytometry [##REF##16360835##22##]. We found that α6⁺/MHCI^-cells were keratinocytes as they expressed keratin 14 (K14). The α6⁺/MHCI^-cells exhibited characteristics attributed to stem cells: they were clonogenic <italic>in vitro</italic>, relatively small, and had low granularity [##REF##16360835##22##]. In the present work we employ microarray technology, to report global transcriptional profiles of two cell populations: the basal cells that express MHCI, α6⁺/MHCI⁺(transient amplifying cells) and the basal cells that have low/negative MHCI expression, α6⁺/MHCI^-cells, (putative stem cells). Cells were isolated using fluorescence-activated cell sorter (FACS) directly from human epidermis. Further comparisons were made with published data of hair follicle stem cell gene expression profiles.</p>", "<p>In addition, using flow cytometry we have analyzed the expression of nuclear proliferation antigen, Ki67. Our data indicate that MHCI^-cells are quiescent <italic>in situ</italic>. Following FACS sorting, α6⁺/MHCI^-and α6⁺/MHCI⁺cells were grown at clonal densities to determine their colony forming efficiency (CFE). The analysis of CFEs in the initial, primary, culture and in the first passage indicate that α6⁺/MHCI^-cells have higher proliferative potential than α6⁺/MHCI⁺cells, another feature attributed to stem cells.</p>" ]

[ "<title>Methods</title>", "<title>Isolation of keratinocytes</title>", "<p>Neonatal foreskins were obtained from routine circumcisions. After washing in PBS, and removing of subcutaneous fat, the tissue was cut into 5 × 5 mm pieces and incubated overnight at 4°C in Dulbecco's modified Eagle's medium containing 2.5 mg/ml Dispase II (Boehringer Mannheim, Indianopolis, IN), penicillin (100 units per ml), and streptomycin (100 μg/ml). Epithelial sheaths were separated from the dermis by gentle peeling with forceps. Keratinocytes were harvested after incubation with trypsin/EDTA solution (0.05% and 0.01%, respectively) for 10 minutes at 37°C [##REF##16360835##22##].</p>", "<title>Immunocytochemistry and flow cytometry</title>", "<p>After trypsin neutralization and blocking with buffer containing BSA and human IgG, keratinocytes were immunolabeled with R-phycoerythrin (PE)-conjugated mouse anti-human β2 microglobulin (BD Biosciences, San Diego, CA, USA), and with a monoclonal anti-human α6 integrin-fluorescein isothiocyanate (FITC) conjugate (Serotec, Oxford, UK). Control samples were incubated with appropriate isotype controls. All incubations were performed at 4°C. Cells were sorted using FACSVantage, or FACSAria (Becton Dickinson, Franklin Lakes, NJ). Flow cytometry data used for sorting cells that donated RNA for microarray experiments, as well as representative controls are shown in Additional file ##SUPPL##6##7##. The data were analyzed using Cell Quest software (BD Biosciences). Selection of basal keratinocytes with anti-human α6 integrin eliminates suprabasal cells (terminally differentiated keratinocytes), and non-keratinocytes, such as melanocytes and dentritic cells) [##REF##15024388##21##,##REF##16822832##60##].</p>", "<p>For the expression of Ki67 isolated keratinocytes were immunolabeled with mouse anti-human antibody against MHCI (BD Biosciences) and with a polyclonal antibody against Ki67 (Zymed San Francisco, CA). Secondary antibodies were goat anti mouse IgG1 PE-conjugated antibody (Southern Biotechnology Associates Inc. Birmingham, AL) and donkey anti rabbit FITC-conjugated antibody. For the setting of gates, secondary control and single color positive controls were used.</p>", "<title>Keratinocyte culture</title>", "<p>Keratinocytes were grown in a keratinocyte medium (3:1 DMEM, F12) supplemented with FBS and additives in 100 mm culture dishes, previously seeded with lethally irradiated 3T3 fibroblasts [##REF##8486621##68##]. The medium was replaced every other day. Colonies were visualized after two weeks in culture following fixation in 10% formalin and staining with 2% rhodamine B (Sigma). Colony forming efficiency (CFE) is the ratio of colony number to plating cell number expressed as a percentage. Results are presented as means ± SD.</p>", "<title>RNA isolation, amplification and microarray analysis</title>", "<p>Following cell sorting, RNA isolation (RNeasy Mini Kit (Qiagen)) and amplification (MessageAmp aRNA Kit (Ambion)) microarray was performed using Affymetrix HGU 133 A+B GeneChip set. Scanned images of Affymetrix GeneChip arrays were quantified using Affymetrix GCOS software, Gene Chip Operating System. The target intensity was set to 500 and the default parameters were used. The results were filtered and probe sets with a \"No Change\" (NC) call were removed. Additionally, probe sets that were scored \"Increased\" (I) or \"Marginal Increase\" (MI), but called absent on the experimental sample, as well as probe sets that were scored Decreased (D) or \"Marginal Decrease\" (MD) and called absent in the baseline sample, were removed. For the resulting list of probe sets a fold change column was calculated. Microarray Gene Expression Data have been deposited, accession number GSE11089 [##UREF##1##69##]. For the genes that are differentially expressed in both arrays see Additional files ##SUPPL##1##2##, ##SUPPL##7##8## and ##SUPPL##8##9##.</p>", "<title>Semi-quantative RT-PCR</title>", "<p>Total RNA was isolated directly after cell sorting using the RNeasy Mini Kit (QIAGEN) according to manufacturer's protocol. Extracted RNA was reverse transcribed by using Sensiscript RT Kit (QIAGEN) according to manufacturer's instructions. It must be noted that RNA used for RT-PCR and for each microarray analysis was isolated from cells derived from multiple skin samples that consisted of different donors. By pooling skin samples we were able to obtain sufficient amount of cells and at the same time average any potential individual differences. Primers for <italic>KRTHA1 </italic>[##REF##9405442##70##], β-catenin [##REF##15972956##71##], <italic>KLF4 </italic>[##REF##15570219##72##], and <italic>DSG3 </italic>[##REF##16015083##73##] were published previously. Other primers used are: ITGA6 F: 5'-TGCTGTTGGTTCCCTCTCAGAT-3'. ITGA6 R: 5'-CTGGCGGAGGTCAATTCTGT-3'. MYCBP F: 5'-ATGGCCCATTACAAAGCCGC-3'. MYCBP R: 5'-CTATTCAGCACGCTTCTCCT-3'. Initial PCR step was 1 minute at 94.0°C, followed by 25, 30, 35 cycles of a 15 seconds melting at 94.0°C, a 15 seconds annealing at 55.0°C and a 15 seconds extension at 72.0°C. The final extension was at 72.0°C for 1 minute.</p>", "<title>Validation of microarray data using immynocytochemistry and flow cytometry</title>", "<p>For the analyses of protein expressions in MHCI^-and MHCI⁺cells the following antibodies were used: Antibody against MHCI, mouse anti-human (IgG1 isotype, BD biosciences); antibody against NF-κB p65, rabbit anti human (Santa Cruz); antibody against cleaved/active Notch1, rabbit anti human (Calbiochem); and antibody against CD71, mouse anti human (IgG1 isotype, Diaclone, France). Secondary antibodies used for these analyses were: FITC-conjugated goat anti-mouse IgM (Sigma), PE-conjugated goat anti mouse IgG1 (SouthernBiotechnology, Birmingham, Al), and FITC-conjugated donkey anti-rabbit (Jackson ImmunoResearch Lab. Inc., West Grove, PA).</p>", "<title>Analysis of NF-κB activity using TransAM NF-κB p50 kit</title>", "<p>Nuclear extracts were obtained using Nuclear Extract Kit purchased from Active Motif (Carlsbad, CA). 100,000 cells (α6⁺/MHCI⁺and α6⁺/MHCI^-each) were directly sorted in PBS buffer that contained phosphatase inhibitors, supplied with the kit. Levels of the active/phospho-NF-κB p50 in the nucleus were assayed using TransAM NF-κB p50 Transcription Factor Assay Kit (Active Motif) according to manufacturer's protocol. Nuclear extract of HeLa cells stimulated with TNF-α for 30 minutes supplied by Active Motif was used as a positive control.</p>", "<title>Microarray data analysis</title>", "<p>Additional file ##SUPPL##0##1## contains all the transcripts, which demonstrated equal or higher than 2 (≥ 2) fold difference between α6⁺/MHCI^-cells and α6⁺/MHCI⁺cells (Additional file ##SUPPL##0##1## shows only values that are ≥ 2 fold in log2 scale). The genes that are consistently downregulated or upregulated in both arrays were shown in tables S6, S7 and S8. All transcripts that belong to MHCI protein family are downregulated in α6⁺/MHCI^-cells as expected, which indicate that our selection process was successful (see Additional file ##SUPPL##0##1##). Several reports were used as the base to screen for c-Myc target genes [##REF##10737792##74##, ####REF##9858526##75##, ##REF##12651860##76##, ##REF##11085504##77##, ##REF##11983916##78##, ##REF##11353853##79##, ##REF##14576301##80##, ##REF##11139609##81####11139609##81##], Wnt target genes [##REF##16207355##82##, ####REF##15777744##83##, ##REF##12154096##84##, ##REF##15961525##85##, ##REF##14623818##86##, ##REF##14707084##87##, ##REF##15140269##88##, ##REF##11832495##89##, ##REF##12408868##90##, ##REF##12095419##91##, ##REF##15794748##92##, ##REF##16575872##93####16575872##93##], TGF-β/BMP target genes [##REF##12919699##94##, ####REF##12718878##95##, ##REF##14635189##96##, ##REF##15689496##97####15689496##97##], and targets of NF-κB pathway [##REF##10602461##59##,##REF##11687580##98##, ####REF##12221085##99##, ##REF##15722553##100####15722553##100##].</p>", "<title>Statistical analysis</title>", "<p>Student's <italic>t</italic>-test was applied for statistical analysis. Error bars represent ± SD.</p>" ]

[ "<title>Results and discussion</title>", "<p>Skin is the largest and most accessible organ in the body. The differentiation axis of the interfollicular epidermis is spatially well defined: the basal layer contains proliferating cells, while suprabasal layers, stratum spinosum, stratum granulosum, and stratum corneum harbor post-mitotic, differentiating keratinocytes [##UREF##0##23##,##REF##2269655##24##]. These features facilitate the analysis of cells at the specific differentiation stage. Like all self-renewing tissues, epidermis contains stem cells, which are located in the stratum basale. Several proteins have been suggested as markers for keratinocyte stem cell enrichment [##REF##8500165##10##, ####REF##9520465##11##, ##REF##11851883##12##, ##REF##12042327##13##, ##REF##12953062##14####12953062##14##]. We have previously described a basal keratinocyte population that lacks gap junction protein Cx43 in human and mouse epidermis [##REF##11851883##12##]. We have shown that Cx43 negative cells co-localize with label-retaining cells, hair follicle bulge stem cells [##REF##2364430##6##, ####REF##10201531##7##, ##REF##10966107##8####10966107##8##]. Cx43 negative keratinocytes comprise about 10% of human basal keratinocytes and are blast like, small and have low granularity as determined by flow cytometry. Cells in the limbus of the eye, the region of the corneal epithelium that contains stem cells, were also shown to lack Cx43 [##REF##9203348##25##]. In searching for additional markers that can be used to obtain viable cells, we isolated a subset of Cx43 negative keratinocytes characterized by low/negative expression of MHCI that comprised up to 2% of basal epidermal cells [##REF##16360835##22##]. It was believed that almost all nucleated cells express MHCI [##REF##4139789##26##]. Recently, however, stem cells of several tissues were shown to lack MHCI expression [##REF##12114532##15##, ####REF##12832694##16##, ##REF##12077603##17##, ##REF##12738887##18##, ##REF##15277692##19##, ##REF##9282991##20##, ##REF##15024388##21####15024388##21##]. Molecules encoded by MHC are involved in self/non-self discrimination in vertebrates. MHCI molecules bind endogenously derived peptides and stimulate a distinct branch of the adaptive immune system mediated by CD8⁺T cells. The human MHC termed HLA (Human Leukocyte Antigen) encodes three classical polymorphic class I genes: HLA A, B, and C. To isolate transient amplifying cells (α6⁺/MHCI⁺) and presumptive stem cells (α6⁺/MHCI^-), we used antibodies against α6 integrin, a basal cell marker, in combination with antibodies against β2 microglobulin, the light chain of MHCI molecule. Previously we have shown that similar results were obtained regardless of whether antibodies to MHCI heavy chain, or antibodies against β2 microglobulin were used [##REF##16360835##22##].</p>", "<title>α6⁺/MHCI^-cells are quiescent <italic>in situ</italic>, yet in culture display higher proliferative potential</title>", "<p>During tissue homeostasis stem cells are infrequently dividing; thus, and one of the characteristics of stem cells is their quiescence <italic>in situ</italic>. To determine the proliferative status of MHCI^-and MHCI⁺populations, we analyzed the expression of nuclear proliferation antigen Ki67, which is a marker for actively cycling cells. The data presented reflect keratinocyte proliferation since in normal epidermis, non keratinocytes are found to be non-cycling [##REF##808553##27##,##REF##8919043##28##]. Although the absolute values of Ki67 may vary depending on the total number of gated cells, the ratio of Ki67 expressing MHCI^-and MHCI⁺cells is held constant. Flow cytometric analysis showed that MHCI⁺cells expressed more than four time higher levels of Ki67 than MHCI^-cells (Figure ##FIG##0##1##). Only 0.9% of MHCI^-cells expressed Ki67, while 3.9% of MHCI⁺cells were in the cell cycle. Given the low percentages of MHCI^-cells in the basal layer, it is clear that the bulk of cell production in the epidermis is accomplished through divisions of transient amplifying cells, a finding which is in accordance with the established view of epidermal homeostasis.</p>", "<p>High proliferative potential is another feature of stem cells. One way of assessing proliferative potential of keratinocytes is to analyze their colony forming efficiency (CFE) [##REF##8500165##10##]. To assess proliferative potential of the two cell populations' colony forming efficiency (CFE) was analyzed in the primary and secondary cultures. Sorted cells were seeded at clonal densities and colonies formed evaluated after two weeks. In primary cultures, α6⁺/MHCI^-cells exhibited lower colony forming efficiency than α6⁺/MHCI⁺cells (Figure ##FIG##1##2##). However, in secondary cultures α6⁺/MHCI^-showed higher CFE than α6⁺/MHCI⁺cells. From primary to secondary culture, an increase of 38 times was observed in the CFE of α6⁺/MHCI^-cells, while the CFE of α6⁺/MHCI⁺cells increased only 3.6 times (Figure ##FIG##1##2##). Previously, lower initial CFE was observed in limbal epithelial stem cell [##REF##16150918##29##]. Our data indicate that α6⁺/MHCI^-cells have higher proliferative potential than α6⁺/MHCI⁺cells, another characteristic attributed to stem cells. The majority of available data regarding CFE are from secondary or tertiary keratinocyte cultures [##REF##8500165##10##,##REF##2436229##30##]. In the secondary cultures, α6⁺/MHCI^-cells did display higher CFE than α6⁺/MHCI⁺cells, which is an indication of a higher proliferative potential of the α6⁺/MHCI^-cells [##REF##8500165##10##]. The lower CFE of α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells that we recorded in our primary cultures may be due to the time needed for stem cells to transit from a quiescent to a proliferative state.</p>", "<title>Gene expression profile indicates that α6⁺/MHCI^-cells exhibit properties of stem cells</title>", "<p>Microarray profiles of stem cells and their progeny provide a global view into differences of expression of a large number of genes and enable analyses of molecular processes involved in self renewal, proliferation and differentiation. Gene expression profiles of hair follicle bulge stem cells were recently reported [##REF##14671312##9##,##REF##15024388##21##,##REF##15339667##31##, ####REF##16395407##32##, ##REF##16809539##33####16809539##33##], yet until now no data are available with regard to human interfollicular keratinocyte stem cells. We report on transcriptional profiles of putative human keratinocyte stem cells and their immediate progeny, transient amplifying cells. Global gene expression profile was obtained from sorted α6⁺/MHCI^-cells and α6⁺/MHCI⁺cells using DNA microarray chips. We identified a comprehensive list of differentially expressed genes. Notably, all of the MHCI genes were downregulated in α6⁺/MHCI^-cells, thus confirming the successful separation of α6⁺/MHCI⁺and α6⁺/MHCI^-cells. The data also show that expression of MHCI proteins in keratinocytes is regulated at the transcriptional level. The HLA-E transcript is downregulated in α6⁺/MHCI^-cells confirming our previous results obtained at the protein level [##REF##16360835##22##]. The expression of non-classical HLA molecules is thought to protect cells that lack classical HLA expression from lysis by NK cells. At present, it is not known what mechanisms protect presumptive stem cells, MHCI^-cells, from attack by NK cells, especially since MHCI^-cells do not express detectable levels of non-classical HLA-E and HLA-G molecules [##REF##16360835##22##].</p>", "<p>We found that most of the mRNAs of genes encoding cellular receptors and other cell surface molecules were more abundant in α6⁺/MHCI^-cells than in α6⁺/MHCI⁺cells (see Additional file ##SUPPL##0##1##). Conversely, mRNAs of genes encoding proteins that take part in ribosome biosynthesis, RNA splicing, translation, protein degradation, and energy metabolism were more abundant in α6⁺/MHCI⁺cells (see Additional file ##SUPPL##0##1##). These findings are consistent with reports by other investigators who demonstrated that stem cells are characterized by few ribosomes and mitochondria (features related to undifferentiated state of stem cells) but contain a large numbers of receptors [##REF##14671312##9##,##REF##16051216##34##].</p>", "<p>In reference to stem cells quiescence, it should also be noted that transcripts of <italic>CDKN1C </italic>and <italic>CDKN2A </italic>whose products are cyclin-dependent kinase inhibitors were enriched in α6⁺/MHCI^-cells (Figure ##FIG##2##3A##). The product of <italic>CDKN1C</italic>, p57^Kip2, one of the Cip/Kip family members, is tightly associated with inhibition of proliferation of human interfollicular keratinocytes [##REF##17112701##35##], and is upregulated in hair follicle bulge [##REF##15339667##31##], while the product of <italic>CDKN2A</italic>, a 16 kDa protein p16^INK4aimposes a G1 cell cycle arrest [##REF##9823374##36##,##REF##7758941##37##]. Furthermore, in addition to the low protein level of Ki67 observed in MHCI negative population (Figure ##FIG##0##1##), the transcript of Ki67 was less abundant in α6⁺/MHCI^-cells (Table ##TAB##0##1##, and Additional file ##SUPPL##0##1##). Conversely, transcripts of genes encoding proteins that are related to cell proliferation such as cyclins, proteins involved in chromosome remodeling, DNA replication and repair were preferentially enriched in α6⁺/MHCI⁺cells (Table ##TAB##0##1##, and Additional file ##SUPPL##0##1##).</p>", "<p>Interestingly, type I IFN (IFN-α, and IFN-β) has been shown to inhibit cell proliferation by inducing G1 cell cycle arrest. It has been reported that interferon α (IFN-α) has antiproliferative effects on bone marrow stromal precursors, hepatic progenitor cells, and mesenchymal stem cells [##REF##17375123##39##, ####REF##16628647##40##, ##REF##1533991##41####1533991##41##]. We observed enrichment of transcripts of the IFN-α family of proteins in α6⁺/MHCI^-cells as well as <italic>STAT2</italic>, the specific transducing activator of IFN-α transcription. This is the first report that suggests involvement of type I IFN in epidermal stem cell quiescence. Further studies are needed to determine whether IFN-α is synthesized by α6⁺/MHCI^-cells, whether its pathway is active and whether it contributes to α6⁺/MHCI^-cell quiescence (Table ##TAB##0##1##).</p>", "<p>It has been reported that the components of the inositol phospholipid signaling system are present and that the system itself is active in murine embryonic stem cells [##REF##16845989##42##]. In the present study, we demonstrate that several components of the inositol phospholipid signaling system are enriched in α6⁺/MHCI^-cells (Table ##TAB##0##1##). Nevertheless, the functional significance of this observation needs further investigation.</p>", "<p>Among the transcripts enriched in α6⁺/MHCI^-cells, there were transcripts previously shown to be upregulated in cell population enriched for human interfollicular stem cells, such as α6 integrin [##REF##9520465##11##], melanoma-associated chondroitin sulfate proteoglycan [##REF##16877544##43##], phosphorylase kinase α2 [##REF##16877544##43##], and the transcript of p53 homolog <italic>p51/p73L/p63/p40 </italic>[##REF##11248048##44##] (Figure ##FIG##2##3A## and Additional file ##SUPPL##0##1##). Using flow cytometry we found that MHCI^-cells expressed lower level of transferrin receptor (<italic>CD71</italic>), a negative marker used to enrich for interfollicular epidermal stem cells [##REF##9520465##11##], when compared to MHCI⁺cells (Figure ##FIG##3##4##). It should also be noted that although differences in the expression of <italic>CD71 </italic>transcript were lower than 2 fold, <italic>CD71 </italic>mRNA was downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells consistently in both arrays (see Additional file ##SUPPL##1##2##).</p>", "<p>Among the transcripts enriched in α6⁺/MHCI⁺cells, there were mRNAs of genes whose products are expressed at low levels in cell population enriched for human interfollicular stem cells, such as desmosomal proteins including desmoglein 3 (<italic>DSG3</italic>) [##REF##12953062##14##], as well as the proliferation associated transcription factor <italic>c-Myc</italic>, found to be expressed at the lower levels in cultured human interfollicular stem cells [##REF##9353256##45##] (Figures ##FIG##2##3A, 3B## and Additional file ##SUPPL##0##1##). Moreover, it has been reported that epidermal growth factor receptor (EGFR) signaling is downregulated in putative human interfollicular stem cells [##REF##16877544##43##]. In accordance with this observation, we found that <italic>EGFR </italic>itself was downregulated in α6⁺/MHCI^-cells (Fig. ##FIG##2##3A## and Additional file ##SUPPL##0##1##).</p>", "<p>Most notably, we present results that demonstrate that putative stem cells have lower expression of mRNAs encoding proteins that take part in energy metabolism, which can explain how stem cells can be quiescent and at the same time maintain small size (see Additional file ##SUPPL##0##1##).</p>", "<title>Comparison of α6⁺/MHCI^-and α6⁺/MHCI⁺microarray databases with the transcriptional profiles of human and mouse hair follicle stem cells</title>", "<p>We compared α6⁺/MHCI^-transcriptional profile with the published transcriptional profile of human hair follicle stem cells [##REF##16395407##32##] and with four different sets of data of murine hair follicle stem cells [##REF##14671312##9##,##REF##15024388##21##,##REF##15339667##31##,##REF##16809539##33##]. We found that eleven genes, which were enriched in human hair follicle stem cells, were also enriched in α6⁺/MHCI^-cells (Table ##TAB##0##1##). Similarly, mRNAs of nine genes, which were downregulated in human hair follicle stem cells, were also downregulated in α6⁺/MHCI^-cells (Table ##TAB##1##2##). Nevertheless, the differences between the gene expression profiles of stem cells from the two human tissues, i.e. interfollicular and follicular, were also observed including the genes that demonstrated two fold difference in the expression (total of nine genes) (see Additional file ##SUPPL##2##3##).</p>", "<p>mRNAs of eighty-two genes, which were enriched in murine hair follicle stem cells, were also enriched in α6⁺/MHCI^-cells, while mRNAs of forty-one genes, which were downregulated in murine hair follicle stem cells, were also downregulated in α6⁺/MHCI^-cells (Tables ##TAB##0##1##, ##TAB##1##2## and Additional file ##SUPPL##3##4##). Transcription factors <italic>LHX2 </italic>and <italic>TCF3 </italic>that were shown to maintain SC features [##REF##16809539##33##,##REF##17018284##46##], were among the genes that were upregulated in α6⁺/MHCI^-cells as well as in murine hair follicle stem cells. Upon screening of our microarray database for <italic>TCF3 </italic>targets [##REF##17018284##46##], we found that transcripts of twelve genes reported to be upregulated by <italic>TCF3 </italic>were more abundant in α6⁺/MHCI^-cells and conversely transcripts of nine genes repressed by <italic>TCF3 </italic>were more abundant in α6⁺/MHCI⁺cells (see Additional file ##SUPPL##4##5##). Interestingly, both arrays that we performed showed that <italic>LHX2 </italic>was among the most upregulated mRNAs in α6⁺/MHCI^-cells, while <italic>WIF-1 </italic>mRNA, which was the most enriched mRNA in murine hair follicle stem cells according to one report [##REF##16809539##33##], was the mRNA that showed the highest difference of expression between α6⁺/MHCI^-cells and α6⁺/MHCI⁺cells (97 fold).</p>", "<title>Uupregulation of transcripts encoding bone morphogenic factors in α6⁺/MHCI^-cells</title>", "<p>It has been shown that TGF-β and the bone morphogenic factors are upregulated in epidermal stem cells [##REF##14671312##9##,##REF##15339667##31##,##REF##16809539##33##]. Consequently, hair follicle stem cells are enriched with TGF-β/BMP targets [##REF##14671312##9##]. In accordance with those findings, we observed that transcripts of several genes whose products are necessary for the activation of TGF-β/phospho-Smad pathway, such as genes necessary for latent TGF-β activation (<italic>LTBP-1 </italic>and <italic>LTBP-2</italic>), secreted activators (<italic>BMP5, BMP8, BMP10, BMP15</italic>), and transcriptional activators of TGF-β responses (<italic>MADH3 </italic>and <italic>MADH6</italic>), were enriched in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells (Table ##TAB##0##1##). In addition, transcripts of forty-nine target genes shown to be upregulated by TGF-β/BMP pathway were more abundant in α6⁺/MHCI^-cells. Conversely, transcripts of the genes whose expression is shown to be suppressed by TGF-β/phospho-Smad pathway, including transcription factors <italic>c-Myc </italic>and <italic>KLF4</italic>, were more abundant in α6⁺/MHCI⁺cells (Figure ##FIG##2##3B##, Table ##TAB##1##2##). Since TGFβ/BMP pathway is tightly associated with stem cell quiescence [##REF##15339667##31##,##REF##17553962##47##,##REF##16960130##48##], upregulation of BMPs in α6⁺/MHCI^-cells might explain why these cells exhibit characteristics of quiescent cells.</p>", "<title>Transcripts of Wnt receptors and Wnt signaling inhibitors are enriched in α6⁺/MHCI^-cells</title>", "<p>Wnt pathway plays an important role in hair follicle morphogenesis and cycling [##REF##16824012##49##, ####REF##10498690##50##, ##REF##12646922##51####12646922##51##]. Researchers found that transcripts of several Wnt genes were downregulated in the mouse hair follicle bulge stem cells. In addition, higher levels of several genes that inhibit Wnt signaling pathway as well as higher levels of transcripts of the Wnt receptors were found in epidermal stem cells [##REF##14671312##9##,##REF##15024388##21##,##REF##16395407##32##,##REF##16809539##33##]. In the same cells, in general, targets of Wnt signaling (such as hair keratin, <italic>KRTHA1</italic>, nuclear proliferation antigen <italic>Ki67 </italic>[##REF##14671312##9##,##REF##16809539##33##,##REF##12646922##51##] are downregulated. Consistent with these observations, we found that <italic>WNT3 </italic>and <italic>WNT4 </italic>were downregulated, while Wnt receptors <italic>FZD1</italic>, <italic>FZD4</italic>, <italic>FZD7</italic>, and the inhibitors of Wnt signaling pathway, <italic>DAB2</italic>, <italic>TCF3</italic>, <italic>CTBP2</italic>, <italic>WIF1</italic>, <italic>DKK1</italic>, and <italic>DKK2 </italic>were upregulated in α6⁺/MHCI^-cells (Table ##TAB##0##1## and ##TAB##1##2##). Furthermore, transcripts of thirty eight genes that are known to be upregulated by Wnt signaling pathway were less abundant in α6⁺/MHCI^-cells, including <italic>MYCBP </italic>and type I hair keratin 1 (<italic>KRTHA1</italic>) (Figure ##FIG##2##3B## and Table ##TAB##1##2##). Also as expected to be found in stem cells, transcripts of genes that are downregulated by Wnt signaling pathway were found to be more abundant in α6⁺/MHCI^-cells. We found transcripts of fourteen such genes in α6⁺/MHCI^-cells (Table ##TAB##1##2##).</p>", "<title>Transcripts of the markers implicated in mammalian growth/differentiation are downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells</title>", "<p>Stem cells are the least differentiated cells in their tissue of origin; therefore, transcription factors and signaling pathways that induce differentiation are expected to be downregulated in these cells. As mentioned above, we found that <italic>MYC </italic>was less abundant in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells (see Additional files ##SUPPL##0##1## and ##SUPPL##5##6##). Upon screening of our microarray database for c-Myc targets, we found that the transcripts of sixty-one genes including <italic>MYCBP </italic>(Figure ##FIG##2##3B## and Additional file ##SUPPL##5##6##), which were reported previously to be upregulated by c-Myc, were upregulated in α6⁺/MHCI⁺cells. Conversely, transcripts of nineteen genes that were reported to be downregulated by c-Myc were upregulated in α6⁺/MHCI^-cells (see Additional file ##SUPPL##5##6##). Since many targets of c-Myc are involved in the ribosomal biogenesis, the downregulation of <italic>MYC </italic>may account for the observed downregulation of large numbers of genes that are involved in ribosomal biogenesis in α6⁺/MHCI^-cells (see Additional file ##SUPPL##0##1##), and may be an additional evidence that the downregulation of <italic>MYC </italic>in stem cells is related to their quiescent/undifferentiated state.</p>", "<p>Similar to <italic>c-MYC </italic>expression, the expression of <italic>KLF4 </italic>(Kruppel-like factor 4), a transcription factor that is mainly expressed in the differentiating layers of epidermis [##REF##10431239##52##], and <italic>BMP2</italic>, a member of bone morphogenetic protein family that is mainly expressed in proliferative basal and differentiated suprabasal keratinocytes [##REF##11788714##53##], were downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells (Table ##TAB##1##2## and Additional file ##SUPPL##0##1##).</p>", "<p>Notch1 signaling has been shown to stimulate differentiation in mammalian skin [##REF##14570040##54##]. Transcripts of the genes, such as fillagrin, integrin alpha 6, loricrin, (<italic>FLG, INTA6, LOR</italic>) whose expression is repressed by Nothch1 signaling [##REF##14570040##54##,##REF##17079689##55##] were more abundant in α6⁺/MHCI^-cells (Figure ##FIG##2##3A##, and Additional file ##SUPPL##0##1##). Furthermore, expression of <italic>KRT5</italic>, which is upregulated by Notch1 signaling [##REF##14570040##54##], was more abundant in α6⁺/MHCI⁺cells (Figure ##FIG##2##3A##, and see Additional file ##SUPPL##0##1##). We investigated whether the levels of Notch1 activity differ in α6⁺/MHCI^-and α6⁺/MHCI⁺cells. By using flow cytometry analysis with an antibody specific for cleaved/active Notch1, we could not detect any significant presence of cleaved/active Notch1 in α6⁺/MHCI^-cells contrary to α6⁺/MHCI⁺cells (Figure ##FIG##4##5##), which indicates that differentiation-inducing pathway Notch1 is downregulated/inhibited in α6⁺/MHCI^-cells.</p>", "<p>It has been shown that Notch1 signaling pathway activates NF-κB pathway and upregulates subunits of NF-κB and its targets [##REF##11591772##56##, ####REF##12107827##57##, ##REF##9528780##58####9528780##58##]. Thus, we investigated whether the levels of activity of NF-κB pathway, which is downstream of Notch1 pathway, differ in MHCI^-and MHCI⁺cells. First, we analyzed the expression of NF-κB subunit RelA in MHCI^-and MHCI⁺cells and found a positive correlation between the expression of NF-κB subunit RelA and MHCI (Figure ##FIG##5##6##). Since both MHCI molecules and NF-κB subunits are targets of NF-κB pathway, this result suggested that similar to the Notch1 pathway, NF-κB pathway is downregulated in MHCI^-cells.</p>", "<p>Search of our cDNA array data base for activators and targets of NF-κB pathway identified transcripts of genes, which either directly activate NF-κB pathway, or play a role in the activation of this pathway, such as <italic>BMP2 </italic>and <italic>MALT1 </italic>[##REF##10602461##59##], that were more abundant in α6⁺/MHCI⁺cells compared to α6⁺/MHCI^-cells. Furthermore, in accordance with this finding, transcripts of thirty genes that are reported to be upregulated by Rel/NF-κB transcription factors [##REF##10602461##59##], were enriched in α6⁺/MHCI⁺cells (Table ##TAB##1##2##). Thus, our microarray data suggest that the NF-κB pathway is downregulated in α6⁺/MHCI^-cells as compared to α6⁺/MHCI⁺cells. To determine whether NF-κB pathway is indeed more active in α6⁺/MHCI⁺cells, we performed TRANS AM NF-κB ELISA assay using nuclear extracts of α6⁺/MHCI⁺and α6⁺/MHCI^-cells. We found that the relative amount of nuclear phospho-NF-κB p50 bound to DNA, an indicator of an active NF-κB pathway, is higher in α6⁺/MHCI⁺cells than α6⁺/MHCI^-cells (Figure ##FIG##6##7##). Collectively, the data demonstrate that Notch1 signaling as well as the downstream NF-κB pathway are both downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells.</p>", "<title>The comparison of our data with the transcriptional profile of the genes that are differentially expressed in the basal and differentiating layers of the epidermis</title>", "<p>To gain further insights of epidermal differentiation, we compared our database with the published database of differentially expressed genes in basal and suprabasal layers of the human epidermis [##REF##16822832##60##]. The comparison of our data with the transcriptional profile of the genes that are differentially expressed in the basal and differentiated layers of the epidermis suggests that TGF-β/phospho-Smad pathway-induced transcription profile fades along the epidermal differentiation axis. The abundance of transcripts, which are upregulated by TGF-β/phospho-Smad pathway (such as <italic>COL6A1</italic>, <italic>LTBP2</italic>, <italic>MMP9</italic>, <italic>PLAT</italic>), decrease during differentiation, from presumptive stem to transient amplifying cells (basal layer) and further to cells of the suprabasal layers, where these transcripts are present at the lowest level. In addition, the transcript of <italic>MADH3</italic>, a transcriptional activator of TGF-β responses, is also increasingly downregulated. Conversely, abundance of transcripts, which are downregulated by TGF-β/phospho-Smad pathway (such as <italic>KLF4 </italic>and <italic>MYC</italic>), appear to positively correlate with the increase of epidermal differentiation (Figure ##FIG##7##8A##).</p>", "<p>On the other hand, the comparison suggests that Notch1 and Wnt pathway-induced transcription profiles strengthen along the epidermal differentiation axis. Transcripts of <italic>MYC </italic>and <italic>ELF1</italic>, which are upregulated by Wnt signaling gradually increase. These findings are in accordance with the previous reports, which demonstrate that TGF-β/phospho-Smad pathway prevents keratinocyte differentiation [##REF##16960130##48##], and, conversely, Notch1 and Wnt pathways induce keratinocyte differentiation [##REF##14570040##54##,##REF##11371349##61##,##REF##11432830##62##]. Thus, it might be possible that while TGF-β/phospho-Smad pathway is gradually downregulated, Wnt pathway becomes increasingly more active during epidermal differentiation. Nevertheless, further investigation is necessary to validate this hypothesis. Similarly, while the transcription of <italic>KRT1</italic>, which is upregulated by Notch1 signaling, increases, <italic>ITGA6 </italic>mRNA, which is downregulated by Notch1 signaling, decreases in keratinocytes during differentiation (Figure ##FIG##7##8A##). Several reports demonstrated that suprabasal cells have higher Notch1 activity than basal cells [##REF##17079689##55##,##REF##14597207##63##]. As already mentioned, employing flow cytometry analysis and antibodies against cleaved/active Notch 1, no detectable levels of cleaved/active Notch 1 was observed in MHCI^-cells indicating the lack of Notch1 activity (Figure ##FIG##4##5##).</p>", "<p>Interestingly, previous reports suggested that strong adherence of stem cells to extracellular matrix-rich basement membrane may be involved in retaining these cells in their natural residence (niche) [##REF##10431239##52##,##REF##12756224##64##]. In accordance with these observations, the comparison of our data with published transcription profiles of the basal and suprabasal cells of the epidermis revealed that during differentiation transcripts of several genes (<italic>MUC1</italic>, <italic>COL6A1</italic>, <italic>ITGA6</italic>, <italic>MMP9</italic>, <italic>PLAT</italic>, <italic>CEACAM1</italic>) whose products are soluble or membrane-bound factors that play a role in the interaction of cells with the microenvironment, decrease gradually during differentiation (Figure ##FIG##7##8A##).</p>", "<p>We also found that twenty-one genes were upregulated in α6⁺/MHCI⁺cells (TA cells) alone, and later become downregulated during terminal differentiation (Figure ##FIG##7##8B##). Among these transcripts there are ones related to cell cycle (<italic>CCNB1</italic>, <italic>CCND2</italic>, <italic>RFC5</italic>) as well as mRNAs of the genes whose products induce cell growth and division (<italic>ZWINT</italic>, <italic>BLCAP</italic>). Since α6⁺/MHCI^-cells are quiescent, and terminally differentiating cells are post-mitotic it is not surprising to find transcripts whose products accelerates cell proliferation among the genes that are upregulated only in α6⁺/MHCI⁺cells during epidermal differentiation. Similarly, genes whose products suppress cell growth and proliferation were enriched both in α6⁺/MHCI^-cells and terminally differentiating suprabasal cells, such as <italic>PLAGL1</italic>, a zinc finger transcription factor that induces cell cycle arrest in the skin and whose expression is diminished in basal cell carcinomas [##REF##16179495##65##], putative tumor suppressors insulin-like growth factor-binding protein 7 (<italic>IGFBP7</italic>), and <italic>DOC1 </italic>[##REF##16428440##66##,##REF##17312390##67##] (see Additional file ##SUPPL##0##1## and reference [##REF##16822832##60##]).</p>" ]

[ "<title>Results and discussion</title>", "<p>Skin is the largest and most accessible organ in the body. The differentiation axis of the interfollicular epidermis is spatially well defined: the basal layer contains proliferating cells, while suprabasal layers, stratum spinosum, stratum granulosum, and stratum corneum harbor post-mitotic, differentiating keratinocytes [##UREF##0##23##,##REF##2269655##24##]. These features facilitate the analysis of cells at the specific differentiation stage. Like all self-renewing tissues, epidermis contains stem cells, which are located in the stratum basale. Several proteins have been suggested as markers for keratinocyte stem cell enrichment [##REF##8500165##10##, ####REF##9520465##11##, ##REF##11851883##12##, ##REF##12042327##13##, ##REF##12953062##14####12953062##14##]. We have previously described a basal keratinocyte population that lacks gap junction protein Cx43 in human and mouse epidermis [##REF##11851883##12##]. We have shown that Cx43 negative cells co-localize with label-retaining cells, hair follicle bulge stem cells [##REF##2364430##6##, ####REF##10201531##7##, ##REF##10966107##8####10966107##8##]. Cx43 negative keratinocytes comprise about 10% of human basal keratinocytes and are blast like, small and have low granularity as determined by flow cytometry. Cells in the limbus of the eye, the region of the corneal epithelium that contains stem cells, were also shown to lack Cx43 [##REF##9203348##25##]. In searching for additional markers that can be used to obtain viable cells, we isolated a subset of Cx43 negative keratinocytes characterized by low/negative expression of MHCI that comprised up to 2% of basal epidermal cells [##REF##16360835##22##]. It was believed that almost all nucleated cells express MHCI [##REF##4139789##26##]. Recently, however, stem cells of several tissues were shown to lack MHCI expression [##REF##12114532##15##, ####REF##12832694##16##, ##REF##12077603##17##, ##REF##12738887##18##, ##REF##15277692##19##, ##REF##9282991##20##, ##REF##15024388##21####15024388##21##]. Molecules encoded by MHC are involved in self/non-self discrimination in vertebrates. MHCI molecules bind endogenously derived peptides and stimulate a distinct branch of the adaptive immune system mediated by CD8⁺T cells. The human MHC termed HLA (Human Leukocyte Antigen) encodes three classical polymorphic class I genes: HLA A, B, and C. To isolate transient amplifying cells (α6⁺/MHCI⁺) and presumptive stem cells (α6⁺/MHCI^-), we used antibodies against α6 integrin, a basal cell marker, in combination with antibodies against β2 microglobulin, the light chain of MHCI molecule. Previously we have shown that similar results were obtained regardless of whether antibodies to MHCI heavy chain, or antibodies against β2 microglobulin were used [##REF##16360835##22##].</p>", "<title>α6⁺/MHCI^-cells are quiescent <italic>in situ</italic>, yet in culture display higher proliferative potential</title>", "<p>During tissue homeostasis stem cells are infrequently dividing; thus, and one of the characteristics of stem cells is their quiescence <italic>in situ</italic>. To determine the proliferative status of MHCI^-and MHCI⁺populations, we analyzed the expression of nuclear proliferation antigen Ki67, which is a marker for actively cycling cells. The data presented reflect keratinocyte proliferation since in normal epidermis, non keratinocytes are found to be non-cycling [##REF##808553##27##,##REF##8919043##28##]. Although the absolute values of Ki67 may vary depending on the total number of gated cells, the ratio of Ki67 expressing MHCI^-and MHCI⁺cells is held constant. Flow cytometric analysis showed that MHCI⁺cells expressed more than four time higher levels of Ki67 than MHCI^-cells (Figure ##FIG##0##1##). Only 0.9% of MHCI^-cells expressed Ki67, while 3.9% of MHCI⁺cells were in the cell cycle. Given the low percentages of MHCI^-cells in the basal layer, it is clear that the bulk of cell production in the epidermis is accomplished through divisions of transient amplifying cells, a finding which is in accordance with the established view of epidermal homeostasis.</p>", "<p>High proliferative potential is another feature of stem cells. One way of assessing proliferative potential of keratinocytes is to analyze their colony forming efficiency (CFE) [##REF##8500165##10##]. To assess proliferative potential of the two cell populations' colony forming efficiency (CFE) was analyzed in the primary and secondary cultures. Sorted cells were seeded at clonal densities and colonies formed evaluated after two weeks. In primary cultures, α6⁺/MHCI^-cells exhibited lower colony forming efficiency than α6⁺/MHCI⁺cells (Figure ##FIG##1##2##). However, in secondary cultures α6⁺/MHCI^-showed higher CFE than α6⁺/MHCI⁺cells. From primary to secondary culture, an increase of 38 times was observed in the CFE of α6⁺/MHCI^-cells, while the CFE of α6⁺/MHCI⁺cells increased only 3.6 times (Figure ##FIG##1##2##). Previously, lower initial CFE was observed in limbal epithelial stem cell [##REF##16150918##29##]. Our data indicate that α6⁺/MHCI^-cells have higher proliferative potential than α6⁺/MHCI⁺cells, another characteristic attributed to stem cells. The majority of available data regarding CFE are from secondary or tertiary keratinocyte cultures [##REF##8500165##10##,##REF##2436229##30##]. In the secondary cultures, α6⁺/MHCI^-cells did display higher CFE than α6⁺/MHCI⁺cells, which is an indication of a higher proliferative potential of the α6⁺/MHCI^-cells [##REF##8500165##10##]. The lower CFE of α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells that we recorded in our primary cultures may be due to the time needed for stem cells to transit from a quiescent to a proliferative state.</p>", "<title>Gene expression profile indicates that α6⁺/MHCI^-cells exhibit properties of stem cells</title>", "<p>Microarray profiles of stem cells and their progeny provide a global view into differences of expression of a large number of genes and enable analyses of molecular processes involved in self renewal, proliferation and differentiation. Gene expression profiles of hair follicle bulge stem cells were recently reported [##REF##14671312##9##,##REF##15024388##21##,##REF##15339667##31##, ####REF##16395407##32##, ##REF##16809539##33####16809539##33##], yet until now no data are available with regard to human interfollicular keratinocyte stem cells. We report on transcriptional profiles of putative human keratinocyte stem cells and their immediate progeny, transient amplifying cells. Global gene expression profile was obtained from sorted α6⁺/MHCI^-cells and α6⁺/MHCI⁺cells using DNA microarray chips. We identified a comprehensive list of differentially expressed genes. Notably, all of the MHCI genes were downregulated in α6⁺/MHCI^-cells, thus confirming the successful separation of α6⁺/MHCI⁺and α6⁺/MHCI^-cells. The data also show that expression of MHCI proteins in keratinocytes is regulated at the transcriptional level. The HLA-E transcript is downregulated in α6⁺/MHCI^-cells confirming our previous results obtained at the protein level [##REF##16360835##22##]. The expression of non-classical HLA molecules is thought to protect cells that lack classical HLA expression from lysis by NK cells. At present, it is not known what mechanisms protect presumptive stem cells, MHCI^-cells, from attack by NK cells, especially since MHCI^-cells do not express detectable levels of non-classical HLA-E and HLA-G molecules [##REF##16360835##22##].</p>", "<p>We found that most of the mRNAs of genes encoding cellular receptors and other cell surface molecules were more abundant in α6⁺/MHCI^-cells than in α6⁺/MHCI⁺cells (see Additional file ##SUPPL##0##1##). Conversely, mRNAs of genes encoding proteins that take part in ribosome biosynthesis, RNA splicing, translation, protein degradation, and energy metabolism were more abundant in α6⁺/MHCI⁺cells (see Additional file ##SUPPL##0##1##). These findings are consistent with reports by other investigators who demonstrated that stem cells are characterized by few ribosomes and mitochondria (features related to undifferentiated state of stem cells) but contain a large numbers of receptors [##REF##14671312##9##,##REF##16051216##34##].</p>", "<p>In reference to stem cells quiescence, it should also be noted that transcripts of <italic>CDKN1C </italic>and <italic>CDKN2A </italic>whose products are cyclin-dependent kinase inhibitors were enriched in α6⁺/MHCI^-cells (Figure ##FIG##2##3A##). The product of <italic>CDKN1C</italic>, p57^Kip2, one of the Cip/Kip family members, is tightly associated with inhibition of proliferation of human interfollicular keratinocytes [##REF##17112701##35##], and is upregulated in hair follicle bulge [##REF##15339667##31##], while the product of <italic>CDKN2A</italic>, a 16 kDa protein p16^INK4aimposes a G1 cell cycle arrest [##REF##9823374##36##,##REF##7758941##37##]. Furthermore, in addition to the low protein level of Ki67 observed in MHCI negative population (Figure ##FIG##0##1##), the transcript of Ki67 was less abundant in α6⁺/MHCI^-cells (Table ##TAB##0##1##, and Additional file ##SUPPL##0##1##). Conversely, transcripts of genes encoding proteins that are related to cell proliferation such as cyclins, proteins involved in chromosome remodeling, DNA replication and repair were preferentially enriched in α6⁺/MHCI⁺cells (Table ##TAB##0##1##, and Additional file ##SUPPL##0##1##).</p>", "<p>Interestingly, type I IFN (IFN-α, and IFN-β) has been shown to inhibit cell proliferation by inducing G1 cell cycle arrest. It has been reported that interferon α (IFN-α) has antiproliferative effects on bone marrow stromal precursors, hepatic progenitor cells, and mesenchymal stem cells [##REF##17375123##39##, ####REF##16628647##40##, ##REF##1533991##41####1533991##41##]. We observed enrichment of transcripts of the IFN-α family of proteins in α6⁺/MHCI^-cells as well as <italic>STAT2</italic>, the specific transducing activator of IFN-α transcription. This is the first report that suggests involvement of type I IFN in epidermal stem cell quiescence. Further studies are needed to determine whether IFN-α is synthesized by α6⁺/MHCI^-cells, whether its pathway is active and whether it contributes to α6⁺/MHCI^-cell quiescence (Table ##TAB##0##1##).</p>", "<p>It has been reported that the components of the inositol phospholipid signaling system are present and that the system itself is active in murine embryonic stem cells [##REF##16845989##42##]. In the present study, we demonstrate that several components of the inositol phospholipid signaling system are enriched in α6⁺/MHCI^-cells (Table ##TAB##0##1##). Nevertheless, the functional significance of this observation needs further investigation.</p>", "<p>Among the transcripts enriched in α6⁺/MHCI^-cells, there were transcripts previously shown to be upregulated in cell population enriched for human interfollicular stem cells, such as α6 integrin [##REF##9520465##11##], melanoma-associated chondroitin sulfate proteoglycan [##REF##16877544##43##], phosphorylase kinase α2 [##REF##16877544##43##], and the transcript of p53 homolog <italic>p51/p73L/p63/p40 </italic>[##REF##11248048##44##] (Figure ##FIG##2##3A## and Additional file ##SUPPL##0##1##). Using flow cytometry we found that MHCI^-cells expressed lower level of transferrin receptor (<italic>CD71</italic>), a negative marker used to enrich for interfollicular epidermal stem cells [##REF##9520465##11##], when compared to MHCI⁺cells (Figure ##FIG##3##4##). It should also be noted that although differences in the expression of <italic>CD71 </italic>transcript were lower than 2 fold, <italic>CD71 </italic>mRNA was downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells consistently in both arrays (see Additional file ##SUPPL##1##2##).</p>", "<p>Among the transcripts enriched in α6⁺/MHCI⁺cells, there were mRNAs of genes whose products are expressed at low levels in cell population enriched for human interfollicular stem cells, such as desmosomal proteins including desmoglein 3 (<italic>DSG3</italic>) [##REF##12953062##14##], as well as the proliferation associated transcription factor <italic>c-Myc</italic>, found to be expressed at the lower levels in cultured human interfollicular stem cells [##REF##9353256##45##] (Figures ##FIG##2##3A, 3B## and Additional file ##SUPPL##0##1##). Moreover, it has been reported that epidermal growth factor receptor (EGFR) signaling is downregulated in putative human interfollicular stem cells [##REF##16877544##43##]. In accordance with this observation, we found that <italic>EGFR </italic>itself was downregulated in α6⁺/MHCI^-cells (Fig. ##FIG##2##3A## and Additional file ##SUPPL##0##1##).</p>", "<p>Most notably, we present results that demonstrate that putative stem cells have lower expression of mRNAs encoding proteins that take part in energy metabolism, which can explain how stem cells can be quiescent and at the same time maintain small size (see Additional file ##SUPPL##0##1##).</p>", "<title>Comparison of α6⁺/MHCI^-and α6⁺/MHCI⁺microarray databases with the transcriptional profiles of human and mouse hair follicle stem cells</title>", "<p>We compared α6⁺/MHCI^-transcriptional profile with the published transcriptional profile of human hair follicle stem cells [##REF##16395407##32##] and with four different sets of data of murine hair follicle stem cells [##REF##14671312##9##,##REF##15024388##21##,##REF##15339667##31##,##REF##16809539##33##]. We found that eleven genes, which were enriched in human hair follicle stem cells, were also enriched in α6⁺/MHCI^-cells (Table ##TAB##0##1##). Similarly, mRNAs of nine genes, which were downregulated in human hair follicle stem cells, were also downregulated in α6⁺/MHCI^-cells (Table ##TAB##1##2##). Nevertheless, the differences between the gene expression profiles of stem cells from the two human tissues, i.e. interfollicular and follicular, were also observed including the genes that demonstrated two fold difference in the expression (total of nine genes) (see Additional file ##SUPPL##2##3##).</p>", "<p>mRNAs of eighty-two genes, which were enriched in murine hair follicle stem cells, were also enriched in α6⁺/MHCI^-cells, while mRNAs of forty-one genes, which were downregulated in murine hair follicle stem cells, were also downregulated in α6⁺/MHCI^-cells (Tables ##TAB##0##1##, ##TAB##1##2## and Additional file ##SUPPL##3##4##). Transcription factors <italic>LHX2 </italic>and <italic>TCF3 </italic>that were shown to maintain SC features [##REF##16809539##33##,##REF##17018284##46##], were among the genes that were upregulated in α6⁺/MHCI^-cells as well as in murine hair follicle stem cells. Upon screening of our microarray database for <italic>TCF3 </italic>targets [##REF##17018284##46##], we found that transcripts of twelve genes reported to be upregulated by <italic>TCF3 </italic>were more abundant in α6⁺/MHCI^-cells and conversely transcripts of nine genes repressed by <italic>TCF3 </italic>were more abundant in α6⁺/MHCI⁺cells (see Additional file ##SUPPL##4##5##). Interestingly, both arrays that we performed showed that <italic>LHX2 </italic>was among the most upregulated mRNAs in α6⁺/MHCI^-cells, while <italic>WIF-1 </italic>mRNA, which was the most enriched mRNA in murine hair follicle stem cells according to one report [##REF##16809539##33##], was the mRNA that showed the highest difference of expression between α6⁺/MHCI^-cells and α6⁺/MHCI⁺cells (97 fold).</p>", "<title>Uupregulation of transcripts encoding bone morphogenic factors in α6⁺/MHCI^-cells</title>", "<p>It has been shown that TGF-β and the bone morphogenic factors are upregulated in epidermal stem cells [##REF##14671312##9##,##REF##15339667##31##,##REF##16809539##33##]. Consequently, hair follicle stem cells are enriched with TGF-β/BMP targets [##REF##14671312##9##]. In accordance with those findings, we observed that transcripts of several genes whose products are necessary for the activation of TGF-β/phospho-Smad pathway, such as genes necessary for latent TGF-β activation (<italic>LTBP-1 </italic>and <italic>LTBP-2</italic>), secreted activators (<italic>BMP5, BMP8, BMP10, BMP15</italic>), and transcriptional activators of TGF-β responses (<italic>MADH3 </italic>and <italic>MADH6</italic>), were enriched in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells (Table ##TAB##0##1##). In addition, transcripts of forty-nine target genes shown to be upregulated by TGF-β/BMP pathway were more abundant in α6⁺/MHCI^-cells. Conversely, transcripts of the genes whose expression is shown to be suppressed by TGF-β/phospho-Smad pathway, including transcription factors <italic>c-Myc </italic>and <italic>KLF4</italic>, were more abundant in α6⁺/MHCI⁺cells (Figure ##FIG##2##3B##, Table ##TAB##1##2##). Since TGFβ/BMP pathway is tightly associated with stem cell quiescence [##REF##15339667##31##,##REF##17553962##47##,##REF##16960130##48##], upregulation of BMPs in α6⁺/MHCI^-cells might explain why these cells exhibit characteristics of quiescent cells.</p>", "<title>Transcripts of Wnt receptors and Wnt signaling inhibitors are enriched in α6⁺/MHCI^-cells</title>", "<p>Wnt pathway plays an important role in hair follicle morphogenesis and cycling [##REF##16824012##49##, ####REF##10498690##50##, ##REF##12646922##51####12646922##51##]. Researchers found that transcripts of several Wnt genes were downregulated in the mouse hair follicle bulge stem cells. In addition, higher levels of several genes that inhibit Wnt signaling pathway as well as higher levels of transcripts of the Wnt receptors were found in epidermal stem cells [##REF##14671312##9##,##REF##15024388##21##,##REF##16395407##32##,##REF##16809539##33##]. In the same cells, in general, targets of Wnt signaling (such as hair keratin, <italic>KRTHA1</italic>, nuclear proliferation antigen <italic>Ki67 </italic>[##REF##14671312##9##,##REF##16809539##33##,##REF##12646922##51##] are downregulated. Consistent with these observations, we found that <italic>WNT3 </italic>and <italic>WNT4 </italic>were downregulated, while Wnt receptors <italic>FZD1</italic>, <italic>FZD4</italic>, <italic>FZD7</italic>, and the inhibitors of Wnt signaling pathway, <italic>DAB2</italic>, <italic>TCF3</italic>, <italic>CTBP2</italic>, <italic>WIF1</italic>, <italic>DKK1</italic>, and <italic>DKK2 </italic>were upregulated in α6⁺/MHCI^-cells (Table ##TAB##0##1## and ##TAB##1##2##). Furthermore, transcripts of thirty eight genes that are known to be upregulated by Wnt signaling pathway were less abundant in α6⁺/MHCI^-cells, including <italic>MYCBP </italic>and type I hair keratin 1 (<italic>KRTHA1</italic>) (Figure ##FIG##2##3B## and Table ##TAB##1##2##). Also as expected to be found in stem cells, transcripts of genes that are downregulated by Wnt signaling pathway were found to be more abundant in α6⁺/MHCI^-cells. We found transcripts of fourteen such genes in α6⁺/MHCI^-cells (Table ##TAB##1##2##).</p>", "<title>Transcripts of the markers implicated in mammalian growth/differentiation are downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells</title>", "<p>Stem cells are the least differentiated cells in their tissue of origin; therefore, transcription factors and signaling pathways that induce differentiation are expected to be downregulated in these cells. As mentioned above, we found that <italic>MYC </italic>was less abundant in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells (see Additional files ##SUPPL##0##1## and ##SUPPL##5##6##). Upon screening of our microarray database for c-Myc targets, we found that the transcripts of sixty-one genes including <italic>MYCBP </italic>(Figure ##FIG##2##3B## and Additional file ##SUPPL##5##6##), which were reported previously to be upregulated by c-Myc, were upregulated in α6⁺/MHCI⁺cells. Conversely, transcripts of nineteen genes that were reported to be downregulated by c-Myc were upregulated in α6⁺/MHCI^-cells (see Additional file ##SUPPL##5##6##). Since many targets of c-Myc are involved in the ribosomal biogenesis, the downregulation of <italic>MYC </italic>may account for the observed downregulation of large numbers of genes that are involved in ribosomal biogenesis in α6⁺/MHCI^-cells (see Additional file ##SUPPL##0##1##), and may be an additional evidence that the downregulation of <italic>MYC </italic>in stem cells is related to their quiescent/undifferentiated state.</p>", "<p>Similar to <italic>c-MYC </italic>expression, the expression of <italic>KLF4 </italic>(Kruppel-like factor 4), a transcription factor that is mainly expressed in the differentiating layers of epidermis [##REF##10431239##52##], and <italic>BMP2</italic>, a member of bone morphogenetic protein family that is mainly expressed in proliferative basal and differentiated suprabasal keratinocytes [##REF##11788714##53##], were downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells (Table ##TAB##1##2## and Additional file ##SUPPL##0##1##).</p>", "<p>Notch1 signaling has been shown to stimulate differentiation in mammalian skin [##REF##14570040##54##]. Transcripts of the genes, such as fillagrin, integrin alpha 6, loricrin, (<italic>FLG, INTA6, LOR</italic>) whose expression is repressed by Nothch1 signaling [##REF##14570040##54##,##REF##17079689##55##] were more abundant in α6⁺/MHCI^-cells (Figure ##FIG##2##3A##, and Additional file ##SUPPL##0##1##). Furthermore, expression of <italic>KRT5</italic>, which is upregulated by Notch1 signaling [##REF##14570040##54##], was more abundant in α6⁺/MHCI⁺cells (Figure ##FIG##2##3A##, and see Additional file ##SUPPL##0##1##). We investigated whether the levels of Notch1 activity differ in α6⁺/MHCI^-and α6⁺/MHCI⁺cells. By using flow cytometry analysis with an antibody specific for cleaved/active Notch1, we could not detect any significant presence of cleaved/active Notch1 in α6⁺/MHCI^-cells contrary to α6⁺/MHCI⁺cells (Figure ##FIG##4##5##), which indicates that differentiation-inducing pathway Notch1 is downregulated/inhibited in α6⁺/MHCI^-cells.</p>", "<p>It has been shown that Notch1 signaling pathway activates NF-κB pathway and upregulates subunits of NF-κB and its targets [##REF##11591772##56##, ####REF##12107827##57##, ##REF##9528780##58####9528780##58##]. Thus, we investigated whether the levels of activity of NF-κB pathway, which is downstream of Notch1 pathway, differ in MHCI^-and MHCI⁺cells. First, we analyzed the expression of NF-κB subunit RelA in MHCI^-and MHCI⁺cells and found a positive correlation between the expression of NF-κB subunit RelA and MHCI (Figure ##FIG##5##6##). Since both MHCI molecules and NF-κB subunits are targets of NF-κB pathway, this result suggested that similar to the Notch1 pathway, NF-κB pathway is downregulated in MHCI^-cells.</p>", "<p>Search of our cDNA array data base for activators and targets of NF-κB pathway identified transcripts of genes, which either directly activate NF-κB pathway, or play a role in the activation of this pathway, such as <italic>BMP2 </italic>and <italic>MALT1 </italic>[##REF##10602461##59##], that were more abundant in α6⁺/MHCI⁺cells compared to α6⁺/MHCI^-cells. Furthermore, in accordance with this finding, transcripts of thirty genes that are reported to be upregulated by Rel/NF-κB transcription factors [##REF##10602461##59##], were enriched in α6⁺/MHCI⁺cells (Table ##TAB##1##2##). Thus, our microarray data suggest that the NF-κB pathway is downregulated in α6⁺/MHCI^-cells as compared to α6⁺/MHCI⁺cells. To determine whether NF-κB pathway is indeed more active in α6⁺/MHCI⁺cells, we performed TRANS AM NF-κB ELISA assay using nuclear extracts of α6⁺/MHCI⁺and α6⁺/MHCI^-cells. We found that the relative amount of nuclear phospho-NF-κB p50 bound to DNA, an indicator of an active NF-κB pathway, is higher in α6⁺/MHCI⁺cells than α6⁺/MHCI^-cells (Figure ##FIG##6##7##). Collectively, the data demonstrate that Notch1 signaling as well as the downstream NF-κB pathway are both downregulated in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells.</p>", "<title>The comparison of our data with the transcriptional profile of the genes that are differentially expressed in the basal and differentiating layers of the epidermis</title>", "<p>To gain further insights of epidermal differentiation, we compared our database with the published database of differentially expressed genes in basal and suprabasal layers of the human epidermis [##REF##16822832##60##]. The comparison of our data with the transcriptional profile of the genes that are differentially expressed in the basal and differentiated layers of the epidermis suggests that TGF-β/phospho-Smad pathway-induced transcription profile fades along the epidermal differentiation axis. The abundance of transcripts, which are upregulated by TGF-β/phospho-Smad pathway (such as <italic>COL6A1</italic>, <italic>LTBP2</italic>, <italic>MMP9</italic>, <italic>PLAT</italic>), decrease during differentiation, from presumptive stem to transient amplifying cells (basal layer) and further to cells of the suprabasal layers, where these transcripts are present at the lowest level. In addition, the transcript of <italic>MADH3</italic>, a transcriptional activator of TGF-β responses, is also increasingly downregulated. Conversely, abundance of transcripts, which are downregulated by TGF-β/phospho-Smad pathway (such as <italic>KLF4 </italic>and <italic>MYC</italic>), appear to positively correlate with the increase of epidermal differentiation (Figure ##FIG##7##8A##).</p>", "<p>On the other hand, the comparison suggests that Notch1 and Wnt pathway-induced transcription profiles strengthen along the epidermal differentiation axis. Transcripts of <italic>MYC </italic>and <italic>ELF1</italic>, which are upregulated by Wnt signaling gradually increase. These findings are in accordance with the previous reports, which demonstrate that TGF-β/phospho-Smad pathway prevents keratinocyte differentiation [##REF##16960130##48##], and, conversely, Notch1 and Wnt pathways induce keratinocyte differentiation [##REF##14570040##54##,##REF##11371349##61##,##REF##11432830##62##]. Thus, it might be possible that while TGF-β/phospho-Smad pathway is gradually downregulated, Wnt pathway becomes increasingly more active during epidermal differentiation. Nevertheless, further investigation is necessary to validate this hypothesis. Similarly, while the transcription of <italic>KRT1</italic>, which is upregulated by Notch1 signaling, increases, <italic>ITGA6 </italic>mRNA, which is downregulated by Notch1 signaling, decreases in keratinocytes during differentiation (Figure ##FIG##7##8A##). Several reports demonstrated that suprabasal cells have higher Notch1 activity than basal cells [##REF##17079689##55##,##REF##14597207##63##]. As already mentioned, employing flow cytometry analysis and antibodies against cleaved/active Notch 1, no detectable levels of cleaved/active Notch 1 was observed in MHCI^-cells indicating the lack of Notch1 activity (Figure ##FIG##4##5##).</p>", "<p>Interestingly, previous reports suggested that strong adherence of stem cells to extracellular matrix-rich basement membrane may be involved in retaining these cells in their natural residence (niche) [##REF##10431239##52##,##REF##12756224##64##]. In accordance with these observations, the comparison of our data with published transcription profiles of the basal and suprabasal cells of the epidermis revealed that during differentiation transcripts of several genes (<italic>MUC1</italic>, <italic>COL6A1</italic>, <italic>ITGA6</italic>, <italic>MMP9</italic>, <italic>PLAT</italic>, <italic>CEACAM1</italic>) whose products are soluble or membrane-bound factors that play a role in the interaction of cells with the microenvironment, decrease gradually during differentiation (Figure ##FIG##7##8A##).</p>", "<p>We also found that twenty-one genes were upregulated in α6⁺/MHCI⁺cells (TA cells) alone, and later become downregulated during terminal differentiation (Figure ##FIG##7##8B##). Among these transcripts there are ones related to cell cycle (<italic>CCNB1</italic>, <italic>CCND2</italic>, <italic>RFC5</italic>) as well as mRNAs of the genes whose products induce cell growth and division (<italic>ZWINT</italic>, <italic>BLCAP</italic>). Since α6⁺/MHCI^-cells are quiescent, and terminally differentiating cells are post-mitotic it is not surprising to find transcripts whose products accelerates cell proliferation among the genes that are upregulated only in α6⁺/MHCI⁺cells during epidermal differentiation. Similarly, genes whose products suppress cell growth and proliferation were enriched both in α6⁺/MHCI^-cells and terminally differentiating suprabasal cells, such as <italic>PLAGL1</italic>, a zinc finger transcription factor that induces cell cycle arrest in the skin and whose expression is diminished in basal cell carcinomas [##REF##16179495##65##], putative tumor suppressors insulin-like growth factor-binding protein 7 (<italic>IGFBP7</italic>), and <italic>DOC1 </italic>[##REF##16428440##66##,##REF##17312390##67##] (see Additional file ##SUPPL##0##1## and reference [##REF##16822832##60##]).</p>" ]

[ "<title>Conclusion</title>", "<p>Most of our knowledge regarding epidermal stem cells comes from murine studies. This is the first report that uncovers the transcriptional profile of human interfollicular epidermal stem cells and their progeny, transient amplifying cells isolated directly from their niches and analyzed.</p>", "<p>In summary, the results presented here show that α6⁺/MHCI^-cells exhibit characteristics attributed to stem cells. Comparison of the transcription profiles of α6⁺/MHCI^-cells and α6⁺/MHCI⁺cells with the existing profiles of hair follicle bulge stem cells further indicate that α6⁺/MHCI^-cells are enriched for stem cells. Our findings may bring new insights into regulatory mechanisms involved in epidermal homeostasis, and bring understanding of deregulations of these mechanisms that take place in skin disorders including cancer, and most importantly may lead to identification of potential therapeutic targets. In addition, as a first comprehensive gene expression profile of putative human epithelial cells isolated directly from tissue, the generated database may be of importance for studies of gene expression profiles of other human epithelial tissues. By defining characteristics of interfollicular epidermal stem cells and by identifying genes whose expression is altered during differentiation, we have opened new roads for better understanding of stem cell characteristics and epidermal differentiation.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Human interfollicular epidermis is sustained by the proliferation of stem cells and their progeny, transient amplifying cells. Molecular characterization of these two cell populations is essential for better understanding of self renewal, differentiation and mechanisms of skin pathogenesis. The purpose of this study was to obtain gene expression profiles of alpha 6⁺/MHCI⁺, transient amplifying cells and alpha 6⁺/MHCI^-, putative stem cells, and to compare them with existing data bases of gene expression profiles of hair follicle stem cells. The expression of Major Histocompatibility Complex (MHC) class I, previously shown to be absent in stem cells in several tissues, and alpha 6 integrin were used to isolate MHCI positive basal cells, and MHCI low/negative basal cells.</p>", "<title>Results</title>", "<p>Transcriptional profiles of the two cell populations were determined and comparisons made with published data for hair follicle stem cell gene expression profiles. We demonstrate that presumptive interfollicular stem cells, alpha 6⁺/MHCI^-cells, are enriched in messenger RNAs encoding surface receptors, cell adhesion molecules, extracellular matrix proteins, transcripts encoding members of IFN-alpha family proteins and components of IFN signaling, but contain lower levels of transcripts encoding proteins which take part in energy metabolism, cell cycle, ribosome biosynthesis, splicing, protein translation, degradation, DNA replication, repair, and chromosome remodeling. Furthermore, our data indicate that the cell signaling pathways Notch1 and NF-κB are downregulated/inhibited in MHC negative basal cells.</p>", "<title>Conclusion</title>", "<p>This study demonstrates that alpha 6⁺/MHCI^-cells have additional characteristics attributed to stem cells. Moreover, the transcription profile of alpha 6⁺/MHCI^-cells shows similarities to transcription profiles of mouse hair follicle bulge cells known to be enriched for stem cells. Collectively, our data suggests that alpha 6⁺/MHCI^-cells may be enriched for stem cells. This study is the first comprehensive gene expression profile of putative human epithelial stem cells and their progeny that were isolated directly from neonatal foreskin tissue. Our study is important for understanding self renewal and differentiation of epidermal stem cells, and for elucidating signaling pathways involved in those processes. The generated data base may serve those working with other human epithelial tissue progenitors.</p>" ]

[ "<title>Abbreviations</title>", "<p>α6: integrin alpha 6; BMP: Bone Morphogenic Protein; CDKN: cyclin dependent kinase inhibitor; CFE: colony forming efficiency; IFN-α: interferon alpha; KRT: keratin; MHCI: Major Histocompatibility Complex I; NF-κB: Nuclear Factor kappa B; α6⁺/MHCI^-; α6⁺/MHCI⁺denote keratinocytes sorted according to integrin alpha 6 and MHCI expressions. MHCI^-and MHCI⁺denote total epidermal cells that express or do not express MHCI.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>MM designed and organized the study, performed cell isolation and labeling for flow cytometry analysis and FACS sorting, isolated RNA for microarray analysis, participated in data analysis, manuscript drafting and revision. SSK analyzed the microarray data, performed RT-PCR reactions and NF-κB determination in isolated nuclei, prepared figures and manuscript draft and was involved in organization of the study and manuscript revision. SRS participated in the design and organization of the study, and manuscript revission. PMD participated in the analysis of the microarray data and manuscript revision. MFB participated in the analysis of the microarray data.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p><bold>Funding</bold>: This research was supported by grants USPHS/NIH/NIAMS K0-1 AR02079 (MM) and R03 AR49936 (MM).</p>", "<p>We thank members of Dr. Simon's lab, Microarray Core Facility at SUNY at Stony Brook, Microarray Core Facility at Cold Spring Harbor Laboratory, and Flow Cytometry Core Facility at SUNY at Stony Brook for their help with this project.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>MHCI negative cells express low levels of Ki67</bold>. (A) A representative flow cytometric analysis of the expression of proliferation antigen Ki67 in MHCI positive, and MHCI negative cells. Gates were set using isotype control antibodies and single color control antibodies. In this experiment 3.9% of MHCI positive cells express Ki67, while only 0.9% of MHCI negative cells express Ki67. Although the exact values of proliferating population may vary from experiment to experiment the ratio of MHCI positive and MHCI negative proliferating populations stay constant. This result demonstrate quiescent nature of MHCI negative cells. (B) Single positive isotype control for PE. (C) Single positive control for FITC. (D) Secondary control.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Comparison of colony forming efficiencies of α6⁺/MHCI⁺and α6⁺/MHCI^-cells</bold>. A representative experiment of the colony forming efficiency (CFE) of sorted α6⁺/MHCI⁺cells and α6⁺/MHCI^-cells cultured on 3T3 fibroblast feeder layer. Primary culture of α6⁺/MHCI^-cells exhibits lower CFE than α6⁺/MHCI⁺cells. However, secondary culture of α6⁺/MHCI^-cells exhibits higher CFE than α6⁺/MHCI⁺cells (*<italic>P </italic>&lt; 0.001). The CFE of α6⁺/MHCI⁺cells increased ~3.6× from primary to secondary culture, while it increased ~38× for α6⁺/MHCI^-. These results indicate higher proliferative potential of α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells. Cells directly sorted by FACS (from niche, i.e. primary cultures) were seeded at the following concentrations: α6⁺/MHCI⁺cells, 3,000 cells per plate; α6⁺/MHCI^-cells, 10,000 cells per plate. It must be noted that even though higher cell number was plated for α6⁺/MHCI^-cells in the primary cultures, the CFE of these cells was lower than the CFE of α6⁺/MHCI⁺cells. In secondary cultures equal numbers of cells were plated (100 cells per plate) for both, α6⁺/MHCI⁺cells, and α6⁺/MHCI^-cells.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Expression of interfollicular epidermal stem cell markers in MHCI negative cells</bold>. (A) Differentially expressed genes identified by microarray analysis that were discussed in the text. The numbers shown are in fold change in log2 scale and are the highest score for the gene. (B) The differentially expressed genes identified by microarray analysis and/or known epidermal SC markers were confirmed using semi-quantitative RT-PCR. PCR was run for 25, 30, and 35 cycles. β-catenin mRNA, which did not show significant change of expression in the two cell populations was used as a control.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>CD71 expression in MHCI negative and MHCI positive cells</bold>. (A) showing A representative flow cytometry analysis of the expression of MHCI and CD71 indicating that epidermal cells that exhibit lack/low expression of MHCI also exhibit lower expression of CD71 than cells that express high level of MHCI. Gates were set using isotype control antibodies and single color antibodies. The geometrical mean channel fluorescence of the populations is indicated. (B) Single positive control for PE. (C) Single positive control for FITC. (D) Isotype control.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>Lack of Notch1 activity in MHCI negative cells</bold>. Figure demonstrating lack of cleaved/active Notch1 expression in α6⁺/MHCI^-cells and its presence in α6⁺/MHCI⁺The data are obtained by flow cytometry analysis using an antibody specific for cleaved/active Notch1.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>NF-κB activity in MHCI negative and MHCI positive cells</bold>. (A) A representative flow cytometry analysis of the expression of MHCI and NF-κB subunit RelA/p65 proteins showing that epidermal cells that exhibit low expression of RelA/p65 also exhibit lack/low expression of MHCI. The geometrical mean channel fluorescence of the populations is indicated. (B) Single positive control for PE. (C) Single positive control for FITC. (D) Secondary control.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p><bold>NF-κB activity in α6⁺/MHCI negative and MHCI positive cells</bold>. Nuclear extracts of sorted cells were analyzed for NF-κB p50-binding activity; data were expressed in optical density (O.D.) units obtained with the TransAM ELISA NF-κB assay for phosopho-p50 (**<italic>P </italic>&lt; 0.017).</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p><bold>Changes in the gene expression during epidermal differentiation</bold>. (A) Genes that are gradually downregulated, or upregulated during epidermal differentiation. The EST* represents the Human clone 23933 mRNA (B) Genes that are upregulated in α6⁺/MHCI⁺cells (TA cells), and subsequently downregulated in suprabasal cells.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Selected genes enriched in α6⁺/MHCI⁺cells compared to α6⁺/MHCI^-cells.</p></caption><table frame=\"hsides\" rules=\"groups\"><tbody><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>Factors downregulated in HHFSC</bold></td><td align=\"left\">CDC2 (1.7), PRC1 (1.6), RRM2 (2.3), ZWINT (1.1), KPNA2 (1.1), FEN1 (1.1), TOP2A (4.5), TYMS (1), RHEB2 (1.2).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>Factors downregulated in MHFSC</bold></td><td align=\"left\">THBD (1.1), RBMS1 (4.2), MYC (1), ABCD4 (1.8), UGP2 (1.6), IGFBP3 (1.1), WNT3 (3.8), WNT4 (3.6), DSC2 (1.8), HSPA1A (2.4), CKMT1 (1), RORA (2.4), ANXA1 (1.4), ANXA2 (1.8), COL17A1 (2.2), IL6ST (1.7), TGFBR2 (1.3), LGALS7 (1.8), KRT5 (1), KRT15 (1.2), SERPINB2 (1), SERPINB7 (1.8), GPR87 (1.3), TGFBI (1.3), VSNL1 (1.2), CLCA2 (1.7), E48 (2), MKI67 (2), CKS2 (1.1), PRC1 (1.6), HLA-B (2.4), HLA-B39 (2), HLA-C (1.1), HLA-Cw*1701 (2.9), HLA-E (1.6), D6S81E (1.7), CCNB1 (2.2), CCNB2 (1.7), CCND1 (1.3), CCND2 (3.3), CHEK1 (1.7), CDC6 (1.6).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>TGFβ/BMP-repressed factors</bold></td><td align=\"left\">MYC (1), MT1F (1.6), MT1G (1.1), UGP2 (1.6), CCND2 (3.3), KRT15 (1.2), VIL2 (3.3), CKMT1 (1), EHF (2.2), NIBAN (1.7), VAMP8 (3.2), TNNI2 (1.4), KLF4#.</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>WNT-induced factors</bold></td><td align=\"left\">MYC (1), CCNB1 (2.2), CCND1 (1.3), CCND2 (3.3), JUN (1), CKS2 (1.1), MKI67 (2), BIRC5 (2.5), TNNT1 (3.5), TNNI2 (1.4), MBNL (1.4), IGFBP3 (1.1), PTTG1 (1.1), EGFR (1.2), EMP1 (2.1), CSPG6 (2.8), CALD1 (1.2), BTEB2 (1.8), DUSP6 (3.5), FOS (2.2), JWA (1.2), HSP70 (1.4), KRT5 (1), GSTM3 (1.3), NCOA3 (1.3), OSF2 (1.1), SDC4 (1), ELF1 (1.6), HMG14 (1), TRA1 (2.3), CDC6 (1.6), DHFR (1.1), ADE2H1 (2), NSAP1 (1.2), MCM4 (3.4), KPNB3 (1.2), KRTHA1 (0.5)#, MYCBP (0.6)#.</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>NF-κB-induced factors</bold></td><td align=\"left\">MYC (1), MT1F (1.6), MT1G (1.1), CCND1 (1.3), FTH1 (1.7), IGFBP1 (2), HMG14 (1), AKR1C2 (1), UGCG (1.1), GBP-1 (1.2), ATF3 (3.3), SDC4 (1), PTGS2 (1.3), BMP2 (1.3), DUSP6 (3.5), MGP (1.1), FOS (2.2), MCP-1 (5.4), PIG7 (2), MIF (1.8), PMAIP1 (1.3), LSR68 (3.6), TNFS10 (2.7), HLA-B (2.4), HLA-B39 (2), HLA-C (1.1), HLA-Cw*1701 (2.9), HLA-E (1.6), D6S81E (1.7), HSPB1 (3).</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Selected genes enriched in α6⁺/MHCI^-cells compared to α6⁺/MHCI⁺cells.</p></caption><table frame=\"hsides\" rules=\"groups\"><tbody><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>mRNAs enriched in HHFSC</bold></td><td align=\"left\">TNRC9 (1.5), PHLDA1 (1), WIF-1 (6.6), RIG (1.6), DPYSL2 (1.9), DPYSL3 (1.4), GPM6B (2.5), FZD1 (1.3), NFATC1 (1.7), FST (2.3), DCT (2.5).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>mRNAs enriched in MHFSC</bold></td><td align=\"left\">LHX2 (5), TCF3 (1.5), WIF-1 (6.6), TRPS1 (2), BACH2 (1.1), LTBP1 (1.1), LTBP2 (1), ID2 (1), ID4 (1), DPYSL2 (1.9), DPYSL3 (1.4), GPR49 (2), GADD45G (1.2), ENPP1 (2.1), FBN2 (1), FOXC1 (1.7), VIM (1.7), DCT (2.5), MERTK (1.5), CRYM (1.5), CNR1 (1.4), SCD (2), TCF7 (1.1), CPE (1.9), EDNRB (2.8), AML1 (2.3), GPM6B(2.5), FGFR1 (2.5), CSPG2 (1.5), CSPG4 (2.1), NFATC1 (1.7), FYN (1.8), PRDM5 (1.5), ARG2 (1.4), MOX2 (2.9), DLX2 (1.8), ADAMTS5 (1.6), PHLDA1 (1), FZD7 (1.5), GUCY1B3 (1.5), TYR (2.1), COL1A2 (2), GPR64 (2.3), GSTM5 (1.2), PPAP2B (1.6), MITF (1.8), SNCAIP (1.2), SOX9 (1.9), MYH10 (1.1), MADH6 (1.6), INSIG1 (1.2), PLAT (1.4), PEG3 (2.8), NFIB (1.2), DAB2 (1.9), IGFBP5 (1.1), IGFBP7 (2.2), ITM2A (1.2), GFRA1 (1.5), ALCAM (1.6), BDNF (2.6), SDF1 (1.1), COL3A1 ((1), COL4A1 (1.9), COL4A2 (1.8), COL5A1 (1.1), COL6A1 (1.1), COL14A1 (1.3), HXB (1.3), ACTN1 (1.1), HPGD (1.1), APP (1.2), CTBP2 (1), MYO1B (1.1), SIAT4C (1.5), EFNB2 (1.3), EDG2 (1.2), CYP1B1 (3.5), PRLR (1.1), ALDH7A1 (1.1), DCAMKL1 (1.6), PAK3 (1.7).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>TGFβ/BMP signaling</bold></td><td align=\"left\">MADH3 (1.3), MADH6 (1.6), MADHIP (1.5), FST (2.3), BMP5 (5.1), BMP8 (2), BMP10 (1.2). BMP15 (1.3), INHBC (1.8).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>TGFβ/BMP-induced factors</bold></td><td align=\"left\">COL3A1 (1), COL4A1 (1.9), COL5A1 (1.1), COL6A1 (1.1), COL9A2 (1.6), COL11A1 (3.7), COL11A2 (1), COL14A1 (1.3), GPR56 (1.2), SOX4 (1.2), CLU (1.2), IQGAP1 (1.3), LMCD1 (2.7), SPRY4 (1.2), ITGB5 (1.1), LTBP1 (1), LTBP2 (1.1), GSN (1), PPAP2B (1.6), PEA15 (1.8), HEF1 (1.2), ID2 (1), ID4 (1), TGFB1l1 (1.3), FRZB (3.9), VCAM (1.2), FST (2), HXB (1.3), GSPG2 (2), AGC1 (1.5), THBS1 (2.3), APOE (1.1), MADH6 (1.6), NFATC1 (1.7), CKB (1.5), MMP9 (1.5), PLAUR (1.3), PLAT (1.4), APP (1.2), PTPRC (1.4), FZD1 (1.3), FYN (1.8), VAV1 (1.3), HCLS1 (1.1), TAL1 (1.1), LPL (1.3), BDNF (2.6), APBA3 (1.3), CDKN1C (2.1).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>WNT signaling</bold></td><td align=\"left\">FRZB (3.9), FZD1 (1.3), FZD4 (1.7), FZD7 (1.5), WIF-1 (6.6), DKK1 (1.9), DKK2 (1.7), TCF3 (1.5), TCF7 (1.1), TCFL2 (1.2), TLE1 (1), DAB2 (1.9), CTBP2 (1).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>WNT-repressed factors</bold></td><td align=\"left\">ACTN3(1.1), AKAP12 (1.1), CTSB (1.4), PLA2G7 (3.1), LTBP2 (1), DAB2 (1.9), FST (2.3), CLU (1.2), TCF3 (1.5), MEG3 (1.7), PPAP2B (1.6), LPL (1.3), ID4 (1), CDKN1C (2.1).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>Interferon signaling</bold></td><td align=\"left\">IFNA5 (2.9), IFNA6 (4), IFNA7 (1.4), STAT2 (1.5), CIS4 (1), SSI-3 (1.6).</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\"><bold>Inositol phospholipid signaling</bold></td><td align=\"left\">INPP4B (1.2), PIGB (1.2), PLCB4 (1), PLCE2 (1.4), KIAA0581 (2), GPLD1 (2.2), PIK3CD (1.2), NUDT4 (1.7), LOC51196 (1.7).</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p>List of the genes that are differentially expressed in α6+/MHCI- cells and in α6+/MHCI+ cells. Entire Affymetrix probe set and their annotated genes that are up-regulated ≥ 2-fold in either α6+/MHCI- cells or α6+/MHCI+ cells sorted according to their functions. Some of the genes are involved in multiple processes in the cell and could be placed in several tables. The table shows the difference in the expression α6+/MHCI+ cells vs. α6+/MHCI- cells. \"-\"sign indicates that the gene is upregulated in α6+/MHCI- cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p>List of the entire genes that are differentially expressed in either α6+/MHCI+ cells or α6+/MHCI- cells and are consistently upregulated or down regulated in both arrays. \"-\"sign indicates that the gene is upregulated in α6+/MHCI- cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p>List of the genes that are differentially expressed in α6⁺/MHCI⁺cells vs. α6⁺/MHCI^-cells and also expressed in human hair follicle SCs (HHFSC). \"-\"sign indicates that the gene is upregulated in α6⁺/MHCI^-cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S4\"><caption><title>Additional file 4</title><p>List of the genes that are differentially expressed in α6⁺/MHCI⁺cells vs. α6⁺/MHCI^-cells and also expressed in murine hair follicle SCs (MHFSC). \"-\"sign indicates that the gene is upregulated in α6⁺/MHCI^-cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S5\"><caption><title>Additional file 5</title><p>Expression of TCF3 targets in α6⁺/MHCI^-and α6⁺/MHCI⁺cells. The table shows the difference in the expression in α6⁺/MHCI⁺cells vs. α6⁺/MHCI^-cells. \"-\"sign indicates that the gene is upregulated in α6⁺/MHCI^-cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S6\"><caption><title>Additional file 6</title><p>Expression of MYC targets in α6⁺/MHCI^-and α6⁺/MHCI⁺cells. The table shows the difference in the expression in α6⁺/MHCI⁺cells vs. α6⁺/MHCI^-cells. \"-\"sign indicates that the gene is upregulated in α6⁺/MHCI^-cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S7\"><caption><title>Additional file 7</title><p>Flow cytometry data of sorted cells that donated RNA for microarray experiments.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S8\"><caption><title>Additional file 8</title><p>List of the genes that are differentially expressed in α6+/MHCI+ cells and α6+/MHCIcells and are consistently upregulated or downregulated ≥ 2 fold in both arrays. \"-\"sign indicates that the gene is upregulated in α6+/MHCI- cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S9\"><caption><title>Additional file 9</title><p>List of the genes that are differentially expressed in at least one array ≥ 2 fold in either α6+/MHCI+ cells or α6+/MHCI- cells and are consistently upregulated or down regulated in both arrays. \"-\"sign indicates that the gene is upregulated in α6+/MHCI- cells. The numbers that show the difference in the level of gene expression are in log2 scale.</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>Functional classification of some of the genes that are selectively upregulated ≥ 2 fold in α6⁺/MHCI⁺cells relative to α6⁺/MHCI^-cells (for the full list see Additional file ##SUPPL##0##1##). The numbers shown in parentheses are in log2 scale and are the highest score for that gene. Expression of the genes marked with (#) have been additionally verified (Figure ##FIG##3##4##).</p></table-wrap-foot>", "<table-wrap-foot><p>Functional classification of some of the genes that are selectively upregulated ≥ 2 fold in α6⁺/MHCI^-cells relative to α6⁺/MHCI⁺cells (for the full list see Additional file ##SUPPL##0##1##). The numbers shown in parentheses are in log2 scale and are the highest scores for that gene.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2164-9-359-1\"/>", "<graphic xlink:href=\"1471-2164-9-359-2\"/>", "<graphic xlink:href=\"1471-2164-9-359-3\"/>", "<graphic xlink:href=\"1471-2164-9-359-4\"/>", "<graphic xlink:href=\"1471-2164-9-359-5\"/>", "<graphic xlink:href=\"1471-2164-9-359-6\"/>", "<graphic xlink:href=\"1471-2164-9-359-7\"/>", "<graphic xlink:href=\"1471-2164-9-359-8\"/>" ]

[ "<media xlink:href=\"1471-2164-9-359-S1.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S2.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S3.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S4.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S5.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S6.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S7.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S8.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-359-S9.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Splicing of precursor mRNA is one of the essential cellular processes in eukaryotic organisms. Although this process has been extensively studied since the discovery of splicing three decades ago [##REF##902310##1##,##REF##269380##2##], resulting in a thorough understanding of the splicing pathway and identification of the numerous components of the splicing machinery, there are still many unanswered questions. For example, while the ability of pre-mRNA to form intramolecular interactions between short complementary segments in long yeast introns was initially suggested 20 years ago [##UREF##0##3##], the role of pre-mRNA secondary structure in splicing is not well understood.</p>", "<p>Introns in <italic>S. cerevisiae </italic>are known to have bimodal length distribution [##REF##10024174##4##] and can be classified into short and long introns based on their length. The distance between the 5' splice site and the branchpoint sequence, also known as the 'lariat length' or 'branchpoint distance' (we also refer to it as linear branchpoint distance), is tightly correlated with intron length (with a Pearson correlation coefficient of <italic>r </italic>= 0.99 [##REF##15470237##5##]) and can also be used to classify introns into long (5'L) and short (5'S) [##UREF##0##3##]. It was hypothesized that 5'L introns, for which the branchpoint distance is greater than 200 nt, can fold into secondary structures to optimize the positioning of the 5' splice site and branchpoint sequence to one that is optimal for spliceosome assembly [##UREF##0##3##]. This hypothesis was confirmed for a limited number of yeast introns by comprehensive biological experiments that demonstrated that the existence of such secondary structure elements is essential for splicing efficiency [##REF##3322814##6##, ####REF##8458083##7##, ##REF##7493320##8##, ##REF##8718681##9##, ##REF##8903339##10##, ##REF##9356473##11####9356473##11##]. Structural elements that exhibit a similar effect on splicing efficiency were also found in introns of <italic>Drosophila melanogaster </italic>and related species [##REF##12972637##12##]. Furthermore, in mammalian cells, folding of long intron sequences is facilitated by protein binding and interactions, which presumably shortens the long distance between essential splicing sequences [##REF##16396608##13##].</p>", "<p>The nature of the base-pairing interactions within introns and their effect on splicing efficiency were most extensively studied in <italic>S. cerevisiae's </italic>ribosomal protein gene <italic>RPS17B</italic>, previously known as <italic>RP51B </italic>(YDR447C). It was shown that secondary structure interaction between two sequence segments located downstream of the 5' splice site and upstream of the branchpoint sequence promotes efficient splicing of the <italic>RPS17B </italic>pre-mRNA [##REF##8458083##7##]. This interaction was further tested by comprehensive mutational and structure-probing analysis to determine the structure of the stem formed in the wildtype intron and the sensitivity of splicing efficiency to the alterations in this stem [##REF##7493320##8##,##REF##8718681##9##]. These studies demonstrated that complementary pairing between two ends of the <italic>RPS17B </italic>intron, but not necessarily the formation of the described stem, is essential for its efficient splicing <italic>in vitro </italic>and <italic>in vivo</italic>.</p>", "<p>While the authors of the previous studies speculated that the function of the complementary pairing is to shorten the branchpoint distance, they did not attempt to determine the secondary structure of the intron and the resulting 'structural' branchpoint distance. Thus a functional relationship between this distance and the splicing efficiency remains unknown.</p>", "<p>In this paper we use computational RNA secondary structure prediction to investigate the secondary structures of wildtype and mutant intron sequences within the <italic>S. cerevisiae RPS17B </italic>pre-mRNA. We present a unique algorithm for measuring 'structural' distance between two bases in an RNA secondary structure and use it to compute the distance between the 5' splice site and the branchpoint sequence based on the predicted secondary structure. Our analysis show that there is a tight correlation between structural branchpoint distances and splicing efficiency levels for all mutants examined.</p>" ]

[ "<title>Methods</title>", "<title>Computational RNA secondary structure prediction</title>", "<p>In this work we used four different RNA secondary prediction tools: mfold [##UREF##1##14##,##REF##10329189##15##], RNAsubopt [##REF##10070264##20##], RNAfold [##UREF##4##22##] and Alifold [##REF##12824340##26##].</p>", "<p>Mfold was used for predicting MFE secondary structures for Libri et al.'s [##REF##7493320##8##] mutants and for predicting suboptimal structures within 5% of the MFE during the mutant design process. Mfold uses dynamic programming to identify the MFE secondary structure and a set of suboptimal structures within a user defined percentage from the MFE for a given RNA sequence. We used both, the web (3.2) and command line (3.0) versions of mfold with default parameters.</p>", "<p>RNAsubopt was used to compute a sample of 1000 suboptimal structures within the 5% from the MFE. Unlike mfold, it computes <italic>all </italic>suboptimal secondary structures within a user defined energy range or percentage from the MFE for a given RNA sequence. It can also draw a random sample of the computed suboptimal structures using their Boltzmann weights. We used the command line version of RNAsubopt with options \"-ep 5 -p 1000 -noLP\", which specify the percentage from the MFE (5%), random sample size (1000) and disable prediction of helices of length 1.</p>", "<p>RNAfold was used to compute partition function and base-pair probabilities. RNAfold uses dynamic programming to compute the MFE secondary structure of a given RNA sequence, but when run with option '-p' it also computes base-pair probabilities.</p>", "<p>Alifold was used to compute consensus secondary structure for <italic>RPS17B </italic>and <italic>RPS6B </italic>introns based on the alignment of introns in <italic>Sensu stricto </italic>species. It uses modified dynamic programming algorithms that add a covariance term to the standard energy model to compute a consensus secondary structure for a set of aligned RNAs.</p>", "<p>RNAsubopt, RNAfold and Alifold are part of the Vienna RNA secondary structure package [##UREF##4##22##] (we used version 1.7). All four algorithms use free energy calculation based on Turner's nearest neighbour energy model [##REF##10329189##15##,##REF##2432595##36##, ####REF##2456874##37##, ##REF##2456074##38####2456074##38##].</p>", "<title>Distance calculation in an RNA secondary structure</title>", "<p>Calculating the spatial distance between two nucleotides in a folded RNA molecule requires knowledge of the tertiary structure of the molecule. Since currently there are no reasonably reliable algorithms for predicting RNA tertiary structure, our distance calculation is based solely on RNA secondary structure. Considering that secondary structure is generally believed to play a crucial role in tertiary structure formation [##REF##9241415##34##,##REF##10550208##35##], this approach should give us a good approximation of the true spatial distance.</p>", "<p>To calculate the structural branchpoint distance <italic>d</italic>_{<italic>s</italic>}, we consider a predicted secondary structure of the intronic pre-mRNA as an undirected graph whose vertices are nucleotide bases and whose edges correspond to the bonds between the nucleotides. These bonds can be either sugar-phosphate bonds between the nucleotides in the RNA chain or the hydrogen bonds between paired bases in a given RNA secondary structure. Figure ##FIG##9##10## shows the conversion from an RNA secondary structure to the secondary structure graph representing it. To compute the distance between two vertices in the graph, we employed Dijkstra's shortest-path algorithm [##UREF##6##39##]. Since Dijkstra's algorithm requires a directed graph, we represent each non-directed edge (<italic>u</italic>, <italic>v</italic>) as two directed edges, (<italic>u</italic>, <italic>v</italic>) and (<italic>v</italic>, <italic>u</italic>). All edges in the RNA secondary structure graph have uniform weight <italic>w</italic>(<italic>u</italic>, <italic>v</italic>) = 1.</p>", "<p>In our implementation of the algorithm, the inputs to the program are a pseudoknot-free RNA secondary structure in dot-bracket notation (Vienna format) and the locations of two bases for which the distance needs to be calculated. These bases are the first nucleotide of the intron and the bulging A in the branchpoint sequence (UACUA<underline>A</underline>C). The output of the program is the shortest distance between these two bases, which we consider as the structural branchpoint distance (<italic>d</italic>_{<italic>s</italic>}) for the given intron secondary structure. The program is available at <ext-link ext-link-type=\"uri\" xlink:href=\"http://cs.ubc.ca/~rogic/splicing.html\"/>.</p>", "<title>Mutant sequences</title>", "<p>We used two basic strategies for designing intron mutants with desired structural characteristics. To obtain mutants with long structural branchpoint distances we aimed to disrupt a zipper stem that was bringing the donor site and the branchpoint sequence close together in the wildtype intron. Conversely, for the mutants designed to have efficient splicing we aimed to stabilize the zipper stem found in the wildtype intron. With these strategies in mind, we used a combination of a trial-and-error approach and secondary structure designs computed by RNA Designer [##REF##15095976##40##] to obtain mutant sequence with desired structural characteristics.</p>", "<p>Most of the intron mutants that we designed have segment substitutions around 20–30 nt long. Sequence segments of this size allowed us to rearrange the secondary structure of a mutant in a desired way. The exception is mutant <italic>rps6b-S5 </italic>which has three short insertions (8 nt in total) in the polypyrimidine tract of <italic>RPS6B </italic>intron. Mutant <italic>rps17b-L3 </italic>is a result of two 3-nucleotide-segment substitutions in Libri et al.'s [##REF##7493320##8##] mutant <italic>8mUB1 </italic>(the middle sequence of lower case letters represents the original <italic>8mUB1 </italic>mutation). Similarly, mutant <italic>rps17b-S3 </italic>is a result of a 4-nucleotide-segment substitution in the <italic>3mDB1 </italic>mutant (the first segment of lower case letters represents the original <italic>3mDB1 </italic>mutation). Table ##TAB##7##8## gives the location and sequence of mutant substitutions and Figure ##FIG##10##11## depicts mutant locations with respect to the secondary structure of the introns we studied.</p>", "<title>Generation and assaying of intron mutants</title>", "<p>Using the <italic>TRP1 </italic>gene as a selectable marker, <italic>RPS17B</italic>, <italic>RPS6B </italic>and <italic>APE2 </italic>were tagged at their genomic locus with a -13MYC fragment to generate C-terminal protein fusions in yeast strains derived from a s288c background [##REF##9717241##41##]. Western blotting with a MYC antibody (Covance Research Products) confirmed expression of the correct size protein product in each strain. The intron of the selected gene plus 5' and 3' flanking sequences were deleted through homologous recombination with the <italic>URA3 </italic>selectable marker in each of these tagged strains. Intron DNA containing sequences homologous to regions 5' and 3' of the <italic>URA3 </italic>insertion plus the selected intron mutations were created by PCR. Transformation of these fragments into the appropriate intron deletion strain results in recombination, removal of the <italic>URA3 </italic>gene, and insertion of the mutant intron sequence. The <italic>URA3 </italic>gene product leads to cell death when placed on 5-fluoroorotic acid (5-FOA) due to the conversion of 5-FOA to a toxic by product [##REF##3323810##42##]. After transformation, cells can be selected on 5-FOA for those strains that have lost <italic>URA3 </italic>via insertion of the mutant intron, and thus can grow in the presence of 5-FOA. PCR was used to confirm that 5-FOA resistant strains were the result of insertion of the mutated intron in place of <italic>URA3</italic>. Each intron mutation was subsequently confirmed by sequencing. Strains containing the correct intron mutations were mated with a strain carrying a 13 MYC epitope tagged protein of a different molecular weight as an internal control and assayed for protein expression levels by western blotting. Western blotting was performed using 20–200 ng of whole cell lysate with a MYC antibody (Covance Research Products) and was quantitated after being developed with ECL Plus Western Blotting Detection reagent (Amersham Bioscience) using a Storm Imaging system (Amersham Bioscience). For each mutant assayed, the internal control was used to normalize protein loading, and the experiments were performed a minimum of 2 times on two independently derived mutant isolates.</p>", "<title>Yeast intron dataset</title>", "<p>In order to obtain a high quality yeast intron dataset we consulted three databases: the Ares lab Yeast Intron Database [##REF##12073325##43##], the Yeast Intron DataBase [##REF##10592188##44##], and the Comprehensive Yeast Genome Database [##REF##11752246##45##]. For additional information, we used the <italic>Saccharomyces </italic>Genome Database (SGD) [##UREF##7##46##]. We constructed our dataset by including introns that have consistent annotations between at least two of the three databases. We considered only introns from single-intron genes (which represent the majority of intron-containing genes in <italic>S. Cerevisiae</italic>) that interrupt the gene's coding region (this excluded introns found in the 5' UTR region). The number of introns found to have a consistent annotation between at least two databases was 214 (there are ~240 introns in the yeast genome). Eleven of these were excluded because they were not supported by the latest comparative genomic study [##REF##12748633##23##], which labeled them as possible misannotations. The final dataset contains 203 yeast introns, 155 of which are experimentally verified and 48 are putative introns. There are 98 long (5'L) and 105 short (5'S) introns. We call this dataset the STRuctural INtron (STRIN) dataset. The STRIN dataset is available at <ext-link ext-link-type=\"uri\" xlink:href=\"http://cs.ubc.ca/~rogic/splicing.html\"/>.</p>" ]

[ "<title>Results</title>", "<title>Secondary structure of <italic>RPS17B </italic>intron and the efficiency of splicing</title>", "<p>The first goal of our study was to determine if the splicing efficiency results previously reported for <italic>RPS17B </italic>intron [##REF##7493320##8##] can be correlated with the computationally predicted secondary structures of wildtype and mutant intron sequences.</p>", "<p>In this study the sensitivity of splicing to alterations in the stem formed in the <italic>RPS17B </italic>intron was tested by introducing mutations in the interacting regions designated UB1 (upstream box 1) and DB1 (downstream box 1). The assumption behind the mutant design was that any mutation within the stem would disrupt it and change the intron secondary structure in such a way that the resulting structural branchpoint distance (<italic>d</italic>_{<italic>s</italic>}) would be greater than for the wildtype intron. The authors created 9 mutant introns within the <italic>RPS17B </italic>gene: <italic>3mUB1 </italic>(3 nt mutation), <italic>4mUB1 </italic>(4 nt), <italic>5mUB1 </italic>(5 nt), <italic>6mUB1 </italic>(6 nt) and <italic>8mUB1 </italic>(8 nt), where mutations fall in the UB1 region; <italic>3mDB1 </italic>(3 nt) and <italic>5mDB1 </italic>(5 nt), where mutations fall in the DB1 region and are designed to restore the base-pairing disrupted by the mutations in the <italic>3mUB1 </italic>and <italic>5mUB1</italic>, respectively; and <italic>3mUB1_3mDB1 </italic>and <italic>5mUB1_5mDB1</italic>, which are double mutants. All of the single mutants are expected to disrupt the secondary structure, while the double mutants are predicted to restore it. The <italic>RPS17B </italic>intron was inserted into the coding region of the copper resistance gene (<italic>CUP1</italic>), which served as a reporter gene. Thus, yeast cells grown on copper containing medium will be viable only if the intron-containing <italic>Cup1 </italic>mRNA is spliced. The results of this assay suggested that for all single mutants except <italic>8mUB1</italic>, splicing was reduced. Surprisingly, <italic>8mUB1 </italic>had a similar growth rate on copper media as the wildtype intron suggesting that splicing was as efficient. Out of two double mutants, <italic>5mUB1_5mDB1 </italic>was able to partially rescue copper resistance, while <italic>3mUB1_3mDB1 </italic>did not. The authors hypothesized that these unexpected results were the result of some secondary structure rearrangements; however, the secondary structure of the mutants <italic>8mUB1 </italic>and <italic>3mUB1_3mDB1 </italic>was not explored.</p>", "<p>In order to investigate if the differences in the splicing efficiency levels are due to the differences in secondary structures, we computed the minimum free energy (MFE) structures of the introns using mfold [##UREF##1##14##,##REF##10329189##15##], one of the most frequently used RNA secondary structure prediction tools. The comparative RNA secondary structure prediction, which is considered more reliable, requires a certain number of orthologous sequences which were available only for the wildtype <italic>RPS17B </italic>intron and not for the mutants created in [##REF##7493320##8##].</p>", "<p>According to the mfold MFE predictions, the introduced mutations have the desired effect of disrupting the stem in all single mutants, but the compensatory mutations fail to restore it in two double mutants. Focusing on the positioning between the donor site and the branchpoint sequence, we compared the part of the structure that contains these two sites across all the mutants. The specified structural domain was almost identical for the <italic>3mUB1</italic>, <italic>5mUB1</italic>, <italic>8mUB1</italic>, <italic>3mDB1</italic>, <italic>3mUB1_3mDB1 </italic>and <italic>5mUB1_5mDB1 </italic>mutants, some of which have very different splicing efficiency levels (see Additional file ##SUPPL##0##1##). Moreover, the full secondary structures of the <italic>3mUB1 </italic>and <italic>3mUB1_3mDB1 </italic>mutants were almost identical with only three base-pairs difference, while the copper resistance experiment suggested significant differences in splicing efficiency. Therefore, it appears that differences in the splicing efficiency of Libri et al.'s [##REF##7493320##8##] mutants cannot be attributed to differences in the computed MFE secondary structures of introns.</p>", "<p>However, considering only a single, minimum free energy secondary structure prediction of an intron might not be the appropriate approach. While functional, non-coding RNAs, such as tRNAs and rRNAs, have a strong evolutionary pressure to maintain their unique, functional structure, it is believed that mRNAs, whose primary role is to carry the protein coding information to the translation apparatus, do not have functional constraints on their global structure. Thus, instead of always folding into unique MFE structure, it is likely that mRNAs exist in a population of structures [##UREF##2##16##, ####REF##7929367##17##, ##REF##16202126##18####16202126##18##]. Another reason for considering suboptimal structures, especially when using computational prediction methods, is that RNA secondary structure prediction algorithms have limited accuracy and sometimes the correct structure is buried among the suboptimal predictions with free energies very close to the MFE [##REF##10329189##15##,##UREF##3##19##,##REF##10070264##20##].</p>", "<title>Structural branchpoint distances of suboptimal secondary structures and the efficiency of splicing</title>", "<p>Based on these considerations, we modified our approach to include not only the optimal, i.e., MFE structure, but also near-optimal predictions whose free energies are within 5% of the optimum. There is an exponential relationship between the free energy of a structure and its probability in the ensemble of all possible structures for a given sequence. The probability of a structure <italic>S</italic>_{<italic>i </italic>}in the Boltzmann ensemble of all possible structures (<italic>S</italic>₁, <italic>S</italic>₂,...) for a given RNA sequence is given by:</p>", "<p></p>", "<p>where Δ<italic>G</italic>(<italic>S</italic>_{<italic>i</italic>}) is the free energy of structure <italic>S</italic>_{<italic>i</italic>}, <italic>Q </italic>= Σ_{<italic>S</italic>}<italic> e</italic>^{-Δ<italic>G</italic>(<italic>S</italic>)/<italic>RT </italic>}the partition function for all possible secondary structures for the given sequence, <italic>R </italic>is the physical gas constant, and <italic>T </italic>is the temperature. The probability of a secondary structure is also called the Boltzmann weight of that structure.</p>", "<p>From the equation we can see that the lower the free energy of a structure the higher its probability, thus, the predictions within 5% from the MFE also represent the most probable structures for a given sequence, with the MFE prediction being the one with the highest probability.</p>", "<p>We used RNAsubopt algorithm [##REF##10070264##20##] to sample 1000 suboptimal structures within 5% of the MFE for each considered intron. RNAsubopt first calculates all suboptimal structures within a user defined energy range and then produces a random sample of structures, drawn with probabilities equal to their Boltzmann weights. Therefore, RNAsubopt computes a representative sample of the secondary structure space within 5% of the MFE.</p>", "<p>Since the pair-wise structure comparison and distance estimation approach that we used for MFE structure predictions were not applicable to large number of structures we had to devise a new way to quantify the structural distance between the donor site and the branchpoint sequence. We designed an algorithm that converts an RNA secondary structure into a graph and then applies a shortest-path algorithm from graph theory to compute the shortest distance between two bases in the secondary structure. To the best of our knowledge this is the first algorithm for structural distance computation. More details are given in Materials and Methods.</p>", "<p>For each secondary structure prediction, we computed the exact distance between the donor site and the branchpoint sequences (<italic>d</italic>_{<italic>s</italic>}) using the shortest-path algorithm. The average structural branchpoint distances are given in Table ##TAB##0##1##. We assigned descriptive splicing efficiency labels based on the gel images in Figure 2A in [##REF##7493320##8##]. The distributions of computed structural branchpoint distances for each of the <italic>RPS17B </italic>mutants are given in Figure ##FIG##0##1##.</p>", "<p>These results suggest an interesting correlation between the average structural branchpoint distance and the splicing efficiency levels: sequences that are more efficiently spliced (wildtype, <italic>3mUB1</italic>, <italic>5mUB1</italic>, <italic>8mUB1</italic>, <italic>5mUB1_5mDB1</italic>, and <italic>4mUB1</italic>) have lower values for the average distance than those that are poorly spliced. After assigning numerical values to the descriptive splicing efficiency labels (efficient = 1, slightly reduced = 2, reduced = 3 and inhibited = 4) we obtain a Pearson correlation coefficient of 0.87.</p>", "<p>The histograms in Figure ##FIG##0##1## offer further insights into the relationship between structural branchpoint distances of introns and their efficiency of splicing; introns that are spliced efficiently or with slightly reduced efficiency have large frequency of suboptimal structures with <italic>d</italic>_{<italic>s </italic>}&lt; 10. Mutant <italic>5mUB1_5mDB1</italic>, which does not have this prominent peak in its distribution histogram and mutant <italic>4mUB1</italic>, which has reduced splicing efficiency, but not completely inhibited, still have higher frequency of structures with <italic>d</italic>_{<italic>s </italic>}&lt; 20 than the remaining, poorly spliced mutants. The correlation coefficient between splicing efficiency level and the proportion of structures with <italic>d</italic>_{<italic>s </italic>}&lt; 20 is 0.85.</p>", "<p>Finally, the cumulative distribution plot of structural branchpoint distances for all mutants, where lines are labeled according to the splicing efficiency levels (efficient – blue, slightly reduced – green, reduced – black and inhibited – red) shows a clear separation of spliced and unspliced mutants (Figure ##FIG##1##2##).</p>", "<p>Upon closer inspection we noticed that most of the structures with <italic>d</italic>_{<italic>s </italic>}&lt; 10 have <italic>d</italic>_{<italic>s </italic>}= 4. Analysis of the secondary structures of these sequences reveals that this distance corresponds to a structural conformation where the donor and branchpoint sequences have two base-pairing interactions between them (see Section 2.1.3). The observed base-pairing interactions are not necessarily inconsistent with established models of the splicing process, according to which spliceosomal snRNAs interact with the donor site and the branchpoint sequence, since the base-pairing can be easily disrupted after the splicing factors have been aligned properly.</p>", "<title>Structural branchpoint distances and the efficiency of splicing for other published <italic>RPS17B </italic>mutants</title>", "<p>In order to test the generality of the observed correlation between splicing efficiency levels and structural branchpoint distances we also analyzed the <italic>RPS17B </italic>intron mutants described in [##REF##8718681##9##]. These are <italic>mut-UB1i</italic>, which has an inverted UB1 sequence; <italic>mut-DB1i</italic>, which has an inverted DB1 sequence; <italic>mut</italic>-<italic>UB1iDB1i</italic>, which has both UB1 and DB1 sequences inverted to make them complementary to each other; <italic>mut-5</italic>, which reduces the consecutive pairing region to 5 base-pairs, <italic>mut-12</italic>; which improves pairing to 12 consecutive base-pairs (eliminating one one-nucleotide bulge); and <italic>mut-18</italic>, which extends pairing to 18 consecutive base-pairs (eliminating all three bulges in the pairing region). The authors compared splicing efficiency of the wildtype and mutant introns by analyzing the formation of spliceosomal complexes. Based on their gel images, we assigned descriptive and numerical splicing efficiency labels to the tested sequences (see Table ##TAB##1##2##). The average structural branchpoint distances of 1000 suboptimal structures sampled from within 5% of the MFE for each mutant are given in Table ##TAB##1##2##.</p>", "<p>The branchpoint distance results for these mutants are similar to those of Libri et al.'s [##REF##7493320##8##] mutants; the average structural branchpoint distances are lower for the sequences that are efficiently spliced (wildtype, <italic>mut-UB1iDB1i</italic>, <italic>mut-12</italic>, and <italic>mut-18</italic>). After assigning numerical values to the descriptive splicing efficiency labels (improved = 1, normal = 2 and reduced = 3), we obtain the correlation coefficient as 0.85. This, again, corresponds to the ability of these sequences to fold in such a way as to bring the donor site and the branchpoint sequences close to each other; each of the efficiently spliced sequences has a large fraction of predicted secondary structures for which <italic>d</italic>_{<italic>s </italic>}&lt; 10 (Figure ##FIG##2##3## and Additional file ##SUPPL##1##2##). The mutants that show reduced splicing have very few of these structures (0.02% for <italic>mut-UB1i </italic>and 0.33% for <italic>mut-5</italic>), except for <italic>mut-DB1i</italic>, which has 11.3% of structures with <italic>d</italic>_{<italic>s </italic>}&lt; 10. However, this is still significantly lower than for the efficiently spliced mutants. Again, the cumulative distribution plot clearly separates mutants based on their splicing efficiency (Figure ##FIG##2##3##).</p>", "<title>Base-pairing probabilities of the RPS17B intron and the efficiency of splicing</title>", "<p>The branchpoint distance analysis of <italic>S. cerevisiae's </italic>RPS17B intron suggests that the ability to form highly probable secondary structures (within 5% of the MFE) with short distance between the donor site and the branchpoint sequence seems to be required for efficient splicing of the intron. The short structural branchpoint distance for the <italic>RPS17B </italic>intron results from two base-pair interactions: between the first intron base (G) and the third base of the branchpoint sequence (C); and between the second base in the intron (U) and the second base of the branchpoint sequence (A) (see Figure ##FIG##3##4##). It is possible to compute the probability of these base-pairing interactions directly using a dynamic programming algorithm that computes the partition function [##REF##1695107##21##]. The base-pair probability reflects a sum of all probability-weighted structures in which the chosen base-pair occurs. Thus, these base-pairing probabilities also take into account the structures that were not within 5% from the MFE, eliminating the necessity to chose an arbitrary percent suboptimality value. The base-pair probabilities can be computed using RNAfold [##UREF##4##22##], another frequently used program for RNA secondary structure prediction.</p>", "<p>The base-pair probability values for the wildtype <italic>RPS17B </italic>intron and all of Libri et al.'s [##REF##7493320##8##] mutants are given in Table ##TAB##2##3##. The probability values for the two base-pairs (G-C and U-A) are identical up to second decimal place for each intron sequence and that is why only one number is shown in the table. It can be observed that all of the efficiently spliced sequences have higher base-pair probabilities than the poorly spliced sequences (<italic>r </italic>= -0.92). The correlation is not strictly linear since, for example, the mutant sequence <italic>8mUB1 </italic>has almost the same base-pair probability value as <italic>3mUB1 </italic>and <italic>5mUB1</italic>, although it is more efficiently spliced than these two. Similarly, the double mutant <italic>5mUB1_5mDB1 </italic>is more efficiently spliced than <italic>4mUB1</italic>, but this is not reflected in the base-pair probability values.</p>", "<p>For Charpentier and Rosbash's mutants, the base-pair probabilities are also higher for the sequences that are more efficiently spliced (Table ##TAB##3##4##): all of the sequences that are efficiently spliced (wildtype, <italic>mut-UB1iDB1i</italic>, <italic>mut-12</italic>, and <italic>mut-18</italic>) have base-pair probabilities of 0.40, while the other sequences have lower values (<italic>r </italic>= -0.85).</p>", "<p>Overall, based on the results for Libri et al.'s [##REF##7493320##8##] and Charpentier and Rosbash's [##REF##8718681##9##] mutants it seems that, at least for <italic>RPS17B </italic>intron, base-pair probabilities for the two base-pairs formed between the first two bases of the intron and the second and third base of the branchpoint sequence are good indicators of splicing efficiency. We will see in the following sections that this is not a general requirement for all genes. Taken together with the observed correlation between the splicing efficiency levels and structural branchpoint distances the results are consistent with the following hypothesis: the existence of highly probable secondary structures that have short branchpoint distance is required for efficient splicing of yeast introns.</p>", "<title>Experimental testing of the hypothesis</title>", "<p>In order to test the validity of the proposed hypothesis, we designed and functionally tested <italic>in vivo </italic>a series of <italic>RPS17B </italic>intron mutants. To assay the effect of these mutations on splicing we opted to introduce the mutated intron sequences at their endogenous locus, instead of within the <italic>CUP1 </italic>gene as was previously done [##REF##7493320##8##,##REF##8718681##9##]. This allows us to analyze the splicing of this intron within its normal context of flanking DNA sequences. We estimated the splicing efficiency directly from protein expression levels, which were quantified using a fluorescence imaging system.</p>", "<p>Using protein expression as a measurement of splicing efficiency requires that: 1) the level of protein abundance is proportional to the mRNA abundance (for a given gene) in the cell and, 2) the abundance of mRNA in the cell reflects any change in splicing efficiency. To demonstrate that <italic>RPS17B </italic>follows these general rules, we analyzed a number of Libri et al.'s [##REF##7493320##8##] mutants that have previously documented changes in mRNA levels for their protein expression levels. The sequences tested were the wildtype <italic>RPS17B </italic>intron, and the <italic>5mUB1</italic>, <italic>3mUB1</italic>, <italic>8mUB1</italic>, <italic>5mDB1</italic>, and <italic>3mDB1 </italic>mutated introns. The levels of protein expression, as shown in Figure ##FIG##4##5##, are proportional to the levels of copper-resistance in the copper growth assay in [##REF##7493320##8##]. Moreover, our approach is able to provide a quantifiable measure for mutants such as <italic>3mDB1 </italic>and <italic>5mDB1</italic>, which did not support any growth in the copper growth assay. Thus, using changes in protein expression levels in the context of different intron sequences to assay the effects of mutations on splicing efficiency is a valid approach.</p>", "<title>New <italic>RPS17B </italic>intron mutants</title>", "<p>We designed 8 new <italic>RPS17B </italic>intron mutants for the purpose of testing our current model of correlation between intronic pre-mRNA secondary structure and splicing efficiency. The most important structural characteristic used for mutant design was structural branchpoint distance (<italic>d</italic>_{<italic>s</italic>}) of its MFE and suboptimal structures. Four mutants that are predicted to splice efficiently were designed to have multiple suboptimal structures with contact conformation (Figure ##FIG##3##4##) and short average structural branchpoint distance (these mutants are labeled with letter 'S', which stands for short <italic>d</italic>_{<italic>s</italic>}). The only exception is mutant <italic>rps17b-S2</italic>, which does not have any suboptimal structures with contact conformation, but still exhibits a short structural branchpoint distance (most of the suboptimal predictions have <italic>d</italic>_{<italic>s </italic>}= 10). This mutant was designed to test whether contact conformation, rather than the resulting short structural branchpoint distance, is important for splicing. Four mutants that are predicted to have reduced splicing were designed not to have any structures with contact conformation or otherwise short structural branchpoint distances (these mutants are labeled with letter 'L', which stands for long <italic>d</italic>_{<italic>s</italic>}).</p>", "<p>The mutant design was based on mfold predictions, while RNAsubopt predictions where used post-experimentally to analyze the results. Mfold also samples the suboptimal space of secondary structures, however it does not compute all possible structures and the sample is much smaller. Although the distribution of <italic>d</italic>_{<italic>s </italic>}computed based on structure predictions by mfold is similar to the one based on RNAsubopt predictions, the average distances for RNAsubopt predictions are not as distinct between 'S' and 'L' mutants as ones based on mfold predictions.</p>", "<p>Table ##TAB##4##5## shows average <italic>d</italic>_{<italic>s </italic>}for newly designed mutants based on RNAsubopt predictions and base-pair probabilities computed by RNAfold. The analogous table based on mfold suboptimal predictions, which was used in the design process is given in Additional file ##SUPPL##2##3##.</p>", "<p>As seen in Figure ##FIG##5##6##, mutants <italic>rps17b-L1</italic>, <italic>rps17b-L2 </italic>and <italic>rps17b-L4 </italic>have reduced protein expression levels when compared to the wildtype as expected. Mutant <italic>rps17b-L3 </italic>has reduced splicing efficiency but not as much as the other three mutants with long structural branchpoint distances. As previously explained, this mutant was designed to have reduced splicing based on suboptimal predictions by mfold, which failed to predict any structures with <italic>d</italic>_{<italic>s </italic>}&lt; 10. However, RNAsubopt, which does a more rigorous sampling of the suboptimal space, detected a small fraction of suboptimal structures that have <italic>d</italic>_{<italic>s </italic>}&lt; 10 (see Additional file ##SUPPL##3##4##). This is in agreement with the relatively high probability of base-pairing interaction between the donor site and the branchpoint sequence (0.21).</p>", "<p>Mutants <italic>rps17b-S1</italic>, <italic>rps17b-S2</italic>, and <italic>rps17b-S3 </italic>are all spliced efficiently, as predicted. The efficient splicing of mutant <italic>rps17b-S2</italic>, which has short structural branchpoint distance (<italic>d</italic>_{<italic>s </italic>}= 10) without contact conformation in many of the predicted structures, suggests that a specific structural arrangement between the donor site and the branchpoint sequence is not required for efficient splicing. Mutant <italic>rps17b-S4 </italic>shows reduced levels of protein abundance, which is in disagreement with our prediction. The mutated sequence for this mutant has the same location as the mutated sequence for the mutant <italic>rps17b-S3</italic>, which is efficiently spliced, thus we can exclude the possibility that the discrepancy in splicing is sequence-based. A possible explanation for this phenomenon may be the existence of a very thermodynamically stable stem (with free energy Δ<italic>G </italic>= -36.6 kcal/mol) that holds the 5' splice site and the branchpoint together (analogous stems in wildtype introns have much higher free energy, see Section 2.3). This stem may be too stable to be disrupted, which might prevent the spliceosome to bind to the splice signals [##REF##7493320##8##]. Overall, the results on the new <italic>RPS17B </italic>intron mutants are consistent with the proposed model of the role of intronic secondary structure in gene splicing in yeast.</p>", "<title>Selecting additional genes for experimental validation</title>", "<p>To further validate our hypothesis regarding the role of intron secondary structure in splicing, we selected additional yeast, intron-containing genes to test our model. The selection criteria were: the linear distance (number of nucleotides) between the donor site and the branchpoint sequence is greater than 200 nt (5'L introns); the intron does not contain an snRNA gene; the gene is not essential (i.e., cells are viable if the gene is mutated or deleted); and the protein product has relatively high abundance in the cell, is amenable to c-terminal tagging, and has molecular weight between 20–120 kDa (to facilitate manipulation).</p>", "<p>From our initial dataset of 98 yeast genes that contain 5'L introns (see Materials and Methods), 18 genes matched the selection criteria (17 of these were ribosomal protein genes). We selected two of these for the experiments: the ribosomal protein gene <italic>RPS6B </italic>(YBR181C) and the amino-peptidase gene <italic>APE2 </italic>(YKL157W).</p>", "<p>The <italic>RPS6B </italic>gene contains one intron of length 352 nt, with a linear branchpoint distance (the distance between the 5' splice site and the branchpoint sequence) of <italic>d </italic>= 329 nt. The computed structural branchpoint distance (<italic>d</italic>_{<italic>s</italic>}) is 18 for the MFE and all the suboptimal computationally predicted secondary structures within 5% of the MFE. Thus for this intron, unlike for the <italic>RPS17B </italic>intron, the donor and branchpoint sequences are not base-paired.</p>", "<p>The <italic>APE2 </italic>gene contains one intron of length 383 nt, with a linear branchpoint distance of <italic>d </italic>= 327 nt. One of the suboptimal structures within 5% of the MFE has a structural branchpoint distance of 6 and the others have greater distances. In the suboptimal prediction that has <italic>d</italic>_{<italic>s </italic>}= 6 there is no base-pairing interactions between the donor and branchpoint sequences.</p>", "<title><italic>RPS6B </italic>intron mutants</title>", "<p>We designed intron mutants for the <italic>RPS6B </italic>gene in a similar manner as for the <italic>RPS17B </italic>gene: the mutants that are supposed to have efficient splicing were designed to have similar structural branchpoint distances as the wildtype intron, and the mutants that are supposed to have reduced splicing were designed to have longer distances (see Additional file ##SUPPL##4##5##). Table ##TAB##5##6## shows average structural branchpoint distances for a sample of 1000 suboptimal predictions within 5% of the MFE and the probability of short branchpoint distance derived form the base-pairing probabilities. The reported probability is the highest base-pair probability between the first donor nucleotide and any nucleotide within 20 bases away from the branchpoint adenosine. This guarantees that the branchpoint distance in a secondary structure that contains that base-pair will be no longer than 20.</p>", "<p>From Figure ##FIG##6##7## we can see that all of the 'S' mutants, which have structural branchpoint distances similar to the wildtype intron, are expressed at levels similar to the wildtype. Mutant <italic>rps6b-L1</italic>, which has <italic>avg</italic>(<italic>d</italic>_{<italic>s</italic>}) = 40 shows a reduction in splicing efficiency. The probability of <italic>d</italic>_{<italic>s </italic>}&lt; 20 also correlates well with the protein expression data except for mutant <italic>rps6b-S5 </italic>for which <italic>d</italic>_{<italic>s </italic>}&gt; 20 for all suboptimal predictions. Thus, for the <italic>RPS6B </italic>gene, structural branchpoint distances slightly longer than 20 seem to be still optimal for splicing. To summarize, the protein expression data for the <italic>RPS6B </italic>gene containing designed intron mutants are compatible with our proposed model of splicing efficiency dependence on the structural branchpoint distance.</p>", "<title><italic>APE2 </italic>intron mutants</title>", "<p>Using the same selection criteria as before, we designed six <italic>APE2 </italic>intron mutants. The values for average <italic>d</italic>_{<italic>s </italic>}and the probabilities of structural branchpoint distance shorter than 20 are given in Table ##TAB##6##7##, and the histograms of structural branchpoint distance distributions are given in Additional file ##SUPPL##5##6##.</p>", "<p>The experimental results are consistent with our prediction for five out of seven mutants: mutants <italic>ape2-S1</italic>, <italic>ape2-S2</italic>, <italic>ape2-S3 </italic>and <italic>ape2-S5 </italic>all have a level of protein abundance similar to the wildtype (Figure ##FIG##7##8##) and mutant <italic>ape2-L1 </italic>shows significantly reduced expression as expected. Mutant <italic>ape2-L2</italic>, which was expected to have reduced protein abundance as a consequence of reduced splicing efficiency, is expressed at the same level as the wildtype. Also, mutant <italic>ape2-S4 </italic>has reduced splicing despite the fact that it has a similar distribution of structural branchpoint distances as the wildtype intron. Since this mutant has the mutation at the same location as <italic>ape2-L1 </italic>(see Materials and Methods), it is possible that the intron segment that we mutated was important for splicing (e.g., contained a splicing enhancer). Overall, the results for <italic>APE2 </italic>mutants support our hypothesis of the role of structural branchpoint distance in gene splicing.</p>", "<title>Shortening of branchpoint distances by zipper stems</title>", "<p>The splicing efficiency study of <italic>RPS17B</italic>, <italic>RPS6B </italic>and <italic>APE2 </italic>genes containing wildtype and mutant introns supports our hypothesis that short structural branchpoint distances are required for efficient splicing. Although these distances are computed in the context of the secondary structure of the entire intron, our hypothesis is still consistent with the original hypothesis [##UREF##0##3##] that attributes the shortening of a long branchpoint distance to a single stem. Such stems, which we will refer to as 'zipper' stems, since they 'zip' the intron, are probably essential for achieving a short structural branchpoint distance. If we analyze the computed secondary structures of the <italic>RPS17B</italic>, <italic>RPS6B </italic>and <italic>APE2 </italic>wildtype introns we can easily identify stable stems whose 3' and 5' constituents are close to the donor site and the branchpoint sequence (Figure ##FIG##8##9##). The zipper stem labeled in the <italic>RPS17B </italic>intron is the same as the one identified in [##REF##8718681##9##] using experimental structure probing.</p>", "<p>To further test the functional importance of the identified zipper stems we performed comparative structure analysis using several closely related yeast species (<italic>S. paradoxus</italic>, <italic>S. mikatae</italic>, and <italic>S. bayanus</italic>, as well as <italic>S. cerevisiae</italic>, all belonging to the <italic>Saccharomyces sensu stricto </italic>group). We used multiple sequence alignments to extract the orthologous intron sequences for our three genes [##REF##12748633##23##,##UREF##5##24##]. Both <italic>RPS17B </italic>and <italic>RPS6B </italic>intron alignments contain three <italic>sensu stricto </italic>sequences. The multiple sequence alignment for <italic>APE2 </italic>contains all four sequences; however, these are not intronic sequences but sequences from the exon 2 of the <italic>APE2 </italic>gene. This error is due to the old <italic>S. cerevisiae </italic>annotation which mapped two genes to the location of the current <italic>APE2 </italic>gene [##REF##10734188##25##].</p>", "<p>We computed the consensus structure of <italic>RPS17B </italic>and <italic>RPS6B </italic>introns using Alifold [##REF##12824340##26##]. The previously indicated zipper stems were found in the consensus structures for both genes (Figure ##FIG##8##9##), thus suggesting evolutionary conservation of these structural elements.</p>" ]

[ "<title>Discussion</title>", "<p>The hypothesis that secondary structure interactions within yeast introns are needed for efficient splicing was proposed two decades ago [##UREF##0##3##]. Since then, experimental evidence in support of this hypothesis was found for several of <italic>S. cerevisiae's </italic>introns [##REF##3322814##6##, ####REF##8458083##7##, ##REF##7493320##8##, ##REF##8718681##9##, ##REF##8903339##10##, ##REF##9356473##11####9356473##11##]. These studies identified complementary segments located downstream of the donor site and upstream of the branchpoint sequence whose base-pairing interactions are essential for splicing. It is conjectured that the function of the formed stem is to bring the donor site and the branchpoint sequence closer together so that they are in optimal alignment for spliceosome assembly.</p>", "<p>In this paper we use computational RNA secondary structure prediction to study structural requirements for efficient splicing in yeast. Our approach considers a representative sample of suboptimal structures with free energies close to the MFE and it also considers the entire secondary structure of an intron, rather than a single stem, both of which are more consistent with the nature of RNA molecules. Furthermore, the approach includes a calculation of the structural branchpoint distance, which is used to quantify the effect of the secondary structure on the distance between the donor site and the branchpoint sequence and can easily be correlated with splicing efficiency measurements. Using this method we were able to identify structural characteristics of the <italic>RPS17B </italic>intron and its mutants that seem to be responsible for their splicing differences. Notably, mutants that are likely to have a short structural branchpoint distance are spliced more efficiently.</p>", "<p>Based on our model of structural requirements for efficient splicing we computationally designed intron mutants for three <italic>S. cerevisiae genes</italic>, <italic>RPS17B</italic>, <italic>RPS6B </italic>and <italic>APE2</italic>, and experimentally tested their splicing efficiency. The results were mostly consistent with our model, with a few exceptions (<italic>rps17b-L3</italic>, <italic>rps17b-S4</italic>, <italic>ape2-L1 </italic>and <italic>ape2-S4</italic>) which may be due to some structural characteristics of mutants that are not considered by the current model or some inherent approximations in the model that are discussed below. Some of the intron mutants that were designed to have different structural characteristics and splicing efficiencies have mutations at the same locations (e.g., <italic>rps17b-L3 </italic>and <italic>8mUB1</italic>; <italic>rps17b-S3 </italic>and <italic>3mDB1</italic>; <italic>rps6b-L1 </italic>and <italic>rps6b-S3</italic>). The experimental results that confirm differences in splicing between these pairs of mutants indicate that the secondary structure of a pre-mRNA, rather than the underlying primary sequence, is responsible for differences in splicing.</p>", "<p>We also tested our model on the <italic>YRA1 </italic>gene intron, whose splicing efficiency had previously been studied by Preker and Guthrie [##REF##16618971##27##]. The published experimental results were in agreement with our model; the efficiently spliced mutants (Δ<italic>L10 </italic>and Δ<italic>TCC</italic>/<italic>GGA</italic>) had higher base-pair probabilities than the poorly spliced sequences (wildtype intron and mutants Δ<italic>R</italic>/<italic>L10</italic>, <italic>TCC </italic>Δ<italic>L10</italic>, <italic>GGA ΔL10 </italic>and <italic>TCC</italic>+<italic>GGA </italic>Δ<italic>L10</italic>) (data not shown).</p>", "<p>Our current model is simplified in the sense that the secondary structure of an intron is computed disregarding its flanking sequences, and the three dimensional branchpoint distance is estimated from secondary structure interactions. However, we believe that folding intronic sequences in isolation is appropriate, partly because of the existence of co-transcriptional splicing, where splicing occurs before the entire pre-mRNA has been synthesized [##REF##8986595##28##, ####REF##12897147##29##, ##REF##15989964##30####15989964##30##]. Therefore, the precise part of the pre-mRNA that serves as the splicing substrate is not known. The region upstream of the transcribed intron, which consists of the 5' UTR and the first exon, is also not precisely defined due to the fact that the transcription start sites have not been unambiguously mapped [##REF##15905473##31##]. In addition, 5'UTRs are known to associate with a number of protein factors [##REF##12244124##32##,##REF##15314027##33##] which are likely to have an effect on the structure formation, but these interactions are not currently modelled by computational RNA secondary structure approaches. A preliminary investigation, in which we considered some of the upstream region yielded inconclusive results (data not shown). Thus, we believe that folding only intronic sequences gives us a reasonable approximation of the secondary structure of an intron at the time of the splicing reaction.</p>", "<p>The approximation of the three dimensional branchpoint distance using pre-mRNA secondary structure is necessary since there are no reasonably reliable algorithms for predicting RNA tertiary structure. However, it is believed that RNA secondary structure plays a crucial role in tertiary structure formation, since most tertiary interactions are thought to arise after the formation of a stable secondary structure, when the molecule is able to bend around the flexible, single-stranded regions [##REF##9241415##34##,##REF##10550208##35##]. Moreover, the tertiary structure interactions that arise in the later stages of folding are usually too weak to disrupt secondary structure that has already formed. Therefore, we believe that the structural branchpoint distance based on the secondary structure interactions provides a reasonable approximation of the true spatial distance.</p>" ]

[ "<title>Conclusion</title>", "<p>Our computational study offers further insights into the role of pre-mRNA secondary structure in gene splicing in yeast. We show that it is necessary to consider near-optimal structure predictions to be able to detect structural differences between intron mutants that have different splicing efficiencies. We also propose a novel method for quantifying a distance between two bases in an RNA secondary structure and apply this to compute structural branchpoint distances in the studied intron mutants. Positive experimental results on three different yeast genes suggest that our model of structural requirements for efficient splicing can be applied universally to all 5'L yeast introns. Additional laboratory experiments are needed to refine the current model by determining the upper bound of the structural branchpoint distance needed for efficient splicing and acceptable thermodynamic stability of the stems adjacent to splicing signals. Considering that several biological studies indicate that shortening of the branchpoint distance, either by formation of secondary structure or by protein interactions, is important for efficient splicing in <italic>Drosophila melanogaster </italic>and some mammalian species [##REF##12972637##12##,##REF##16396608##13##], it might be possible to extend our model to define structural requirements for efficient splicing in other eukaryotes. Another possible application of our findings is in gene-finding, where structural characteristics of identified long introns can be used to distinguish between real and false positive predictions.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Secondary structure interactions within introns have been shown to be essential for efficient splicing of several yeast genes. The nature of these base-pairing interactions and their effect on splicing efficiency were most extensively studied in ribosomal protein gene <italic>RPS17B </italic>(previously known as <italic>RP51B</italic>). It was determined that complementary pairing between two sequence segments located downstream of the 5' splice site and upstream of the branchpoint sequence promotes efficient splicing of the <italic>RPS17B </italic>pre-mRNA, presumably by shortening the branchpoint distance. However, no attempts were made to compute a shortened, 'structural' branchpoint distance and thus the functional relationship between this distance and the splicing efficiency remains unknown.</p>", "<title>Results</title>", "<p>In this paper we use computational RNA secondary structure prediction to analyze the secondary structure of the <italic>RPS17B </italic>intron. We show that it is necessary to consider suboptimal structure predictions and to compute the structural branchpoint distances in order to explain previously published splicing efficiency results. Our study reveals that there is a tight correlation between this distance and splicing efficiency levels of intron mutants described in the literature. We experimentally test this correlation on additional <italic>RPS17B </italic>mutants and intron mutants within two other yeast genes.</p>", "<title>Conclusion</title>", "<p>The proposed model of secondary structure requirements for efficient splicing is the first attempt to specify the functional relationship between pre-mRNA secondary structure and splicing. Our findings provide further insights into the role of pre-mRNA secondary structure in gene splicing in yeast and also offer basis for improvement of computational methods for splice site identification and gene-finding.</p>" ]

[ "<title>Authors' contributions</title>", "<p>SR conceived the study, performed computational experiments and drafted the manuscript. HHH and AKM participated in the design and coordination of the study and together with BFO supervised the research project. BM designed and performed laboratory experiments and helped draft the manuscript. PH helped design laboratory experiments. All authors analyzed the results and reviewed drafts of the manuscript. All authors read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>We gratefully acknowledge valuable feedback from the four anonymous reviewers, which helped us to improve several aspects of our study.</p>", "<p>S. Rogic was supported by an NSERC (the Natural Sciences and Engineering Research Council of Canada) Discovery Grant to A. Mackworth, who also holds a Canada Research Chair in Artificial Intelligence. H. Hoos was partly supported by the Mathematics of Information Technology and Complex Systems (MITACS) Network of Centres of Excellence. B. Montpetit was supported by awards from NSERC and the Michael Smith Foundation for Health Research. P. Hieter is supported by the Canadian Institutes of Health Research (grant MOP-38096) and the U.S. National Institutes of Health (grant P01-CA0161519).</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Distribution histograms of structural branchpoint distances for (a) wt, (b) <italic>3mUB1</italic>, (c) <italic>5mUB1</italic>, (d) <italic>8mUB1</italic>, (e) <italic>3mDB1</italic>, (f) <italic>5mDB1</italic>, (g) <italic>3mUB1_3mDB1</italic>, (h) <italic>5mUB1_5mDB1</italic>, (i) <italic>6mUB1</italic>, and (j) <italic>4mUB1 </italic>introns.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Cumulative distributions of structural branchpoint distances for all Libri et al.'s [##REF##7493320##8##] intron mutants.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Cumulative distributions of structural branchpoint distances for all Charpentier and Rosbash's [##REF##8718681##9##] intron mutants.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>A part of the wildtype <italic>RPS17B </italic>intron secondary structure that shows base-pairing between the donor site and the branchpoint sequence. The highlighted stem is the same as the one identified in [##REF##8718681##9##] using experimental structure probing.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>Protein expression levels for the <italic>RPS17B </italic>gene containing some of Libri et al.'s [##REF##7493320##8##] mutant introns. Expression levels are normalized with respect to the internal loading control and plotted as a fraction of the wildtype expression level. Shaded boxes represent the mean value for several different samples and error bars represent +/- 1 standard deviation for these samples. The error bar for the wildtype intron comes from the comparison of two different wildtype samples.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>Protein expression results for the <italic>RPS17B </italic>gene containing the newly designed mutant introns.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>Protein expression results for the <italic>RPS6B </italic>gene containing the newly designed mutant introns.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>Protein expression results for the <italic>APE2 </italic>gene containing the newly designed mutant introns. Protein expression level is normalized with respect to wildtype expression level. Shaded boxes represent the mean value for several different samples and error bars represent +/- 1 standard deviation for these samples.</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p>Portions of the <italic>RPS17B</italic>, <italic>RPS6B </italic>and <italic>APE2 </italic>introns containing computationally identified zipper stems. The free energy values (Δ<italic>G</italic>) for the shaded zipper stem are given in parentheses. Stems conserved between <italic>Saccharomyces sensu stricto </italic>group are also labeled.</p></caption></fig>", "<fig position=\"float\" id=\"F10\"><label>Figure 10</label><caption><p>Conversion from the RNA secondary structure to the graph representing it. (a) Graphical representation of the secondary structure of an intron produced by mfold (filled-in circles represent base-pairing interactions, i.e., hydrogen bonds). (b) Graph representing the RNA structure in (a). The bolded path between the source and target vertices is the one found by the algorithm to be the shortest (<italic>d</italic>_{<italic>s </italic>}= 11).</p></caption></fig>", "<fig position=\"float\" id=\"F11\"><label>Figure 11</label><caption><p>Location of mutations with respect to the secondary structure for (a) <italic>RPS17B</italic>, (b) <italic>RPS6B</italic>, and (c) <italic>APE2 </italic>introns. The two lines for each mutant indicate the beginning and end of the sequence segment that was modified.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Average structural branchpoint distances for the wildtype (wt) <italic>RPS17B </italic>intron and Libri et al.'s [##REF##7493320##8##] intron mutants.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>mutant</bold></td><td align=\"center\"><bold>average <italic>d</italic>_{<italic>s</italic>}</bold></td><td align=\"center\"><bold>splicing efficiency</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>wt</bold></td><td align=\"center\">26.67</td><td align=\"center\">efficient (1)</td></tr><tr><td align=\"center\"><bold><italic>3mUB1</italic></bold></td><td align=\"center\">27.67</td><td align=\"center\">slightly reduced (2)</td></tr><tr><td align=\"center\"><bold><italic>5mUB1</italic></bold></td><td align=\"center\">28.42</td><td align=\"center\">slightly reduced (2)</td></tr><tr><td align=\"center\"><bold><italic>8mUB1</italic></bold></td><td align=\"center\">27.94</td><td align=\"center\">efficient (1)</td></tr><tr><td align=\"center\"><bold><italic>3mDB1</italic></bold></td><td align=\"center\">37.55</td><td align=\"center\">inhibited (4)</td></tr><tr><td align=\"center\"><bold><italic>5mDB1</italic></bold></td><td align=\"center\">39.19</td><td align=\"center\">inhibited(4)</td></tr><tr><td align=\"center\"><bold><italic>3mUB1_3mDB1</italic></bold></td><td align=\"center\">37.44</td><td align=\"center\">inhibited (4)</td></tr><tr><td align=\"center\"><bold><italic>5mUB1_5mDB1</italic></bold></td><td align=\"center\">33.81</td><td align=\"center\">slightly reduced (2)</td></tr><tr><td align=\"center\"><bold><italic>6mUB1</italic></bold></td><td align=\"center\">46.31</td><td align=\"center\">inhibited (4)</td></tr><tr><td align=\"center\"><bold><italic>4mUB1</italic></bold></td><td align=\"center\">32.08</td><td align=\"center\">reduced (3)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Average structural branchpoint distances for the wildtype (wt) <italic>RPS17B </italic>intron and Charpentier and Rosbash's [##REF##8718681##9##] intron mutants.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>mutant</bold></td><td align=\"center\"><bold>average <italic>d</italic>_{<italic>s</italic>}</bold></td><td align=\"center\"><bold>splicing efficiency</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>wt</bold></td><td align=\"center\">26.67</td><td align=\"center\">normal (2)</td></tr><tr><td align=\"center\"><bold><italic>mut-UB1i</italic></bold></td><td align=\"center\">42.51</td><td align=\"center\">reduced (3)</td></tr><tr><td align=\"center\"><bold><italic>mut-DB1i</italic></bold></td><td align=\"center\">35.95</td><td align=\"center\">reduced (3)</td></tr><tr><td align=\"center\"><bold><italic>mut-UB1iDB1i</italic></bold></td><td align=\"center\">26.39</td><td align=\"center\">improved (1)</td></tr><tr><td align=\"center\"><bold><italic>mut-5</italic></bold></td><td align=\"center\">32.14</td><td align=\"center\">reduced (3)</td></tr><tr><td align=\"center\"><bold><italic>mut-12</italic></bold></td><td align=\"center\">24.82</td><td align=\"center\">improved (1)</td></tr><tr><td align=\"center\"><bold><italic>mut-18</italic></bold></td><td align=\"center\">25.30</td><td align=\"center\">improved (1)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Base-pairing probabilities of contact conformation (Figure 4) for the wildtype (wt) <italic>RPS17B </italic>intron and Libri et al.'s [##REF##7493320##8##] intron mutants.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>mutant</bold></td><td align=\"center\"><bold>base-pairing probability</bold></td><td align=\"center\"><bold>splicing efficiency</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>wt</bold></td><td align=\"center\">0.40</td><td align=\"center\">efficient</td></tr><tr><td align=\"center\"><bold><italic>3mUB1</italic></bold></td><td align=\"center\">0.33</td><td align=\"center\">slightly reduced</td></tr><tr><td align=\"center\"><bold><italic>5mUB1</italic></bold></td><td align=\"center\">0.31</td><td align=\"center\">slightly reduced</td></tr><tr><td align=\"center\"><bold><italic>8mUB1</italic></bold></td><td align=\"center\">0.34</td><td align=\"center\">efficient</td></tr><tr><td align=\"center\"><bold><italic>3mDB1</italic></bold></td><td align=\"center\">0.01</td><td align=\"center\">inhibited</td></tr><tr><td align=\"center\"><bold><italic>5mDB1</italic></bold></td><td align=\"center\">&lt; 0.01</td><td align=\"center\">inhibited</td></tr><tr><td align=\"center\"><bold><italic>3mUB1_3mDB1</italic></bold></td><td align=\"center\">0.01</td><td align=\"center\">inhibited</td></tr><tr><td align=\"center\"><bold><italic>5mUB1_5mDB1</italic></bold></td><td align=\"center\">0.11</td><td align=\"center\">slightly reduced</td></tr><tr><td align=\"center\"><bold><italic>6mUB1</italic></bold></td><td align=\"center\">0.05</td><td align=\"center\">inhibited</td></tr><tr><td align=\"center\"><bold><italic>4mUB1</italic></bold></td><td align=\"center\">0.18</td><td align=\"center\">reduced</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Base-pairing probabilities of contact conformation for the wildtype (wt) <italic>RPS17B </italic>intron and Charpentier and Rosbash's [##REF##8718681##9##] intron mutants.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>mutant</bold></td><td align=\"center\"><bold>base-pairing probability</bold></td><td align=\"center\"><bold>splicing efficiency</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>wt</bold></td><td align=\"center\">0.40</td><td align=\"center\">normal</td></tr><tr><td align=\"center\"><bold><italic>mut-UB1i</italic></bold></td><td align=\"center\">0.04</td><td align=\"center\">reduced</td></tr><tr><td align=\"center\"><bold><italic>mut-DB1i</italic></bold></td><td align=\"center\">0.25</td><td align=\"center\">reduced</td></tr><tr><td align=\"center\"><bold><italic>mut-UB1iDB1i</italic></bold></td><td align=\"center\">0.40</td><td align=\"center\">improved</td></tr><tr><td align=\"center\"><bold><italic>mut-5</italic></bold></td><td align=\"center\">0.04</td><td align=\"center\">reduced</td></tr><tr><td align=\"center\"><bold><italic>mut-12</italic></bold></td><td align=\"center\">0.40</td><td align=\"center\">improved</td></tr><tr><td align=\"center\"><bold><italic>mut-18</italic></bold></td><td align=\"center\">0.40</td><td align=\"center\">improved</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Characteristics of newly designed <italic>RPS17B </italic>mutants.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>mutant</bold></td><td align=\"center\"><bold>avg(<italic>d</italic>_{<italic>s</italic>})</bold></td><td align=\"center\"><bold>bp prob</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>wt</bold></td><td align=\"center\">26.67</td><td align=\"center\">0.40</td></tr><tr><td align=\"center\"><bold><italic>rps17b-L1</italic></bold></td><td align=\"center\">43.63</td><td align=\"center\">0.0</td></tr><tr><td align=\"center\"><bold><italic>rps17b-L2</italic></bold></td><td align=\"center\">41.11</td><td align=\"center\">0.0</td></tr><tr><td align=\"center\"><bold><italic>rps17b-L3</italic></bold></td><td align=\"center\">34.05</td><td align=\"center\">0.21</td></tr><tr><td align=\"center\"><bold><italic>rps17b-L4</italic></bold></td><td align=\"center\">32.98</td><td align=\"center\">0.04</td></tr><tr><td align=\"center\"><bold><italic>rps17b-S1</italic></bold></td><td align=\"center\">24.55</td><td align=\"center\">0.40</td></tr><tr><td align=\"center\"><bold><italic>rps17b-S2</italic></bold></td><td align=\"center\">29.62</td><td align=\"center\">0.03</td></tr><tr><td align=\"center\"><bold><italic>rps17b-S3</italic></bold></td><td align=\"center\">12.65</td><td align=\"center\">0.80</td></tr><tr><td align=\"center\"><bold><italic>rps17b-S4</italic></bold></td><td align=\"center\">9.27</td><td align=\"center\">0.70</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T6\"><label>Table 6</label><caption><p>Characteristics of newly designed <italic>RPS6B </italic>mutants.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>mutant</bold></td><td align=\"center\"><bold>avg(<italic>d</italic>_{<italic>s</italic>})</bold></td><td align=\"center\"><bold>bp prob</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>wt</bold></td><td align=\"center\">18.06</td><td align=\"center\">0.84</td></tr><tr><td align=\"center\"><bold><italic>rps6b-L1</italic></bold></td><td align=\"center\">36.74</td><td align=\"center\">0</td></tr><tr><td align=\"center\"><bold><italic>rps6b-S1</italic></bold></td><td align=\"center\">19.08</td><td align=\"center\">0.65</td></tr><tr><td align=\"center\"><bold><italic>rps6b-S2</italic></bold></td><td align=\"center\">18.04</td><td align=\"center\">0.84</td></tr><tr><td align=\"center\"><bold><italic>rps6b-S3</italic></bold></td><td align=\"center\">18.04</td><td align=\"center\">0.83</td></tr><tr><td align=\"center\"><bold><italic>rps6b-S4</italic></bold></td><td align=\"center\">18.09</td><td align=\"center\">0.84</td></tr><tr><td align=\"center\"><bold><italic>rps6b-S5</italic></bold></td><td align=\"center\">22.00</td><td align=\"center\">0</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T7\"><label>Table 7</label><caption><p>Characteristics of newly designed <italic>APE2 </italic>mutants.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>mutant</bold></td><td align=\"center\"><bold>avg(<italic>d</italic>_{<italic>s</italic>})</bold></td><td align=\"center\"><bold>bp prob</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>wt</bold></td><td align=\"center\">27.90</td><td align=\"center\">0.37</td></tr><tr><td align=\"center\"><bold><italic>ape2-L1</italic></bold></td><td align=\"center\">75.73</td><td align=\"center\">0</td></tr><tr><td align=\"center\"><bold><italic>ape2-L2</italic></bold></td><td align=\"center\">69.68</td><td align=\"center\">0</td></tr><tr><td align=\"center\"><bold><italic>ape2-S1</italic></bold></td><td align=\"center\">8.93</td><td align=\"center\">0.82</td></tr><tr><td align=\"center\"><bold><italic>ape2-S2</italic></bold></td><td align=\"center\">23.33</td><td align=\"center\">0.50</td></tr><tr><td align=\"center\"><bold><italic>ape2-S3</italic></bold></td><td align=\"center\">24.60</td><td align=\"center\">0.45</td></tr><tr><td align=\"center\"><bold><italic>ape2-S4</italic></bold></td><td align=\"center\">25.14</td><td align=\"center\">0.42</td></tr><tr><td align=\"center\"><bold><italic>ape2-S5</italic></bold></td><td align=\"center\">4.10</td><td align=\"center\">0.99</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T8\"><label>Table 8</label><caption><p>Specifications for the new intron mutants used in our study.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>mutant</bold></td><td align=\"left\"><bold>segment location</bold></td><td align=\"left\"><bold>original sequence</bold></td><td align=\"left\"><bold>substitution/insertion</bold></td></tr></thead><tbody><tr><td align=\"left\"><bold><italic>rps17b-L1</italic></bold></td><td align=\"left\">258–268 (11 nt)</td><td align=\"left\">UGAAGAGAGGU</td><td align=\"left\">augagacaacu</td></tr><tr><td align=\"left\"><bold><italic>rps17b-L2</italic></bold></td><td align=\"left\">138–157 (20 nt)</td><td align=\"left\">GAUUAGAAAACUCCAUUACU</td><td align=\"left\">cuuaaguuaguaaauaccuc</td></tr><tr><td align=\"left\"><bold><italic>rps17b-L3</italic></bold></td><td align=\"left\">22–47 (26 nt)</td><td align=\"left\">UGAAGCCGGAUAUGAUGGACUGGGC</td><td align=\"left\">uuaAGCCGcuacuacuUGGACUGucg</td></tr><tr><td align=\"left\"><bold><italic>rps17b-L4</italic></bold></td><td align=\"left\">167–189 (23 nt)</td><td align=\"left\">AGAAGAGCGCUCAAUGAAGUAGU</td><td align=\"left\">uggcuuggguuaguaggugccuc</td></tr><tr><td align=\"left\"><bold><italic>rps17b-S1</italic></bold></td><td align=\"left\">217–231 (15 nt)</td><td align=\"left\">AAUUGCUUUCGAAUG</td><td align=\"left\">uuucauguguucagc</td></tr><tr><td align=\"left\"><bold><italic>rps17b-S2</italic></bold></td><td align=\"left\">280–286 (7 nt)</td><td align=\"left\">UAAGUUG</td><td align=\"left\">uacguac</td></tr><tr><td align=\"left\"><bold><italic>rps17b-S3</italic></bold></td><td align=\"left\">246–253 (8 nt)</td><td align=\"left\">AUCCAAUG</td><td align=\"left\">uagCggcu</td></tr><tr><td align=\"left\"><bold><italic>rps17b-S4</italic></bold></td><td align=\"left\">244–253 (10 nt)</td><td align=\"left\">UUAUCCAAUG</td><td align=\"left\">cuucaucaac</td></tr><tr><td align=\"left\"><bold><italic>rps6b-L1</italic></bold></td><td align=\"left\">21–54 (34 nt)</td><td align=\"left\">CCUUAGAAUUCUAAUGAAUCAGCACGCGCUAACC</td><td align=\"left\">guauuuugggugugucccuguuauaaauaauacc</td></tr><tr><td align=\"left\"><bold><italic>rps6b-S1</italic></bold></td><td align=\"left\">19–29 (11 nt)</td><td align=\"left\">AUCCUUAGAAU</td><td align=\"left\">uuuguuaguaa</td></tr><tr><td align=\"left\"><bold><italic>rps6b-S2</italic></bold></td><td align=\"left\">87–113 (27 nt)</td><td align=\"left\">CACAAAUUAGUGCACUAUAAUAAAAAU</td><td align=\"left\">uuauaaauagugauaccauuugguaaa</td></tr><tr><td align=\"left\"><bold><italic>rps6b-S3</italic></bold></td><td align=\"left\">21–57 (37 nt)</td><td align=\"left\">CCUUAGAAUUCUAAUGAAUCAGCACGCGCUAACCGGC</td><td align=\"left\">aaauuccaacguuucccugcaacaugccuuucuuccg</td></tr><tr><td align=\"left\"><bold><italic>rps6b-S4</italic></bold></td><td align=\"left\">38–55 (18 nt)</td><td align=\"left\">AUCAGCACGCGCUAACCG</td><td align=\"left\">auucccaacagacugucc</td></tr><tr><td align=\"left\"><bold><italic>rps6b-S5</italic></bold></td><td align=\"left\">337–345</td><td align=\"left\">GUAUUAUUU</td><td align=\"left\">GgUguucAUUAUUacaU</td></tr><tr><td align=\"left\"><bold><italic>ape2-L1</italic></bold></td><td align=\"left\">159–175 (17 nt)</td><td align=\"left\">UGUUACCCUCAUAUUCU</td><td align=\"left\">ggguacaauuaauagag</td></tr><tr><td align=\"left\"><bold><italic>ape2-L2</italic></bold></td><td align=\"left\">237–252 (16 nt)</td><td align=\"left\">GCAAUAGCUUAGGUAA</td><td align=\"left\">ccuucguacuuuuggg</td></tr><tr><td align=\"left\"><bold><italic>ape2-S1</italic></bold></td><td align=\"left\">23–37 (15 nt)</td><td align=\"left\">CAAAGAAACAAGGAA</td><td align=\"left\">agggcagaaauagaa</td></tr><tr><td align=\"left\"><bold><italic>ape2-S2</italic></bold></td><td align=\"left\">43–57 (15 nt)</td><td align=\"left\">AUACAUAAUAUAAAU</td><td align=\"left\">aacugguagguacgu</td></tr><tr><td align=\"left\"><bold><italic>ape2-S3</italic></bold></td><td align=\"left\">237–252 (16 nt)</td><td align=\"left\">GCAAUAGCUUAGGUAA</td><td align=\"left\">caaugaaugagaacuc</td></tr><tr><td align=\"left\"><bold><italic>ape2-S4</italic></bold></td><td align=\"left\">159–175 (17 nt)</td><td align=\"left\">UGUUACCCUCAUAUUCU</td><td align=\"left\">aaauauuaccuaagcua</td></tr><tr><td align=\"left\"><bold><italic>ape2-S5</italic></bold></td><td align=\"left\">300–322 (23 nt)</td><td align=\"left\">CUCGUUACCGACCUUUGAGUUCU</td><td align=\"left\">uuaagcuuuuguguuugagaaca</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2164-9-355-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>P</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:msup><mml:mi>e</mml:mi><mml:mrow><mml:mo>−</mml:mo><mml:mi>Δ</mml:mi><mml:mi>G</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>S</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>/</mml:mo><mml:mi>R</mml:mi><mml:mi>T</mml:mi></mml:mrow></mml:msup></mml:mrow><mml:mi>Q</mml:mi></mml:mfrac></mml:mrow></mml:semantics></mml:math></disp-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p>Minimum free energy structures for Libri et al.'s [##REF##7493320##8##] mutants predicted by mfold.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p>Distribution histograms of structural branchpoint distances for (a) wt, (b) <italic>UB1i</italic>, (c) <italic>DB1i</italic>, (d) <italic>UB1iDB1i</italic>, (e) <italic>mut-5</italic>, (f) <italic>mut-12</italic>, and (g) <italic>mut-18 </italic>introns.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p>Structural characteristics of newly designed <italic>RPS17B </italic>mutants based on mfold predictions: <bold><italic>d</italic>_{<italic>s </italic>}</bold>– structural branchpoint distances for MFE and all suboptimal predictions within 5% from the MFE; <bold>avg </bold>– average <bold><italic>d</italic>_{<italic>s</italic>}</bold>; <bold>bp prob </bold>– base-pairing probability of interaction between the donor site and the branchpoint sequence based on the partition function.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S4\"><caption><title>Additional file 4</title><p>Distribution histograms of structural branchpoint distances for (a) <italic>rps17b-L1</italic>, (b) <italic>rps17b-L2</italic>, (c) <italic>rps17b-L3</italic>, (d) <italic>rps17b-L4</italic>, (e) <italic>rps17b-S1</italic>, (f) <italic>rps17b-S2</italic>, (g) <italic>rps17b-S3</italic>, and (h) <italic>rps17b-S4 </italic>mutants.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S5\"><caption><title>Additional file 5</title><p>Distribution histograms of structural branchpoint distances for (a) <italic>RPS6B </italic>wildtype intron, (b) <italic>rps6b-L1</italic>, (c) <italic>rps6b-S1</italic>, (d) <italic>rps6b-S2</italic>, (e) <italic>rps6b-S3</italic>, (f) <italic>rps6b-S4</italic>, and (g) <italic>rps6b-S5 </italic>mutants.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S6\"><caption><title>Additional file 6</title><p>Distribution histograms of structural branchpoint distances for (a) <italic>APE2 </italic>wildtype intron, (b) <italic>ape2-L1</italic>, (c) <italic>ape2-L2</italic>, (d) <italic>ape2-S1</italic>, (e) <italic>ape2-S2</italic>, (f) <italic>ape2-S3</italic>, (g) <italic>ape2-S4</italic>, and (h) <italic>ape2-S5 </italic>mutants.</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>Levels of splicing efficiency were approximated from the gel images in Figure 2A in [##REF##7493320##8##]. The numbers within parentheses correspond to numerical values assigned to descriptive splicing efficiency labels.</p></table-wrap-foot>", "<table-wrap-foot><p>We inferred levels of splicing efficiency based on Figures 2 and 3 and Table 1 in [##REF##8718681##9##]. The numbers within parentheses correspond to numerical values assigned to descriptive splicing efficiency labels.</p></table-wrap-foot>", "<table-wrap-foot><p><bold><italic>avg</italic>(<italic>d</italic>_{<italic>s</italic>}) </bold>– average structural branchpoint distances of 1000 suboptimal structures predicted by RNAsubopt; <bold>bp prob </bold>– base-pairing probability of interaction between the donor site and the branchpoint sequence based on the partition function.</p></table-wrap-foot>", "<table-wrap-foot><p><bold><italic>avg</italic>(<italic>d</italic>_{<italic>s</italic>}) </bold>– average structural branchpoint distance; <bold>bp prob </bold>– probability of <bold><italic>d</italic>_{<italic>s </italic>}&lt; 20</bold>.</p></table-wrap-foot>", "<table-wrap-foot><p><bold><italic>avg</italic>(<italic>d</italic>_{<italic>s</italic>}) </bold>– average structural branchpoint distance; <bold>bp prob </bold>– probability of <bold>d_{<italic>s </italic>}&lt; 20</bold>.</p></table-wrap-foot>", "<table-wrap-foot><p>The upper case letters represent the original sequences and the lower case letters represent substitution or insertion sequences. The first base of an intron is numbered 1.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2164-9-355-1\"/>", "<graphic xlink:href=\"1471-2164-9-355-2\"/>", "<graphic xlink:href=\"1471-2164-9-355-3\"/>", "<graphic xlink:href=\"1471-2164-9-355-4\"/>", "<graphic xlink:href=\"1471-2164-9-355-5\"/>", "<graphic xlink:href=\"1471-2164-9-355-6\"/>", "<graphic xlink:href=\"1471-2164-9-355-7\"/>", "<graphic xlink:href=\"1471-2164-9-355-8\"/>", "<graphic xlink:href=\"1471-2164-9-355-9\"/>", "<graphic xlink:href=\"1471-2164-9-355-10\"/>", "<graphic xlink:href=\"1471-2164-9-355-11\"/>" ]

[ "<media xlink:href=\"1471-2164-9-355-S1.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-355-S2.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-355-S3.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-355-S4.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-355-S5.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-355-S6.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Efforts at extracting biologically meaningful information from the genome sequence of <italic>Plasmodium sp</italic>. are fueled by the necessity to find new methods of malaria control. <italic>Plasmodium falciparum </italic>resistance to most common antimalarials such as chloroquine, sulfadoxine or pyrimethamine, is now-days widespread, and resistance to other antimalarials is increasing alarmingly. There is recent evidence for emerging in vitro resistance to one of the major groups of drugs, artemisinins [##REF##16325698##1##] and treatment failures to recently introduced drug combinations including artemisinin derivatives [##REF##17292769##2##,##REF##16704735##3##]. Research priorities include the discovery of new drugs, the understanding of drugs' mode of action and resistance mechanisms. Whatever the drug's mode of action, a better appraisal of the biological processes taking place in a parasite dying within an erythrocyte is needed to uncover additional drug targets and potentiate the effect of existing weapons. Towards reaching such goals, much hope rests on high through-put technologies such as DNA microarrays aimed at the study of gene expression at the genome scale and proteomics. Both approaches are complementary, and each has its own limitations. Proteomic approaches are complicated by solubility issues, detection of low abundance proteins and post-translational alterations. Both proteome and transcriptome approaches investigate steady state levels, a net sum of synthesis and decay of the gene product. Transcriptome studies are limited by the uncertain correlation of steady state RNA with protein levels. They nevertheless deliver a comprehensive and sensitive, genome-wide exploration of expression profiling. This proved invaluable in deciphering the developmental cycle gene expression patterns in malaria parasites [##REF##12929205##4##,##REF##12893887##5##].</p>", "<p>Few studies have shown that the transcriptome of <italic>P. falciparum </italic>can be altered by exposure of the parasite to a drug. Analysis of the parasite transcriptome under doxycyclin showed that apicoplast gene expression was deeply altered [##REF##16940111##6##]. This study is so far the only published example of a specific transcriptome response to a drug related to the drug's known target-pathway. No significant changes of the mRNA levels for enzymes involved in the lipid biosynthesis pathway could be evidenced upon exposure for 24–36 hours to the antimalarial choline analog T4, which targets the inhibition of phosphatidyl choline biosynthesis [##REF##17690387##7##]. In a recent study of the transcriptome of asynchronous <italic>P. falciparum </italic>cultures under chloroquine pressure [##REF##17475254##8##], around 600 genes were differentially expressed in presence of the drug, but only 38 of these were observed in two different experiments. Such poor reproducibility of results along with the low expression ratios observed for these 38 genes was interpreted as reflecting limited reactivity of the parasite transcriptome to environmental stimuli and the possible importance of post-transcriptional gene expression regulation in <italic>Plasmodium</italic>. However, the observed discrepancies may relate to different proportions of developmental stages in the 2 populations explored. Indeed, parasite maturation stage is an essential variable to control, as a majority of <italic>P. falciparum </italic>genes are differentially expressed during the 48 hour erythrocytic cycle [##REF##12929205##4##]. Such developmentally regulated control of gene expression hinders the interpretation of results if drug treatment exerts a retardation effect on the parasite erythrocytic cycle.</p>", "<p>We have used here a novel strategy and analysed the alterations of the parasite transcriptome in synchronous parasite cultures shortly after exposure to a lethal dose of artesunate, a rapidly acting drug. Artemisinin derivatives are potent antimalarials that are the cornerstone of currently recommended drug combination to treat <italic>P. falciparum </italic>malaria [##UREF##0##9##]. Artemisinins, which are endoperoxide-containing sesquiterpene lactones are structurally distinct from all other anti-malarials. Importantly they are the most rapidly acting antimalarials known to date, which moreover are active on young as well mature erythrocytic stages [##REF##15555542##10##]. These unique properties allowed the kinetic analysis of the events occurring in synchronous cultures of parasites exposed to a lethal drug concentration at different time points along the intraerythrocytic developmental cycle. We explored the dynamics of transcriptome alteration during the process leading to parasite death in different parasite stages. This identified 398 genes with dynamic transcriptome alterations after exposure to artesunate.</p>" ]

[ "<title>Methods</title>", "<title>Parasite cultures</title>", "<p>The FCR3 strain (FUP/CB line) [##REF##1944415##62##] of <italic>P. falciparum </italic>was used throughout the experiments. Parasites were cultured using the method described by Trager and Jensen with the modifications described in Ralph et al. [##REF##16277748##63##]. Cultures were tightly synchronized by lysing maturing forms through treatment with 0.3 M Alanine at two successive cycles, starting 3 hours after rings appeared in the culture [##REF##16277748##63##]. Artesunate treatment was performed the following cycle (see below).</p>", "<title>Drug treatment</title>", "<p>Artesunate powder (Arenco, Belgium) was dissolved in PBS, and the 1 μg/μL aliquoted stock solution was kept at -20°C. To detect possible loss of activity during storage of stock solution, the IC₅₀of the drug was tested periodically. The IC₅₀remained stable at around 3.4 nM (2.2–4.6 nM, SD = 1.2). This also confirmed the artesunate sensitivity of the FCR3 strain.</p>", "<p>Determination of minimal drug concentration irreversibly damaging 100% of the parasites: The lethal artesunate dose to be used throughout this investigation was determined using synchronized cultures incubated with increasing concentrations of artesunate (30-100-300-500-3000 ng/mL) for 3 hours, washed twice with RPMI and reincubated at 37°C for 24–30 hours, until next reinvasion. Reinvasion was assessed by examining Giemsa stained blood smears. The lethal dose was the minimal artesunate concentration needed to prevent 100% of reinvasion.</p>", "<p>Parasite treatment for studies of the transcriptome: artesunate was added at five different time points during the cycle following parasite synchronization, staggered between 20 hours and 30 hours after appearance of the first rings in the culture (at approximately 20, 22, 26, 28 and 30 hours), while untreated cultures were cultured in parallel in the absence of drug. Parasites were extracted in Trizol after 90 minutes and 3 hours of incubation.</p>", "<title>Total RNA preparation</title>", "<p>Parasite cultures were harvested by centrifugation and lysed in 5 pellet volumes of Trizol (Gibco) before freezing at -80°C. Total RNA was prepared from thawed samples following the manufacturer's instructions. RNA Quality was assessed with an Agilent 2100 Bioanalyser.</p>", "<title>Microarrays and hybridizations</title>", "<p>The microarrays used here have been described elsewhere [##REF##16277748##63##]. Briefly, glass slides were spotted with 7392 70-mer oligonucleotides originating from the Malaria Oligo Set (Qiagen-Operon) and custom design oligonucleotides, covering most of all <italic>P. falciparum </italic>genes. Of the 5542 identified <italic>P. falciparum </italic>genes, the array covers 4741 genes (85%). RNA labelling and hybridization were performed as described [##REF##16277748##63##]. For the comparison of drug versus no drug at each time point, dye swaps with two technical replicates were performed to compensate dye effect and to assess technical reproducibility, leading to 4 hybridized slides per experiment.</p>", "<p>In order to determine the genes differentially expressed between time 0 and 3 hours due to parasite development, three series of hybridizations were performed with cDNAs prepared at times 20/23 hours, 23/26 hours and 25/28 hours from the control cultures left without artesunate. Microarray data were deposited in ArrayExpress [ArrayExpress:E-MEXP-1435].</p>", "<title>Statistical analysis</title>", "<p>All the slides were analyzed pooled together using R [##UREF##2##64##] and Bioconductor [##REF##15461798##65##] with the limma package [##REF##16646809##66##,##UREF##3##67##]. After logarithm transformation of ratio of median of the intensities (without background substraction) in the two channels, data were weighted to penalize spots of poor quality, flagged, or spots for which intensities in the two channels were close to background (median of intensity less than median of background plus 2 standard deviations for both Cy3 and Cy5 signals). In a second step, an intensity-dependent normalization was applied to each slide (loess), followed by quantile normalization [##UREF##4##68##] applied across the arrays to ensure that the log ratios had the same empirical distribution across arrays and across channels. After fitting a linear model to the expression data for each probe using the least square method, comparisons of interest were extracted with a contrast matrix. A moderated t-statistic and a log-odds of differential expression were computed for each contrast for each gene. Finally, the raw p-values were adjusted using the Holm procedure and a type I error rate of 0.05 was applied. When for a given gene significant log ratios were obtained for several oligos, the retained log ratio was one with the maximum absolute value.</p>", "<title>Reverse trancription and qPCR</title>", "<p>RNA samples were treated with DNAse I (Invitrogen) for 15 minutes at 37°C in presence of RNAseOUT™ inhibitor (Invitrogen). RNA was reverse transcribed with Superscript II RT (Invitrogen) using random hexamers with pre-extension step of 10 minutes at 25°C followed by extension for 50 minutes at 42°C. Then, cDNA was treated with RNAse H for 20 minutes at 37°C.</p>", "<p>Quantitative real time PCR was performed on cDNA using an ABI Prism 7900 HT thermal cycler system (Applied Biosystems) for 40 cycles (95°C for 15 s, 55°C for 15 s and 60°C for 45 s). Reactions were done in 20 μL volumes using SYBR Green PCR master Mix (Applied Biosystems) and 0.5 μM primers in triplicate with 6 concentrations of cDNA (2500 pg/μL, 625 pg/μL, 156 pg/μL, 39 pg/μL and 10 pg/μL) to assess primer amplification efficiency. Absence of DNA contamination was check using RNA sample treated as cDNA without RT (2500 pg/μL).</p>", "<p>Primers were designed using eprimer3 [##UREF##5##69##] [See Additional file ##SUPPL##4##5## for the primer sequences]. Transcript abundance was compared using mean of ΔΔCt values calculated for all cDNA dilutions using PFI0425w (putative transporter) as endogenous normalizer and the control condition as reference. This gene was chosen as it remained non differentially expressed in all our experiments and showed very moderate levels of variation in whole-cycle transcriptome analysis [##REF##12929205##4##].</p>" ]

[ "<title>Results</title>", "<title>Choice of experimental design</title>", "<p>In order to work with homogeneous parasite populations, tightly synchronized parasites were used in all experiments, along with an artesunate concentration which irreversibly damages 100% of the parasites within 3 hours. The lethal dose of artesunate, defined here as the minimal concentration needed for a 3 hour exposure to completely prevent reinvasion (see \"Materials and methods\" section), was 780 nM. Importantly, it corresponds to 100–1000 times the IC₅₀, this dose remains physiological, being in the range of peak plasma concentration reached in patients administered artesunate (1300–8400 nM) [##REF##15504846##11##].</p>", "<p>Each experiment consisted in an artesunate-free culture (control) and an artesunate treated culture. For each time point explored, cDNA from parasites with and without the drug were labeled with different cyanines, mixed and hybridized. To analyse dynamic transcriptome alterations, 2 pilot experiments were performed in which RNA was harvested at 30, 90, 180 minutes and 10 hours of incubation with the drug. No significant changes in gene expression could be detected at 30 minutes. In contrast, after 10 hour incubation time, major transcriptome alterations occurred under artesunate, but their analysis was un-interpretable due to drug-induced massive slowing-down of parasite development (data not shown). We thus decided to focus on 2 drug-exposure time, namely 90 minutes and 3 hours after drug addition.</p>", "<p>Since artesunate is active on ring stages as well as on older trophozoite stages, we decided to explore different time points along the erythrocytic cycle and search for genes affected at each time point. Therefore, 5 drug treatment experiments, staggered between 20 hours and 30 hours of parasite development, were performed for the comparison of drug versus no drug at 90 minutes and 3 hours.</p>", "<title>Transcript changes under artesunate exposure</title>", "<p>We looked for genes displaying similar changes across all experiments, with the hypothesis that responses to artesunate would not be stage-dependent within the explored 10 hour window of the developmental cycle. The results of 5 experiments performed at time points staggered between 20–30 hours post-invasion, were analyzed using ANOVA with a Holm-Bonferroni method of p-value adjustment, p &lt; 0.05. This led to a list of genes with similar ratios of expression (i. e. low variance). From this list, only genes differentially expressed at 3 hours under artesunate and not after 3 hours in the absence of the drug were retained (p = 1). It is important to stress that what we refer to, for convenience, as over- or under-expression in fact refers to higher or lower steady state RNA levels.</p>", "<p>Due to the high number of microarrays analyzed (32 at 3 hours), power of the statistical analysis was such that expression log ratios as low as +/- 0.3 could be found for genes that displayed significant changes by ANOVA [for distribution of log ratios, see Additional file ##SUPPL##0##1##]. This appeared questionable from a biological standpoint and we chose to use a cut-off log-ratio of +/- 0.8. This value is lower than what is usually (and often arbitrarily) chosen in most transcriptome studies, i.e. +/- 1, but we felt it was justified given the large amount of data analyzed here and the low amplitude of fold-changes reported in previous <italic>P. falciparum </italic>transcriptome studies [##REF##17475254##8##,##REF##17283083##12##].</p>", "<p>In addition, as genes of potential interest could be differentially expressed in the controls but more-so or differently under artesunate, we needed to identify these and filter out the developmentally regulated gene expression profiles. Indeed, the extent to which the drug slowed-down parasite development is unknown. Thus, different expression levels in artesunate-exposed vs. control (unexposed) cultures could reflect the direct effect of artesunate together with a possible difference in growth rate/developmental stage. To identify the developmentally regulated genes in the artesunate-free control culture, the 3 hours RNA was hybridized against the time 0 RNA (3 control experiments). Even within such a short time window, 81 genes showed differential expression in the control culture. Filtering-out this potential confounding \"slowing-down effect\" established a second list of genes differentially expressed in the artesunate experiments and in the control experiments, but with significantly different expression-ratios under these two conditions (p &lt; 0.05) [see Additional file ##SUPPL##1##2## for the list of differentially expressed genes].</p>", "<p>This resulted in the following: out of the 4703 genes analyzed, 398 (approx 8.5%) were differentially expressed after 3 hours in presence of artesunate, 244 over-expressed (approx 5%) and 154 (approx 3.5%) under-expressed. Of these 398 genes differentially expressed at 3 hours, 42 showed significant expression changes at 90 minutes, presumably reflecting an early response to the drug.</p>", "<p>Our technically demanding and relatively stringent approach may have generated false negatives, but we wanted to be as confident as possible about the list of genes selected as being differentially expressed. This confidence was supported by the results of qRT-PCR performed with 12 over-expressed and 4 under-expressed genes, which confirmed the microarray data (Spearman correlation coefficient: 0.81) [see Additional file ##SUPPL##2##3## for the comparison between microarray and qRT-PCR results].</p>", "<title>Classification of differentially expressed genes</title>", "<p>Interesting features were observed in the group of differentially expressed genes. For genes expressed differentially at 3 hours, there was a strong positive bias in the sub-set of over-expressed genes towards polymorphic genes located in sub-telomeric position and carrying PEXEL/HCT motifs [##REF##16540187##13##] used for export beyond the parasitophorous vacuole into the erythrocyte cytoplasm, or to the red blood cell membrane/surface (Figure ##FIG##0##1##). This was determined after excluding from the analysis the strain-specific <italic>var, stevor </italic>and <italic>rifin </italic>multigene families – which are not adequately represented in our case since the parasite strain used differs from 3D7. Concomitant with this positive bias, there was a negative bias towards conserved genes in the over-expressed genes. In contrast, the group of under-expressed genes displayed a positive bias towards conserved genes and a negative bias towards polymorphic genes.</p>", "<p>Functions (in most cases putative) were assigned to 213 of the 398 differentially expressed genes [##REF##17283083##12##,##REF##12519984##14##, ####REF##14681429##15##, ##UREF##1##16##, ##REF##15256513##17##, ##REF##15023358##18##, ##REF##17307260##19##, ##REF##17428722##20##, ##REF##17941702##21####17941702##21##] [see Additional file ##SUPPL##1##2##]. Apart from lipid and purine/pyrimidine metabolism, few genes belonged to the metabolic pathways of the parasite (as defined by Ginsburg [##UREF##1##16##]). Most genes with altered expression under artesunate pressure were related to chaperones, transporters, cell cycle, kinases, Zn finger proteins, transcription activating proteins, proteins involved in proteasome degradation, oxidative stress and in cell cycle regulation. A positive bias towards over-expression could be observed for genes related to oxidative stress, kinases, transcription associated proteins and chaperones, and towards under-expression for genes related to proteasome degradation and transporters (Figure ##FIG##1##2##). Slowing down of parasite development may be linked to the differential expression of three genes involved in cell cycle regulation [see Additional file ##SUPPL##3##4##]. Two genes were over-expressed, namely PFF1125 encoding a putative RNA binding protein mei2 homologue and PFL1855 encoding a putative cell cycle control protein, while PFA0345w, encoding a putative centrin, was under-expressed.</p>", "<title>Oxidative stress</title>", "<p>Oxidative stress is likely to occur in parasites submitted to artesunate, whether linked to the direct action of the drug or to the reduced capacity of the damaged parasite to inactivate free radicals. The antioxidant defense in <italic>P. falciparum </italic>involves both the glutathione and the thioredoxin system, the relative contribution of each remaining unknown [##REF##15037104##22##]. Four genes involved in the parasite antioxidant defense [##REF##15037104##22##,##REF##15245577##23##], showed altered transcription profiles (Table ##TAB##0##1##) : gamma glutamyl cystein synthetase and glutathione synthetase (both involved in the glutathione system) were over-expressed. In contrast, the genes encoding protein disulfide isomerase related protein (potentially involved in oxidative protein folding, catalyzing both the oxidation and isomerization of disulfides on nascent polypeptides [##REF##10597631##24##]) and glutathione peroxidase (shown to rather encode a thioredoxin peroxidase [##REF##11087748##25##] and thus involved in the thioredoxin system) were under-expressed. The consecutive action of gamma glutamyl synthetase and glutathione synthetase leads to the production of reduced glutathione (GSH), which plays a pivotal role in the antioxidant defense through maintenance of the red-ox state of protein -SH moeties, the reduction of the noxious hydrogen and lipid peroxides and the extrusion of toxic compounds (including drugs) [##REF##15037104##22##]. The predicted outcome of the over-expression of these 2 genes would be an increased supply of GSH in response to the oxidative stress induced by artesunate.</p>", "<title>Chaperones, protein trafficking and related genes</title>", "<p>Genome mining on parasite chaperone analogues has led to the conclusion that approximately 2% of the <italic>P. falciparum </italic>genes encoded potential or confirmed chaperones [##REF##17307260##19##]. Chaperone proteins have diverse functions, such as serving as transcription factors, regulating the cell cycle, being involved in protein degradation, binding damaged proteins, protein transport across membranes and protein folding. As expected in a parasite under severe injury, 13 genes encoding chaperone or chaperone-related proteins (Table ##TAB##1##2##) were found up-regulated at 3 hours of artesunate pressure, among which PfHsp70, PfHsp90 and 6 genes belonging to the Hsp40 chaperone machinery of <italic>P. falciparum </italic>[##REF##17428722##20##]. The first four were already over-expressed at 90 minutes, along with PFD0080c (a member of the <italic>resa</italic>-like DnaJ family) and thus can be considered part of the early response to the drug. Interestingly, most (9/13), chaperone-encoding genes up-regulated under artesunate have potential export signals, suggesting possible export of their corresponding proteins from the parasite to the erythrocyte cytoplasm or membrane. Chaperone involvement in trafficking of membrane or exported proteins has been highlighted in a recent analysis of the parasite chaperone network [##REF##17941702##21##]. In particular, analysis of the interactome suggested that a member of the Hsp40 family (PF14_0700), which interacts with PfHsp70 (PF08_0054) and PfHsp90 (PF07_0029) could function as a co-chaperone in the Hsp90 complex of the parasite to transport eythrocyte membrane exported proteins such as Antigen 332 (PF11_0507). It is remarkable to note that PF14_0700, PF08_0054, PF07_0029 and PF11_0507 were all over-expressed under artesunate pressure [see Additional file ##SUPPL##1##2##]. This suggests artesunate-triggered altered intracellular trafficking, contributing to remodeling of the parasitized erythrocyte membrane, a conclusion further substantiated by the observed increased steady state levels of genes encoding proteins interacting with the erythrocyte cytoskeleton such as MESA (PFE0040c), several members of the RESA family and number of genes carrying PEXEL/HT motifs. Alterations of the intracellular trafficking is also suggested by altered transcription profiles of genes such as signal peptidase (MAL13P1.167) as well as transporters, including organellar importers (see below).</p>", "<title>Transporters</title>", "<p>Eleven genes involved in transport were under-expressed (Table ##TAB##2##3##). The putative UDP galactose antiporter (PF11_0141) down-regulated at 3 hours, was the only transporter found down-regulated at 90 minutes, suggesting that alteration of its transcript levels is an early event. Four subunits of the vacuolar ATP synthase were under-expressed. This constitutes one of the few cases of concommittant regulation of individual components of a specific machinery or pathway in this study, suggesting that indeed the activity of the vacuolar ATP synthase was severely impaired. This enzyme contributes to acidification of vacuolar and organelle contents.</p>", "<p>Only 4 genes encoding transporters were over-expressed, amongst which <italic>pfmdr1 </italic>(PFE1150w), the P-glycoprotein homologue of <italic>P. falciparum</italic>, which modulates susceptibility to antimalarial drugs such as quinine, mefloquine and artemisinins [##REF##10706290##26##, ####REF##16845638##27##, ##REF##16794577##28####16794577##28##]. An association has been shown in the field between sensitivity to arylaminoalcohols or endoperoxides, <italic>pfmdr1 </italic>allelic status and gene copy number [##REF##10582887##29##], suggesting that an over-expressed functional <italic>pfmdr1 </italic>conferred a multidrug resistance like phenotype [##REF##15288742##30##,##REF##15876420##31##]. Over-expression of <italic>pfmdr1 </italic>we show to occur under artesunate pressure may have an impact on susceptibility to the drug(s) associated to artemisinins upon ACT administration. It also suggests that not only gene copy number but also gene transcription rates and mRNA stability should be monitored when exploring mechanisms of field parasite drug resistance.</p>", "<title>Lipid metabolism and the Apicoplast</title>", "<p>Eight genes involved in lipid metabolism were differentially expressed (Table ##TAB##3##4##).</p>", "<p><italic>P. falciparum </italic>contains an unusually high number of Acyl CoA synthases (ACS) and binding proteins that might play a role in fatty acid salvage and transport from the host cell [##REF##17715365##32##,##REF##16860410##33##]. Amongst the 13 different ACS genes, a non subtelomeric acyl CoA synthetase gene PFB0685c has expanded into a family of 9 duplicated genes mainly located in the subtelomeric regions of the genome and transcribed during the erythrocytic stage of the parasite. At least two ACS proteins PfACS1 (PF14_0761) and PfACS3 (PFL2570w) were shown to interact with ankyrin through their lysine-rich C-terminal domain [##REF##12850263##34##]. Similar domains are only present in the genes of this family, possibly targeting these enzymes to the erythrocyte cytoplasm and/or membrane, an interesting possibility as all the ACS proteins have a signal peptide but are devoid of the PEXEL/HT motif. This would contribute to the capacity of <italic>P. falciparum </italic>to activate fatty acid scavenged from the plasma or the erythrocyte membrane, a critical metabolic function for the survival of the parasite. All 4 ACS genes over-expressed under artesunate (PFB0685c, PFB695c, MAL13P1_485 and PF14_0761) belong to this family.</p>", "<p>Two acyl-CoA binding protein genes, PF08_0099 and PF10_0016, were under-expressed. PF14_0664, an Acetyl CoA carboxylase which generates Malonyl-CoA, likely to be located in the apicoplast [##REF##17715365##32##], was already found over-expressed at 90 minutes. Interestingly, the triose phosphate transporter located in the outermost membrane of the apicoplast, PFE0410w, was under-expressed under artesunate. This transporter, also called PFoTPT, fuels the apicoplast by providing carbon, reducing power and ATP [##REF##16760253##35##], but also quite importantly, dihydroxyacetone phosphate (DHAP), the precursor needed for isoprenoid and for phospholipids biosynthesis [##REF##15083156##36##]. Altogether these data suggest an alteration of fatty acid metabolism both intracellularly and in the apicoplast.</p>", "<title>Mitochondrial genes and genes encoding proteins targeted to the mitochondrion</title>", "<p>The main metabolic function allowed by the active mitochondrial electron transport chain maintained by <italic>Plasmodium </italic>is regeneration of ubiquinone, which is required as the electron acceptor for dihydroorotate dehydrogenase, an essential enzyme for pyrimidine biosynthesis [##REF##17330044##37##]. Indeed, 3 important genes involved in mitochondrial electron transport are under-expressed in presence of artesunate: 2 subunits of the cytochrome c oxydase (downstream of the cytochrome bc1 complex inhibited by atovaquone) coxI and coI and ubiquinol-cytochrome c reductase hinge protein, PF14_0248 (Table ##TAB##4##5##). This interference with the electron transport chain could in turn affect pyrimidine biosynthesis.</p>", "<p>PFC0975c encoding a peptidyl-prolyl cis-trans isomerase cyclophilin-type belonging to the chaperone network of the mitochondrial matrix [##UREF##1##16##] is under-expressed in response to artesunate. Cyclophilin D/peptidylprolyl isomerase activity is a regulator of mitochondrial permeability transition, a non selective inner-membrane permeabilization occuring in response to increased calcium load and redox stress. These 2 conditions are induced by artesunate. Such a phenomenon can lead to necrosis through activation of phospholipases, proteases and nucleases [##REF##16935572##38##].</p>", "<title>Purine/pyrimidine metabolism</title>", "<p>In addition to the possibly impaired respiratory chain interfering with pyrimidine synthesis, expression of other genes related to purine/pyrimidine biosynthesis was also modified by artesunate pressure (Table ##TAB##5##6##): cytidine and deoxycytidylate deaminase (PF13_0259) hypoxanthine phophorybosyl transferase (PF10_0121), deoxyuridine 5'triphosphate nucleotidohydrolase (PF11_0282) and uridine phosphorylase (PFE0660c) are all under-expressed, while dihydroorotase (PF14_0697) is over-expressed. Interference with such key metabolic processes may readily cause parasite death.</p>", "<title>Signaling/kinases</title>", "<p>The kinome of <italic>P. falciparum </italic>is atypical, with an atypical MAP kinase family and a particular R45-FIKK kinase family [##REF##15479470##39##].</p>", "<p>Fourteen genes encoding kinases, some of which potentially related to signal transduction, were differentially expressed in presence of artesunate (Table ##TAB##6##7##). Four of the 11 over-expressed genes were already differentially expressed at 90 minutes, among which the calcium calmodulin dependent protein kinase 2, PfPK2 (PFL1885c). PfPK2 belongs to the CamK group of <italic>P. falciparum </italic>kinases, a family of kinases which can be activated in response to increased calcium levels, possibly resulting from specific inhibition of the SERCA PfATPase6 by the drug [##REF##12435441##40##]. The other 3 genes with increased expression levels at 90 minutes were members of the R45-FIKK multigene family [##REF##15023358##18##,##REF##15752424##41##]. This Apicomplexa-specific kinase gene family is composed of 20 genes all presenting sub-telomeric positions, scattered over 11 chromosomes. It has recently been shown that these kinases localize in different compartments of the infected erythrocyte, some being associated with the erythrocyte membrane [##REF##17181785##42##]. Under artesunate pressure, 3 members of the R45-FIKK family were over-expressed at 90 minutes (PFI0095c, MAL7P1.144 and PFL0040c) and remained at elevated levels subsequently, while two additional R45-FIKK Kinases displayed increased transcript levels at 3 hours, namely PFD1165w and PF11_0510. All 5 present export motifs. PFD1165w and PFL0040c are associated with Maurer's clefts, and PFL0040c with the infected erythrocyte membrane as well [##REF##17181785##42##].</p>", "<title>Transcription associated proteins; zinc finger genes</title>", "<p>Transcriptome studies of parasites during the erythrocytic stages of development have shown that in many cases, transcription is soon followed by translation, when the parasite \"needs\" the considered proteins to further develop [##REF##12929205##4##]. In contrast, comparison between transcriptomic and proteomic studies have shown the existence of discrepancies in RNA and protein expression for about 50% of the genes expressed during the erythrocytic development [##REF##15520293##43##,##REF##15637271##44##]. This has been clearly confirmed for certain genes, such as <italic>var </italic>genes transcribed at the early ring stage and translated several hours later [##REF##16848786##45##] or a series of sexual stage proteins the corresponding genes of which are transcribed long before translation occurs [##REF##9364968##46##]. Genome mining revealed that relatively few genes (156, 2/3 less than expected) encode transcription associated proteins [##REF##15256513##17##]. Six non-Zn finger transcription associated protein genes, as defined by Coulson [##REF##15256513##17##], were over-expressed under artesunate, 3 were under-expressed (Table ##TAB##7##8##). This relatively low number suggests that transcriptional regulation may not play a major role in gene-expression modulation under drug-induced stress. Altered decay rates may modulate gene expression in drug-exposed parasites and contribute to the dynamic changes of mRNA levels observed here. In this regard, it is worth noting the recently outlined decreasing RNA decay rate as the parasite matures. A list of genes encoding putative decay components and possibly contributing to such a mechanism of post-transcriptional regulation in <italic>P.falciparum </italic>has been proposed [##REF##17612404##47##]. However, there was no observed effect of artesunate on expression of these genes.</p>", "<p>The low number of TAP identified by Coulson [##REF##15256513##17##] contrasted with an over-representation of proteins with Zn finger motifs [##REF##17210253##48##] potentially involved in regulating RNA stability. In the present study, 12 genes identified as putative Zn finger bearing proteins were differentially expressed in the presence of artesunate; 10/12 were over-expressed, two of which (PFI0470w and PFE1245w) already at 90 minutes (Table ##TAB##8##9##).</p>" ]

[ "<title>Discussion</title>", "<p>The methodology adopted here allowed us to find reproducible alterations of the transcriptome in <italic>P. falciparum </italic>parasites under artesunate pressure. To overcome the major hurdle of developmentally regulated genes, we combined the use of a fast acting drug at high dose with limited incubation time at 5 successive but synchronized developmental stages. Most observed alterations were of low amplitude, possibly reflecting particular mechanisms of gene regulation in malaria parasites. This resulted in a relatively broad pertubation, that nevertheless was highly significant and consistent across experiments. We did not detect altered expression of a single specific metabolic pathway or of the putative drug target molecules but several functionally important groups of genes exhibited interesting dynamic changes. Within the intrinsic limitations of all transcriptome studies (in which steady-state RNA levels are studied and not protein quantities, post-translational modifications or biological activities), the overall picture that emerges from our current analysis of the transcriptome changes suggests i) pertubation of the intracellular trafficking and organisation, including changes in chaperones, transporters, remodeling the erythrocyte space beyond the parasitophorous vacuolar membrane up to the membrane of the infected erythrocyte; ii) probable down-phasing of metabolism as deduced from down-regulation of key enzymes/transporters implicated in purine, pyrimidine and isoprenoid synthesis; iii) an altered mitochondrion; iv) an altered redox homeostasis and possibly altered protein turnover. In addition, the transcriptome analysis highlighted numerous genes with unknown function, annotated as coding for hypothetical proteins [see Additional file ##SUPPL##1##2##]. Some display consistent high and early up-regulation and clearly deserve further investigation.</p>", "<p>The physiological significance of the observed perturbations need to be assessed in future studies. Transcriptome analysis is the first of a multi-step process, to be followed by scrutiny of the role of the individual genes identified here. In particular, it will be essential to establish whether or not the genes in question are implicated in a rate-limiting process and to analyse the dynamics of the corresponding protein products, including their turnover, post-translational modifications and cellular localization. An important question is to clarifiy which transcriptome alterations are specific to the response to artemisinins and which are related to a parasite lethally injured, as this has numerous implications for drug development.</p>", "<p>We will discuss the transcriptome data in the frame of existing knowledge on mechanisms of action of artemisinin derivatives and mechanisms resulting in reduced susceptibility/resistance. We will next discuss possible common changes in steady state RNA levels after different stresses inflicted to the parasite such as exposure to 41°C or treatment with chloroquine, that can cause parasite death.</p>", "<p>The mechanism of action of artemisinins is debated. Artemisinin are endoperoxide-containing sesquiterpene lactones. Fe⁺⁺-dependent activation of the endoperoxide bridge is required for the drug to be active [##REF##17640025##49##]. Cleavage of the endoperoxide moiety forms highly reactive oxyl radicals that rearrange to more stable carbon-centered radical intermediates. These in turn form covalent adducts with parasite products thought to be responsible for a pleiotropic effect of the drug. Such pleiotropicity may explain the variety of genes we show here to be affected in their expression. Recent evidence suggest more specific mechanisms, such as inhibition of specific targets, with specific inhibition of the <italic>P. falciparum </italic>SERCA-type Ca++ pump (PfATPase6) [##REF##12931192##50##]. Importantly, field isolates with markedly reduced <italic>in vitro </italic>susceptibility to artemether presented a mutant SERCA-type PfATPase6 [##REF##16325698##1##,##REF##17954693##51##]. Artemisinin has been shown to inhibit the endoplasmic reticulum – located SERCA-type in <italic>Toxoplasma gondii</italic>, a related Apicomplexan parasite as well [##REF##17766463##52##]. We did not observe altered steady state levels of <italic>PfATPase6 </italic>mRNA at any time point investigated, indicating that over-expression of this gene is not part of the parasite response to artesunate.</p>", "<p>Ultrastructural alterations of the morphology and their kinetics are debated as well. Early-stage alterations of the mitochondrion, the endothelial reticulum and the digestive vacuole were detected in some studies [##REF##8214279##53##] but not others [##REF##17938190##54##]. Some of the transcriptome alterations may account for a disorganisation of the digestive vacuole and the mitochondrion. We observed increased levels of <italic>Pfmdr1 </italic>expression concomitant with decreased mRNA levels of four of the subunits of the H⁺vacuolar ATPase, which is implicated in regulating calcium intracellular stores of acidic compartments. Downproduction of the H⁺vacuolar ATPase and/or alteration of its subunit ratio is predicted to negatively impact on intracellular calcium homeostasis. This is most probably on the critical path, since disruption of calcium homeostasis is central to cellular death, necrosis endophagy and apoptosis. Our transcriptome analysis also shows evidence that artesunate interferes with expression of genes related to the mitochondrion, in most cases by down-regulating their steady state RNA levels. This calls for additional studies on the mitochondrial activity after artesunate exposure.</p>", "<p>Additional factors are to be considered in light of the transcriptome data, in particular the intracellular partitioning of the drug. Several studies have demonstrated the selective uptake of artemisinin derivatives. It has been shown that artemisinin and dihydroartemisinin, two sesquiterpene endoperoxide drugs closely related to artesunate, are transported by the tubulovesicular network (TVN) formed within the erythrocyte cytoplasm for nutrient import to the parasite, and that these drugs disrupted the protein organization of the TVN [##REF##10413044##55##]. Numerous genes implicated in intracellular space modelling, trafficking, chaperones and transport display altered transcription profiles in artesunate-treated parasites. The markedly bias for such genes was indeed the most salient observation of our transcriptome study. It is tempting to speculate that this results in remodelling the intracellular space, including remodelling of the infected red blood cell membrane and TVN. This could either be a response aimed at correcting drug-inflicted damage, preventing further intake of drug or reflect the dysfunction resulting from such damage. Protein trafficking is all the more important for parasite growth and survival that the parasite develops inside a parasitophorous vacuole within the erythrocyte cytoplasm. Interactions between the parasite and the erythrocyte are multiple, with an increasing number of parasite proteins recognized as being involved (reviewed in [##REF##17223924##56##]). Parasite-induced modifications of the red blood cell through interactions of parasite proteins exported to the host cell membrane or cytoskeleton play a major role in parasite survival and virulence, through induction of infected red cell cytoadhesive properties or decreased deformability. Altered transcript levels of genes involved in host cell remodelling and intracellular trafficking have been reported in parasites lethally damaged by exposure to febrile temperatures [##REF##17283083##12##]. This suggests that perturbing intracellular trafficking/remodeling is either on the path to parasite death or an attempt to overcome a lethal injury. This also suggests that alterations of intracellular trafficking by artesunate may be one of the mechanisms through which the drug can be active on a wide range of parasite developmental stages.</p>", "<p>Analysis of the proteome of artemether-treated <italic>P. falciparum </italic>parasites showed that 30 of the 101 proteins identified using a MALDI-TOF-MS-based analysis [##REF##15832369##57##], displayed a greater than three-fold expression level compared to untreated control cultures. Two of the over-expressed proteins, namely hsp90 (PF07_0029), PF14_0425 (fructose-biphosphate aldolase) also presented an increased transcript level in our study. Direct comparison of both studies is precluded due to limitations such as different sensitivities of the transcriptome and proteome approaches, more comprehensive coverage of the expressed products in the transcriptome approach, different drug exposure protocols and/or transcription/translation uncoupling, which has been described for approx 50% of the proteins [##REF##15520293##43##,##REF##15637271##44##].</p>", "<p>To put the alterations observed under artesunate in perspective with changes induced by unrelated forms of stress, we analyzed the data in the light of the transcriptome alterations reported after chloroquine [##REF##17475254##8##] treatment or incubation at 41°C [##REF##17283083##12##], both treatments reported as having led to parasite death. The transcriptome studies are not directly comparable, as they were performed under different experimental conditions (synchronous/asynchronous parasites, glass slide microarrays/Affymetrix chips) and corresponded to different natures of stress (rapid acting vs.slow acting). However, identification of a few genes with altered expression under such different experimental conditions provides interesting insight. Such joint analysis is illustrated in the Venn diagram (Figure ##FIG##2##3## see also tables ##TAB##0##1##, ##TAB##1##2##, ##TAB##2##3##, ##TAB##3##4##, ##TAB##4##5##, ##TAB##5##6##, ##TAB##6##7##, ##TAB##7##8##, ##TAB##8##9##, 10 for different functional groups). There were only 4 genes identified as differentially expressed under all 3 types of stress: PFB0095c encoding PfEMP3, PFE1245w, encoding a putative zinc finger protein, and PF14_0151 encoding a putative RNA-binding protein and MAL7P1.171 encoding a hypothetical protein. PfEMP3 and MAL7P1.171 have an export motif. PfEMP3 is exported to the cytoplasmic face of the erythrocyte membrane in the mature stages. It associates with the erythrocyte cytoskeleton, which it destabilizes [##REF##17626011##58##]. Both PfEMP3 and MAL7P1.171 have been shown to contribute to trafficking of PfEMP1 [##REF##16262789##59##,##REF##18614010##60##]. Their role in the response to lethal stresses deserves to be investigated. Characterization of the products encoded by the remaining 2 common genes certainly warrants particular attention, as these genes may play a crucial role in the general response to stress.</p>", "<p>The transcriptome of chloroquine-treated and artesunate-treated parasites shared three additional altered gene profiles, namely PFE1455w encoding a putative sugar transporter, PFI0095c and PFL0040c both encoding putative kinases belonging to the R45-FIKK family. Interestingly again, the three predicted proteins have an export motif. However, since there were large inter-experiment variations in the transcriptome analysis of chloroquine-treated parasites, commonalities are probably largely underestimated.</p>", "<p>Ninety four genes were differentially regulated both under heat shock and artesunate, 63 over-expressed and 31 under-expressed under both conditions, amongst which members of the R45-FIKK kinase family, chaperones, molecules involved in lipid metabolism and number of hypothetical proteins -including as discussed above numerous genes coding for proteins with an export motif. Two of the 5 R45-FIKK kinase genes over-expressed under artesunate were also over-expressed under chloroquine pressure [##REF##17475254##8##]. Of the 3 R45/FIKK kinase genes found over-expressed under heat shock, one was also differentially expressed under chloroquine. These data suggest that expression of genes belonging to the R45-FIKK kinase family reflects the response of the parasite to different types of environmental modifications and that the parasite response to different natures of stress may include common pathways. The same could be said about genes belonging to 2 other families. Eight of the 13 chaperone/chaperone-related encoding genes over-expressed under artesunate were also over-expressed under heat-shock. Of the 30 Hsp40 type <italic>P. falciparum </italic>chaperones, 6 were over-expressed under artesunate, 4 in common with heat-shock.</p>" ]

[ "<title>Conclusion</title>", "<p>The rapid parasiticidal activity of artesunate at high doses allowed us to study the effects of the drug on the transcriptome of synchronized parasite cultures over a short 3 hour time window. A positive bias was observed in favor of subtelomeric localization, polymorphism and presence of potential export sequences for the subset of upregulated genes. The corresponding genes, related to protein trafficking, kinases or membrane remodeling, often belong to multi-gene families [##REF##17631186##61##]. Low amplitude transcript level alterations were observed, but the large number of affected genes likely induces substantial perturbations of the interface between the parasite and its host cellular environment. Whether these constitute parasite survival mechanisms or in contrast metabolic cytotoxic alterations remains to be established. The transcriptome analysis identified what may represent pathways leading to parasite death, such as inhibition of purine/pyrimidine metabolism, interference with the mitochondrial electron transport chain, with protein turn-over or with the integrity of the food vacuole and calcium stores.</p>", "<p>The high proportion of over-expressed genes encoding proteins exported from the parasite highlight the importance of extra-parasitic compartments as fields for exploration in drug research which, to date, has mostly focused on the parasite within its plasma membrane rather than within its intra- and extra-erythrocytic environment.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Translation of the genome sequence of <italic>Plasmodium sp</italic>. into biologically relevant information relies on high through-put genomics technology which includes transcriptome analysis. However, few studies to date have used this powerful approach to explore transcriptome alterations of <italic>P. falciparum </italic>parasites exposed to antimalarial drugs.</p>", "<title>Results</title>", "<p>The rapid action of artesunate allowed us to study dynamic changes of the parasite transcriptome in synchronous parasite cultures exposed to the drug for 90 minutes and 3 hours. Developmentally regulated genes were filtered out, leaving 398 genes which presented altered transcript levels reflecting drug-exposure. Few genes related to metabolic pathways, most encoded chaperones, transporters, kinases, Zn-finger proteins, transcription activating proteins, proteins involved in proteasome degradation, in oxidative stress and in cell cycle regulation. A positive bias was observed for over-expressed genes presenting a subtelomeric location, allelic polymorphism and encoding proteins with potential export sequences, which often belonged to subtelomeric multi-gene families. This pointed to the mobilization of processes shaping the interface between the parasite and its environment. In parallel, pathways were engaged which could lead to parasite death, such as interference with purine/pyrimidine metabolism, the mitochondrial electron transport chain, proteasome-dependent protein degradation or the integrity of the food vacuole.</p>", "<title>Conclusion</title>", "<p>The high proportion of over-expressed genes encoding proteins exported from the parasite highlight the importance of extra-parasitic compartments as fields for exploration in drug research which, to date, has mostly focused on the parasite itself rather than on its intra and extra erythrocytic environment. Further work is needed to clarify which transcriptome alterations observed reflect a specific response to overcome artesunate toxicity or more general perturbations on the path to cellular death.</p>" ]

[ "<title>Abbreviations</title>", "<p>GSH: reduced glutathione; ACS: Acyl CoA synthetase; TAP: transcription associated protein; CQ: chloroquine; HS: heat-shock; ART: artesunate; SNP: single nucleotide polymorphism; HMM: hidden Markov model</p>", "<title>Authors' contributions</title>", "<p>ON establishment of experimental protocoles, parasite cultures, drug exposure, RNA extraction, q-PCR, genome mining, interpretation of results, manuscript preparation. EB establishment of experimental protocoles, microarray spotting, q-PCR, genome mining, statistical analysis, interpretation of results, manuscript preparation. GD parasite cultures, drug exposure, RNA extractions, manuscript preparation. CP microarray spotting, hybridizations. MAD quality control of arrays, statistical analysis. OS microarray spotting, quality control of arrays. GG quality control of arrays, statistical analysis. SB interpretation of results, genome mining. JP interpretation of results. OMP interpretation of results, manuscript preparation. JYC spotting of microarrays, statistical analysis, interpretation of results. PHD establishment of experimental protocoles, parasite cultures, RNA extraction, genome mining, interpretation of results, manuscript preparation. All authors have read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>The work received financial support from the Délégation Générale pour l'Armement (DGA n°22120/DSP/SREAF and n°04 34 025), the Programme PAL+/Fonds National pour la Science, the Institut Pasteur, the Fonds dédié \"Combattre les Maladies parasitaires\" (Sanofi-Aventis/Ministère de l'Enseignement supérieur et de la Recherche) and the Programme Génopole. ON was supported by the Thailand Research Fund through the Royal Golden Jubilee PhD programme (Grant No. PHD/0157/2542) to JP and EB by the DGA. The authors are grateful to Dr Zbynek Bozdech for invaluable advice during the early stages of the study and to Dr Genevieve Milon for helpful comments and critical review of the manuscript.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Association plots of gene expression with gene chromosomal localization, predicted export sequences and polymorphism</bold>. A : Bias in gene expression versus gene chromosomal localization (subtel = subtelomeric position of gene defined as &lt;150 kb from telomere). B: Bias in gene expression versus presence of predicted export signals [##REF##16540187##13##]. C: Bias in gene expression versus gene polymorphism (as defined by presence or absence of non synonymous SNPs, from analysis of 3D7, Dd2, HB3, Ghana1 and IT genomes). Bar width is proportional to the number of genes per category and bar height to the Pearson residuals for an independence model. Blue and red colors indicate significant bias.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Differential gene expression under artesunate exposure: transcript distribution among different gene functional categories</bold>. \"Total\" pie-chart: genes present in the 8 different color-coded functional categories as distributed in the genome represented on the DNA array. \"over-exp.\" and \"under-exp.\" pie-charts: distribution of over-expressed and under-expressed genes respectively, among the 8 different functional categories. p-value: p-value of chi-square test of over-expressed gene and under-expressed gene compared to the total genes in the genome represented on the DNA array.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Venn diagram of genes differentially expressed under artesunate treatment chloroquine treatment, and heat-shock</bold>. Purple circle: 398 genes differentially expressed under artesunate (ART). Orange circle: 30 genes differentially expressed under chloroquine (CQ) [##REF##17475254##8##]. Green circle: 334 genes differentially expressed under heat-shock (HS) [##REF##17283083##12##]. ID of genes differentially expressed in ART and CQ, HS and CQ, ART, CQ and HS are shown.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Differentially expressed genes involved in antioxidant defence in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"left\">PFI0925w</td><td align=\"left\">gamma-glutamylcysteine synthetase</td><td/><td align=\"center\">1,41</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PFE0605c</td><td align=\"left\">glutathione synthetase</td><td/><td align=\"center\">0,94</td><td/><td/></tr><tr><td align=\"left\">PFL0595c</td><td align=\"left\">glutathione peroxidase</td><td/><td align=\"center\">-0,95</td><td/><td/></tr><tr><td align=\"left\">PF11_0352</td><td align=\"left\">protein disulfide isomerase related protein</td><td/><td align=\"center\">-0,89</td><td align=\"center\">-</td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Differentially expressed chaperone and chaperone-related genes in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"center\"><bold>Pexel</bold></td><td align=\"left\"><bold>family</bold></td><td align=\"center\"><bold>ref</bold></td><td align=\"center\"><bold>ART 90 mn</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"left\">PFE1605w</td><td align=\"left\">DNAJ protein</td><td align=\"center\">yes</td><td align=\"left\">resa like/dnaJ</td><td align=\"center\">1, 3</td><td/><td align=\"center\">1,01</td><td/><td/></tr><tr><td align=\"left\">PFD0080c</td><td align=\"left\">hypothetical protein, conserved in P.falciparum</td><td align=\"center\">yes</td><td align=\"left\">resa like</td><td align=\"center\">3</td><td align=\"center\">1,31</td><td align=\"center\">1,82</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF10_0378</td><td align=\"left\">DnaJ protein, putative</td><td align=\"center\">yes</td><td align=\"left\">Hsp40, type III</td><td align=\"center\">1, 2, 4</td><td align=\"center\">1,33</td><td align=\"center\">1,53</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PFE0040c</td><td align=\"left\">mature parasite-infected erythrocyte surface antigen (MESA) or PfEMP2</td><td align=\"center\">yes</td><td align=\"left\">Hsp40, type IV</td><td align=\"center\">2, 4</td><td align=\"center\">0,92</td><td align=\"center\">1,87</td><td/><td/></tr><tr><td align=\"left\">PFA0660w</td><td align=\"left\">protein with DNAJ domain, dnj1/sis1 family</td><td align=\"center\">yes</td><td align=\"left\">Hsp40, type II</td><td align=\"center\">1, 2, 4</td><td align=\"center\">1,01</td><td align=\"center\">1,34</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">MAL7P1.7</td><td align=\"left\">RESA-like protein</td><td align=\"center\">yes</td><td align=\"left\">resa like</td><td align=\"center\">3</td><td/><td align=\"center\">0,83</td><td/><td/></tr><tr><td align=\"left\">PFL0050c</td><td align=\"left\">hypothetical protein</td><td align=\"center\">yes</td><td align=\"left\">resa like</td><td align=\"center\">3</td><td/><td align=\"center\">0,86</td><td/><td/></tr><tr><td align=\"left\">PF07_0029</td><td align=\"left\">heat shock protein 86</td><td align=\"center\">no</td><td align=\"left\">hsp90</td><td align=\"center\">1</td><td/><td align=\"center\">0,9</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF07_0030</td><td align=\"left\">heat shock protein 86 family protein</td><td align=\"center\">no</td><td align=\"left\">Hsp86 family</td><td align=\"center\">5</td><td/><td align=\"center\">1,63</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF08_0054</td><td align=\"left\">heat shock 70 kDa protein</td><td align=\"center\">no</td><td align=\"left\">hsp70</td><td align=\"center\">1</td><td/><td align=\"center\">0,88</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF11_0034</td><td align=\"left\">hypothetical protein</td><td align=\"center\">yes</td><td align=\"left\">Hsp40, type IV</td><td align=\"center\">1, 2, 4</td><td/><td align=\"center\">0,81</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PFA0675w</td><td align=\"left\">P. falciparum RESA-like protein with DnaJ domain</td><td align=\"center\">yes</td><td align=\"left\">Hsp40, type IV</td><td align=\"center\">1, 2, 4</td><td/><td align=\"center\">1,96</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF14_0700</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"center\">no</td><td align=\"left\">Hsp40, type III</td><td align=\"center\">1, 2, 4</td><td/><td align=\"center\">0,81</td><td/><td/></tr><tr><td align=\"left\">PF11_0509</td><td align=\"left\">ring-infected erythrocyte surface antigen putative</td><td align=\"center\">yes</td><td align=\"left\">Hsp40, type IV</td><td align=\"center\">1, 2, 4</td><td/><td align=\"center\">-1,14</td><td/><td/></tr><tr><td align=\"left\">PF13_0102</td><td align=\"left\">DNAJ-like Sec63 homologue</td><td align=\"center\">no</td><td align=\"left\">Hsp40, type III</td><td align=\"center\">1, 2, 4</td><td/><td align=\"center\">-1,04</td><td/><td/></tr><tr><td align=\"left\">PF11_0216</td><td align=\"left\">hypothetical protein</td><td align=\"center\">no</td><td align=\"left\">Heat shock factor binding protein 1</td><td align=\"center\">6</td><td/><td align=\"center\">-0,93</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">PFC0975c</td><td align=\"left\">PFCYP19, cyclophilin, peptidyl-prolyl cis-trans isomerase</td><td align=\"center\">no</td><td align=\"left\">cyclophillin</td><td align=\"center\">1</td><td/><td align=\"center\">-0,97</td><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Differentially expressed transport associated genes in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"left\">PF07_0121</td><td align=\"left\">NMD3 protein, putative</td><td/><td align=\"center\">0,85</td><td/><td/></tr><tr><td align=\"left\">PFB0435c</td><td align=\"left\">amino acid transporter, putative</td><td/><td align=\"center\">0,89</td><td/><td/></tr><tr><td align=\"left\">PFE1455w</td><td align=\"left\">sugar transporter, putative</td><td/><td align=\"center\">0,92</td><td/><td align=\"center\">+</td></tr><tr><td align=\"left\">PFE1150w</td><td align=\"left\">multidrug resistance protein, <italic>Pfmdr1</italic></td><td/><td align=\"center\">1,10</td><td/><td/></tr><tr><td align=\"left\">PF11_0141</td><td align=\"left\">UDP-galactose transporter, putative</td><td align=\"center\">-0,86</td><td align=\"center\">-0,82</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">PFC0725c</td><td align=\"left\">formate-nitrate transporter, putative</td><td/><td align=\"center\">-1,16</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">MAL8P1.13</td><td align=\"left\">integral membrane protein, conserved/folate-Biopterin transporter</td><td/><td align=\"center\">-1,13</td><td/><td/></tr><tr><td align=\"left\">MAL13P1.206</td><td align=\"left\">Na+ -dependent Pi transporter, sodium-dependent phosphate transporter <italic>PfPiT</italic></td><td/><td align=\"center\">-0,90</td><td/><td/></tr><tr><td align=\"left\">PF13_0252</td><td align=\"left\">nucleoside transporter 1 <italic>PfNT1/PfENT1</italic></td><td/><td align=\"center\">-1,32</td><td/><td/></tr><tr><td align=\"left\">PFL0170w</td><td align=\"left\">Transporter, major facilitator superfamily</td><td/><td align=\"center\">-0,86</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">PFE0410w</td><td align=\"left\">triose or hexose phosphate/phosphate translocator, putative</td><td/><td align=\"center\">-0,88</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">PF13_0227</td><td align=\"left\">vacuolar ATP synthase subunit D, putative</td><td/><td align=\"center\">-0,92</td><td/><td/></tr><tr><td align=\"left\">PF11_0412</td><td align=\"left\">Vacuolar ATP synthase subunit F, putative</td><td/><td align=\"center\">-0,81</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">PF13_0130</td><td align=\"left\">vacuolar ATP synthase subunit G, putative</td><td/><td align=\"center\">-0,82</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">PFE0965c</td><td align=\"left\">vacuolar ATP synthetase, subunit C putative</td><td/><td align=\"center\">-0,85</td><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Differentially expressed genes involved in lipid metabolism in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"left\">MAL13P1.485</td><td align=\"left\">acyl-coa ligase antigen</td><td/><td align=\"center\">0,94</td><td/><td/></tr><tr><td align=\"left\">PFB0685c</td><td align=\"left\">acyl-CoA synthetase, PfACS9</td><td/><td align=\"center\">0,83</td><td/><td/></tr><tr><td align=\"left\">PFB0695c</td><td align=\"left\">acyl-CoA synthetase</td><td align=\"center\">0,88</td><td align=\"center\">1,02</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF14_0761</td><td align=\"left\">fatty acyl CoA synthetase 1 PfACS1</td><td/><td align=\"center\">0,99</td><td/><td/></tr><tr><td align=\"left\">PF14_0664</td><td align=\"left\">biotin carboxylase subunit of acetyl CoA carboxylase, putative</td><td align=\"center\">0,97</td><td align=\"center\">1,31</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF08_0099</td><td align=\"left\">acyl CoA binding protein, putative</td><td/><td align=\"center\">-0,89</td><td/><td/></tr><tr><td align=\"left\">PF10_0016</td><td align=\"left\">acyl CoA binding protein, putative</td><td/><td align=\"center\">-1,04</td><td align=\"center\">-</td><td/></tr><tr><td align=\"left\">PFE0410w</td><td align=\"left\">triose or hexose phosphate/phosphate translocator, putative</td><td/><td align=\"center\">-0,88</td><td align=\"center\">-</td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Differentially expressed genes related to the mitochondrion in artesunate treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"center\">PF08_0054</td><td align=\"left\">heat shock 70 kDa protein</td><td/><td align=\"center\">0,88</td><td align=\"center\">+</td><td/></tr><tr><td align=\"center\">coxI</td><td align=\"left\">putative cytochrome oxidase I</td><td/><td align=\"center\">-1,07</td><td/><td/></tr><tr><td align=\"center\">coI</td><td align=\"left\">putative cytochrome oxidase I</td><td/><td align=\"center\">-1,15</td><td/><td/></tr><tr><td align=\"center\">PF14_0248</td><td align=\"left\">ubiquinol-cytochrome c reductase hinge protein, putative</td><td/><td align=\"center\">-0,99</td><td align=\"center\">-</td><td/></tr><tr><td align=\"center\">PFC0975c</td><td align=\"left\">PFCYP19, cyclophilin, peptidyl-prolyl cis-trans isomerase</td><td/><td align=\"center\">-0,97</td><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T6\"><label>Table 6</label><caption><p>Differentially expressed genes related to purine/pyrimidine metabolism in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"center\">PF14_0697</td><td align=\"left\">dihydroorotase, putative</td><td align=\"center\">0,96</td><td align=\"center\">0,93</td><td/><td/></tr><tr><td align=\"center\">PF13_0259</td><td align=\"left\">cytidine and deoxycytidylate deaminase family, putative</td><td/><td align=\"center\">-1,22</td><td/><td/></tr><tr><td align=\"center\">PF10_0121</td><td align=\"left\">hypoxanthine phosphoribosyltransferase</td><td/><td align=\"center\">-1,24</td><td/><td/></tr><tr><td align=\"center\">PF11_0282</td><td align=\"left\">deoxyuridine 5'-triphosphate nucleotidohydrolase, putative</td><td align=\"center\">-0,88</td><td align=\"center\">-0,95</td><td/><td/></tr><tr><td align=\"center\">PFE0660c</td><td align=\"left\">purine nucleotide phosphorylase, putative</td><td align=\"center\">-1,07</td><td align=\"center\">-1,53</td><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T7\"><label>Table 7</label><caption><p>Differentially expressed genes encoding kinases in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"left\"><bold>Kinase type</bold></td><td align=\"left\"><bold>Name</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"left\">PFL0040c</td><td align=\"left\">protein kinase, R45/FIKK family</td><td align=\"left\">R45/FIKK</td><td align=\"left\">FIKK12</td><td align=\"center\">0,88</td><td align=\"center\">1,16</td><td/><td/></tr><tr><td align=\"left\">MAL7P1.144</td><td align=\"left\">protein kinase, R45/FIKK family</td><td align=\"left\">R45/FIKK</td><td align=\"left\">FIKK7.1</td><td align=\"center\">0,90</td><td align=\"center\">1,82</td><td/><td/></tr><tr><td align=\"left\">PFI0095c</td><td align=\"left\">protein kinase, R45/FIKK family</td><td align=\"left\">R45/FIKK</td><td align=\"left\">FIKK9.1</td><td align=\"center\">0,83</td><td align=\"center\">1,28</td><td/><td/></tr><tr><td align=\"left\">PF11_0510</td><td align=\"left\">protein kinase, R45/FIKK family</td><td align=\"left\">R45/FIKK</td><td align=\"left\">FIKK11</td><td/><td align=\"center\">0,95</td><td/><td/></tr><tr><td align=\"left\">PFD1165w</td><td align=\"left\">protein kinase, R45/FIKK family</td><td align=\"left\">R45/FIKK</td><td align=\"left\">FIKK4.1</td><td/><td align=\"center\">1,05</td><td/><td/></tr><tr><td align=\"left\">PFL1885c</td><td align=\"left\">calcium/calmodulin-dependent protein kinase 2, putative</td><td align=\"left\">kinase</td><td align=\"left\">PfPK2</td><td align=\"center\">1,35</td><td align=\"center\">1,21</td><td/><td/></tr><tr><td align=\"left\">PFL2280w</td><td align=\"left\">cyclin g-associated kinase, putative</td><td align=\"left\">kinase</td><td/><td/><td align=\"center\">1,10</td><td/><td/></tr><tr><td align=\"left\">PF14_0346</td><td align=\"left\">cGMP-dependent protein kinase 1, beta isozyme, putative</td><td align=\"left\">kinase</td><td/><td/><td align=\"center\">1,03</td><td/><td/></tr><tr><td align=\"left\">PF11_0242</td><td align=\"left\">protein kinase</td><td align=\"left\">kinase</td><td/><td/><td align=\"center\">0,95</td><td/><td/></tr><tr><td align=\"left\">PF14_0294</td><td align=\"left\">mitogen-activated protein kinase 1, PfMAP1</td><td align=\"left\">kinase</td><td align=\"left\">PfMAP1</td><td/><td align=\"center\">1,05</td><td/><td/></tr><tr><td align=\"left\">PFD0975w</td><td align=\"left\">ROI kinase-like protein</td><td align=\"left\">RIO1</td><td/><td/><td align=\"center\">0,90</td><td/><td align=\"center\">+</td></tr><tr><td align=\"left\">PF11_0377</td><td align=\"left\">casein kinase 1, PfCK1</td><td align=\"left\">kinase</td><td/><td/><td align=\"center\">-0,91</td><td/><td align=\"center\">+</td></tr><tr><td align=\"left\">PF11_0227</td><td align=\"left\">serine/threonine protein kinase, puative</td><td align=\"left\">kinase</td><td/><td/><td align=\"center\">-0,98</td><td/><td/></tr><tr><td align=\"left\">PFB0150c</td><td align=\"left\">protein kinase, putative</td><td align=\"left\">kinase</td><td/><td/><td align=\"center\">-1,05</td><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T8\"><label>Table 8</label><caption><p>Differentially expressed genes encoding transcription-associated proteins in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"left\"><bold>Pfam</bold></td><td align=\"left\"><bold>annotation</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"left\">PF11_0053</td><td align=\"left\">PfSNF2L</td><td align=\"left\">TAP clustering</td><td align=\"left\">Nucleosome remodeling::SNF2L</td><td/><td align=\"center\">0,82</td><td/><td/></tr><tr><td align=\"left\">PF11_0264</td><td align=\"left\">DNA-dependent RNA polymerase</td><td align=\"left\">TAP clustering</td><td align=\"left\">MITOCHONDRIAL RNA POLYMERASE</td><td/><td align=\"center\">1,02</td><td/><td/></tr><tr><td align=\"left\">PFL0560c</td><td align=\"left\">minichromosome maintenance protein, putative</td><td align=\"left\">TAP clustering</td><td align=\"left\">Minichromosome maintenance protein</td><td/><td align=\"center\">0,85</td><td/><td/></tr><tr><td align=\"left\">PFE0090w</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">TAP clustering</td><td align=\"left\">Histone trancription regulator</td><td/><td align=\"center\">0,82</td><td/><td/></tr><tr><td align=\"left\">PF11_0297</td><td align=\"left\">hypothetical protein</td><td align=\"left\">TAP clustering</td><td align=\"left\">CCR4-NOT complex::NOT2</td><td/><td align=\"center\">0,86</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PFE1245w</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">PF00642</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (and similar)(PF00642)</td><td align=\"center\">1,28</td><td align=\"center\">0,91</td><td align=\"center\">+</td><td align=\"center\">+</td></tr><tr><td align=\"left\">PF14_0236</td><td align=\"left\">hypothetical protein</td><td align=\"left\">PF00642</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (and similar)(PF00642)</td><td/><td align=\"center\">0,89</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PFC0680w</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">PF00642</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (and similar)(PF00642)</td><td/><td align=\"center\">1,55</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF10_0186</td><td align=\"left\">hypothetical protein</td><td align=\"left\">PF00642</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (and similar)(PF00642)</td><td/><td align=\"center\">0,98</td><td/><td/></tr><tr><td align=\"left\">MAL13P1.122</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">PF00856</td><td align=\"left\">SET domain (PF00856)</td><td/><td align=\"center\">0,81</td><td/><td/></tr><tr><td align=\"left\">PFI0470w</td><td align=\"left\">FHA domain protein, putative</td><td align=\"left\">PF00097</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td align=\"center\">0,89</td><td align=\"center\">1,05</td><td/><td/></tr><tr><td align=\"left\">PF10_0046</td><td align=\"left\">hypothetical protein</td><td align=\"left\">PF00097</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td/><td align=\"center\">1,17</td><td/><td/></tr><tr><td align=\"left\">PFF0165c</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">PF00097</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td/><td align=\"center\">0,94</td><td/><td/></tr><tr><td align=\"left\">PFL0440c</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">PF00097</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td/><td align=\"center\">0,95</td><td/><td/></tr><tr><td align=\"left\">PF11_0315</td><td align=\"left\">hypothetical protein</td><td align=\"left\">TAP clustering</td><td align=\"left\">APICOPLAST RNA beta, beta, &amp; beta subunits</td><td/><td align=\"center\">1,64</td><td/><td/></tr><tr><td align=\"left\">PFL0145c</td><td align=\"left\">high mobility group protein</td><td align=\"left\">PF00505</td><td align=\"left\">HMG (high mobility group) box (PF00505)</td><td/><td align=\"center\">-0,97</td><td/><td/></tr><tr><td align=\"left\">MAL13P1.76</td><td align=\"left\">TFIIH basal transcription factor subunit</td><td align=\"left\">TAP clustering</td><td align=\"left\">TFIIH::p44</td><td/><td align=\"center\">-0,83</td><td/><td/></tr><tr><td align=\"left\">PF14_0718</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">TAP clustering</td><td align=\"left\">RNA polymerase II-associated factor SOH1</td><td/><td align=\"center\">-0,86</td><td/><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T9\"><label>Table 9</label><caption><p>Differentially expressed genes encoding proteins with Zn-finger motifs in artesunate-treated cells</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Gene ID</bold></td><td align=\"left\"><bold>Description</bold></td><td align=\"left\"><bold>Zn-finger</bold></td><td align=\"center\"><bold>ART 90 min</bold></td><td align=\"center\"><bold>ART 3 h</bold></td><td align=\"center\"><bold>HS</bold></td><td align=\"center\"><bold>CQ</bold></td></tr></thead><tbody><tr><td align=\"left\">PFE1245w</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (PF00642)</td><td align=\"center\">1,28</td><td align=\"center\">0,91</td><td align=\"center\">+</td><td align=\"center\">+</td></tr><tr><td align=\"left\">PF14_0236</td><td align=\"left\">hypothetical protein</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (PF00642)</td><td/><td align=\"center\">0,89</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PFC0680w</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (PF00642)</td><td/><td align=\"center\">1,55</td><td align=\"center\">+</td><td/></tr><tr><td align=\"left\">PF10_0186</td><td align=\"left\">hypothetical protein</td><td align=\"left\">Zinc finger C-x8-C-x5-C-x3-H type (PF00642)</td><td/><td align=\"center\">0,98</td><td/><td/></tr><tr><td align=\"left\">MAL13P1.122</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">PHD-finger (IPR001965)</td><td/><td align=\"center\">0,81</td><td/><td/></tr><tr><td align=\"left\">PF14_0197</td><td align=\"left\">hypothetical protein</td><td align=\"left\">DNL zinc finger (PF05180)</td><td/><td align=\"center\">0,87</td><td/><td/></tr><tr><td align=\"left\">PFI0470w</td><td align=\"left\">FHA domain protein, putative</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td align=\"center\">0,89</td><td align=\"center\">1,05</td><td/><td/></tr><tr><td align=\"left\">PF10_0046</td><td align=\"left\">hypothetical protein</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td/><td align=\"center\">1,17</td><td/><td/></tr><tr><td align=\"left\">PFF0165c</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td/><td align=\"center\">0,94</td><td/><td/></tr><tr><td align=\"left\">PFL0440c</td><td align=\"left\">hypothetical protein, conserved</td><td align=\"left\">C3HC4 type (RING finger) (PF00097)</td><td/><td align=\"center\">0,95</td><td/><td/></tr><tr><td align=\"left\">PFB0140w</td><td align=\"left\">hypothetical protein</td><td align=\"left\">DHHC zinc finger domain (PF01529)</td><td/><td align=\"center\">-0,91</td><td/><td/></tr><tr><td align=\"left\">MAL13P1.76</td><td align=\"left\">TFIIH basal transcription factor subunit</td><td align=\"left\">TFIIH C1-like domain (PF07975)</td><td/><td align=\"center\">-0,83</td><td/><td/></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional File 1</title><p><bold>Distribution of log ratios for statistically significant genes</bold>. Distribution of log ratios for genes differentially expressed upon 3 hour incubation with artesunate. Grey bars: genes selected (log ratio cut-off: +/- 0.8).</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional File 2</title><p><bold>Mean log ratio of gene expression in presence/absence of artesunate at 90 minutes and 3 hours</bold>. List is made-up of genes displaying significant changes at 3 hour incubation time, across 5 experiments performed at time points staggered between 20–30 h of parasite development. Values shown express the mean log ratios calculated from all experiments; when for a given gene significant log ratios were obtained for several oligos, the retained log ratio was one with the maximum absolute value. In the 3 hour column, only values significant at 3 hours as defined by ANOVA are shown, with a mean log-ratio cut-off of +/- 0.8. Values in bold: log ratios of controls (0 hours vs 3 hours in absence of artesunate) have been substracted from the log ratios obtained after 3 hours in artesunate. Red cell-background: over-expression. Green cell-background: under-expression. Gene ID and description: from PlasmoDB.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional File 3</title><p><bold>Confirmation by real-time quantitative rt-PCR of differential gene expression induced by 3 hour incubation with artesunate</bold>. Transcripts abundances were compared by ΔΔCt values calculated using PFI0425w (putative transporter) as endogenous control. Values for microarray and qPCR are expression fold changes between parasites in presence and in absence of artesunate (Pearson correlation coefficient between the two techniques: 0.81).</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S4\"><caption><title>Additional File 4</title><p><bold>Genes related to cell cycle regulation</bold>. Cell cycle associated genes were defined according to the GO biological process G0:0007049 (cell cycle). Red cell-background: over-expression. Green cell-background: under-expression.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S5\"><caption><title>Additional File 5</title><p>Sequences of primers used for q RT-PCR validation.</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>Genes coding for proteins involved in antioxidant defence defined in [##REF##15245577##23##].</p><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>HS column: over (+) or under (-) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>HS column: over (+) or under (-) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Chaperone- and chaperone-related genes were defined based on: 1) Acharya et al. [##REF##17307260##19##], 2) Botha et al. [##REF##17428722##20##], 3) resa-like annotation from GeneDB, 4) genes fitting the HMM profile of DNAJ (PF00226) from Pfam, 5) The Hsp86 family annotation from GeneDB, 6) genes fitting the HMM profile of heat shock factor binding protein 1 (PF06825) from Pfam.</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>HS and CQ columns: over (+) or under (-) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Genes involved in transport were selected from the pathway maps of Ginsburg [##UREF##1##16##].</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>HS column: over (+) or under (-) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Genes were selected based on [##UREF##1##16##].</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>HS column: over (+) or under (-) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Genes were selected based on [##UREF##1##16##].</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Genes were selected based on [##UREF##1##16##].</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>CQ column: over (+) or under (-) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Kinases were defined as genes fitting the HMM profil kinase PF00069 (update of Ward et al. [##REF##15479470##39##]) and ROI Kinases as genes fiiting the HMM profil PF01163. FIKK kinases were defined as described in Schneider and Puijalon [##REF##15752424##41##].</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>HS and CQ columns: over (+) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Genes encoding transcription-associated proteins (TAPs) were selected from the TAP clustering of Coulson et al. [##REF##15256513##17##] and completed with an update of genes fitting the following HMM profiles defined by Coulson and colleagues [##UREF##1##16##] : PF00097, PF00642, PF00096, PF00856, PF00249, PF00439, PF00505, PF00628, PF00098, PF00808, PF01753, PF00313, PF00569, PF00643, PF08523, PF02146, PF03366.</p></table-wrap-foot>", "<table-wrap-foot><p>Gene ID: identifier as found in PlasmoDB.</p><p>Art columns: log ratios of gene expression under artesunate exposure at 90 minutes and 3 hours.</p><p>HS and CQ columns: over (+) expression.</p><p>Results were compared with transcriptome modifications induced by heat-shock (HS) [##REF##17283083##12##] and chloroquine (CQ) [##REF##17475254##8##].</p><p>Genes containing Zn-motifs were defined as genes fitting the following Zn-finger HMM profiles: B-box zinc finger (PF00643), C3HC zinc finger-like (PF07967), C3HC4 type (RING finger) (PF00097), CHY zinc finger (PF05495), CSL zinc finger (PF05207), CW-type Zinc Finger (PF07496), DHHC zinc finger domain (PF01529), DNL zinc finger (PF05180), FYVE zinc finger (PF01363), HIT zinc finger (PF04438), MIZ/SP-RING zinc finger (PF02891), MYND finger (PF01753), PHD-finger (PF00628), Putative zinc finger motif, C2HC5-type (PF06221), Sec23/Sec24 zinc finger (PF04810), SWIM zinc finger (PF04434), TFIIH C1-like domain (PF07975), Tim10/DDP family zinc finger (PF02953), U1 zinc finger (PF06220), Zinc finger C-x8-C-x5-C-x3-H type (and similar)(PF00642), Zinc finger found in FPG and IleRS (PF06827), Zinc finger, C2H2 type (PF00096), Zinc finger, ZZ type (PF00569), ZN-finger, Zn-finger in Ran binding protein and others (PF00641), Zn-finger in ubiquitin-hydrolases and other protein (PF02148), ZPR1 zinc-finger domain (PF03367), AN1-like Zinc finger (PF01428), C2H2 and C2HC zinc fingers (SSF57667), PHD-finger (IPR001965).</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2164-9-388-1\"/>", "<graphic xlink:href=\"1471-2164-9-388-2\"/>", "<graphic xlink:href=\"1471-2164-9-388-3\"/>" ]

[ "<media xlink:href=\"1471-2164-9-388-S1.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-388-S2.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-388-S3.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-388-S4.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-388-S5.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Understanding the mode of action of antitumor drugs is considered an absolute prerequisite for the advancement on the design of new drugs. It is generally believed that antitumor activity is mediated by the capacity of certain drugs to induce DNA damage and trigger apoptosis. However, there are many indications that this mechanism, whatever relevant may it be, does not account for all therapeutic effects of some antitumor drugs [##REF##10075079##1##,##REF##11991684##2##].</p>", "<p>The anthracycline antibiotic daunorubicin is widely used in cancer chemotherapy [##REF##15169927##3##]. It accumulates in the nuclei of living cells and intercalates into DNA quantitatively [##UREF##0##4##,##REF##1459004##5##], a property associated to some of the most relevant effects of the drug: inhibition of DNA replication and gene transcription [##REF##10075079##1##,##REF##11172687##6##,##REF##12656675##7##], displacement of protein factors from the transcription complex [##REF##10446226##8##] and topoisomerase II poisoning [##REF##17897022##9##]. Daunorubicin has the property of arresting cell growth at drug concentrations not sufficient for promoting noticeable DNA damage, and through mechanisms that differ from the apoptotic pathway [##REF##12656675##7##]. These findings impelled to define new mechanisms of daunorubicin antiproliferative activity at clinically relevant concentrations.</p>", "<p>Daunorubicin shows remarkable sequence specificity for 5'-WCG-3' DNA tracts [##REF##2207063##10##]. This property has led to the suggestion that daunorubicin may compete with transcription factors with overlapping recognition sites for binding to DNA. This model would explain several effects of daunorubicin, such as inhibition of RNA polymerase II [##REF##10075079##1##,##REF##11172687##6##,##REF##12656675##7##] and the suppression of the co-ordinate initiation of DNA replication in <italic>Xenopus </italic>oocyte extracts [##REF##9057105##11##].</p>", "<p>To test the capacity of daunorubicin to displace key transcription factors from their binding sites in chromatin <italic>in vivo</italic>, and, therefore, to inhibit their action [##REF##11172687##6##], we used the yeast <italic>Saccharomyces cerevisiae </italic>as a model. In a previous work [##REF##12164785##12##], we showed that yeast strains deficient in ergosterol synthesis <italic>(Δerg6 </italic>strains) are particularly sensitive to daunorubicin, overcoming one of the main setbacks to the use of yeast in pharmacological studies, which is their resistance to many anti-tumour drugs [##REF##11121507##13##,##REF##2845409##14##].</p>", "<p>We demonstrated that daunorubicin treatment in Δ<italic>erg6 </italic>cells precluded activation of several genes required for galactose utilization (GAL genes) and, consequently, treated cells were unable to growth in galactose. This effect was related to the presence of CpG steps in the cognate DNA binding sequence of Gal4p, the key transcription factor for activation of GAL genes [##REF##12164785##12##,##REF##8668194##15##]. The present work aims to extend this type of analysis to the totality of the yeast genome, in order to assess the generality of this model.</p>" ]

[ "<title>Methods</title>", "<title>Yeast growth and daunorubicin treatment</title>", "<p>Daunorubicin (Sigma, St. Louis, MO, U.S.A.) was freshly prepared as a 2 mM stock solution in sterile 150 mM NaCl solution, and diluted to each final concentrations before use. A single colony of <italic>S. cerevisiae </italic>(BY4741 <italic>erg6Δ </italic>(MAT<bold>a</bold>, <italic>his3Δ1, leu2Δ0, met15Δ0, ura3Δ0</italic>, YML008c::KanMX4, from EUROSCARF, Frankfurt, Germany) was inoculated into 25 ml of YPD medium (10 g/L yeast extract, 20 g/L peptone and 20 g/L dextrose) and grown overnight at 30°C in an environmental shaker (250 rpm) until exponential phase. This yeast culture was used to inoculate 500 ml of YPD to an initial A₆₀₀of 0.1 and further incubated at the same conditions until A₆₀₀= 0.4. This culture was then divided into three aliquots and diluted four times with fresh YPD medium. Daunorubicin was then added to each culture at a final concentration of 12 mM and cultures were allowed to grow for 1 or 4 hours. The whole procedure was repeated for Real-Time quantitative PCR (qRT-PCR) validation; in this case, only two biological replicas were obtained.</p>", "<title>RNA Preparation</title>", "<p>Cultures were centrifuged for 5 min at 3000 rpm, washed with 5 ml MilliQ water and subsequently centrifuged (repeated twice). Total RNA was extracted with the RiboPure Yeast kit (Ambion, Austin, TX, USA). Total RNA was quantified by spectrophotometry in a NanoDrop ND-1000 (NanoDrop Technologies, Wilmintong DE, USA) and its integrity checked on TBE-agarose gels. The resulting total RNA was then treated with DNAseI I (F. Hoffmann-La Roche, Basel Switzerland) to remove contaminating genomic DNA.</p>", "<title>DNA Microarray Analysis</title>", "<p>Microarrays used in this work were produced at the Genomics Unit of the Scientific Park of Madrid (Spain). They consist of 13,824 spots, each one corresponding to a synthetic oligonucleotide (70-mer, Yeast Genome Oligo Set, OPERON, Cologne, Germany) encompassing the complete set of 6306 ORFs coded by the <italic>S. cerevisiae </italic>genome. Each ORF was printed at least twice; 600 spots were used as negative controls, either void or printed with random oligonucleotides; a small subset of genes (<italic>ACT1</italic>, <italic>HSP104</italic>, <italic>NUP159</italic>, <italic>NUP82</italic>, <italic>RPL32</italic>, <italic>RPS6B</italic>, <italic>SWI1</italic>, <italic>TDH1</italic>, <italic>TDH2</italic>, <italic>TUB4 </italic>and <italic>UBI1</italic>) were printed between 6 and 12 times for testing reproducibility.</p>", "<p>Fifteen <italic>μ</italic>g of total RNA were used for cDNA synthesis and labelling with Cy3-dUTP and Cy5-dUTP fluorescent nucleotides, following indirect labelling protocol (CyScribe post-labelling kit, GE-Healthcare, New York, NY, USA). Labelling efficiency was evaluated by measuring Cy3 or Cy5 absorbance in Nanodrop Spectrophotometer. Microarray prehybridization was performed in 5× SSC (SSC: 150 mM NaCl, 15 mM Na-citrate, pH 7.0), 0.1% SDS, 1%BSA at 42°C for 45 min. (Fluka, Sigma-Aldrich, Buchs SG, Switzerland). Labelled cDNA was dried in a vacuum trap and used as probe after resuspension in 110 <italic>μ</italic>l of hybridization solution (50% Formamide, 5×SSC, 0.1% SDS, 100 <italic>μ</italic>g/ml salmon sperm from Invitrogen, Carlsbad, CA, USA). Hybridization and washing were performed in a Lucidea Slide Pro System (GE Healthcare, Uppsala, Sweden). Arrays were scanned with a GenePix 4000B fluorescence scanner and analyzed by Genepix 5.0 Pro software (Axon Instruments, MDS Analytical Technologies, Toronto, Canada). Data was filtered according to spot quality. Only those spots whose intensity was twice background signal and, at least 75% of pixels had intensities above background plus two standard deviations were selected for further calculations. In average, about 60 to 70% of spots in each array were considered suitable for further analysis following these criteria.</p>", "<title>Quantitative Real Time RT-PCR Assay</title>", "<p>An aliquot of RNA preparations from untreated and treated samples, used in the microarray experiments, was saved for qRT-PCR follow-up studies. First strand cDNA was synthesized from 2 <italic>μ</italic>g of total DNAseI-treated RNA in a 20 <italic>μ</italic>l reaction volume using Omniscript RT Kit (Qiagen, Valencia, CA, USA) following manufacture's instructions. qRT-PCR reactions were performed by triplicate using the ABI-PRISM 7000 Sequence Detection System (Applied Biosystems, Foster City, CA, USA) using the SYBR Green PCR Master Mix (Applied Biosystems). Gene-specific primers (listed in Table ##TAB##3##4##) were designed using Primer Express software (Applied Biosystems). Amplified fragments were confirmed by sequencing in a 3730 DNA Analyzer (Applied Biosystems) and sequences were compared with the published genomic data at SGD. Real time PCR conditions included an initial denaturation step at 95°C for 10 min, followed by 40 cycles of a two steps amplification protocol: denaturation at 95°C for 15 s and annealing/extension at 60°C for 1 min. Relative expression values of different genes were calculated following the ΔΔ<italic>C</italic>_{<italic>T </italic>}method [##REF##11846609##31##,##REF##11328886##32##], using <italic>RPO21 </italic>as reference gene.</p>", "<title>Clustering and statistical analysis</title>", "<p>Our experimental design allowed to obtain up to 6 determinations for each gene and condition: three biological replicates per condition, two replicated spots for each gene in the array. Statistical analyses only considered genes for which a minimum of nine (out of 18) data values passed the microarray quality standards (3458 genes). Data were calculated as binary logarithms (log₂) of fluorescence ratios (treated versus untreated samples). Significant changes on expression values between the starting point (time 0) and samples taken at 1 and 4 hours of daunorubicin treatment were determined by the Student's T-test. The whole dataset, combining data from the three time points, was analyzed with the TIGR MeV program [##REF##12613259##33##]. Data were normalised by experiments and clustered by hierarchical clustering (Euclidean distance), treating duplicated spots as independent data series. Genes showing significant variations between time points were identified by ANOVA with the Bonferroni correction (p &lt; 0.05). These genes were grouped by their expression patterns in a two-dimensional map grid by SOM (Self-Organizing Maps) [##REF##12416686##34##], to generate hypotheses on the relationships and the function of genes. Classification of genes by gene ontology (GO) in biological process categories [##REF##11752257##35##] was performed in the SDG page. Documented regulators of both affected and non-affected genes were retrieved from YEASTRACT [##REF##16381908##16##]. Statistical analyses on the frequency of regulated genes in different subsets of data were performed using hypergeometric distribution tests with the Bonferroni correction (see SGD page, and <ext-link ext-link-type=\"uri\" xlink:href=\"http://mathworld.wolfram.com/HypergeometricDistribution.html\"/>)</p>" ]

[ "<title>Results</title>", "<title>Effects of daunorubicin on the yeast transcriptome</title>", "<p>The effects of daunorubicin on the yeast transcriptome were studied after 1 h and 4 h of treatment (Figure ##FIG##0##1##). The results indicate a general inhibitory effect of daunorubicin at both time points, as down regulated genes predominate over up regulated ones, and this trend was especially significant when considering genes whose expression changed by more than four-fold (lines \"4X\" and \"0.25X\" in Figure ##FIG##0##1##). Multi-array analysis of the expression changes in the whole dataset confirmed these trends. ANOVA analysis of normalized data showed statistically significant differences in expression upon daunorubicin treatment for 475 genes (14%) at least in one of the time points analysed. Affected genes were grouped in four clusters by a Self-Organising Maps (SOM) algorithm, according to their differential expression at the three time points analysed (Figure ##FIG##1##2##, list of genes for each cluster in Table ##TAB##0##1##). Clusters A to C (280 genes in total) corresponded to genes whose transcription decreased upon daunorubicin treatment, whereas all genes that became activated by the treatment (195 genes) were grouped in Cluster D. Genes in Clusters C and D showed very little or no difference in expression between one and four hours of treatment (see the horizontal median line in the corresponding plots between time points 1 h and 4 h in Figure ##FIG##1##2##), whereas genes in Cluster A were the only ones in which the effect (an inhibition, in this case) after four hours of treatment was significantly stronger than the observed after one hour (Figure ##FIG##1##2##). Cluster B, consisting only in three genes, was the only one in which the effect was stronger at one hour than at four hours. Our data thus indicated that most daunorubicin-related changes in gene expression were already significant after only one hour of treatment and that these effects either increased or remained stable after four hours for essentially all analysed genes.</p>", "<p>Gene Ontology (GO) analysis of genes activated and repressed by daunorubicin treatment showed a very different distribution of GO categories for both groups. Up-regulated genes fell into three main functional categories: Genes related to ribosome assembly and metabolism, Ty transposition, and proteolytic processes (Table ##TAB##1##2##). Whereas the two last categories may indicate a certain level of stress, up regulation of ribosome assembling-related genes usually correlates with a positive effect in cell growth. In contrast, GO analysis of genes down regulated by daunorubicin showed a general decrease of energy-producing metabolism, including genes involved in fermentation and in the tricarboxylic acid cycle. A significant proportion of down-regulated genes appeared involved in the metabolism of nitrogen compounds, including amino acids (Table ##TAB##2##3##). The dissociation between expression of ribosomal and glycolytic genes upon daunorubicin treatment can be observed in Figure ##FIG##2##3##, which shows up-regulation of most ribosomal protein genes and down-regulation of sugar and alcohol-metabolism related genes at one and four hours of daunorubicin treatment. Figure ##FIG##3##4## shows a scheme of the glycolytic pathway, highlighting genes down regulated by daunorubicin. These genes codify the enzymes responsible for no less than 9 consecutive steps of the pathway. Therefore, the data suggests that the fermentation capacity should be depressed in daunorubicin-treated yeast cells.</p>", "<p>The effects of daunorubicin treatment in gene expression of 15 selected genes were validated by qRT-PCR (list of genes and primers in Table ##TAB##3##4##, results in Table ##TAB##4##5##). The results, presented as ratios between treated and untreated cells at 0 h and 4 h of treatment, include data from up to 5 biological replicates, showed a general good agreement with microarray data. Most (8 out of 9) sugar and alcohol-metabolism related genes showed a 2 to 4 fold decrease on expression of after 4 h of treatment, a behaviour comparable to the one observed in the microarray analysis. Similarly, two out of the three amino acid metabolism genes analysed showed a 3 to 4 fold decrease on expression. In contrast, a small, but significant, increase on the expression of the ribosomal protein genes <italic>RPS28A </italic>was also observed, also in agreement with the general trend observed for ribosomal-protein genes in the microarray data. We added to this analysis the heat-shock protein <italic>HSP26</italic>, as a representative of a small group of HSP genes (<italic>HSP12</italic>, <italic>HSP26</italic>, <italic>HSP42 </italic>and <italic>HSP104</italic>) with appeared down regulated by daunorubicin in the microarray analysis (Table ##TAB##0##1##). These results were corroborated by qRT-PCR quantitation, which showed 8-fold reduction of HSP26 transcription after four hours of daunorubicin treatment (Table ##TAB##4##5##). These results confirmed the general decrease in genes related with glucose utilisation while transcription of ribosomal protein gene was either not affected or slightly increased.</p>", "<title>Identification of transcription factors associated to daunorubicin-repressed genes</title>", "<p>Transcription factors reported to bind to the promoters of daunorubicin-repressed genes were identified using the on-line bioinformatics tools available at the YEASTRACT web page (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.yeastract.com/\"/>, [##REF##16381908##16##]). From the 170 transcription factors included in the YEASTRACT database, 32 of them were found to bind to daunorubicin-repressed gene promoters in a significantly higher proportion than expected only by chance (Table ##TAB##5##6##). The table indicates the total number of genes associated to each transcription factor present in the whole dataset (that is, the 3458 ORF analysed), the number of these genes showing down-regulation by daunorubicin, the expected number by a random distribution (over 280 down regulated genes) and the \"enrichment factor\", that is, the ratio between observed and expected absolute frequencies for each factor.</p>", "<p>Some of these factors (Yap1p, Msn2p, Msn4p) are intimately related to stress response, whereas others, such as Gcr2p, Adr1p, Mig1p and Rgt1p, are associated to carbohydrate and alcohol metabolism. In addition, Gcn4p and Met4p are known regulators of amino acids biosynthetic pathways. In this regard, the transcription factor list recapitulates the functional distribution of daunorubicin down regulated genes in Table ##TAB##2##3##. Fourteen transcription factors showed enrichment factors over 3 fold, indicating that their associated genes were found in the daunorubicin down regulated dataset at 3 to 5 times higher frequencies than expected (Table ##TAB##6##7##). Many of these factors are known regulators of glycolytic genes, such as Rgt1p, Mig1p, Gcr2p or Adr1p; therefore, their inclusion in the list may merely reflect the general decrease of transcription of the regulated genes. In addition, this list includes a strikingly high proportion (10 out 14) of transcription factors encompassing CpG steps in their DNA binding sites, irrespectively their relationship with the glycolytic pathway. This observation is consistent with a preferential effect of daunorubicin on the expression of genes regulated by transcription factors with CpG steps in their DNA recognition sequences, in keeping with previous results [##REF##10446226##8##]. This specific inhibition of transcriptional activation by daunorubicin suggests that it may compete with some transcription factors for DNA binding in CpG-reach sequences in gene promoters.</p>", "<title>Correlation of daunorubicin effects and deletions of transcription factor genes</title>", "<p>A direct prediction of the DNA-binding competition model for daunorubicin action is that its presence in the cell should produce a phenocopy of genetic deletion of these factors [##REF##12164785##12##], or their partial depletion [##REF##12656675##7##]. To test this prediction, we compared the effects of daunorubicin shown here with a large dataset of null deletions of 42 transcription factors, many of them coincident with the set in Table ##TAB##5##6##[##REF##16880382##17##]. Table ##TAB##7##8## shows the correlation between microarray data from six deletion strains [##REF##16880382##17##] and the corresponding figures from the 4 h daunorubicin-treatment dataset. For these calculations, ratios between deleted and wild type strains were compared to 4 h to 0 h ratios, only for those genes that showed significant variations in expression (positive or negative) due to daunorubicin treatment. The six strains shown in Table ##TAB##7##8## are the only ones in the dataset [##REF##16880382##17##] showing positive and significant correlation (p &lt; 0.001, Bonferroni) with daunorubicin-treatment data. The best correlation values corresponded to three strains deleted for factors Adr1p, Cst6p and Sok2p; graphs in Figure ##FIG##4##5## show expression ratios for these three strains plotted against the corresponding values from daunorubicin treatment. These plots strongly suggest that at least part of the changes in transcription ratios induced by daunorubicin may be due to competition of the drug with these and other transcription factors for binding to consensus DNA sequences.</p>" ]

[ "<title>Discussion</title>", "<p>The yeast <italic>Saccharomyces cerevisiae </italic>is a favourite tool for testing drugs that interact and/or modify gene regulation, since it shares many common regulatory mechanisms with vertebrates, ranging from cell cycle to transcriptional regulation [##REF##11121507##13##,##REF##15170450##18##, ####REF##17140446##19##, ##REF##17462974##20####17462974##20##]. In a previous paper [##REF##12164785##12##], we showed that daunorubicin specifically inhibited genes required for galactose utilisation, a phenotype we proposed linked to the presence of CpG steps in the recognition sequence of the main regulator for these genes, Gal4p. Here we extended these studies to the whole yeast transcriptome, in conditions of mild inhibition of cell growth.</p>", "<p>Daunorubicin treatment affected transcription of a relative small proportion of genes. We chose a relatively mild treatment, slightly under the IC₅₀, in order to minimise general toxic effects in cell membranes and widespread DNA damage. A conclusion from our analysis is the selective repression by daunorubicin of genes involved in the glycolytic pathway, whereas other genes involved in growth, like ribosomal protein genes, were either not affected or slightly activated. This pattern is very rarely observed in yeast, as glucose utilisation is required for fast growth. Figure ##FIG##5##6## shows ratios of expression changes for 32 glycolysis-related genes (gly genes) and 123 ribosomal protein genes (rpg genes) in 146 stress conditions, including DNA damage (both chemical and by irradiation), oxidative and osmotic stress, amino acid and nitrogen starvation, entering in stationary phase, and temperature shifts ([##REF##11102521##21##,##REF##11598186##22##]; list of genes and conditions in Table ##TAB##8##9##). The graph shows both the ratio between both sets of genes and p-values associated to their differential response to each stress. Low p-values (upper part of the graph, note the reversed Y-axis) correspond to data sets in which the response of both sets of genes showed little or no overlap, whereas high p-values (lower part of the graph) implicate that both sets of genes responded similarly to that specific stress condition. The graph shows that ribosomal protein genes are preferentially inhibited in many stress conditions compared to glycolysis-related genes (right portion of the graph), whereas daunorubicin treatment datasets (1 h and 4 h) differentiate clearly from the rest by specifically depressing glycolytic gene transcription without a parallel decrease of ribosomal synthesis (upper left part of the graph). We concluded that daunorubicin effects couldn't be ascribed to any of the tested stresses, including DNA damage and oxidative stress. This conclusion is further supported by the fact that many stress-related genes, like HSPs, were down regulated, rather than up regulated upon daunorubicin treatment.</p>", "<p>Inspection of promoters of daunorubicin-inhibited genes showed that they present a significant high proportion of DNA binding sites for a defined subset of transcription factors, most of them related to sugar metabolism. These data have to be interpreted not necessarily as an indication of direct interaction of the drug with these transcription factors, but only as a hint of the regulatory networks, or regulons, particularly affected by the drug. Due to the complexity of eukaryotic promoters, several factors may appear in any particular affected promoter, although the putative direct effect of the drug may affect to only one or two of them. A particularly relevant example is Mig1p, a transcriptional repressor central in the catabolite repression by glucose and that binds to many glycolytic gene promoters [##REF##9618445##23##]. Therefore, it appears on the lists of transcription factors preferentially associated to daunorubicin-inhibited genes (Tables ##TAB##5##6## and ##TAB##6##7##), although the hypothetical suppression of its binding to DNA would result in activation, rather than inhibition, of the affected gene. This is the most reasonable explanation by the appearance in these lists of some transcription factors that do not encompass daunorubicin-preferred sites in their recognition sequences (Table ##TAB##6##7##).</p>", "<p>Data mining identified several microarray datasets with patterns resembling to the ones observed in daunorubicin-treated cells. Best correlations were observed for strains deleted for some glucose-related transcription factor genes, especially <italic>ADR1</italic>, <italic>CST6 </italic>and <italic>SOK2</italic>. Deletion of these genes results in a general decrease on transcription of glycolytic genes with relatively mild effects on transcription of genes related to cell growth, like ribosomal protein genes -exactly the pattern observed in daunorubicin-treated cells. Two of these three factors (Adr1p and Cst6p) were identified as preferentially associated to genes down regulated by daunorubicin (Table ##TAB##5##6##, Figure ##FIG##3##4##). This list also includes a high proportion of factors whose DNA recognition sequences include CpG steps, the preferred binding site for daunorubicin [##UREF##0##4##]. Therefore, we concluded that daunorubicin inhibition of yeast growth might be mediated by its interaction with DNA at sequences also recognized by some transcription factors, resulting in a transcriptional repression of glycolytic genes, among others. These results corroborate the interest in using yeast mutants as an <italic>in vivo </italic>system to identify the determinants of chemosensitivity [##REF##11121507##13##].</p>", "<p>The amazing conservation of regulatory elements among opisthokonta (taxon that includes fungi and animals, among other groups) allows identification of pathways and transcription factors common to yeast and humans. For example, Cst6p is a basic leucine zipper transcription factor of the ATF/CREB family, which includes <italic>bona fide </italic>orthologues in mammals, not only in functional terms (targets for the cAMP regulatory pathway), but also by their binding to identical DNA sequences, 5'-TGACGTCA-3' [##REF##10825197##24##]. This sequence includes a high affinity site for daunorubicin, providing an explanation for several of the effects observed in this work. Sok2p is also known to participate in the cAMP regulatory pathway [##REF##8524252##25##], and, therefore, many cAMP-regulated promoters encompass binding sites for both factors. This circumstance provides a good explanation for the good correlation between the changes in gene expression due to the deletion of the corresponding gene and those observed upon daunorubicin treatment, although the DNA recognition sequence for Sok2p (5'-TGCAGNNA-3', [##REF##15343339##26##]) does not include high affinity sites for daunorubicin. Therefore, our data suggest that daunorubicin may target the cAMP signalling pathway of yeast, inhibiting expression of many regulated genes and particularly those under control of Cst6p, ant that may be explained by binding of the drug to the Cst6p DNA recognition site. The question of whether daunorubicin may have similar effects in the cAMP-mediated regulation of proliferation of mammalian cells is still open.</p>", "<p>Extrapolation of these results to tumour cells can be undertaken at several levels. First, as a general model, they demonstrate that DNA-intercalating drugs can block cell growth by selectively reducing the efficiency of different transcription factors. If these factors are required for cell growth, this would prevent tumour propagation at effective concentration of the drug much below the ones required for the massive DNA damage required to trigger apoptosis [##REF##15948668##27##,##REF##18226599##28##]. In addition, the specific effects of daunorubicin on the glycolysis pathway may be relevant to its antitumor effect. One of the most outstanding alterations in cancer cells is their dependence on glycolytic pathways for the generation of ATP [##REF##16892078##29##], and there is compelling evidence that mitochondrial defects in tumour cells under hypoxia are remarkably sensitive to glycolysis inhibition [##REF##16892078##29##]. Besides, it has been recently reported that some inhibitors of glucose uptake sensitize tumour cells to daunorubicin [##REF##16906425##30##]. Our data would suggest that daunorubicin might work not only as a DNA-damaging agent but also as an inhibitor of glycolytic pathways, a combined effect that might have broad therapeutic implications against cancer cells growing under hypoxic conditions.</p>" ]

[ "<title>Conclusion</title>", "<p>The yeast <italic>Saccharomyces cerevisiae </italic>is a powerful tool for the study the effects of drugs on eukaryotic cells. We showed that the antitumor drug daunorubicin alters transcription of some very specific subsets of genes, in a pattern in which sugar- metabolising pathways become down-regulated whereas proliferation-related genes, like ribosomal protein genes, are unaffected or even activated. This pattern is very similar to the one observed in yeast strains deleted for some transcription factors related to the regulation of the glycolytic pathway, like Adr1p, Cst6p and Sok2p. This results are consistent with the hypothesis that daunorubicin impairs binding of different transcription factors by competing for their DNA binding sequences, therefore limiting their effectiveness and affecting the corresponding regulatory networks. This proposed mechanism might have broad therapeutic implications in cancer therapeutics.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>The antitumor drug daunorubicin exerts some of its cytotoxic effects by binding to DNA and inhibiting the transcription of different genes. We analysed this effect <italic>in vivo </italic>at the transcriptome level using the budding yeast <italic>Saccharomyces cerevisiae </italic>as a model and sublethal (IC₄₀) concentrations of the drug to minimise general toxic effects.</p>", "<title>Results</title>", "<p>Daunorubicin affected a minor proportion (14%) of the yeast transcriptome, increasing the expression of 195 genes and reducing expression of 280 genes. Daunorubicin down-regulated genes included essentially all genes involved in the glycolytic pathway, the tricarboxylic acid cycle and alcohol metabolism, whereas transcription of ribosomal protein genes was not affected or even slightly increased. This pattern is consistent with a specific inhibition of glucose usage in treated cells, with only minor effects on proliferation or other basic cell functions. Analysis of promoters of down-regulated genes showed that they belong to a limited number of transcriptional regulatory units (regulons). Consistently, data mining showed that daunorubicin-induced changes in expression patterns were similar to those observed in yeast strains deleted for some transcription factors functionally related to the glycolysis and/or the cAMP regulatory pathway, which appeared to be particularly sensitive to daunorubicin.</p>", "<title>Conclusion</title>", "<p>The effects of daunorubicin treatment on the yeast transcriptome are consistent with a model in which this drug impairs binding of different transcription factors by competing for their DNA binding sequences, therefore limiting their effectiveness and affecting the corresponding regulatory networks. This proposed mechanism might have broad therapeutic implications against cancer cells growing under hypoxic conditions.</p>" ]

[ "<title>Authors' contributions</title>", "<p>MR: Growth effects, microarray analysis, qRT-PCR. MC: qRT-PCR analysis, technical assistance. JP &amp; BP: co-direction, data mining and analysis, preparation and writing of the manuscript. All co-authors read and approved the manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This work has been supported by the Spanish Ministry for Education and Science (MEC, grants BIO2005-00840, BFU2007-60998/BMC and AGL2000-0133-P4-03). The contribution of the Centre de Referència en Biotecnologia de la Generalitat de Catalunya is also acknowledged.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Effects of daunorubicin to the yeast transcriptome.</bold> Expression data from treated and untreated cells (expressed as binary logs) were compared before and after one and four hours of incubation with daunorubicin. Data are represented as log₂of the ratios of gene expression values after 1 h (left) and 4 h (right) of daunorubicin treatment versus the initial values (Time 0). Only genes whose expression was significantly altered by the treatment (T-test, brown dots, p &lt; 10^-5, yellow squares, p &lt; 10^-2) are shown. Discontinuous lines in the plots indicate the calculated positions of genes changed by 4-, 2-, 0.5- and 0.25-fold; they are included as references to compare with the changes in expression of different genes.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Transcriptional profiles for genes classified into clusters by SOM.</bold> Data are shown as logarithmic values of the ratio of fluorescence between treated and untreated cells before (0 h) and 1 and 4 hours after treatment. No correction was performed to compensate differences in labelling or detection of the two fluorochromes. The thick solid lines in the middle of the graphs correspond to median values, coloured areas correspond to the intervals between 1st and 3rd quartiles (dark gray) and the total distribution (light gray). Averaged values for Cluster B (3 genes, discontinuous line) in included in the Cluster A plot.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Transcriptional rate changes for Ribosomal Protein genes (solid dots) and Glycolytic genes (diamonds) after 1 (Y-axis) and 4 h (X-axis) of daunorubicin treatment.</bold> Data are expressed as logarithmic values of expression ratios between treated and untreated cells.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Scheme of the glycolytic pathway.</bold> Genes codifying for the enzymes implicated in each step are detailed; green labels indicate genes whose expression was reduced upon daunorubicin treatment.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>Transcription ratios between daunorubicin-treated cells and three strains deleted for different transcription factors.</bold> The X-axis corresponds to microarray data for cells treated with daunorubicin for four hours (treated vs. untreated, log₂values). The Y-axis corresponds to data from reference [##REF##16880382##17##]. Only data for the 475 genes affected by daunorubicin were considered.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Differential expression for glycolytic genes (gly) and ribosomal protein genes (rpg) in yeast cells subjected to different treatments.</bold> Fold induction or repression values were calculated for 32 glycolytic genes and 123 ribosomal protein genes for each of the 146 stress conditions, plus the two daunorubicin treatments. The X-axis values correspond to ratios between the average of fold induction/repression for glycolitic and ribosomal protein genes for in each experiment; Y-axis indicates the probability of both sets of genes being equally affected by each treatment. Note the reverse scale of the Y-axis. Each dot represent a single stress dataset for a particular stress condition; they are grouped in several categories: Daunorubicin treatment (DNR, 1 h and 4 h, red squares), DNA damaging agents (DD, 15 conditions, blue diamonds), osmotic stress (OS, 12 conditions, green triangles), oxidative stress (Ox, 45 conditions, yellow diamond), temperature stress (T, 37 conditions, orange circle), amino acid and nitrogen starvation (N, 15 conditions, dark brown circle) and maintenance in stationary phase for long periods of time (22 conditions, red triangles). Two vertical, discontinuous lines indicate 2-fold induction or repression; note that ratio values are expressed as log₂transformants. Except for daunorubicin-treatment, all data are from references [##REF##11102521##21##,##REF##11598186##22##]. Genes and conditions analysed are listed in Table 9.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Gene clusters defined by SOM analysis</p></caption><table frame=\"hsides\" rules=\"groups\"><tbody><tr><td align=\"center\" colspan=\"4\">Cluster A</td><td align=\"left\">Cluster B</td><td align=\"left\">Cluster C</td><td align=\"center\" colspan=\"4\">Cluster D</td></tr><tr><td colspan=\"4\"><hr/></td><td colspan=\"1\"><hr/></td><td colspan=\"1\"><hr/></td><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>AAH1</italic></td><td align=\"left\"><italic>GPI12</italic></td><td align=\"left\"><italic>PPM1</italic></td><td align=\"left\"><italic>YDR428C</italic></td><td align=\"left\"><italic>URA2</italic></td><td align=\"left\"><italic>ACT1</italic></td><td align=\"left\"><italic>ACC1</italic></td><td align=\"left\"><italic>RPC31</italic></td><td align=\"left\"><italic>YBL051C</italic></td><td align=\"left\"><italic>YMR074C</italic></td></tr><tr><td align=\"left\"><italic>AAT2</italic></td><td align=\"left\"><italic>GPM1</italic></td><td align=\"left\"><italic>PRB1</italic></td><td align=\"left\"><italic>YDR453C</italic></td><td align=\"left\"><italic>YJU3</italic></td><td align=\"left\"><italic>ARG8</italic></td><td align=\"left\"><italic>ANB1</italic></td><td align=\"left\"><italic>RPC40</italic></td><td align=\"left\"><italic>YBL057C</italic></td><td align=\"left\"><italic>YMR085W</italic></td></tr><tr><td align=\"left\"><italic>ACO1</italic></td><td align=\"left\"><italic>GPM2</italic></td><td align=\"left\"><italic>PRY1</italic></td><td align=\"left\"><italic>YDR516C</italic></td><td align=\"left\"><italic>YML056C</italic></td><td align=\"left\"><italic>ARO4</italic></td><td align=\"left\"><italic>ARL1</italic></td><td align=\"left\"><italic>RPG1</italic></td><td align=\"left\"><italic>YBR012W-B</italic></td><td align=\"left\"><italic>YMR130W</italic></td></tr><tr><td align=\"left\"><italic>ADE12</italic></td><td align=\"left\"><italic>GRE2</italic></td><td align=\"left\"><italic>PRY3</italic></td><td align=\"left\"><italic>YDR539W</italic></td><td/><td align=\"left\"><italic>AYR1</italic></td><td align=\"left\"><italic>BFR1</italic></td><td align=\"left\"><italic>RPL13B</italic></td><td align=\"left\"><italic>YCL019W</italic></td><td align=\"left\"><italic>YMR158C-B</italic></td></tr><tr><td align=\"left\"><italic>ADE17</italic></td><td align=\"left\"><italic>GRE3</italic></td><td align=\"left\"><italic>PSA1</italic></td><td align=\"left\"><italic>YFR017C</italic></td><td/><td align=\"left\"><italic>CAR2</italic></td><td align=\"left\"><italic>CAF20</italic></td><td align=\"left\"><italic>RPL32</italic></td><td align=\"left\"><italic>YCR082W</italic></td><td align=\"left\"><italic>YNL054W-B</italic></td></tr><tr><td align=\"left\"><italic>ADH1</italic></td><td align=\"left\"><italic>GSF2</italic></td><td align=\"left\"><italic>PST1</italic></td><td align=\"left\"><italic>YGL121C</italic></td><td/><td align=\"left\"><italic>CDC91</italic></td><td align=\"left\"><italic>CBF5</italic></td><td align=\"left\"><italic>RPL34B</italic></td><td align=\"left\"><italic>YDL076C</italic></td><td align=\"left\"><italic>YNL296W</italic></td></tr><tr><td align=\"left\"><italic>ADH2</italic></td><td align=\"left\"><italic>GSY2</italic></td><td align=\"left\"><italic>RAD51</italic></td><td align=\"left\"><italic>YGL157W</italic></td><td/><td align=\"left\"><italic>DAK1</italic></td><td align=\"left\"><italic>CCT5</italic></td><td align=\"left\"><italic>RPL6A</italic></td><td align=\"left\"><italic>YDL157C</italic></td><td align=\"left\"><italic>YNR046W</italic></td></tr><tr><td align=\"left\"><italic>ADH5</italic></td><td align=\"left\"><italic>GTT1</italic></td><td align=\"left\"><italic>RHR2</italic></td><td align=\"left\"><italic>YGP1</italic></td><td/><td align=\"left\"><italic>ERG10</italic></td><td align=\"left\"><italic>CDC20</italic></td><td align=\"left\"><italic>RPL6B</italic></td><td align=\"left\"><italic>YDL166C</italic></td><td align=\"left\"><italic>YOL026C</italic></td></tr><tr><td align=\"left\"><italic>ALD4</italic></td><td align=\"left\"><italic>GYP7</italic></td><td align=\"left\"><italic>RIB1</italic></td><td align=\"left\"><italic>YGR045C</italic></td><td/><td align=\"left\"><italic>FAS1</italic></td><td align=\"left\"><italic>CDC33</italic></td><td align=\"left\"><italic>RPN10</italic></td><td align=\"left\"><italic>YDR034C-D</italic></td><td align=\"left\"><italic>YOL092W</italic></td></tr><tr><td align=\"left\"><italic>ALD6</italic></td><td align=\"left\"><italic>HHO1</italic></td><td align=\"left\"><italic>RIB4</italic></td><td align=\"left\"><italic>YGR161C</italic></td><td/><td align=\"left\"><italic>GDH1</italic></td><td align=\"left\"><italic>CDC60</italic></td><td align=\"left\"><italic>RPO26</italic></td><td align=\"left\"><italic>YDR060W</italic></td><td align=\"left\"><italic>YOL124C</italic></td></tr><tr><td align=\"left\"><italic>AMS1</italic></td><td align=\"left\"><italic>HMT1</italic></td><td align=\"left\"><italic>RIP1</italic></td><td align=\"left\"><italic>YHL021C</italic></td><td/><td align=\"left\"><italic>GLT1</italic></td><td align=\"left\"><italic>COP1</italic></td><td align=\"left\"><italic>RPS11B</italic></td><td align=\"left\"><italic>YDR084C</italic></td><td align=\"left\"><italic>YOR021C</italic></td></tr><tr><td align=\"left\"><italic>ARA1</italic></td><td align=\"left\"><italic>HOR2</italic></td><td align=\"left\"><italic>RME1</italic></td><td align=\"left\"><italic>YHM1</italic></td><td/><td align=\"left\"><italic>NUP82</italic></td><td align=\"left\"><italic>CPR6</italic></td><td align=\"left\"><italic>RPS19A</italic></td><td align=\"left\"><italic>YDR098C-B</italic></td><td align=\"left\"><italic>YOR262W</italic></td></tr><tr><td align=\"left\"><italic>ARG1</italic></td><td align=\"left\"><italic>HSP104</italic></td><td align=\"left\"><italic>RNR1</italic></td><td align=\"left\"><italic>YHR087W</italic></td><td/><td align=\"left\"><italic>PFK1</italic></td><td align=\"left\"><italic>DIB1</italic></td><td align=\"left\"><italic>RPS26A</italic></td><td align=\"left\"><italic>YDR154C</italic></td><td align=\"left\"><italic>YOR343C-A</italic></td></tr><tr><td align=\"left\"><italic>ARG4</italic></td><td align=\"left\"><italic>HSP12</italic></td><td align=\"left\"><italic>SCM4</italic></td><td align=\"left\"><italic>YIL011W</italic></td><td/><td align=\"left\"><italic>PHB1</italic></td><td align=\"left\"><italic>DPB4</italic></td><td align=\"left\"><italic>RPS4B</italic></td><td align=\"left\"><italic>YDR210C-D</italic></td><td align=\"left\"><italic>YOR343C-B</italic></td></tr><tr><td align=\"left\"><italic>ARG5</italic></td><td align=\"left\"><italic>HSP26</italic></td><td align=\"left\"><italic>SCS7</italic></td><td align=\"left\"><italic>YIL056W</italic></td><td/><td align=\"left\"><italic>PYC2</italic></td><td align=\"left\"><italic>DST1</italic></td><td align=\"left\"><italic>RPS8A</italic></td><td align=\"left\"><italic>YDR210W-D</italic></td><td align=\"left\"><italic>YOR382W</italic></td></tr><tr><td align=\"left\"><italic>ARO3</italic></td><td align=\"left\"><italic>HSP42</italic></td><td align=\"left\"><italic>SCW11</italic></td><td align=\"left\"><italic>YIL077C</italic></td><td/><td align=\"left\"><italic>QCR10</italic></td><td align=\"left\"><italic>FCY1</italic></td><td align=\"left\"><italic>RPT3</italic></td><td align=\"left\"><italic>YDR261C-D</italic></td><td align=\"left\"><italic>YPL199C</italic></td></tr><tr><td align=\"left\"><italic>ASH1</italic></td><td align=\"left\"><italic>HXK1</italic></td><td align=\"left\"><italic>SDS24</italic></td><td align=\"left\"><italic>YJL016W</italic></td><td/><td align=\"left\"><italic>QCR2</italic></td><td align=\"left\"><italic>FKB2</italic></td><td align=\"left\"><italic>RRP4</italic></td><td align=\"left\"><italic>YDR261W-B</italic></td><td align=\"left\"><italic>YPL225W</italic></td></tr><tr><td align=\"left\"><italic>BAP2</italic></td><td align=\"left\"><italic>HXK2</italic></td><td align=\"left\"><italic>SGE1</italic></td><td align=\"left\"><italic>YJL094C</italic></td><td/><td align=\"left\"><italic>RFC5</italic></td><td align=\"left\"><italic>FPR1</italic></td><td align=\"left\"><italic>RRP5</italic></td><td align=\"left\"><italic>YDR316W-B</italic></td><td align=\"left\"><italic>YPR137C-B</italic></td></tr><tr><td align=\"left\"><italic>BAP3</italic></td><td align=\"left\"><italic>HXT1</italic></td><td align=\"left\"><italic>SHM2</italic></td><td align=\"left\"><italic>YJR008W</italic></td><td/><td align=\"left\"><italic>RNR4</italic></td><td align=\"left\"><italic>FRQ1</italic></td><td align=\"left\"><italic>RRP9</italic></td><td align=\"left\"><italic>YDR361C</italic></td><td align=\"left\"><italic>YPR158W-B</italic></td></tr><tr><td align=\"left\"><italic>BAT2</italic></td><td align=\"left\"><italic>HXT2</italic></td><td align=\"left\"><italic>SNO1</italic></td><td align=\"left\"><italic>YKL151C</italic></td><td/><td align=\"left\"><italic>STI1</italic></td><td align=\"left\"><italic>HCH1</italic></td><td align=\"left\"><italic>RRS1</italic></td><td align=\"left\"><italic>YDR365W-B</italic></td><td align=\"left\"><italic>YPS7</italic></td></tr><tr><td align=\"left\"><italic>CAP2</italic></td><td align=\"left\"><italic>IDH1</italic></td><td align=\"left\"><italic>SNQ2</italic></td><td align=\"left\"><italic>YKR067W</italic></td><td/><td align=\"left\"><italic>STP3</italic></td><td align=\"left\"><italic>HIR1</italic></td><td align=\"left\"><italic>RSC6</italic></td><td align=\"left\"><italic>YDR449C</italic></td><td align=\"left\"><italic>YPT31</italic></td></tr><tr><td align=\"left\"><italic>CBP4</italic></td><td align=\"left\"><italic>IDH2</italic></td><td align=\"left\"><italic>SNZ1</italic></td><td align=\"left\"><italic>YLL012W</italic></td><td/><td align=\"left\"><italic>TEF1</italic></td><td align=\"left\"><italic>HIS7</italic></td><td align=\"left\"><italic>RVB2</italic></td><td align=\"left\"><italic>YER007C-A</italic></td><td/></tr><tr><td align=\"left\"><italic>CHA1</italic></td><td align=\"left\"><italic>ILV5</italic></td><td align=\"left\"><italic>SPI1</italic></td><td align=\"left\"><italic>YLR110C</italic></td><td/><td align=\"left\"><italic>TKL1</italic></td><td align=\"left\"><italic>HRP1</italic></td><td align=\"left\"><italic>SAS10</italic></td><td align=\"left\"><italic>YER092W</italic></td><td/></tr><tr><td align=\"left\"><italic>CHS1</italic></td><td align=\"left\"><italic>INO1</italic></td><td align=\"left\"><italic>SRL3</italic></td><td align=\"left\"><italic>YLR111W</italic></td><td/><td align=\"left\"><italic>TSA1</italic></td><td align=\"left\"><italic>HRR25</italic></td><td align=\"left\"><italic>SBH1</italic></td><td align=\"left\"><italic>YER126C</italic></td><td/></tr><tr><td align=\"left\"><italic>CIT1</italic></td><td align=\"left\"><italic>IPT1</italic></td><td align=\"left\"><italic>SRY1</italic></td><td align=\"left\"><italic>YLR122C</italic></td><td/><td align=\"left\"><italic>TTR1</italic></td><td align=\"left\"><italic>HRT1</italic></td><td align=\"left\"><italic>SEC21</italic></td><td align=\"left\"><italic>YER138C</italic></td><td/></tr><tr><td align=\"left\"><italic>CLN2</italic></td><td align=\"left\"><italic>IRA2</italic></td><td align=\"left\"><italic>SSA1</italic></td><td align=\"left\"><italic>YLR231C</italic></td><td/><td align=\"left\"><italic>UGA1</italic></td><td align=\"left\"><italic>ILS1</italic></td><td align=\"left\"><italic>SEC65</italic></td><td align=\"left\"><italic>YER160C</italic></td><td/></tr><tr><td align=\"left\"><italic>COQ1</italic></td><td align=\"left\"><italic>KNS1</italic></td><td align=\"left\"><italic>SSA2</italic></td><td align=\"left\"><italic>YLR331C</italic></td><td/><td align=\"left\"><italic>URA4</italic></td><td align=\"left\"><italic>IMP4</italic></td><td align=\"left\"><italic>SEC72</italic></td><td align=\"left\"><italic>YER183C</italic></td><td/></tr><tr><td align=\"left\"><italic>COS1</italic></td><td align=\"left\"><italic>LAP4</italic></td><td align=\"left\"><italic>SSD1</italic></td><td align=\"left\"><italic>YLR352W</italic></td><td/><td align=\"left\"><italic>YBR070C</italic></td><td align=\"left\"><italic>KAP123</italic></td><td align=\"left\"><italic>SER3</italic></td><td align=\"left\"><italic>YFH1</italic></td><td/></tr><tr><td align=\"left\"><italic>COS7</italic></td><td align=\"left\"><italic>LSC2</italic></td><td align=\"left\"><italic>SUN4</italic></td><td align=\"left\"><italic>YLR414C</italic></td><td/><td align=\"left\"><italic>YDR214W</italic></td><td align=\"left\"><italic>KRI1</italic></td><td align=\"left\"><italic>SES1</italic></td><td align=\"left\"><italic>YFL002W-A</italic></td><td/></tr><tr><td align=\"left\"><italic>COX20</italic></td><td align=\"left\"><italic>MCR1</italic></td><td align=\"left\"><italic>TAT2</italic></td><td align=\"left\"><italic>YLR454W</italic></td><td/><td align=\"left\"><italic>YDR476C</italic></td><td align=\"left\"><italic>KRR1</italic></td><td align=\"left\"><italic>SIT1</italic></td><td align=\"left\"><italic>YFL004W</italic></td><td/></tr><tr><td align=\"left\"><italic>CPA1</italic></td><td align=\"left\"><italic>MDH1</italic></td><td align=\"left\"><italic>TDH1</italic></td><td align=\"left\"><italic>YML128C</italic></td><td/><td align=\"left\"><italic>YER134C</italic></td><td align=\"left\"><italic>LOS1</italic></td><td align=\"left\"><italic>SKP1</italic></td><td align=\"left\"><italic>YGR038C-B</italic></td><td/></tr><tr><td align=\"left\"><italic>CTS1</italic></td><td align=\"left\"><italic>MDH2</italic></td><td align=\"left\"><italic>TDH2</italic></td><td align=\"left\"><italic>YMR090W</italic></td><td/><td align=\"left\"><italic>YER182W</italic></td><td align=\"left\"><italic>LYS7</italic></td><td align=\"left\"><italic>SMD3</italic></td><td align=\"left\"><italic>YGR081C</italic></td><td/></tr><tr><td align=\"left\"><italic>CYC3</italic></td><td align=\"left\"><italic>MEP1</italic></td><td align=\"left\"><italic>TDH3</italic></td><td align=\"left\"><italic>YMR173W-A</italic></td><td/><td align=\"left\"><italic>YGL047W</italic></td><td align=\"left\"><italic>MGM101</italic></td><td align=\"left\"><italic>SNF8</italic></td><td align=\"left\"><italic>YGR161W-B</italic></td><td/></tr><tr><td align=\"left\"><italic>CYT1</italic></td><td align=\"left\"><italic>MEP3</italic></td><td align=\"left\"><italic>THO1</italic></td><td align=\"left\"><italic>YMR181C</italic></td><td/><td align=\"left\"><italic>YGR201C</italic></td><td align=\"left\"><italic>NAT3</italic></td><td align=\"left\"><italic>SNT309</italic></td><td align=\"left\"><italic>YHR052W</italic></td><td/></tr><tr><td align=\"left\"><italic>DDR2</italic></td><td align=\"left\"><italic>MET6</italic></td><td align=\"left\"><italic>TIR2</italic></td><td align=\"left\"><italic>YMR315W</italic></td><td/><td align=\"left\"><italic>YHR049W</italic></td><td align=\"left\"><italic>NIP7</italic></td><td align=\"left\"><italic>SPB1</italic></td><td align=\"left\"><italic>YHR214C-B</italic></td><td/></tr><tr><td align=\"left\"><italic>DDR48</italic></td><td align=\"left\"><italic>MMD1</italic></td><td align=\"left\"><italic>TPI1</italic></td><td align=\"left\"><italic>YNL200C</italic></td><td/><td align=\"left\"><italic>YIL087C</italic></td><td align=\"left\"><italic>NMD3</italic></td><td align=\"left\"><italic>SPE3</italic></td><td align=\"left\"><italic>YHR214C-C</italic></td><td/></tr><tr><td align=\"left\"><italic>DED1</italic></td><td align=\"left\"><italic>MOG1</italic></td><td align=\"left\"><italic>TPS2</italic></td><td align=\"left\"><italic>YNL212W</italic></td><td/><td align=\"left\"><italic>YIR035C</italic></td><td align=\"left\"><italic>NOP12</italic></td><td align=\"left\"><italic>SPE4</italic></td><td align=\"left\"><italic>YIL127C</italic></td><td/></tr><tr><td align=\"left\"><italic>DYN1</italic></td><td align=\"left\"><italic>MRPL35</italic></td><td align=\"left\"><italic>TRR2</italic></td><td align=\"left\"><italic>YOL101C</italic></td><td/><td align=\"left\"><italic>YLL023C</italic></td><td align=\"left\"><italic>NOP58</italic></td><td align=\"left\"><italic>SSF1</italic></td><td align=\"left\"><italic>YJR027W</italic></td><td/></tr><tr><td align=\"left\"><italic>EHT1</italic></td><td align=\"left\"><italic>MSF1'</italic></td><td align=\"left\"><italic>TSL1</italic></td><td align=\"left\"><italic>YOR009W</italic></td><td/><td align=\"left\"><italic>YLR112W</italic></td><td align=\"left\"><italic>NPI46</italic></td><td align=\"left\"><italic>SSP120</italic></td><td align=\"left\"><italic>YJR029W</italic></td><td/></tr><tr><td align=\"left\"><italic>ENO1</italic></td><td align=\"left\"><italic>MTF2</italic></td><td align=\"left\"><italic>TUF1</italic></td><td align=\"left\"><italic>YOR022C</italic></td><td/><td align=\"left\"><italic>YLR356W</italic></td><td align=\"left\"><italic>NPT1</italic></td><td align=\"left\"><italic>STS1</italic></td><td align=\"left\"><italic>YKL014C</italic></td><td/></tr><tr><td align=\"left\"><italic>ENO2</italic></td><td align=\"left\"><italic>NCE102</italic></td><td align=\"left\"><italic>UGP1</italic></td><td align=\"left\"><italic>YOR062C</italic></td><td/><td align=\"left\"><italic>YMR178W</italic></td><td align=\"left\"><italic>NRD1</italic></td><td align=\"left\"><italic>SUI1</italic></td><td align=\"left\"><italic>YKL054C</italic></td><td/></tr><tr><td align=\"left\"><italic>ERG11</italic></td><td align=\"left\"><italic>NCR1</italic></td><td align=\"left\"><italic>URA1</italic></td><td align=\"left\"><italic>YOR081C</italic></td><td/><td align=\"left\"><italic>YNL100W</italic></td><td align=\"left\"><italic>OLI1</italic></td><td align=\"left\"><italic>SUI2</italic></td><td align=\"left\"><italic>YKR081C</italic></td><td/></tr><tr><td align=\"left\"><italic>ERG26</italic></td><td align=\"left\"><italic>OAC1</italic></td><td align=\"left\"><italic>UTR2</italic></td><td align=\"left\"><italic>YOR258W</italic></td><td/><td align=\"left\"><italic>YNL305C</italic></td><td align=\"left\"><italic>OST3</italic></td><td align=\"left\"><italic>SXM1</italic></td><td align=\"left\"><italic>YKT6</italic></td><td/></tr><tr><td align=\"left\"><italic>ERG5</italic></td><td align=\"left\"><italic>OPI3</italic></td><td align=\"left\"><italic>VAP1</italic></td><td align=\"left\"><italic>YOR280C</italic></td><td/><td align=\"left\"><italic>YPL101W</italic></td><td align=\"left\"><italic>PCL1</italic></td><td align=\"left\"><italic>TIF11</italic></td><td align=\"left\"><italic>YLR009W</italic></td><td/></tr><tr><td align=\"left\"><italic>ERG6</italic></td><td align=\"left\"><italic>PBI2</italic></td><td align=\"left\"><italic>VID24</italic></td><td align=\"left\"><italic>YOR289W</italic></td><td/><td align=\"left\"><italic>YPR098C</italic></td><td align=\"left\"><italic>PFS2</italic></td><td align=\"left\"><italic>TIF34</italic></td><td align=\"left\"><italic>YLR035C-A</italic></td><td/></tr><tr><td align=\"left\"><italic>EXG1</italic></td><td align=\"left\"><italic>PCL7</italic></td><td align=\"left\"><italic>YAL053W</italic></td><td align=\"left\"><italic>YOR338W</italic></td><td/><td align=\"left\"><italic>YSA1</italic></td><td align=\"left\"><italic>PHO11</italic></td><td align=\"left\"><italic>TIF35</italic></td><td align=\"left\"><italic>YLR065C</italic></td><td/></tr><tr><td align=\"left\"><italic>FBA1</italic></td><td align=\"left\"><italic>PDC1</italic></td><td align=\"left\"><italic>YBL049W</italic></td><td align=\"left\"><italic>YPL004C</italic></td><td/><td/><td align=\"left\"><italic>PHO12</italic></td><td align=\"left\"><italic>TIP1</italic></td><td align=\"left\"><italic>YLR106C</italic></td><td/></tr><tr><td align=\"left\"><italic>FUN14</italic></td><td align=\"left\"><italic>PDC5</italic></td><td align=\"left\"><italic>YBL064C</italic></td><td align=\"left\"><italic>YPL066W</italic></td><td/><td/><td align=\"left\"><italic>PRE10</italic></td><td align=\"left\"><italic>TPM1</italic></td><td align=\"left\"><italic>YLR157C-B</italic></td><td/></tr><tr><td align=\"left\"><italic>GCV1</italic></td><td align=\"left\"><italic>PDH1</italic></td><td align=\"left\"><italic>YBR006W</italic></td><td align=\"left\"><italic>YPL134C</italic></td><td/><td/><td align=\"left\"><italic>PRE2</italic></td><td align=\"left\"><italic>TPM2</italic></td><td align=\"left\"><italic>YLR159W</italic></td><td/></tr><tr><td align=\"left\"><italic>GCV2</italic></td><td align=\"left\"><italic>PDR5</italic></td><td align=\"left\"><italic>YBR053C</italic></td><td align=\"left\"><italic>YPL156C</italic></td><td/><td/><td align=\"left\"><italic>PRE3</italic></td><td align=\"left\"><italic>TRP1</italic></td><td align=\"left\"><italic>YLR221C</italic></td><td/></tr><tr><td align=\"left\"><italic>GCY1</italic></td><td align=\"left\"><italic>PET8</italic></td><td align=\"left\"><italic>YBR230C</italic></td><td align=\"left\"><italic>YPR153W</italic></td><td/><td/><td align=\"left\"><italic>PRE9</italic></td><td align=\"left\"><italic>UBA1</italic></td><td align=\"left\"><italic>YLR227W-B</italic></td><td/></tr><tr><td align=\"left\"><italic>GLK1</italic></td><td align=\"left\"><italic>PEX11</italic></td><td align=\"left\"><italic>YDC1</italic></td><td align=\"left\"><italic>YPR172W</italic></td><td/><td/><td align=\"left\"><italic>PUP2</italic></td><td align=\"left\"><italic>UBC1</italic></td><td align=\"left\"><italic>YLR410W-B</italic></td><td/></tr><tr><td align=\"left\"><italic>GLO1</italic></td><td align=\"left\"><italic>PGK1</italic></td><td align=\"left\"><italic>YDL124W</italic></td><td align=\"left\"><italic>YRA1</italic></td><td/><td/><td align=\"left\"><italic>RDI1</italic></td><td align=\"left\"><italic>UBC13</italic></td><td align=\"left\"><italic>YML039W</italic></td><td/></tr><tr><td align=\"left\"><italic>GLY1</italic></td><td align=\"left\"><italic>PGM2</italic></td><td align=\"left\"><italic>YDR041W</italic></td><td align=\"left\"><italic>YTP1</italic></td><td/><td/><td align=\"left\"><italic>RLP7</italic></td><td align=\"left\"><italic>UBC4</italic></td><td align=\"left\"><italic>YML093W</italic></td><td/></tr><tr><td align=\"left\"><italic>GND1</italic></td><td align=\"left\"><italic>PHO3</italic></td><td align=\"left\"><italic>YDR233C</italic></td><td align=\"left\"><italic>ZRT1</italic></td><td/><td/><td align=\"left\"><italic>RNA14</italic></td><td align=\"left\"><italic>UBC6</italic></td><td align=\"left\"><italic>YML125C</italic></td><td/></tr><tr><td align=\"left\"><italic>GPA2</italic></td><td align=\"left\"><italic>PIR1</italic></td><td align=\"left\"><italic>YDR319C</italic></td><td align=\"left\"><italic>ZRT2</italic></td><td/><td/><td align=\"left\"><italic>RNH70</italic></td><td align=\"left\"><italic>URA5</italic></td><td align=\"left\"><italic>YMR045C</italic></td><td/></tr><tr><td align=\"left\"><italic>GPD2</italic></td><td align=\"left\"><italic>PLB1</italic></td><td align=\"left\"><italic>YDR387C</italic></td><td/><td/><td/><td align=\"left\"><italic>RPA49</italic></td><td align=\"left\"><italic>VAR1</italic></td><td align=\"left\"><italic>YMR046W-A</italic></td><td/></tr><tr><td align=\"left\"><italic>GPH1</italic></td><td align=\"left\"><italic>PPA2</italic></td><td align=\"left\"><italic>YDR391C</italic></td><td/><td/><td/><td align=\"left\"><italic>RPC10</italic></td><td align=\"left\"><italic>YBL005W-B</italic></td><td align=\"left\"><italic>YMR050C</italic></td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>GO Term finder results for genes up-regulated by daunorubicin</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\" colspan=\"4\"><bold>Gen Ontology Term clustering</bold></td></tr></thead><tbody><tr><td align=\"center\">Functional categories</td><td align=\"left\">GOID</td><td align=\"left\">GOID- associated functions</td><td/></tr><tr><td colspan=\"3\"><hr/></td><td/></tr><tr><td align=\"center\">A</td><td align=\"left\">32196; 32197</td><td align=\"left\">Transposition, Ty metabolism</td><td/></tr><tr><td align=\"center\">B</td><td align=\"left\">27; 460; 466; 6364; 6396; 6996; 16043; 16070; 16072; 22613; 22618; 42254; 42255; 42257; 42273; 43170; 65003</td><td align=\"left\">Ribosome assembling (Protein and rRNA) Proteolysis. Ubiquitin-</td><td/></tr><tr><td align=\"center\">C</td><td align=\"left\">6508; 6511; 19941; 30163; 43632; 44257; 51603</td><td align=\"left\">mediated preoteolysis.</td><td/></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><bold>Gene Clustering</bold></td><td/><td/><td/></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\">Distribution among functional categories</td><td align=\"left\">Genes</td><td align=\"left\">Main gene functions</td><td align=\"center\">Number of genes</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\">A only</td><td align=\"left\">FCY1; FRQ1; HIS7; PCL1; PHO11; SER3; SIT1; SPE3; SPE4; TRP1; URA5; YBL005W-B; YBR012W-B; YCL019W; YDR034C-D; YDR098C-B; YDR210C-D; YDR261C-D; YDR261W-B; YDR316W-B; YDR365W-B; YER138C; YER160C; YFL002W-A; YGR038C-B; YGR161W-B; YHR214C-B; YHR214C-C; YJR027W; YJR029W; YLR035C A; YLR157C-B; YLR227W-B; YLR410W-B; YML039W; YMR045C; YMR050C; YNL054W-B; YPR137C-B; YPR158W-B</td><td align=\"left\">Ty genes</td><td align=\"center\">40</td></tr><tr><td align=\"center\">B only</td><td align=\"left\">ACC1; ANB1; ARL1; BFR1; CAF20; CBF5; CCT5; CDC33; CDC60; COP1; CPR6; DIB1; DPB4; DST1; FPR1; HCH1; HIR1; HRP1; HRR25; ILS1; IMP4; KAP123; KRI1; KRR1; LOS1; MGM101; NAT3; NIP7; NMD3; NOP12; NOP58; NPT1; NRD1; OST3; PFS2; RDI1; RLP7; RNA14; RNH70; RPA49; RPC10; RPC31; RPC40; RPG1; RPL13B; RPL32; RPL34B; RPL6A; RPL6B; RPO26; RPS11B; RPS19A; RPS26A; RPS4B; RPS8A; RRP4; RRP5; RRP9; RRS1; RSC6; RVB2; SAS10; SEC21; SEC65; SEC72; SES1; SMD3; SNT309; SPB1; SSF1; SSP120; SUI1; SUI2; SXM1; TIF11; TIF34; TIF35; TIP1; TPM1; TPM2; UBA1; UBC13; YFH1; YIL127C; YKT6; YNL296W; YOR021C; YPT31</td><td align=\"left\">Ribosomal protein genes, rRNA metabolism, translation.</td><td align=\"center\">87</td></tr><tr><td align=\"center\">C&gt;B</td><td align=\"left\">CDC20; HRT1; PRE10; PRE2; PRE3; PRE9; PUP2; RPN10; RPT3; SKP1; SNF8; STS1; UBC1; UBC4; UBC6</td><td align=\"left\">Endopeptidases, ubiquitin-protein ligases</td><td align=\"center\">15</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\">No GO Term</td><td/><td/><td align=\"center\">53</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>GO Term finder results for genes down-regulated by daunorubicin</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\" colspan=\"4\"><bold>Gen Ontology Term clustering</bold></td></tr></thead><tbody><tr><td align=\"center\">Functional categories</td><td align=\"left\">GOID</td><td align=\"left\">GOID- associated functions</td><td/></tr><tr><td colspan=\"3\"><hr/></td><td/></tr><tr><td align=\"center\">A</td><td align=\"left\">5975; 5996; 6006; 6007; 6066; 6067; 6082; 6090; 6094; 6096; 6113; 6766; 6767; 9056; 9063; 15980; 16051; 16052; 19318; 19319; 19320; 19752; 32787; 44248; 44262; 44275; 46164; 46165; 46364; 46365;</td><td align=\"left\">Alcohol and carbohydrate metabolism (including glycolysis). Vitamin and organic acid metabolism.</td><td/></tr><tr><td align=\"center\">B</td><td align=\"left\">6091; 6099; 6100; 6519; 6520; 6536; 6537; 6807; 8652; 9064; 9084; 9308; 9309; 44271; 46356</td><td align=\"left\">Amino acid metabolic process. Tricarboxilic acid cycle.</td><td/></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Gene Clustering</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\">Distribution among functional categories</td><td align=\"left\">Genes</td><td align=\"left\">Main gene functions</td><td align=\"center\">Number of genes</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\">A&gt;&gt;B</td><td align=\"left\">GPD2; PDC1; PDC5; PCL7; UGP1; DAK1; GLO1; INO1; PGM2; MDH2; PSA1; GRE3; GCY1; GLK1; TPI1; HXK1; HXK2; PFK1; VID24; GND1; TKL1; PYC2; PGK1; TDH3; ENO1; ENO2; TDH1; TDH2; FBA1; GPM1</td><td align=\"left\">Glycolysis</td><td align=\"center\">30</td></tr><tr><td align=\"center\">A&gt;B</td><td align=\"left\">AAH1; ADH1; ADH2; ADH5; ALD4; ALD6; AMS1; ARA1; AYR1; CTS1; EHT1; ERG10; ERG11; ERG26; ERG5; EXG1; FAS1; GPH1; GSY2; HOR2; LAP4; MDH1; PDH1; PEX11; PHO3; PRB1; RHR2; RIB1; RIB4; SCS7; SNO1; SNZ1; TPS2; TSL1</td><td align=\"left\">Alcohol, lipid and sterol metabolism</td><td align=\"center\">34</td></tr><tr><td align=\"center\">A ≈ B</td><td align=\"left\">AAT2; BAT2; CAR2; CHA1; COX20; GCV1; GCV2; GLY1; LSC2; MCR1; PPA2; QCR10; QCR2; RIP1; SRY1; UGA1</td><td align=\"left\">Amino acid metabolism. Respiration</td><td align=\"center\">16</td></tr><tr><td align=\"center\">A&lt;B</td><td align=\"left\">ACO1; ARG1; ARG4; ARG8; ARO3; ARO4; CIT1; CPA1; CYT1; GDH1; GLT1; IDH1; IDH2; ILV5; MEP1; MEP3; MET6; URA2</td><td align=\"left\">Nitrogen compound (including amino acids) metabolism. Tricarboxilic acid cylce</td><td align=\"center\">18</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\">No GO term</td><td/><td/><td align=\"center\">181</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Primers used in this study</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>GENE</bold></td><td align=\"left\"><bold>Primer Sequence</bold></td><td align=\"left\"><bold>Function</bold></td></tr></thead><tbody><tr><td align=\"center\"><italic>ACO1</italic></td><td align=\"left\">for: 5'-GTGGTGCTGATGCCGTTG-3'</td><td align=\"left\">Aconitase</td></tr><tr><td/><td align=\"left\">rev: 5'-CCTTCAATTCCCATGGACGA-3'</td><td/></tr><tr><td align=\"center\"><italic>ACT1</italic></td><td align=\"left\">for: 5'-TGTGTAAAGCCGGTTTTGCC-3'</td><td align=\"left\">Actin</td></tr><tr><td/><td align=\"left\">rev: 5'-TTGACCCATACCGACCATGAT-3'</td><td/></tr><tr><td align=\"center\"><italic>ARG1</italic></td><td align=\"left\">for: 5'-GCCCACATTTCTTACGAGGC-3'</td><td align=\"left\">Arginosuccinate synthetase</td></tr><tr><td/><td align=\"left\">rev: 5'-TGGTCCGGAGCATCCATT-3'</td><td/></tr><tr><td align=\"center\"><italic>ARG4</italic></td><td align=\"left\">for: 5'-AAATTTGTCCGTCATCCAAACG-3'</td><td align=\"left\">Argininosuccinate lyase</td></tr><tr><td/><td align=\"left\">rev: 5'-CCGGTGTGGACTTTACCAGC-3'</td><td/></tr><tr><td align=\"center\"><italic>CAR2</italic></td><td align=\"left\">for: 5'-CATCGCCCAATTGAAAGCTC-3'</td><td align=\"left\">L-ornithine transaminase</td></tr><tr><td/><td align=\"left\">rev: 5'-CCTTGGATGGGTCGATTACG-3'</td><td/></tr><tr><td align=\"center\"><italic>CDC19</italic></td><td align=\"left\">for: 5'-TGGCCATTGCTTTGGACAC-3'</td><td align=\"left\">Pyruvate kinase</td></tr><tr><td/><td align=\"left\">rev: 5'-GGTGAAGATCATTTCGTGGTTTG-3'</td><td/></tr><tr><td align=\"center\"><italic>FBA1</italic></td><td align=\"left\">for: 5'-AATGCTTCCATCAAGGGTGC-3'</td><td align=\"left\">Fructose 1,6-bisphosphate aldolase</td></tr><tr><td/><td align=\"left\">rev: 5'-CAACTGGGATACCGTAAGCTG-3'</td><td/></tr><tr><td align=\"center\"><italic>GPM1</italic></td><td align=\"left\">for: 5'-TCACCGGTTGGGTTGATGTTA-3'</td><td align=\"left\">Glycerate Phosphomutase</td></tr><tr><td/><td align=\"left\">rev: 5'-TCCTTCAACAATTCACCGGC-3'</td><td/></tr><tr><td align=\"center\"><italic>HSP26</italic></td><td align=\"left\">for: 5'-AGAGGCTACGCACCAAGACG-3'</td><td align=\"left\">Heat Shock Protein</td></tr><tr><td/><td align=\"left\">rev: 5'-AGAATCCTTTGCGGGTGTGT-3'</td><td/></tr><tr><td align=\"center\"><italic>HXK1</italic></td><td align=\"left\">for: 5'-GTTGACAGCGAGACCTTGAGAA-3'</td><td align=\"left\">Hexokinase isoenzyme 1</td></tr><tr><td/><td align=\"left\">rev: 5'-CAACCGGGAATCATTGGAAT-3'</td><td/></tr><tr><td align=\"center\"><italic>PGI1</italic></td><td align=\"left\">for: 5'-CTCAAAGAACTTGGTCAACGAT-3'</td><td align=\"left\">Phosphoglucoisomerase</td></tr><tr><td/><td align=\"left\">rev: 5'-CAAACCGGTGACGTTAGCCT-3'</td><td/></tr><tr><td align=\"center\"><italic>PGK1</italic></td><td align=\"left\">for: 5'-CCCAGGTTCCGTTCTTTTGTTG-3'</td><td align=\"left\">3-phosphoglycerate kinase</td></tr><tr><td/><td align=\"left\">rev: 5'-TTGACCATCGACCTTTCTGGA-3'</td><td/></tr><tr><td align=\"center\"><italic>RPO21</italic></td><td align=\"left\">for: 5'-AGGTTTGCTGCAATTTGGACTT-3'</td><td align=\"left\">RNA polymerase II largest subunit B220</td></tr><tr><td/><td align=\"left\">rev: 5'-CAACCTCCCCTTGATACGAGC-3'</td><td/></tr><tr><td align=\"center\"><italic>RPS28A</italic></td><td align=\"left\">for: 5'-AGCCAAGGTCATCAAAGTTTTAGG-3'</td><td align=\"left\">Ribosomal Protein of the Small subunit</td></tr><tr><td/><td align=\"left\">rev: 5'-TTCCAAGAATTCGACACGGAC</td><td/></tr><tr><td align=\"center\"><italic>TDH(1-3)</italic></td><td align=\"left\">for: 5'-AGACTGTTGACGGTCCATCCC-3'</td><td align=\"left\">Glyceraldehyde-3-phosphate dehydrogenase</td></tr><tr><td/><td align=\"left\">rev: 5'-AAGCGGTTCTACCACCTCTCC-3'</td><td/></tr><tr><td align=\"center\"><italic>HOR2</italic></td><td align=\"left\">for: 5'-GTGCAACGCTTTGAACGCT-3'</td><td align=\"left\">Glicerol-1-phosphatase</td></tr><tr><td/><td align=\"left\">rev: 5'-GAAGTTGCCACAGCCCATTT-3'</td><td/></tr><tr><td align=\"center\"><italic>TPS2</italic></td><td align=\"left\">for: 5'-TCATGCCCCATGGCCTAGTA-3'</td><td align=\"left\">Trehalose-6-phosphate phosphatase</td></tr><tr><td/><td align=\"left\">rev: 5'-TTTCTACGTGGCAAACAACGAA-3'</td><td/></tr><tr><td align=\"center\"><italic>GLO1</italic></td><td align=\"left\">for: 5'-AGGATCCAGCAAGGACCGTT-3'</td><td align=\"left\">Glyoxalase</td></tr><tr><td/><td align=\"left\">rev: 5'-GCTTCATACCGAAGTGTTCGG-3'</td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Differential expression in daunorubicin-treated versus non-treated cells, measured by RT-qPCR</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"4\">Treated/Non treated^a)</td><td/><td/><td/><td/></tr><tr><td/><td colspan=\"4\"><hr/></td><td/><td/><td/><td/></tr><tr><td align=\"left\">Function</td><td align=\"center\">ORF</td><td align=\"right\">Time 0</td><td align=\"right\">Time 4 h</td><td align=\"right\">Fold variation <break/>(4 h/0 h)</td><td align=\"right\"><italic>p</italic>^{<italic>b</italic>)}</td><td align=\"right\">Corrected <italic>p </italic><break/>(Bonferroni)</td><td align=\"right\"><italic>n </italic><break/>(technical replicates)</td><td align=\"center\"><italic>n </italic><break/>(biological replicates)</td></tr></thead><tbody><tr><td/><td align=\"center\"><italic>ACO1</italic></td><td align=\"right\">0.001</td><td align=\"right\">-1.090</td><td align=\"right\">0.470</td><td align=\"right\">0.001</td><td align=\"right\">0.020</td><td align=\"right\">60</td><td align=\"center\">5</td></tr><tr><td/><td align=\"center\"><italic>CDC19</italic></td><td align=\"right\">0.034</td><td align=\"right\">-1.132</td><td align=\"right\">0.446</td><td align=\"right\">6.3 × 10^-13</td><td align=\"right\">9.5 × 10^-12</td><td align=\"right\">108</td><td align=\"center\">5</td></tr><tr><td/><td align=\"center\"><italic>FBA1</italic></td><td align=\"right\">0.005</td><td align=\"right\">-1.207</td><td align=\"right\">0.432</td><td align=\"right\">1.0 × 10^-13</td><td align=\"right\">1.5 × 10^-12</td><td align=\"right\">60</td><td align=\"center\">5</td></tr><tr><td/><td align=\"center\"><italic>GPM1</italic></td><td align=\"right\">-0.005</td><td align=\"right\">-0.948</td><td align=\"right\">0.520</td><td align=\"right\">3.0 × 10^-8</td><td align=\"right\">4.5 × 10^-7</td><td align=\"right\">60</td><td align=\"center\">5</td></tr><tr><td align=\"left\">Energy metabolism</td><td align=\"center\"><italic>HOR2</italic></td><td align=\"right\">-0.010</td><td align=\"right\">-1.413</td><td align=\"right\">0.378</td><td align=\"right\">9.0 × 10^-4</td><td align=\"right\">0.014</td><td align=\"right\">24</td><td align=\"center\">2</td></tr><tr><td/><td align=\"center\"><italic>HXK1</italic></td><td align=\"right\">0.315</td><td align=\"right\">-1.935</td><td align=\"right\">0.210</td><td align=\"right\">1.6 × 10^-21</td><td align=\"right\">2.5 × 10^-20</td><td align=\"right\">72</td><td align=\"center\">5</td></tr><tr><td/><td align=\"center\"><italic>PGI1</italic></td><td align=\"right\">0.005</td><td align=\"right\">-0.061</td><td align=\"right\">0.956</td><td align=\"right\">0.80</td><td align=\"right\">&gt; 0.05</td><td align=\"right\">60</td><td align=\"center\">5</td></tr><tr><td/><td align=\"center\"><italic>PGK1</italic></td><td align=\"right\">0.005</td><td align=\"right\">-1.228</td><td align=\"right\">0.425</td><td align=\"right\">8.9 × 10^-19</td><td align=\"right\">1.3 × 10^-17</td><td align=\"right\">60</td><td align=\"center\">5</td></tr><tr><td/><td align=\"center\"><italic>TDH</italic></td><td align=\"right\">-0.015</td><td align=\"right\">-1.428</td><td align=\"right\">0.375</td><td align=\"right\">5.9 × 10^-12</td><td align=\"right\">8.8 × 10^-11</td><td align=\"right\">60</td><td align=\"center\">5</td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\"><italic>ARG1</italic></td><td align=\"right\">-0.010</td><td align=\"right\">-2.032</td><td align=\"right\">0.246</td><td align=\"right\">2.1 × 10^-7</td><td align=\"right\">3.2 × 10^-6</td><td align=\"right\">24</td><td align=\"center\">2</td></tr><tr><td align=\"left\">Amino acid metabolism</td><td align=\"center\"><italic>ARG4</italic></td><td align=\"right\">-0.001</td><td align=\"right\">-1.413</td><td align=\"right\">0.376</td><td align=\"right\">5.3 × 10^-6</td><td align=\"right\">8.0 × 10^-5</td><td align=\"right\">23</td><td align=\"center\">2</td></tr><tr><td/><td align=\"center\"><italic>CAR2</italic></td><td align=\"right\">-0.011</td><td align=\"right\">-0.294</td><td align=\"right\">0.822</td><td align=\"right\">0.09</td><td align=\"right\">&gt; 0.05</td><td align=\"right\">35</td><td align=\"center\">3</td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\"><italic>ACT1</italic></td><td align=\"right\">-0.480</td><td align=\"right\">-1.440</td><td align=\"right\">0.514</td><td align=\"right\">0.126</td><td align=\"right\">&gt; 0.05</td><td align=\"right\">8</td><td align=\"center\">3</td></tr><tr><td align=\"left\">Others</td><td align=\"center\"><italic>HSP26</italic></td><td align=\"right\">0.081</td><td align=\"right\">-2.921</td><td align=\"right\">0.125</td><td align=\"right\">5.1 × 10^-8</td><td align=\"right\">7.6 × 10^-7</td><td align=\"right\">24</td><td align=\"center\">2</td></tr><tr><td/><td align=\"center\"><italic>RPS28A</italic></td><td align=\"right\">-0.005</td><td align=\"right\">0.476</td><td align=\"right\">1.396</td><td align=\"right\">0.002</td><td align=\"right\">0.028</td><td align=\"right\">60</td><td align=\"center\">5</td></tr><tr><td/><td align=\"center\"><italic>TPS2</italic></td><td align=\"right\">0.002</td><td align=\"right\">0.120</td><td align=\"right\">1.086</td><td align=\"right\">0.42</td><td align=\"right\">&gt; 0.05</td><td align=\"right\">22</td><td align=\"center\">2</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T6\"><label>Table 6</label><caption><p>Transcription factors preferently associated to DNR-inhibited genes</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Factor</td><td align=\"right\">Total regulated genes^a)</td><td align=\"center\" colspan=\"3\">DNR-down regulated genes</td><td align=\"center\" colspan=\"2\"><italic>p</italic></td></tr><tr><td colspan=\"1\"><hr/></td><td colspan=\"1\"><hr/></td><td colspan=\"3\"><hr/></td><td colspan=\"2\"><hr/></td></tr><tr><td/><td/><td align=\"right\">Observed</td><td align=\"right\">Expected (out of 280)</td><td align=\"center\">Observed/Expected</td><td align=\"center\">Hypergeometric</td><td align=\"center\">Bonferroni</td></tr></thead><tbody><tr><td align=\"left\">Sok2p</td><td align=\"right\">561</td><td align=\"right\">118</td><td align=\"right\">45.45</td><td align=\"center\">2.6</td><td align=\"center\">5.6 × 10^-27</td><td align=\"center\">7.2 × 10^-25</td></tr><tr><td align=\"left\">Msn2p</td><td align=\"right\">316</td><td align=\"right\">72</td><td align=\"right\">25.58</td><td align=\"center\">2.8</td><td align=\"center\">2.0 × 10^-17</td><td align=\"center\">2.6 × 10^-15</td></tr><tr><td align=\"left\">Msn4p</td><td align=\"right\">286</td><td align=\"right\">67</td><td align=\"right\">23.13</td><td align=\"center\">2.9</td><td align=\"center\">8.3 × 10^-17</td><td align=\"center\">1.1 × 10^-14</td></tr><tr><td align=\"left\">Gis1p</td><td align=\"right\">91</td><td align=\"right\">35</td><td align=\"right\">7.35</td><td align=\"center\">4.8</td><td align=\"center\">1.5 × 10^-16</td><td align=\"center\">1.9 × 10^-14</td></tr><tr><td align=\"left\">Cst6p</td><td align=\"right\">104</td><td align=\"right\">36</td><td align=\"right\">8.44</td><td align=\"center\">4.3</td><td align=\"center\">4.0 × 10^-15</td><td align=\"center\">5.1 × 10^-13</td></tr><tr><td align=\"left\">Pdr3p</td><td align=\"right\">84</td><td align=\"right\">29</td><td align=\"right\">6.8</td><td align=\"center\">4.3</td><td align=\"center\">2.4 × 10^-12</td><td align=\"center\">3.1 × 10^-10</td></tr><tr><td align=\"left\">Yap1p</td><td align=\"right\">1025</td><td align=\"right\">133</td><td align=\"right\">83</td><td align=\"center\">1.6</td><td align=\"center\">2.1 × 10^-11</td><td align=\"center\">2.8 × 10^-9</td></tr><tr><td align=\"left\">Met4p</td><td align=\"right\">746</td><td align=\"right\">105</td><td align=\"right\">60.42</td><td align=\"center\">1.7</td><td align=\"center\">8.8 × 10^-11</td><td align=\"center\">1.1 × 10^-8</td></tr><tr><td align=\"left\">Adr1p</td><td align=\"right\">148</td><td align=\"right\">36</td><td align=\"right\">11.97</td><td align=\"center\">3.0</td><td align=\"center\">3.6 × 10^-10</td><td align=\"center\">4.6 × 10^-8</td></tr><tr><td align=\"left\">Xbp1p</td><td align=\"right\">84</td><td align=\"right\">26</td><td align=\"right\">6.8</td><td align=\"center\">3.8</td><td align=\"center\">5.3 × 10^-10</td><td align=\"center\">6.9 × 10^-8</td></tr><tr><td align=\"left\">Rox1p</td><td align=\"right\">202</td><td align=\"right\">44</td><td align=\"right\">16.33</td><td align=\"center\">2.7</td><td align=\"center\">6.2 × 10^-10</td><td align=\"center\">7.9 × 10^-8</td></tr><tr><td align=\"left\">Aft1p</td><td align=\"right\">397</td><td align=\"right\">66</td><td align=\"right\">32.11</td><td align=\"center\">2.1</td><td align=\"center\">9.5 × 10^-10</td><td align=\"center\">1.2 × 10^-7</td></tr><tr><td align=\"left\">Crz1p</td><td align=\"right\">155</td><td align=\"right\">37</td><td align=\"right\">12.52</td><td align=\"center\">3.0</td><td align=\"center\">1.4 × 10^-9</td><td align=\"center\">1.8 × 10^-7</td></tr><tr><td align=\"left\">Pdr1p</td><td align=\"right\">205</td><td align=\"right\">42</td><td align=\"right\">16.6</td><td align=\"center\">2.5</td><td align=\"center\">3.9 × 10^-9</td><td align=\"center\">5.1 × 10^-7</td></tr><tr><td align=\"left\">Skn7p</td><td align=\"right\">215</td><td align=\"right\">44</td><td align=\"right\">17.42</td><td align=\"center\">2.5</td><td align=\"center\">5.4 × 10^-9</td><td align=\"center\">7.0 × 10^-7</td></tr><tr><td align=\"left\">Gcn4p</td><td align=\"right\">309</td><td align=\"right\">54</td><td align=\"right\">25.04</td><td align=\"center\">2.2</td><td align=\"center\">7.8 × 10^-9</td><td align=\"center\">1.0 × 10^-6</td></tr><tr><td align=\"left\">Stp2p</td><td align=\"right\">131</td><td align=\"right\">32</td><td align=\"right\">10.61</td><td align=\"center\">3.0</td><td align=\"center\">1.5 × 10^-8</td><td align=\"center\">2.0 × 10^-6</td></tr><tr><td align=\"left\">Hsf1p</td><td align=\"right\">266</td><td align=\"right\">48</td><td align=\"right\">21.5</td><td align=\"center\">2.2</td><td align=\"center\">5.2 × 10^-8</td><td align=\"center\">6.7 × 10^-6</td></tr><tr><td align=\"left\">Mig1p</td><td align=\"right\">74</td><td align=\"right\">21</td><td align=\"right\">5.99</td><td align=\"center\">3.5</td><td align=\"center\">1.1 × 10^-7</td><td align=\"center\">1.4 × 10^-5</td></tr><tr><td align=\"left\">Ino2p</td><td align=\"right\">81</td><td align=\"right\">22</td><td align=\"right\">6.53</td><td align=\"center\">3.4</td><td align=\"center\">1.2 × 10^-7</td><td align=\"center\">1.6 × 10^-5</td></tr><tr><td align=\"left\">Gcr2p</td><td align=\"right\">97</td><td align=\"right\">25</td><td align=\"right\">7.89</td><td align=\"center\">3.2</td><td align=\"center\">2.8 × 10^-7</td><td align=\"center\">3.6 × 10^-5</td></tr><tr><td align=\"left\">Mga1p</td><td align=\"right\">151</td><td align=\"right\">31</td><td align=\"right\">12.25</td><td align=\"center\">2.5</td><td align=\"center\">4.6 × 10^-7</td><td align=\"center\">5.9 × 10^-5</td></tr><tr><td align=\"left\">Mbp1p</td><td align=\"right\">242</td><td align=\"right\">42</td><td align=\"right\">19.59</td><td align=\"center\">2.1</td><td align=\"center\">4.6 × 10^-7</td><td align=\"center\">5.9 × 10^-5</td></tr><tr><td align=\"left\">Rfx1p</td><td align=\"right\">87</td><td align=\"right\">23</td><td align=\"right\">7.08</td><td align=\"center\">3.2</td><td align=\"center\">6.0 × 10^-7</td><td align=\"center\">7.7 × 10^-5</td></tr><tr><td align=\"left\">Stp1p</td><td align=\"right\">91</td><td align=\"right\">23</td><td align=\"right\">7.35</td><td align=\"center\">3.1</td><td align=\"center\">1.1 × 10^-6</td><td align=\"center\">1.4 × 10^-4</td></tr><tr><td align=\"left\">Rtg3p</td><td align=\"right\">108</td><td align=\"right\">24</td><td align=\"right\">8.71</td><td align=\"center\">2.8</td><td align=\"center\">1.9 × 10^-6</td><td align=\"center\">2.4 × 10^-4</td></tr><tr><td align=\"left\">Swi4p</td><td align=\"right\">302</td><td align=\"right\">47</td><td align=\"right\">24.49</td><td align=\"center\">1.9</td><td align=\"center\">2.5 × 10^-6</td><td align=\"center\">3.3 × 10^-4</td></tr><tr><td align=\"left\">Rgt1p</td><td align=\"right\">44</td><td align=\"right\">14</td><td align=\"right\">3.54</td><td align=\"center\">4.0</td><td align=\"center\">2.9 × 10^-6</td><td align=\"center\">3.7 × 10^-4</td></tr><tr><td align=\"left\">Ino4p</td><td align=\"right\">333</td><td align=\"right\">50</td><td align=\"right\">26.94</td><td align=\"center\">1.9</td><td align=\"center\">3.1 × 10^-6</td><td align=\"center\">4.0 × 10^-4</td></tr><tr><td align=\"left\">Sut1p</td><td align=\"right\">34</td><td align=\"right\">12</td><td align=\"right\">2.72</td><td align=\"center\">4.4</td><td align=\"center\">4.1 × 10^-6</td><td align=\"center\">5.3 × 10^-4</td></tr><tr><td align=\"left\">Gat4p</td><td align=\"right\">64</td><td align=\"right\">18</td><td align=\"right\">5.17</td><td align=\"center\">3.5</td><td align=\"center\">4.5 × 10^-6</td><td align=\"center\">5.8 × 10^-4</td></tr><tr><td align=\"left\">Nrg1p</td><td align=\"right\">168</td><td align=\"right\">31</td><td align=\"right\">13.61</td><td align=\"center\">2.3</td><td align=\"center\">4.7 × 10^-6</td><td align=\"center\">6.1 × 10^-4</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T7\"><label>Table 7</label><caption><p>Transcription factors selectively enriched in daunorubicin-down regulated gene promoters</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Factor</td><td align=\"center\">Found/expected</td><td align=\"center\"><italic>p^a)</italic></td><td align=\"left\">Binding sequences</td><td align=\"center\">CpG steps</td><td align=\"left\">Characteristics/Function</td></tr></thead><tbody><tr><td align=\"left\">Gis1p</td><td align=\"center\">4.76</td><td align=\"center\">1.9 × 10^-14</td><td align=\"left\">TWAGGGAT, AGGGG</td><td/><td align=\"left\">JmjC domain-containing histone demethylase; transcription factor involved in the expression of genes during nutrient limitation; also involved in the negative regulation of DPP1 and PHR1</td></tr><tr><td align=\"left\">Sut1p</td><td align=\"center\">4.41</td><td align=\"center\">5.3 × 10^-4</td><td align=\"left\">CGCG</td><td align=\"center\">*</td><td align=\"left\">Transcription factor of the Zn [II]2Cys6 family involved in sterol uptake; involved in induction of hypoxic gene expression</td></tr><tr><td align=\"left\">Cst6p</td><td align=\"center\">4.27</td><td align=\"center\">5.1 × 10^-13</td><td align=\"left\">TGACGTCA, TTACGTAA</td><td align=\"center\">*</td><td align=\"left\">Basic leucine zipper (bZIP) transcription factor of the ATF/CREB family, activates transcription of genes involved in utilization of non-optimal carbon sources; involved in telomere maintenance</td></tr><tr><td align=\"left\">Pdr3p</td><td align=\"center\">4.26</td><td align=\"center\">3.1 × 10^-10</td><td align=\"left\">TCCGCGGA</td><td align=\"center\">*</td><td align=\"left\">Transcriptional activator of the pleiotropic drug resistance network, regulates expression of ATP-binding cassette (ABC) transporters through binding to cis-acting sites known as PDREs (PDR responsive elements)</td></tr><tr><td align=\"left\">Rgt1p</td><td align=\"center\">3.95</td><td align=\"center\">3.7 × 10^-4</td><td align=\"left\">CGGANNA</td><td align=\"center\">*</td><td align=\"left\">Glucose-responsive transcription factor that regulates expression of several glucose transporter (HXT) genes in response to glucose; binds to promoters and acts both as a transcriptional activator and repressor</td></tr><tr><td align=\"left\">Xbp1p</td><td align=\"center\">3.82</td><td align=\"center\">6.9 × 10^-8</td><td align=\"left\">GCCTCGARMGA</td><td align=\"center\">*</td><td align=\"left\">Transcriptional repressor that binds to promoter sequences of the cyclin genes, CYS3, and SMF2; expression is induced by stress or starvation during mitosis, and late in meiosis; member of the Swi4p/Mbp1p family; potential Cdc28p substrate</td></tr><tr><td align=\"left\">Mig1p</td><td align=\"center\">3.51</td><td align=\"center\">1.4 × 10^-5</td><td align=\"left\">W(4-5)GCGGGG</td><td align=\"center\">*</td><td align=\"left\">Transcription factor involved in glucose repression; sequence specific DNA binding protein containing two Cys2His2 zinc finger motifs; regulated by the SNF1 kinase and the GLC7 phosphatase</td></tr><tr><td align=\"left\">Gat4p</td><td align=\"center\">3.48</td><td align=\"center\">5.8 × 10^-4</td><td align=\"left\">GATA</td><td/><td align=\"left\">Protein containing GATA family zinc finger motifs</td></tr><tr><td align=\"left\">Ino2p</td><td align=\"center\">3.37</td><td align=\"center\">1.6 × 10^-5</td><td align=\"left\">WYTTCAYRTGS</td><td align=\"center\">*</td><td align=\"left\">Component of the heteromeric Ino2p/Ino4p basic helix-loop-helix transcription activator that binds inositol/choline-responsive elements (ICREs), required for derepression of phospholipid biosynthetic genes in response to inositol depletion</td></tr><tr><td align=\"left\">Rfx1p</td><td align=\"center\">3.25</td><td align=\"center\">7.7 × 10^-5</td><td align=\"left\">TCRYYRYRGCAAC</td><td align=\"center\">*</td><td align=\"left\">Protein involved in DNA damage and replication checkpoint pathway; recruits repressors Tup1p and Cyc8p to promoters of DNA damage-inducible genes; similar to a family of mammalian DNA binding RFX1-4 proteins</td></tr><tr><td align=\"left\">Gcr2p</td><td align=\"center\">3.17</td><td align=\"center\">3.6 × 10^-5</td><td align=\"left\">CTTCC, CWTCC (Gcr1p)</td><td/><td align=\"left\">Transcriptional activator of genes involved in glycolysis; interacts and functions with the DNA binding protein Gcr1p</td></tr><tr><td align=\"left\">Stp1p</td><td align=\"center\">3.13</td><td align=\"center\">1.4 × 10^-4</td><td align=\"left\">CGGCN(6)CGGC</td><td align=\"center\">*</td><td align=\"left\">Transcription factor, activated by proteolytic processing in response to signals from the SPS sensor system for external amino acids; activates transcription of amino acid permease genes and may have a role in tRNA processing</td></tr><tr><td align=\"left\">Stp2p</td><td align=\"center\">3.02</td><td align=\"center\">2.0 × 10^-6</td><td align=\"left\">CGGGGTGN(7)CGCACCG</td><td align=\"center\">*</td><td align=\"left\">Transcription factor, activated by proteolytic processing in response to signals from the SPS sensor system for external amino acids; activates transcription of amino acid permease genes</td></tr><tr><td align=\"left\">Adr1p</td><td align=\"center\">3.01</td><td align=\"center\">4.6 × 10^-8</td><td align=\"left\">TTGGRGN(6-38)CYCCAA</td><td/><td align=\"left\">Carbon source-responsive zinc-finger transcription factor, required for transcription of the glucose-repressed gene ADH2, of peroxisomal protein genes, and of genes required for ethanol, glycerol, and fatty acid utilization</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T8\"><label>Table 8</label><caption><p>Correlation coefficient and associated <italic>p </italic>values between daunorubicin-treated and Transcription-factor deleted strains^a)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Deletion strain</td><td align=\"center\">r</td><td align=\"center\"><italic>p </italic>(T-test)</td><td align=\"center\">Bonferroni</td></tr></thead><tbody><tr><td align=\"left\">Δ<italic>sok2</italic></td><td align=\"center\">0.428</td><td align=\"center\">3.1 × 10^-19</td><td align=\"center\">3.1 × 10^-17</td></tr><tr><td align=\"left\">Δ<italic>adr1</italic></td><td align=\"center\">0.427</td><td align=\"center\">3.8 × 10^-19</td><td align=\"center\">3.8 × 10^-17</td></tr><tr><td align=\"left\">Δ<italic>cst6</italic></td><td align=\"center\">0.344</td><td align=\"center\">1.5 × 10^-12</td><td align=\"center\">1.5 × 10^-10</td></tr><tr><td align=\"left\">Δ<italic>pho4</italic></td><td align=\"center\">0.256</td><td align=\"center\">2.1 × 10^-7</td><td align=\"center\">2.1 × 10^-5</td></tr><tr><td align=\"left\">Δ<italic>ste12</italic></td><td align=\"center\">0.239</td><td align=\"center\">1.3 × 10^-6</td><td align=\"center\">1.3 × 10^-4</td></tr><tr><td align=\"left\">Δ<italic>hap4</italic></td><td align=\"center\">0.236</td><td align=\"center\">1.9 × 10^-6</td><td align=\"center\">1.9 × 10^-4</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T9\"><label>Table 9</label><caption><p>Genes and conditions used for the graph in Figure 6.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Gly genes</td><td align=\"left\">rpg genes</td><td align=\"left\">rpg genes</td><td align=\"center\" colspan=\"3\">Experiments/conditions</td></tr></thead><tbody><tr><td align=\"left\"><italic>ADH1</italic></td><td align=\"left\"><italic>RPL10</italic></td><td align=\"left\"><italic>RPL6A</italic></td><td align=\"left\"><bold>DNA damage^a</bold></td><td align=\"left\"><bold>Osmotic stress^b</bold></td><td align=\"left\"><bold>Oxidative stress^b</bold></td></tr><tr><td align=\"left\"><italic>ADH2</italic></td><td align=\"left\"><italic>RPL11A</italic></td><td align=\"left\"><italic>RPL6B</italic></td><td align=\"left\">DES460 + 0.02% MMS - 120 min</td><td align=\"left\">1M sorbitol - 120 min</td><td align=\"left\">1 mM Menadione (10 min)redo</td></tr><tr><td align=\"left\"><italic>ADH3</italic></td><td align=\"left\"><italic>RPL11B</italic></td><td align=\"left\"><italic>RPL7A</italic></td><td align=\"left\">DES460 + 0.02% MMS - 15 min</td><td align=\"left\">1M sorbitol - 15 min</td><td align=\"left\">1 mM Menadione (105 min) redo</td></tr><tr><td align=\"left\"><italic>ADH5</italic></td><td align=\"left\"><italic>RPL12A</italic></td><td align=\"left\"><italic>RPL7B</italic></td><td align=\"left\">DES460 + 0.02% MMS - 30 min</td><td align=\"left\">1M sorbitol - 30 min</td><td align=\"left\">1 mM Menadione (120 min)redo</td></tr><tr><td align=\"left\"><italic>CDC19</italic></td><td align=\"left\"><italic>RPL12B</italic></td><td align=\"left\"><italic>RPL8A</italic></td><td align=\"left\">DES460 + 0.02% MMS - 5 min</td><td align=\"left\">1M sorbitol - 45 min</td><td align=\"left\">1 mM Menadione (160 min) redo</td></tr><tr><td align=\"left\"><italic>ENO1</italic></td><td align=\"left\"><italic>RPL13A</italic></td><td align=\"left\"><italic>RPL8B</italic></td><td align=\"left\">DES460 + 0.02% MMS - 60 min</td><td align=\"left\">1M sorbitol - 5 min</td><td align=\"left\">1 mM Menadione (20 min) redo</td></tr><tr><td align=\"left\"><italic>ENO2</italic></td><td align=\"left\"><italic>RPL13B</italic></td><td align=\"left\"><italic>RPL9A</italic></td><td align=\"left\">DES460 + 0.02% MMS - 90 min</td><td align=\"left\">1M sorbitol - 60 min</td><td align=\"left\">1 mM Menadione (30 min) redo</td></tr><tr><td align=\"left\"><italic>FBA1</italic></td><td align=\"left\"><italic>RPL14B</italic></td><td align=\"left\"><italic>RPL9B</italic></td><td align=\"left\">DES460 + 0.2% MMS - 45 min</td><td align=\"left\">1M sorbitol - 90 min</td><td align=\"left\">1 mM Menadione (50 min)redo</td></tr><tr><td align=\"left\"><italic>GLK1</italic></td><td align=\"left\"><italic>RPL15B</italic></td><td align=\"left\"><italic>RPS0A</italic></td><td align=\"left\">wt_plus_gamma_10_min</td><td align=\"left\">Hypo-osmotic shock - 15 min</td><td align=\"left\">1 mM Menadione (80 min) redo</td></tr><tr><td align=\"left\"><italic>GPM1</italic></td><td align=\"left\"><italic>RPL16A</italic></td><td align=\"left\"><italic>RPS0B</italic></td><td align=\"left\">wt_plus_gamma_120_min</td><td align=\"left\">Hypo-osmotic shock - 30 min</td><td align=\"left\">1.5 mM diamide (10 min)</td></tr><tr><td align=\"left\"><italic>GPM2</italic></td><td align=\"left\"><italic>RPL16B</italic></td><td align=\"left\"><italic>RPS10A</italic></td><td align=\"left\">wt_plus_gamma_20_min</td><td align=\"left\">Hypo-osmotic shock - 45 min</td><td align=\"left\">1.5 mM diamide (20 min)</td></tr><tr><td align=\"left\"><italic>GPM3</italic></td><td align=\"left\"><italic>RPL17A</italic></td><td align=\"left\"><italic>RPS10B</italic></td><td align=\"left\">wt_plus_gamma_30_min</td><td align=\"left\">Hypo-osmotic shock - 5 min</td><td align=\"left\">1.5 mM diamide (30 min)</td></tr><tr><td align=\"left\"><italic>HXK1</italic></td><td align=\"left\"><italic>RPL17B</italic></td><td align=\"left\"><italic>RPS11A</italic></td><td align=\"left\">wt_plus_gamma_45_min</td><td align=\"left\">Hypo-osmotic shock - 60 min</td><td align=\"left\">1.5 mM diamide (40 min)</td></tr><tr><td align=\"left\"><italic>HXK2</italic></td><td align=\"left\"><italic>RPL18A</italic></td><td align=\"left\"><italic>RPS11B</italic></td><td align=\"left\">wt_plus_gamma_5_min</td><td/><td align=\"left\">1.5 mM diamide (5 min)</td></tr><tr><td align=\"left\"><italic>LAT1</italic></td><td align=\"left\"><italic>RPL18B</italic></td><td align=\"left\"><italic>RPS12</italic></td><td align=\"left\">wt_plus_gamma_60_min</td><td align=\"left\"><bold>AA/N starvation^b</bold></td><td align=\"left\">1.5 mM diamide (50 min)</td></tr><tr><td align=\"left\"><italic>PDA1</italic></td><td align=\"left\"><italic>RPL19A</italic></td><td align=\"left\"><italic>RPS13</italic></td><td align=\"left\">wt_plus_gamma_90_min</td><td align=\"left\">aa starv 0.5 h</td><td align=\"left\">1.5 mM diamide (60 min)</td></tr><tr><td align=\"left\"><italic>PDB1</italic></td><td align=\"left\"><italic>RPL19B</italic></td><td align=\"left\"><italic>RPS14A</italic></td><td/><td align=\"left\">aa starv 1 h</td><td align=\"left\">1.5 mM diamide (90 min)</td></tr><tr><td align=\"left\"><italic>PDC1</italic></td><td align=\"left\"><italic>RPL1A</italic></td><td align=\"left\"><italic>RPS14B</italic></td><td/><td align=\"left\">aa starv 2 h</td><td align=\"left\">1 mM Menadione (40 min) redo</td></tr><tr><td align=\"left\"><italic>PDC5</italic></td><td align=\"left\"><italic>RPL1B</italic></td><td align=\"left\"><italic>RPS15</italic></td><td align=\"left\"><bold>Temperature^b</bold></td><td align=\"left\">aa starv 4 h</td><td align=\"left\">2.5 mM DTT 005 min dtt-1</td></tr><tr><td align=\"left\"><italic>PDX1</italic></td><td align=\"left\"><italic>RPL20A</italic></td><td align=\"left\"><italic>RPS16A</italic></td><td align=\"left\">17 deg growth ct-1</td><td align=\"left\">aa starv 6 h</td><td align=\"left\">2.5 mM DTT 015 min dtt-1</td></tr><tr><td align=\"left\"><italic>PFK1</italic></td><td align=\"left\"><italic>RPL20B</italic></td><td align=\"left\"><italic>RPS16B</italic></td><td align=\"left\">21 deg growth ct-1</td><td align=\"left\">Nitrogen Depletion 1 d</td><td align=\"left\">2.5 mM DTT 030 min dtt-1</td></tr><tr><td align=\"left\"><italic>PFK2</italic></td><td align=\"left\"><italic>RPL21A</italic></td><td align=\"left\"><italic>RPS17A</italic></td><td align=\"left\">25 deg growth ct-1</td><td align=\"left\">Nitrogen Depletion 1 h</td><td align=\"left\">2.5 mM DTT 045 min dtt-1</td></tr><tr><td align=\"left\"><italic>PGI1</italic></td><td align=\"left\"><italic>RPL21B</italic></td><td align=\"left\"><italic>RPS17B</italic></td><td align=\"left\">29 deg growth ct-1</td><td align=\"left\">Nitrogen Depletion 12 h</td><td align=\"left\">2.5 mM DTT 060 min dtt-1</td></tr><tr><td align=\"left\"><italic>PGK1</italic></td><td align=\"left\"><italic>RPL22A</italic></td><td align=\"left\"><italic>RPS18A</italic></td><td align=\"left\">29C to 33C - 15 minutes</td><td align=\"left\">Nitrogen Depletion 2 d</td><td align=\"left\">2.5 mM DTT 090 min dtt-1</td></tr><tr><td align=\"left\"><italic>PGM1</italic></td><td align=\"left\"><italic>RPL22B</italic></td><td align=\"left\"><italic>RPS18B</italic></td><td align=\"left\">29C to 33C - 30 minutes</td><td align=\"left\">Nitrogen Depletion 2 h</td><td align=\"left\">2.5 mM DTT 120 min dtt-1</td></tr><tr><td align=\"left\"><italic>PGM2</italic></td><td align=\"left\"><italic>RPL23A</italic></td><td align=\"left\"><italic>RPS19A</italic></td><td align=\"left\">29C to 33C - 5 minutes</td><td align=\"left\">Nitrogen Depletion 3 d</td><td align=\"left\">2.5 mM DTT 180 min dtt-1</td></tr><tr><td align=\"left\"><italic>STO1</italic></td><td align=\"left\"><italic>RPL23B</italic></td><td align=\"left\"><italic>RPS19B</italic></td><td align=\"left\">33C vs. 30C - 90 minutes</td><td align=\"left\">Nitrogen Depletion 30 min.</td><td align=\"left\">constant 0.32 mM H2O2 (10 min) redo</td></tr><tr><td align=\"left\"><italic>TDH1</italic></td><td align=\"left\"><italic>RPL24A</italic></td><td align=\"left\"><italic>RPS1A</italic></td><td align=\"left\">37 deg growth ct-1</td><td align=\"left\">Nitrogen Depletion 4 h</td><td align=\"left\">constant 0.32 mM H2O2 (100 min) redo</td></tr><tr><td align=\"left\"><italic>TDH2</italic></td><td align=\"left\"><italic>RPL24B</italic></td><td align=\"left\"><italic>RPS1B</italic></td><td align=\"left\">DBY7286 37 degree heat - 20 min</td><td align=\"left\">Nitrogen Depletion 5 d</td><td align=\"left\">constant 0.32 mM H2O2 (120 min) redo</td></tr><tr><td align=\"left\"><italic>TDH3</italic></td><td align=\"left\"><italic>RPL25</italic></td><td align=\"left\"><italic>RPS2</italic></td><td align=\"left\">DBYmsn2/4 (real strain) + 37 degrees (20 min)</td><td align=\"left\">Nitrogen Depletion 8 h</td><td align=\"left\">constant 0.32 mM H2O2 (160 min) redo</td></tr><tr><td align=\"left\"><italic>TPI1</italic></td><td align=\"left\"><italic>RPL26A</italic></td><td align=\"left\"><italic>RPS20</italic></td><td align=\"left\">DBYmsn2-4- 37 degree heat - 20 min</td><td/><td align=\"left\">constant 0.32 mM H2O2 (20 min) redo</td></tr><tr><td align=\"left\"><italic>TYE7</italic></td><td align=\"left\"><italic>RPL26B</italic></td><td align=\"left\"><italic>RPS21A</italic></td><td align=\"left\">Heat Shock 005 minutes hs-2</td><td align=\"left\"><bold>Stationary phase^b</bold></td><td align=\"left\">constant 0.32 mM H2O2 (30 min) redo</td></tr><tr><td/><td align=\"left\"><italic>RPL27A</italic></td><td align=\"left\"><italic>RPS22A</italic></td><td align=\"left\">Heat Shock 015 minutes hs-2</td><td align=\"left\">YPD 1 d ypd-2</td><td align=\"left\">constant 0.32 mM H2O2 (40 min) rescan</td></tr><tr><td/><td align=\"left\"><italic>RPL27B</italic></td><td align=\"left\"><italic>RPS22B</italic></td><td align=\"left\">Heat Shock 030inutes hs-2</td><td align=\"left\">YPD 10 h ypd-2</td><td align=\"left\">constant 0.32 mM H2O2 (50 min) redo</td></tr><tr><td/><td align=\"left\"><italic>RPL28</italic></td><td align=\"left\"><italic>RPS23A</italic></td><td align=\"left\">Heat Shock 05 minutes hs-1</td><td align=\"left\">YPD 12 h ypd-2</td><td align=\"left\">constant 0.32 mM H2O2 (60 min) redo</td></tr><tr><td/><td align=\"left\"><italic>RPL2A</italic></td><td align=\"left\"><italic>RPS23B</italic></td><td align=\"left\">Heat Shock 060 minutes hs-2</td><td align=\"left\">YPD 2 d ypd-2</td><td align=\"left\">constant 0.32 mM H2O2 (80 min) redo</td></tr><tr><td/><td align=\"left\"><italic>RPL3</italic></td><td align=\"left\"><italic>RPS24A</italic></td><td align=\"left\">Heat Shock 10 minutes hs-1</td><td align=\"left\">YPD 2 h ypd-2</td><td align=\"left\">DBY7286 + 0.3 mM H2O2 (20 min)</td></tr><tr><td/><td align=\"left\"><italic>RPL30</italic></td><td align=\"left\"><italic>RPS24B</italic></td><td align=\"left\">Heat Shock 15 minutes hs-1</td><td align=\"left\">YPD 3 d ypd-2</td><td align=\"left\">DBYmsn2/4 (real strain) + 0.32 mM H2O2 (20 min)</td></tr><tr><td/><td align=\"left\"><italic>RPL31A</italic></td><td align=\"left\"><italic>RPS25A</italic></td><td align=\"left\">heat shock 17 to 37, 20 minutes</td><td align=\"left\">YPD 4 h ypd-2</td><td align=\"left\">DBYmsn2msn4 (good strain) + 0.32 mM H2O2</td></tr><tr><td/><td align=\"left\"><italic>RPL31B</italic></td><td align=\"left\"><italic>RPS25B</italic></td><td align=\"left\">Heat Shock 20 minutes hs-1</td><td align=\"left\">YPD 5 d ypd-2</td><td align=\"left\">dtt 000 min dtt-2</td></tr><tr><td/><td align=\"left\"><italic>RPL32</italic></td><td align=\"left\"><italic>RPS26A</italic></td><td align=\"left\">heat shock 21 to 37, 20 minutes</td><td align=\"left\">YPD 6 h ypd-2</td><td align=\"left\">dtt 015 min dtt-2</td></tr><tr><td/><td align=\"left\"><italic>RPL33A</italic></td><td align=\"left\"><italic>RPS26B</italic></td><td align=\"left\">heat shock 25 to 37, 20 minutes</td><td align=\"left\">YPD 8 h ypd-2</td><td align=\"left\">dtt 030 min dtt-2</td></tr><tr><td/><td align=\"left\"><italic>RPL33B</italic></td><td align=\"left\"><italic>RPS27A</italic></td><td align=\"left\">heat shock 29 to 37, 20 minutes</td><td align=\"left\">YPD stationary phase 1 d ypd-1</td><td align=\"left\">dtt 060 min dtt-2</td></tr><tr><td/><td align=\"left\"><italic>RPL34B</italic></td><td align=\"left\"><italic>RPS27B</italic></td><td align=\"left\">Heat Shock 30 minutes hs-1</td><td align=\"left\">YPD stationary phase 12 h ypd-1</td><td align=\"left\">dtt 120 min dtt-2</td></tr><tr><td/><td align=\"left\"><italic>RPL35A</italic></td><td align=\"left\"><italic>RPS28A</italic></td><td align=\"left\">heat shock 33 to 37, 20 minutes</td><td align=\"left\">YPD stationary phase 13 d ypd-1</td><td align=\"left\">dtt 240 min dtt-2</td></tr><tr><td/><td align=\"left\"><italic>RPL35B</italic></td><td align=\"left\"><italic>RPS28B</italic></td><td align=\"left\">Heat Shock 40 minutes hs-1</td><td align=\"left\">YPD stationary phase 2 d ypd-1</td><td align=\"left\">dtt 480 min dtt-2</td></tr><tr><td/><td align=\"left\"><italic>RPL36A</italic></td><td align=\"left\"><italic>RPS29A</italic></td><td align=\"left\">Heat Shock 60 minutes hs-1</td><td align=\"left\">YPD stationary phase 2 h ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL37A</italic></td><td align=\"left\"><italic>RPS29B</italic></td><td align=\"left\">Heat Shock 80 minutes hs-1</td><td align=\"left\">YPD stationary phase 22 d ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL37B</italic></td><td align=\"left\"><italic>RPS3</italic></td><td align=\"left\">steady state 15 dec C ct-2</td><td align=\"left\">YPD stationary phase 28 d ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL38</italic></td><td align=\"left\"><italic>RPS30A</italic></td><td align=\"left\">steady state 17 dec C ct-2</td><td align=\"left\">YPD stationary phase 3 d ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL39</italic></td><td align=\"left\"><italic>RPS30B</italic></td><td align=\"left\">steady state 21 dec C ct-2</td><td align=\"left\">YPD stationary phase 4 h ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL40A</italic></td><td align=\"left\"><italic>RPS31</italic></td><td align=\"left\">steady state 25 dec C ct-2</td><td align=\"left\">YPD stationary phase 5 d ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL40B</italic></td><td align=\"left\"><italic>RPS4A</italic></td><td align=\"left\">steady state 29 dec C ct-2</td><td align=\"left\">YPD stationary phase 7 d ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL41A</italic></td><td align=\"left\"><italic>RPS4B</italic></td><td align=\"left\">steady state 33 dec C ct-2</td><td align=\"left\">YPD stationary phase 8 h ypd-1</td><td/></tr><tr><td/><td align=\"left\"><italic>RPL42A</italic></td><td align=\"left\"><italic>RPS6A</italic></td><td align=\"left\">steady state 36 dec C ct-2</td><td/><td/></tr><tr><td/><td align=\"left\"><italic>RPL42B</italic></td><td align=\"left\"><italic>RPS6B</italic></td><td align=\"left\">steady state 36 dec C ct-2 (repeat hyb)</td><td/><td/></tr><tr><td/><td align=\"left\"><italic>RPL43A</italic></td><td align=\"left\"><italic>RPS7A</italic></td><td/><td/><td/></tr><tr><td/><td align=\"left\"><italic>RPL43B</italic></td><td align=\"left\"><italic>RPS7B</italic></td><td/><td/><td/></tr><tr><td/><td align=\"left\"><italic>RPL4A</italic></td><td align=\"left\"><italic>RPS8A</italic></td><td/><td/><td/></tr><tr><td/><td align=\"left\"><italic>RPL4B</italic></td><td align=\"left\"><italic>RPS8B</italic></td><td/><td/><td/></tr><tr><td/><td align=\"left\"><italic>RPL5</italic></td><td align=\"left\"><italic>RPS9A</italic></td><td/><td/><td/></tr><tr><td/><td/><td align=\"left\"><italic>RPS9B</italic></td><td/><td/><td/></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>a) Data expressed as dual logarithmic values of expression ratios, treated versus untreated. Corrected by <italic>RPO21 </italic>expression.</p><p>b) Student's T-Test, time 0 <italic>versus </italic>time 4 h ratios</p></table-wrap-foot>", "<table-wrap-foot><p>a) Number of genes associated to each factor, following YEASTRACT. Only genes used in the microarray analysis (3458) were considered.</p></table-wrap-foot>", "<table-wrap-foot><p>a) Hypergeometric distribution with Bonferroni correction</p></table-wrap-foot>", "<table-wrap-foot><p>a) Only genes significantly altered by daunorubicin treatment were considered (n = 445).</p></table-wrap-foot>", "<table-wrap-foot><p>a) Data from reference [##REF##11102521##21##]</p><p>b) Data from reference [##REF##11598186##22##]</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2164-9-358-1\"/>", "<graphic xlink:href=\"1471-2164-9-358-2\"/>", "<graphic xlink:href=\"1471-2164-9-358-3\"/>", "<graphic xlink:href=\"1471-2164-9-358-4\"/>", "<graphic xlink:href=\"1471-2164-9-358-5\"/>", "<graphic xlink:href=\"1471-2164-9-358-6\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>We engineered the first respiratory <italic>Saccharomyces cerevisiae </italic>strain, KOY.TM6*P, by integrating the gene encoding a chimeric hexose transporter, Tm6*, into the genome of an <italic>hxt </italic>null yeast [##REF##15071495##1##]. Subsequently we demonstrated the transferability of this respiratory phenotype to a yeast strain, V5 <italic>hxt1-7</italic>Δ, in which only <italic>HXT1-7 </italic>had been deleted [##REF##16204537##2##]. The resulting V5.TM6*P strain produced only minor amounts of ethanol when cultured on external glucose concentrations as high as 5% [##REF##16204537##2##]. Despite the fact that the glucose flux was reduced to 30% in the V5.TM6*P strain compared with that of its parental V5 strain, the V5.TM6*P strain produced biomass at a specific rate as high as 85% of that produced by the V5 wild-type strain, the yield itself increasing by 50% in the mutant compared to its parental strain.</p>", "<p>Having performed this initial physiological characterization of the V5.TM6*P strain, we wanted to examine more thoroughly the basis of its respiratory phenotype by comparing its transcriptome with that of its wild-type parent strain, V5. This approach has been widely used to obtain a global picture of the cellular responses of yeast to a wide range of physiological changes, including those experienced at the diauxic shift [##REF##15987902##3##, ####REF##15758028##4##, ##REF##9381177##5####9381177##5##]. Recently Ohlmeier and colleagues [##REF##14597615##6##] demonstrated that major changes at the transcriptional level are not reflected at the protein level. Indeed, focusing on the mitochondrial proteome following a diauxic shift, they showed that the levels of only 18 out of 253 identified proteins had changed (17 increased, and 1 decreased). Among them were proteins involved in the tricarboxylic acid (TCA) cycle (Sdh1, Sdh2, and Sdh4) and the respiratory chain (Cox4, Cyb2, and Qcr7). This seeming disagreement with their observations of heterogeneous changes in the transcriptome (where more than 4,000 changes were recorded) is also consistent with our own prior data where we showed that the protein patterns obtained from a 2D-PAGE analysis of the two strains (grown under the same conditions used in this study) were not substantially different [##REF##16204537##2##]. We did note, however, that the levels of the upper-part glycolytic intermediates and ATP did differ between the two strains, and that Cdc19 is present in lower amounts in V5.TM6*P. This discrepancy between proteome and transcriptome regulation has also been observed by others [##REF##15642185##7##] and is also consistent with the fact that there is an acknowledged lack of relationship between the levels of glycolytic proteins and the glycolytic flux [##REF##1533788##8##, ####REF##2528863##9##, ##REF##10862878##10####10862878##10##]. Nonetheless, researchers continue to depend on mRNA as an indicator of cellular state, which is a situation that will continue while methods for the global analysis of protein expression are improved [##REF##15642185##7##].</p>", "<p>In this respect, Brauer and colleagues [##REF##15987902##3##, ####REF##15758028##4##, ##REF##9381177##5####9381177##5##] have used transcriptome analysis to examine adjustments and remodelling of the metabolism of glucose-limited yeast cultures. In that study, the authors constructed metabolic models consistent with their cDNA array data. In this work, the V5.TM6*P strain was therefore characterized at the transcriptome level in comparison with the parental V5 strain in order to obtain a more global understanding of any genetic remodelling underlying the physiological differences between them. Both strains were cultured in 50 g/L glucose with six samples being taken at defined points between 38 and 6 g/L glucose for cDNA array analysis in two independent fermentations for each strain. Consequently, we generated a highly redundant dataset (n = 12) where 922 genes were either induced or repressed when comparing V5 and V5.TM6*P under all glucose concentrations. Using <italic>in vivo </italic>studies together with documented binding specificities of three transcription factors known to have a role in the regulation of respiratory genes (the Hap complex, Cat8 and Mig1), it was possible to correlate the up-regulation of all but 18 of the 146 genes induced by a factor ≥2.0 with the presence of one or more of their binding sites. This means that 75% of the transcriptional response of the genes in our dataset could be related to the activities of just three transcription factors, which in turn would account for 88% of the response of the induced set. Finally, we made the unexpected observation that both the induction and repression of many of the genes in our dataset had a clear glucose dependence in the parent V5 strain that was not present in V5.TM6*P. This important result indicates that the relief of glucose repression in the wild-type is already operable at much higher glucose concentrations than is currently widely accepted and that the phenotype of the respiratory strain is glucose insensitive by comparison.</p>" ]

[ "<title>Methods</title>", "<title>Yeast strains</title>", "<p>The auxotrophic wine yeast strain V5 (<italic>MAT</italic>a <italic>ura3 gal</italic>) and the V5 <italic>hxt1–hxt7</italic>Δ mutant strain (<italic>MAT</italic>a <italic>ura3 gal hxt514</italic>Δ::loxP <italic>hxt367</italic>Δ::lox P <italic>hxt2</italic>Δ::loxP) were kindly provided by the Professor Bruno Blondin, Montpellier, France. The construction of V5.TM6*P was previously described by Henricsson and colleagues [##REF##16204537##2##] where the expression cassette from KOY.TM6*P [##REF##15071495##1##] was transferred into the genome of the V5 <italic>hxt1–hxt7</italic>Δ strain by using the primers PROHXT3 (TCAAATGGCGGTGTAGTTTGAAAAG) and TERHXT7 (TTAAGTGACGGGCGATGAGTAAGAA). Transformants were selected by growth on glucose. The resulting <italic>ura3 </italic>strain is referred to as V5.TM6*. The auxotrophic V5 wild type and V5.TM6* used in this study were made prototrophic by integration of <italic>URA3 </italic>[##REF##10618490##37##] and are referred to as the V5 wild-type strain and V5.TM6*P, respectively.</p>", "<title>Growth conditions</title>", "<p>For both pre-cultures and main cultures, 5 × defined minimal medium [##REF##1523884##38##] with 5% glucose as the sole carbon and energy source was used. A two-step pre-cultivation was performed on a rotary shaker. First a 10 mL culture was grown for 72 h, and used to inoculate 100 mL medium, which after 30 h at 30°C was inoculated to give a final OD₆₁₀of 0.05 in a bioreactor (BRO2, Belach Bioteknik AB, Sweden) in a volume of 2.5 L. Polypropylene glycol P2000 was added as antifoam (100 <italic>μ</italic>L/L). The cultivation conditions were 30°C, 1,000 rpm, pH 5.0 with an airflow of 1.25 L/min. Gas evolution was monitored on-line (type CP460 O₂/CO₂, Belach Bioteknik AB). Cultivations were performed at least in duplicate.</p>", "<title>RNA preparation</title>", "<p>Yeast cells (2 – 20 mL) were directly transferred into twice their volume of ice-cold water. The cells were collected by centrifugation at 0°C for 5 min at 3000 × g, frozen in liquid nitrogen and stored at -80°C. Total RNA was then prepared using the RNeasy kit from Qiagen, following the manufacturer's instructions.</p>", "<title>cDNA array production</title>", "<p>PCR products for the yeast ORFs were generated from yeast DNA using the ORF-specific primer set from Research Genetics and then re-amplified with the Resgen universal forward and reverse primers (Catalogue number 40612). PCR products were prepared by Randy Strich (Fox Chase Cancer Centre, USA) as part of a collaboration with the Wistar Institute, Philadelphia, USA. PCR products were spotted with a BioRobotics Microgrid TASII (Genomic Solutions) by the Wistar Genomics Facility. Each array contained 6,319 individual yeast genes and 1,169 gene repeats. The complete list of arrayed genes on array YA04 is listed under GEO Series accession number GPL4423.</p>", "<title>RNA amplification, hybridization, and scanning</title>", "<p>Amplified RNA (aRNA) was prepared from 1 <italic>μ</italic>g of total RNA for each sample tested using the RiboAmp kit (Arcturus). Labelled targets were prepared from 1.6 <italic>μ</italic>g aRNA with Superscript II reverse transcriptase (Invitrogen), in the presence of 3,000–5,000 Ci/mM [α-³³P] dCTP (Amersham Pharmacia Biotech), 1 mmol/L dATP, 1 mmol/L dTTP, 1 mmol/L dGTP, 67 ng/<italic>μ</italic>g oligo-dT (Promega Biosciences), and 0.65 × random decamer primers (Ambion). Labelled targets were hybridized to individual arrays at 42°C for 18 h in 3 mL Microhyb buffer (Invitrogen). Arrays were washed twice in 2 × standard saline citrate (SSC)/1% SDS solution for 30 min at 50°C, once in 0.5 × SSC/1% SDS and once in 0.1 × SSC/0.5% SDS for 30 min at 55°C. The arrays were exposed to phosphor screens (Amersham Biosciences) for 6 days and scanned in a Storm 820 PhosphorImager (Molecular Dynamics). Quantitation of each spot was assessed by Imagene 5.0 software (BioDiscovery Incorporated) using manual spot alignment, measuring median pixel intensity for each spot and subtracting the local background. The data have been deposited in the NCBI Gene Expression Omnibus and are accessible through GEO Series accession number GSE11799.</p>", "<title>cDNA array data analysis</title>", "<p>Observed values less than 0.1 (5% (8752 values) of the 176640 raw data values) were set to 0.1, in accordance with information from Wistar that variations in values below this threshold should be regarded as measurement noise. After converting to a logarithmic scale, average expression values were computed over spots with the same ORF, and arrays were scale normalized by subtracting the median for each of the 24 arrays. Averages were computed over repeated measurements with the same conditions producing 6 values at different glucose concentrations for each of the V5 (38.4, 35, 25.5, 24.4, 10.5, 7.4 g/L glucose) and V5.TM6*P (36.5, 34.8, 26.3, 25.1, 13.5, 5.6 g/L glucose) strains for each gene. T-tests were then performed, comparing the two strains, using Empirical Bayes robustification of variance estimates [##REF##16904203##39##,##UREF##0##40##].</p>", "<p>The T-tests were adjusted for multiple testing using the false discovery rate method (disregarding correlations between genes). Essentially, then, p-values were adjusted so that when selecting all genes with p-values less than a threshold q, a proportion of q false positives would be expected amongst these genes. The estimated effects are shown as factors in Additional file ##SUPPL##0##1##, as the linear modelling was performed on a log scale. The p-values in Additional file ##SUPPL##0##1## are the adjusted values, so that selecting genes with p-values &lt; 0.05 gives an expected false discovery rate of 5%. Calculations were performed using the BioConductor package limma [##UREF##0##40##]. To present the data, a tabular format was chosen over heatmaps in an attempt to make it as accessible as possible to the scientific community.</p>", "<p>When producing data for Figure ##FIG##3##4##, a similar linear model was used, but with a different design matrix: Essentially, for each gene, 2 straight lines were simultaneously fitted, one for each of the strains, to plots of the 6 log-expression values versus glucose concentration. Empirical Bayes robustification and adjustment for multiple testing was included even in this analysis, using the limma tool [##UREF##0##40##]. Lines with slopes significantly different from zero were found for 20 genes for the V5 strain, but for no genes from the V5.TM6*P strain.</p>" ]

[ "<title>Results</title>", "<p>We compared the transcriptional profiles of the V5 and V5.TM6*P strains cultured in 50 g/L glucose, with samples extracted between 38 and 6 g/L residual glucose. We identified 190 genes with adjusted p-values &lt; 0.05 that changed their expression levels either up by a factor of ≥2.0 (146 genes) or down by a factor of ≤0.5 (44 genes) when comparing the two strains at all three sampling points. These genes are presented by functional category of their gene products in Additional file ##SUPPL##0##1## and are discussed below. Additional file ##SUPPL##1##2## provides the information in Excel format and also includes other relevant literature data. A further 636 genes changed their expression levels either up by a factor of &lt;2.0 (317 genes) or down by a factor &gt; 0.5 (319 genes) and are presented as part of Additional file ##SUPPL##2##3##. The remaining 96 genes were either dubious, Ty-transposable elements, or were detected due to cross-hybridisation, as indicated in the legend to Additional file ##SUPPL##0##1##.</p>", "<title>The majority of genes that have an altered expression in V5.TM6*P compared with its parental strain also have an altered expression following diauxic shift</title>", "<p>We compared our dataset with three sets of expression data for yeast undergoing a diauxic shift, where the samples had been taken at least 2 hours after the glucose is depleted [##REF##15987902##3##, ####REF##15758028##4##, ##REF##9381177##5####9381177##5##]. Of the 190 genes in our dataset that changed their expression levels either up or down by a factor of 2 or more, 154 of these could be identified as changing in the same direction in at least one of these three datasets, with 124 having at least two matches. Only 8 genes (<italic>ERG3</italic>, <italic>RPS27A</italic>, <italic>YHB1</italic>, <italic>TRX1</italic>, <italic>YER188w</italic>, <italic>YKL066w</italic>, <italic>YAR068w </italic>and <italic>Y0L150c</italic>) had changes that were in the opposite direction in at least two of the datasets. This significant overlap between our data set and the data sets of the three previous studies is entirely consistent with the fact that there is a transition from a respiro-fermentative metabolism in V5 to a respiratory metabolism in V5.TM6*P.</p>", "<title>Glycolytic and gluconeogenic genes have substantially altered expression levels in V5.TM6*P</title>", "<p>Four of the most strongly induced genes in Additional file ##SUPPL##0##1## are the glycolytic genes, <italic>HXK1 </italic>(factor 9.0) and <italic>GLK1 </italic>(3.8), and the gluconeogenic genes, <italic>PCK1 </italic>(6.9) and <italic>FBP1 </italic>(5.3). Indeed, as shown in Figure ##FIG##0##1##, the induction in V5.TM6*P of several glycolytic and gluconeogenic genes could be consistent with these cells accommodating for a lack of glucose; the hypothesized mechanism behind the strain's reduced ethanol-producing, respiratory phenotype. The final step in glycolysis which produces pyruvate, is catalyzed by pyruvate kinase. In V5.TM6*P the pyruvate kinase genes, <italic>CDC19 </italic>and <italic>PYK2</italic>, respond in an opposing manner: <italic>CDC19 </italic>is repressed by a factor of 0.4 (consistent with our earlier proteome analysis where Cdc19 is present in lower amounts in V5.TM6*P than V5) while <italic>PYK2 </italic>is induced and is relieved of its glucose-repressed state as a result of the low glucose flux in the respiratory strain. Reduced ethanol production in V5.TM6*P compared to the parental strain is also consistent with the two most strongly repressed genes <italic>PDC1 </italic>(0.2) and <italic>PDC5 </italic>(0.2) encoding the main pyruvate decarboxylases.</p>", "<title>Alternative carbon source usage genes are induced in V5.TM6*P compared with V5, consistent with a reduced glucose flux</title>", "<p>An adaptation to enable growth on alternative carbon sources is seen in the induction of the plasma membrane transporter genes <italic>MAL31 </italic>and <italic>STL1 </italic>as well as the genes <italic>SUC2</italic>, <italic>GUT1</italic>, <italic>GUT2</italic>, <italic>ALD6 </italic>and <italic>BDH1 </italic>encoding proteins for the utilization of sucrose, glycerol, acetaldehyde and butanediol, respectively. <italic>GUT1 </italic>in particular is strongly induced by a factor of 6.2, and <italic>STL1 </italic>is the most strongly induced of all genes on the cDNA array (21.7), suggesting that V5.TM6* has undergone a substantial genetic adjustment compared to its parent. Furthermore, <italic>CYB2 </italic>(10.1) and <italic>DLD3 </italic>(0.5), whose gene products are involved in the conversion of lactate to pyruvate, and in the case of <italic>CYB2 </italic>feeding the released electrons into the electron transport chain, are oppositely regulated. Biochemical data and sequence patterns for <italic>YGL157w </italic>(0.3) [##REF##14574691##11##], <italic>YAL061w </italic>(2.9) and <italic>DSF1 </italic>(9.4, homologous to <sc>D</sc>-mannitol 2-dehydrogenase from <italic>Rhodobacter sphaeroides) </italic>suggest that these ORFs encode an oxidoreductase, a polyol dehydrogenase and a putative sugar dehydrogenase, respectively, which could be involved in channelling alternative carbon sources into the TCA cycle. The carbon-deficient state of V5.TM6*P [##REF##15071495##1##] could also be consistent with the induction of <italic>CRC1 </italic>and <italic>CAT2 </italic>whose gene products are involved in <italic>β</italic>-oxidation and carnitine-dependent transport of acetyl-CoA from peroxisomes to mitochondria as a way of using carbon from sources other than glucose [##REF##11329169##12##]. The induction of <italic>SFC1</italic>, <italic>JEN1</italic>, <italic>FBP1 </italic>and <italic>ATO2 </italic>seen in our study has previously been reported to be a specific identifier of carbon-limited growth [##REF##12414795##13##].</p>", "<title>All TCA- and glyoxylate cycle genes have increased expression in V5.TM6*P compared with V5</title>", "<p>The genes encoding the enzymes of the TCA cycle (<italic>CIT1</italic>, <italic>ACO1, IDH1</italic>, <italic>KGD1</italic>, <italic>KGD2</italic>, <italic>LSC2</italic>, <italic>SDH1-4</italic>, <italic>FUM1</italic>, <italic>MDH1</italic>) all have a significant (adjusted p-value &lt; 0.05) increase in expression when comparing V5 and V5.TM6*P. In addition, the gene encoding an isoform of aconitase (<italic>ACO2</italic>) has its expression level reduced by a factor of 0.5. Fermentation is almost completely abolished in the V5.TM6*P strain, and we have previously shown that the TCA cycle has a higher carbon flux [##REF##16204537##2##] to harness the electron flow into the respiratory chain and generate a proton gradient using F₀F₁-ATPase for ATP synthesis (several respiratory chain genes are induced, as discussed below). Interestingly, all genes of the TCA-cycle are induced by greater than a factor of 3.5 except for <italic>CIT1</italic>/<italic>CIT2 </italic>and <italic>KGD1</italic>/<italic>KGD2 </italic>where the factor change is 2.2 – 3.5 for each gene of the pair, presumably yielding the same overall transcriptional response. Consistent with this, Ohlmeier and colleagues have shown that Sdh1, 2 and 4 all have increased expression after a diauxic shift [##REF##14597615##6##].</p>", "<p>The glyoxylate cycle is anabolic, yielding gluconeogenic precursors, and all its genes (<italic>CIT2</italic>, <italic>ACO1, ICL1 </italic>(induced by a factor of 1.5; Additional file ##SUPPL##2##3##), <italic>MLS1 </italic>and <italic>MDH2</italic>) are induced in V5.TM6*P compared with V5 (Additional file ##SUPPL##0##1##, Fig. ##FIG##0##1##). This could be consistent with the correspondingly increased biomass yield of the V5.TM6*P strain (Table ##TAB##0##1##). We note that V5.TM6*P has a lower biomass yield than expected from a totally respiratory phenotype. In chemostat culture under aerobic, fully respiratory conditions, a yield of 0.5 g/g is typical at low dilution rates, whereas the observed low yield in V5.TM6*P has already been published using a range of sugars and concentrations [##REF##16204537##2##]. Furthermore, numerous experiments on KOY.TM6*P, our original respiratory strain based on CEN.PK2-1C, [##REF##15071495##1##] have yielded identical results. These observations are in agreement with our findings in Table ##TAB##0##1## of 0.32 g/g in V5.TM6*P and 0.21 g/g in V5 from batch cultures.</p>", "<title>Genes encoding respiratory chain enzymes are induced in V5.TM6*P compared with V5</title>", "<p>33 key genes encoding mitochondrial enzymes involved in respiration – as well as F₀F₁-ATPase and its chaperones – are all induced by a factor of 2 – 4 in V5.TM6*P (Additional file ##SUPPL##0##1##). The gene encoding Ndi1 (Fig. ##FIG##1##2##), which transfers electrons from NADH to ubiquinone (Q in Fig. ##FIG##1##2##) is induced the most (by a factor of 4.1), while all other components of the respiratory chain are induced by a factor of 2 – 2.5. Genes encoding heme synthesis and incorporation (<italic>HEM2</italic>, <italic>CYT2</italic>), complex assembly (<italic>MBA1</italic>) and heme degradation (<italic>HMX1</italic>) are also induced by a factor of 2 – 2.5. Cytochrome c oxidase also has different chaperones that are induced to form an active structure by proteolysis (<italic>COX20</italic>), as well as delivery of copper (<italic>COX17</italic>). In spite of the higher respiratory rate in V5.TM6*P, several genes involved in detoxification of oxygen radicals are repressed, <italic>e.g. </italic>genes encoding peroxiredoxin (<italic>AHP1</italic>) and thioredoxin (<italic>TRX2</italic>) are repressed by a factor of 0.4 – 0.5. Conversely, the capacity for nitrous oxide detoxification is increased by the induction of <italic>YHB1 </italic>by a factor of 2.4. It is possible that the respiratory phenotype of V5.TM6*P could induce a higher level of oxygen radical formation inside its mitochondria and thus oxidation of guanine could occur on the mitochondrial DNA. Such detrimental effects could be counteracted, <italic>via </italic>the observed induction of <italic>OGG1 </italic>by a factor of 2.2.</p>", "<title>Changes of expression in genes encoding mitochondrial and plasma membrane transporters are consistent with a respiratory phenotype in V5.TM6*P</title>", "<p>The genes encoding the hexose transporters, Hxt1 to Hxt7, have very high sequence identities [##REF##8929273##14##] and thus cross reactivity between them is expected on the array. This explains the observed signal for <italic>HXT1, 4 </italic>and <italic>6</italic>, which have been deleted in V5.TM6*P. <italic>HXT13</italic>, <italic>HXT15 </italic>and <italic>HXT16 </italic>are known to be repressed by high levels of glucose and are consequently induced as expected in V5.TM6*P. A carbon source deficiency triggers an up-regulation of genes encoding plasma membrane proteins responsible for the import of alternative carbon sources, as already noted: <italic>STL1 </italic>(21.7; glycerol); <italic>JEN1 </italic>(10.2; lactate), <italic>PUT4 </italic>(5.8; proline – the main nitrogen source in grape juice) and <italic>MAL31 </italic>(3.1; maltose). A gene associated with nutrient deficiency in yeast [##REF##8139573##15##], <italic>YGP1</italic>, is induced by a factor of 2.6, again supporting the notion that V5.TM6*P has undergone a genetic adjustment compared to its parent.</p>", "<p>It seems that V5.TM6*P also has an altered requirement for metal ions in that genes encoding three metal ion channels have altered expression: <italic>CTR3 </italic>(copper transport) is induced by a factor of 3.0, while <italic>ZRT1 </italic>(zinc transport) and <italic>ENB1 </italic>(iron transport) are repressed by a factor of 0.4 – 0.5. It is likely that copper import is increased so that copper can be transferred to cytochrome c oxidase, which is the main copper-requiring enzyme in Additional file ##SUPPL##0##1##, with 3 copper ions per complex and a copy number of 5,000–12,000 per cell. There are only 45 genes in Additional file ##SUPPL##0##1## with greater than 5,000 copies per cell and none of these has copper as a prosthetic group except cytochrome c oxidase.</p>", "<p>The amino acid uptake systems encoded by <italic>BAP3 </italic>and <italic>GNP1 </italic>are repressed while the proline permease encoded by <italic>PUT4 </italic>is induced. <italic>PUT4 </italic>has a carbon source responsive element (CSRE) [##REF##11024040##16##] that is expected to be induced at the low glucose flux experienced in the V5.TM6*P strain, while <italic>BAP3 </italic>and <italic>GNP1 </italic>are regulated by a sensor of extracellular amino acid concentration, where Ssy5 [##REF##15611869##17##] is a vital component. It is likely that in V5.TM6*P the repression of <italic>BAP3 </italic>and <italic>GNP1 </italic>is a result of the repression (by a factor 0.65; Additional file ##SUPPL##2##3##) of <italic>SSY5 </italic>seen on the cDNA array. Interestingly, induction of genes encoding the ammonia transport system in the V5.TM6*P strain (<italic>ATO2</italic>, <italic>MEP1</italic>) could suggest that the strain is experiencing general nutrient limitation. It has previously been reported for example that the induction of another <italic>MEP </italic>family member, <italic>MEP2</italic>, is associated with nitrogen limitation [##REF##12414795##13##].</p>", "<title>88% of genes induced in V5.TM6*P compared to its parent have transcription factor binding sites for one or more of the Hap complex, Cat8 or Mig1</title>", "<p>Genome-wide studies have yielded a wealth of information on which promoters are bound by different transcription factors [##REF##15343339##18##]. In order to rationalise these resources, we began by looking for biologically-verified data on transcription factors that might control the repression of the subset of 44 genes of Additional file ##SUPPL##0##1##. Since only limited information was publicly available, we did not pursue this part of the dataset further. In contrast, for the 146 induced genes in Additional file ##SUPPL##0##1##, our observation of the similarity between the transcriptional changes in the V5 to V5.TM6*P transition and those in the transition from glucose to ethanol growth lead us to analyse the roles of Hap4, Cat8 and Mig1 which have been previously been verified to be crucial to the expression of genes in the diauxic shift [##REF##11024040##16##,##REF##12537548##19##,##REF##8756637##20##]: Hap4 (in complex with Hap2, 3 and 5) [##REF##12537548##19##] and Cat8 (optionally in complex with Sip4 and Adr1) [##REF##14685767##21##] are transcriptional activators, whereas Mig1 (optionally with Mig2/3) is a transcriptional repressor that works in concert with Tup1-Cyc8 [##REF##7724528##22##].</p>", "<p>The Hap complex is comprised of proteins Hap2, 3, 4 and 5, and has separate subunits for DNA binding and transcriptional activation. DNA binding is mediated by Hap2, 3 and 5 [##REF##7828851##23##] while Hap4 functions as an activation domain. The expression of <italic>HAP4 </italic>is repressed by glucose and is responsible for regulation of Hap complex target genes. We began by scrutinising data for a <italic>HAP4 </italic>over-expression strain, in which 246 genes had their expression affected by at least a factor of 2 [##REF##12537548##19##], which is 5.3% of the genome. 56 of these genes were found in our dataset and are listed in Table ##TAB##1##2##. Only three of these (<italic>GLK1</italic>, <italic>CTR3 </italic>and <italic>ZRT1</italic>) are down-regulated in the <italic>HAP4 </italic>over-expression strain, whereas <italic>GLK1 </italic>and <italic>CTR3 </italic>are induced in the V5.TM6*P strain. Interestingly, a typical Hap complex binding site was not seen in the 500 nucleotide region upstream of the <italic>GLK1, CTR3 </italic>and <italic>ZRT1 </italic>start codon (where the first A in ATG is position 1) while all other genes in Table ##TAB##1##2## except <italic>AGX1</italic>, <italic>MNP1 </italic>and <italic>YNK1 </italic>were found to have at least one Hap complex binding site. These Hap complex-binding-site-containing genes comprise a significant subset (26% of our dataset of 190 genes in Additional file ##SUPPL##0##1##) with a verified biological dependence on Hap4 activation. Clearly, then, Hap4 has a statistically-significant role to play in the metabolism of V5.TM6*P as there is an association between Hap4 dependence and our dataset (χ²= 182.75, degrees of freedom = 1, p = 0.00).</p>", "<p>Since six of the genes from the <italic>HAP4 </italic>over-expression study did not contain Hap complex binding sites, we did a further <italic>in silico </italic>analysis of the induced genes of Additional file ##SUPPL##0##1##. We noted that the genes of Table ##TAB##1##2##, contained motifs that could be sub-divided into 2 distinct families: <bold>CCAAT</bold>G (which we denoted as a Hap4_1 site) and (G/<bold>C)CAA</bold>(G/<bold>T</bold>)CAA (a Hap4_2 site). The bold sequences are the conserved sequences previously identified in the <italic>HAP4 </italic>over-expression strain study [##REF##12537548##19##]. The genes of Additional file ##SUPPL##0##1## were then examined 500 nucleotides upstream and 200 nucleotides downstream from the start codon using WebMOTIF [##REF##17584794##24##] for these two biologically-relevant sites since it has been previously demonstrated that Hap complex binding sites are statistically over-represented up to 400 nucleotides upstream from the start codon [##REF##12537548##19##]: we found that 112/146 (77%) induced genes and 127/190 (67%) of the complete dataset in Additional file ##SUPPL##0##1## had one or more Hap complex binding site (Additional files ##SUPPL##0##1## and ##SUPPL##3##4##). Our analysis included presumed binding sites such as CAAATC and CCAAAC which arose from the biologically-verified data on transcription factors (Table ##TAB##1##2##). For example, <italic>AQY2 </italic>is identified as Hap complex-dependent, and has two CCAAAC sites, but no CCAATNA sites. There are only two other genes that do not contain CCAATNA sites in Table ##TAB##1##2##, namely <italic>CRC1 </italic>and <italic>ALD6</italic>. In addition it was notable that Hap complex consensus sequences are predominantly found upstream of mitochondrial genes with 84/127 (66%) of such genes having at least a partial mitochondrial location. Furthermore, of the induced genes with a known cellular location and at least two Hap complex binding sites, 69/80 (86%) are localized to the mitochondria. It appears, then, that Hap4 activation is preferentially directed at mitochondrially-related functions especially within the TCA-cycle and respiratory chain, but cytosolic support via the glyoxylate cycle is also potentially regulated by Hap4 activation (Additional file ##SUPPL##0##1##). It was further noted that multiple Hap complex binding sites are associated with induction (Fig. ##FIG##2##3A##). A Fisher's exact test showed that there was a statistically-significant association between the change in gene expression and the number of Hap complex binding sites in the gene (p = 0.00). This supported the fact that genes with two or more Hap complex binding sites are likely to be induced, whereas those with one site may be either induced or repressed.</p>", "<p>In order to adopt a similar approach for Cat8, we examined available biological data for Cat8 and Adr1 (as they are co-regulators [##REF##15743812##25##]) from chromatin immunoprecipitations [##REF##15743812##25##], and mRNA expression ratios in wt/Δcat8 strains [##REF##11024040##16##]. We identified 20 genes out of 66 from these studies that were also in our dataset (Table ##TAB##2##3##). Only <italic>ATO2 </italic>was oppositely regulated when comparing our dataset. For all other genes, the change in expression followed the same trend of up-regulation (Table ##TAB##2##3##). As for the Hap4 analysis, we examined these 20 genes for Cat8 binding sites, and identified a typical Cat8 motif [##REF##14685767##21##] in 13 of them, nine being of the Cat8-Sip4 type, TCCATTSRTCCGR (Additional file ##SUPPL##3##4##). The remaining six induced genes (<italic>CYB2</italic>, <italic>GUT1</italic>, <italic>GUT2</italic>, <italic>LSC2</italic>, <italic>SUE1 </italic>and <italic>ODC1</italic>) were further scrutinised, and it was apparent that all except <italic>ODC1 </italic>had been identified from immunoprecipitation data for Adr1 alone, consistent with their lack of Cat8-Sip4 motif. The induced genes of Additional file ##SUPPL##0##1## were therefore searched from the start codon to 1,500 nucleotides upstream for the generic Cat8 motif, CC------CCG motif giving 54 genes (Additional file ##SUPPL##3##4##). The non-coding intergenic regions of these genes were then searched using WebMOTIF [##REF##17584794##24##] and <italic>HXK1</italic>, <italic>COX4</italic>, <italic>ODC1</italic>, <italic>OM14</italic>, <italic>OM45 </italic>and <italic>YOR019w </italic>were found to share the consensus sequence GCCSSTSS(W/Y))CMS in common with <italic>IDP2 </italic>and are thus also listed in Additional file ##SUPPL##3##4##. We noted that genes with these Cat8 binding sites are induced in V5.TM6*P by an average factor of 6.2 while the average for induced genes in Additional file ##SUPPL##0##1## is 3.3. The highly-induced, Cat8-site-containing genes include <italic>HXK1 </italic>(factor change 9.0), <italic>PCK1 </italic>(6.9), <italic>STL1 </italic>(21.7), <italic>JEN1 </italic>(10.3), <italic>ODC1 </italic>(6.3) and <italic>SFC1 </italic>(18.3).</p>", "<p>We observed that some strongly-induced genes in Additional file ##SUPPL##0##1## did not have Hap complex or Cat8 binding sites. Indeed, the expression of many genes such as <italic>SUC2 </italic>is already known to be repressed by both Mig1 and Mig2 in glucose media, although Mig1 is the main factor in this response [##REF##8756637##20##,##REF##9832517##26##]. We therefore examined biological data on gene expression in <italic>Δmig1 </italic>and <italic>Δmig2 </italic>strains [##REF##8756637##20##,##REF##9832517##26##]. In these studies, 11 genes were presented as being Mig dependent, of which 5 were found in Additional file ##SUPPL##0##1## (<italic>HXK1</italic>, <italic>EMI2</italic>, <italic>HXT13, HXT15 </italic>and <italic>DSF1</italic>) and 3 in Additional file ##SUPPL##2##3## (<italic>REG2</italic>, <italic>DOG2 </italic>and <italic>YLR042C</italic>). We then examined protein binding microarray data [##REF##15543148##27##] and tabulated the combined outputs as Table ##TAB##3##4##, which lists 15 Mig-dependent genes. When searching the induced set in Additional file ##SUPPL##0##1##<italic> in silico </italic>for genes containing Mig1 binding sites, 31 were identified, predominantly containing consensus sequences of the SUC2B type (CCCCGGAT), but also in some cases sites that were more homologous to SUC2A (sharing the consensus motif CCCC(G/A)(G/C)AT [##REF##8114729##28##]). We noted that the presence of a Mig1 site in a gene correlated with high induction (Fig. ##FIG##2##3B##) and that overall, our analysis provides a highly complete description of how three transcriptions factors might regulate the 146 induced genes of Additional file ##SUPPL##0##1## (Fig. ##FIG##2##3C##).</p>", "<p>We noted that <italic>HAP4 </italic>(factor change 1.7, Additional file ##SUPPL##2##3##) is induced and <italic>CAT8 </italic>(0.5, adjusted p value &lt; 0.06; Additional file ##SUPPL##4##5##), <italic>MIG2 </italic>(0.5, Additional file ##SUPPL##0##1##) and <italic>MIG3 </italic>(0.3, Additional file ##SUPPL##0##1##) are down-regulated in V5.TM6*P compared to V5. In our dataset, the specific change in array signal for <italic>MIG1 </italic>had an adjusted p-value &gt; 0.05 over the complete experimental range. The result for <italic>CAT8 </italic>was unexpected as it contains an upstream Mig1 binding site similar to <italic>HAP4 </italic>(which is induced) and Cat8 target genes such as <italic>FBP1 </italic>and <italic>PCK1 </italic>are themselves induced. It is possible that the transcript was not probed accurately on the array either due to cross-hybridisation between closely related DNA sequences or on account of its low base expression value (0.24). A future real time Q-PCR experiment would probe the individual transcript.</p>", "<title>Transcripts dependent on external glucose concentration are not found in the respiratory V5.TM6*P strain, even though several glucose- or carbon-source-associated transcripts are glucose dependent in the parental wild-type strain</title>", "<p>The transcriptome of the respiratory V5.TM6*P strain was not found to be responsive to changes in glucose concentration in the culture medium. However, in the wild-type parent strain (V5) the transcriptome was found to vary with glucose availability according to a least squares fit of the logarithmic array values for RNA extracted at the six different glucose concentrations. This dependence on external glucose concentrations is seen in the V5 strain for 20 genes (including those without functional annotations) with an adjusted p-value &lt; 0.05 (Fig. ##FIG##3##4##). The genes with the strongest dependence on external glucose concentrations were found to be <italic>HXK1</italic>, <italic>RGS2</italic>, <italic>ADH7</italic>, <italic>CHA4 </italic>and <italic>GLK1 </italic>(Fig. ##FIG##3##4##). The sugar kinases <italic>HXK1 </italic>and <italic>GLK1 </italic>are induced as expected as they have their maximal expression during growth on other carbon sources [##REF##11024040##16##]. The gene product of <italic>RGS2 </italic>inhibits Gpa2 in the PKA pathway, which is one of the pathways induced by glucose [##REF##12694616##29##]. <italic>ADH7 </italic>appears to be induced on carbon sources other than glucose [##REF##12423374##30##]. In a study by DeRisi and colleagues, batch yeast cultures were harvested for cDNA array analysis at gradually decreasing glucose concentration [##REF##9381177##5##]. Our examination of the supplementary data provided by the authors showed that as glucose was consumed between 18.7 and 7.5 g/L, three of the genes in Figure ##FIG##3##4## (<italic>GLK1</italic>, <italic>HXK1</italic>, and <italic>FAL1</italic>) were found in the top 22 genes with a changed expression from that study. Overall, the results in Figure ##FIG##3##4## show that glucose repression is gradually relieved on going from 37 g/L to 9 g/L glucose. Furthermore relief of repression is already apparent at a surprisingly high glucose concentration, in contrast to the assumption that it is typically triggered over a range of low glucose concentrations [##REF##12694616##29##], the trigger being very low for some genes [##REF##8793872##31##].</p>", "<p>In order to examine whether the glucose dependence of genes in V5 affected our data in Additional file ##SUPPL##0##1##, we excluded the data points at 13.5 and 5.6 g/L glucose for V5.TM6*P, and 10.5 and 7.4 g/L glucose for V5 (which gave the highest array signals) and compared the cDNA array data in this case with each other exactly as for the full data set. Following this procedure, only a small subset of genes that are highly induced had their factor changes increased substantially, but this did not affect the fact that they remain within the top 30 most induced genes in Additional file ##SUPPL##0##1##. They are <italic>HXK1 </italic>(9.0 in Additional file ##SUPPL##0##1##, 19.3 following re-calculation; the gene contains binding sites for the Hap complex, Cat8 and Mig1); <italic>STL1 </italic>(21.7, 27.5; Hap complex, Cat8); <italic>DSF1 </italic>(9.4, 15.1; Hap complex, Mig1); <italic>CYB2 </italic>(10.1, 13.4; Hap complex, Mig1) and <italic>FBP1 </italic>(5.3, 7.4; Cat8, Mig1). Fourteen of the seventeen genes that were most affected by this recalculation were found to contain a Mig1 binding site. It is therefore possible that Mig1 is also involved in relieving glucose repression in the 10 to 5 g/L range, and not only at the depletion of glucose at the diauxic shift. Overall, we found that the amplitudes of the factor changes varied by only 9.3% in this new calculation compared with the original one, which indicated that the glucose dependence of the V5 strain does not have a major influence on the way we generated the data presented in Additional file ##SUPPL##0##1##.</p>" ]

[ "<title>Discussion</title>", "<p>The main physiological change in both the diauxic shift and the transition from a respiro-fermentative to a respiratory metabolism is that glycolysis is channelled to support respiration: this occurs upon glucose depletion during the diauxic shift and throughout the growth curve in glucose medium in V5.TM6*P. Although previously-published genomic analyses of yeast's response to the diauxic shift transition have all been sampled under different conditions – rich (YPD) medium with 2% glucose [##REF##9381177##5##], minimal defined medium in chemostat cultures at different dilution rates [##REF##15758028##4##], and minimal defined media in batch cultures sampled 4 h after the diauxic shift [##REF##15987902##3##] – common themes emerge with changes on going from V5 to V5.TM6*P. Overall 81% of genes which are induced or repressed on going from V5 to V5.TM6*P in our study were found in at least one of these three datasets for the diauxic shift. Furthermore, we also find that for all proteins induced in a study of the diauxic shift by 2D-gel electrophoresis [##REF##14597615##6##], the corresponding genes are induced on going from V5 to V5.TM6*P except for the alcohol dehydrogenase, Adh2. There is a hence close agreement between induction of the following genes and the levels of the corresponding proteins: <italic>FBP1</italic>, <italic>ICL1</italic>, <italic>PCK1</italic>, <italic>IDP2</italic>, <italic>MLS1</italic>, <italic>DLD1</italic>, <italic>ALD6</italic>, <italic>SDH1 </italic>and <italic>CIT2</italic>. This strong correlation with previously-published data and our own observed correlation of data for Cdc19 and <italic>CDC19 </italic>supports the phenotypic similarity between the V5 to V5.TM6*P transition and the diauxic shift. It is worth noting that there is no clear correlation of our data in Additional file ##SUPPL##0##1## with any possible differences in growth rate between the two strains. In a study by Regenberg and colleagues [##REF##17105650##32##], 180 genes changed in expression by more than a factor of 4 when comparing growth rates. Of these 180, 3% of all genes are found in Additional file ##SUPPL##0##1## providing no significant overlap in data.</p>", "<p>It is also possible to see parallels between our data set and a study of nitrogen deprivation and stationary phase growth for wild-type yeast by Gasch and colleagues [##REF##11102521##33##]. This is supported by the fact that ten of the 15 genes that were found to be strongly induced upon nitrogen deprivation and growth in stationary phase in that study are also induced in the V5 to V5.TM6*P transition. Only two genes, <italic>YGR067c </italic>(likely to encode a transcription factor as it has high partial sequence homology with <italic>MIG3 </italic>and <italic>ADR1 </italic>and it fine-tunes the response to nutrient limitation) and <italic>ECM13</italic>, are not induced in the V5.TM6*P strain, but were highly induced in the Gasch study [##REF##11102521##33##] upon nitrogen deprivation and stationary phase growth. In addition, we find that the genes encoding Mep1 and Ato2, which are involved in ammonia transport are also strongly induced in V5.TM6*P in our study. Such an induction is seen in both the diauxic shift studies performed in defined media [##REF##15758028##4##,##REF##11406594##34##], but not in rich medium [##REF##9381177##5##], suggestive of nutrient limitation. In fact, V5.TM6*P also appears carbon limited, not least as a result of the fact that amongst the most strongly induced genes are <italic>FBP1 </italic>and <italic>PCK1</italic>, which encode gluconeogenic enzymes. <italic>STL1</italic>, <italic>MAL31 </italic>and <italic>SUC2 </italic>(encoding proteins involved in usage and/or transport of alternative carbon sources: glycerol, maltose and sucrose respectively) are induced in V5.TM6*P, also consistent with its low glycolytic flux and suggestive of an adaptation of metabolism ('discontinuous modelling) as seen by Brauer and colleagues [##REF##15758028##4##,##REF##11406594##34##] and supported by our previous work [##REF##15071495##1##]. Several of these genes including <italic>FBP1 </italic>and <italic>STL1</italic>, as well as <italic>JEN1</italic>, <italic>ODC1, SFC1 </italic>and <italic>ATO2 </italic>have already been identified as being either induced under carbon limitation or being specific indicators of carbon limitation in yeast cultures [##REF##12414795##13##], consistent with our own data.</p>", "<p>The analysis of transcription factor binding sites present in the genes of Additional file ##SUPPL##0##1## revealed that 127 genes had Hap complex binding sites, the majority of which encode mitochondrial proteins. Of genes that only had Hap complex binding sites present, thirteen had only one and of these only five were induced leading us to hypothesise that multiple Hap complex binding sites are correlated with induction of the Additional file ##SUPPL##0##1## genes (Fig. ##FIG##2##3A##). Interestingly, in a microarray screen of a <italic>hap1Δ </italic>mutant for Hap1-dependent genes [##REF##12112237##35##], 7/24 genes that were induced in the <italic>hap1Δ </italic>mutant are also induced in Additional file ##SUPPL##0##1## (<italic>CYB2</italic>, <italic>MLS1</italic>, <italic>HMX1</italic>, <italic>PUT4</italic>, <italic>STL1</italic>, <italic>HPA2</italic>, <italic>PUT1</italic>), and in each case also have a Hap complex binding site. In addition, 19 genes containing a Cat8 site and 31 genes containing a Mig1 site were predicted, with many of these genes being experimentally-verified in previous <italic>in vivo </italic>studies. Our analysis of the regulation of 88% of the induced genes of our unique respiratory strain (Fig. ##FIG##2##3##) is further supported by a novel computational method [##REF##17224918##36##] that predicts the Hap complex and Mig1 control the genes of Additional file ##SUPPL##0##1##; Cat8 is not included in that database.</p>", "<p>A quite unexpected result from comparing the transcriptional profiles of V5 with V5.TM6*P is the data set in Figure ##FIG##3##4##, showing genes from the parent strain that are clearly glucose dependent and that the glucose dependence observed begins at concentrations (6 – 36 g/L glucose) higher than the range generally regarded as the glucose-repressed region. Strain DBY7286 [##REF##9381177##5##], which is unrelated to V5, previously showed a gradual relief of glucose repression starting at 20 g/L glucose for <italic>GLK1</italic>, <italic>HXK1 </italic>and <italic>FAL1</italic>. These results for V5 and DBY7286 suggest that gradual relief is a general <italic>S. cerevisiae </italic>phenomenon. Future studies will show whether the gradient of the slope is different for different strains. That such glucose dependence requires an intact glucose uptake system is suggested by the fact that it is not seen in the V5.TM6*P strain: the array signal is not changed significantly for any of the genes in the V5.TM6*P strain at the different glucose concentrations sampled. This implies that in V5.TM6*P, glucose sensing is independent of external glucose concentration. The wild-type yeast, on the other hand, shows an extensive modulation of its expression profile at high glucose concentrations despite the fact that its physiological response is largely unaffected. The insensitivity of the V5.TM6*P strain therefore makes it an extremely valuable biotechnological tool as it can be cultured in a wide range of external glucose concentrations, whilst maintaining the same respiratory phenotype [##REF##16204537##2##].</p>" ]

[ "<title>Conclusion</title>", "<p>In this study, we have been able to characterize the transcriptome of a unique respiratory yeast strain. We have been able to identify highly complete collections of known genes in the TCA cycle, glyoxylate cycle and respiratory chain that are consistent with a respiratory metabolism. Our results suggest that there has been genetic remodelling predominantly through the activity of Hap4, Cat8 and Mig1, and that the gene expression profile of V5.TM6*P during growth on glucose resembles wild-type <italic>S. cerevisiae </italic>cells in the diauxic shift.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>We previously described the first respiratory <italic>Saccharomyces cerevisiae </italic>strain, KOY.TM6*P, by integrating the gene encoding a chimeric hexose transporter, Tm6*, into the genome of an <italic>hxt </italic>null yeast. Subsequently we transferred this respiratory phenotype in the presence of up to 50 g/L glucose to a yeast strain, V5 <italic>hxt1-7</italic>Δ, in which only <italic>HXT1-7 </italic>had been deleted. In this study, we compared the transcriptome of the resultant strain, V5.TM6*P, with that of its wild-type parent, V5, at different glucose concentrations.</p>", "<title>Results</title>", "<p>cDNA array analyses revealed that alterations in gene expression that occur when transitioning from a respiro-fermentative (V5) to a respiratory (V5.TM6*P) strain, are very similar to those in cells undergoing a diauxic shift. We also undertook an analysis of transcription factor binding sites in our dataset by examining previously-published biological data for Hap4 (in complex with Hap2, 3, 5), Cat8 and Mig1, and used this in combination with verified binding consensus sequences to identify genes likely to be regulated by one or more of these. Of the induced genes in our dataset, 77% had binding sites for the Hap complex, with 72% having at least two. In addition, 13% were found to have a binding site for Cat8 and 21% had a binding site for Mig1. Unexpectedly, both the up- and down-regulation of many of the genes in our dataset had a clear glucose dependence in the parent V5 strain that was not present in V5.TM6*P. This indicates that the relief of glucose repression is already operable at much higher glucose concentrations than is widely accepted and suggests that glucose sensing might occur inside the cell.</p>", "<title>Conclusion</title>", "<p>Our dataset gives a remarkably complete view of the involvement of genes in the TCA cycle, glyoxylate cycle and respiratory chain in the expression of the phenotype of V5.TM6*P. Furthermore, 88% of the transcriptional response of the induced genes in our dataset can be related to the potential activities of just three proteins: Hap4, Cat8 and Mig1. Overall, our data support genetic remodelling in V5.TM6*P consistent with a respiratory metabolism which is insensitive to external glucose concentrations.</p>" ]

[ "<title>Authors' contributions</title>", "<p>NB initiated the study, prepared the RNA and participated in the data analysis and interpretation, particularly of the transcription factor binding sites. CF carried out the fermentations. PM performed the array analysis. MW performed the statistical analyses of the array data. CC and LS generated the arrays and performed the arraying. LG and CL participated in the experimental design and helped to draft the manuscript. RB participated in the study design, coordinated the data analysis and interpretation, and drafted the manuscript. All authors contributed to the final version of the manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>This work was supported by the European Commission via contracts LSHG-CT-2004-504601 (E-MeP) and LSHG-CT-2006-037793 (OptiCryst) to RMB, and QLG2-CT-2002-00988 (SPINE) to LG and RMB. The Wistar Genomics facility is supported in part by NCI P30 CA10815-34S3. The BBSRC supports bioreactors in the RMB laboratory through an REI award.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>An overview of our dataset in the context of key yeast metabolic pathways</bold>. Genes with altered expression in V5.TM6*P compared with V5 are marked as follows: induced genes are denoted in large, bold font; down-regulated genes are crossed out. Genes that were previously found to be altered in studies of respiratory-state yeast [##REF##14597615##6##], but which are not found in our dataset are denoted in small, grey text. The chimeric Tm6*p transporter, which comprises the amino-terminal half of Hxt1 and the carboxy-terminal half of Hxt7, is responsible for the respiratory phenotype of V5.TM6*P at high glucose concentrations and is present in the plasma membrane. The grey boundaries represent the plasma membrane (top) and the mitochondrial membrane (left). ^†Note that <italic>ICL</italic>(1.5), <italic>NDE1 </italic>(1.8) and <italic>GPD1 </italic>(1.8) are induced in our study (see Additional file ##SUPPL##2##3##). ^‡<italic>LPD1 </italic>is a component of pyruvate dehydrogenase.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>An overview of our dataset in the context of the respiratory chain in yeast</bold>. Genes with altered expression in V5.TM6*P compared with V5 are marked as follows: genes induced by a factor ≥2.0 are denoted in large, bold font; genes induced by a factor &lt;2.0 are denoted in smaller, grey font.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>(A) The influence of multiple Hap complex sites on the change in gene expression of the genes in Additional file </bold>##SUPPL##0##1##. A Fisher's exact test was performed for genes containing predicted binding sites for the Hap complex and their association with induction or repression. Genes were grouped into either those with one Hap complex binding site or those with two or more according to Additional file ##SUPPL##0##1##. <bold>(B) The influence of Hap4, Cat8 and Mig1 on the magnitude of induction of the genes in Additional file </bold>##SUPPL##0##1##. The average fold-induction of genes in Additional file ##SUPPL##0##1##, grouped according to the transcription factor binding sites they contain, was calculated. It was observed that those containing binding sites for each of the Hap complex (where Hap4 is the activator), Cat8 and Mig1, had the highest average fold induction with Mig1 being a dominant factor in high induction. A bar for Cat8 is not included as only one gene (<italic>YOR019W</italic>, factor change 2.1) has a Cat8 site alone. <bold>(C) A Venn diagram of the distribution of binding sites for the Hap complex, Cat8 and Mig1 in the genes in Additional file </bold>##SUPPL##0##1##. 88% of the genes in Additional file ##SUPPL##0##1## contain a binding site for one or more of the Hap complex, Cat8 and Mig1. The range of induction for the 190 genes from Additional file ##SUPPL##0##1## is 2.0 to 21.7.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Transcript dependence on external glucose concentration for 20 genes in the V5 strain</bold>. A clear glucose-dependence (grey bars) was observed for all genes in the V5 strain with only a small variation in array signal (white bars) for the same genes in V5.TM6*P. Genes marked with a dot also have a significant change of expression on going from V5 to V5.TM6*P and as such are listed in Additional file ##SUPPL##0##1##. The factor change in gene expression was obtained as the quotient between the expression values at 37 and 9 g/L glucose given by the line fitted to the expression data for each gene. R²values are for the linear fit for the V5 strain. The R²values for V5.TM6*P were between 0.16 and 0.78 indicating no significant linear relationship.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Growth characteristics of V5 and V5.TM6*P at 30°C starting at 50 g/L glucose.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"left\"><bold>V5</bold></td><td align=\"left\"><bold>V5.TM6*P</bold></td></tr></thead><tbody><tr><td align=\"left\"><bold>Generation time during glucose growth (h)</bold></td><td align=\"left\">3.7 (0.2; n = 2)</td><td align=\"left\">4.5 (0.3; n = 4)</td></tr><tr><td align=\"left\"><bold>Biomass yield (g/g)</bold></td><td align=\"left\">0.21 (0.02 ; n = 2)</td><td align=\"left\">0.32 (0.01; n = 4)</td></tr><tr><td align=\"left\"><bold>Glucose consumption (mmolg dry weight)^-1h^-1</bold></td><td align=\"left\">9.00 (0.70; n = 2)</td><td align=\"left\">2.70 (0.20; n = 4)</td></tr><tr><td align=\"left\"><bold>Ethanol yield (g/g)</bold></td><td align=\"left\">0.33 (0.01; n = 2)</td><td align=\"left\">0.05 (0.01; n = 4)</td></tr><tr><td align=\"left\"><bold>Glycerol yield (g/g)</bold></td><td align=\"left\">0.04 (0; n = 2)</td><td align=\"left\">0</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Biologically-verified data on transcription factors that act on genes also found in Additional file ##SUPPL##0##1##: transcript dependence on <italic>HAP4 </italic>over-expression.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>ORF</bold></td><td align=\"left\"><bold>GENE</bold></td><td align=\"left\"><bold>FACTOR CHANGE IN GENE EXPRESSION FROM V5 TO V5.TM6*P</bold></td><td align=\"left\"><bold>FACTOR CHANGE IN GENE EXPRESSION FROM WT TO A <italic>HAP4 </italic>OVER-EXPRESSION STRAIN </bold>[##REF##12537548##19##]</td></tr></thead><tbody><tr><td align=\"left\" colspan=\"4\"><bold>Glycolysis</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YCL040W</italic></td><td align=\"left\"><italic>GLK1</italic></td><td align=\"left\">3.8</td><td align=\"left\">0.4</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Alternative carbon source utilization</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YML054C</italic></td><td align=\"left\"><italic>CYB2</italic></td><td align=\"left\">10.1</td><td align=\"left\">5.0</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>TCA cycle</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YIL125W</italic></td><td align=\"left\"><italic>KGD1</italic></td><td align=\"left\">2.2</td><td align=\"left\">2.2</td></tr><tr><td align=\"left\"><italic>YDR148C</italic></td><td align=\"left\"><italic>KGD2</italic></td><td align=\"left\">3.5</td><td align=\"left\">4.4</td></tr><tr><td align=\"left\"><italic>YFL018C</italic></td><td align=\"left\"><italic>LPD1</italic></td><td align=\"left\">2.0</td><td align=\"left\">2.2</td></tr><tr><td align=\"left\"><italic>YKL085W</italic></td><td align=\"left\"><italic>MDH1</italic></td><td align=\"left\">3.6</td><td align=\"left\">2.2</td></tr><tr><td align=\"left\"><italic>YKL148C</italic></td><td align=\"left\"><italic>SDH1</italic></td><td align=\"left\">4.6</td><td align=\"left\">4.7</td></tr><tr><td align=\"left\"><italic>YLL041C</italic></td><td align=\"left\"><italic>SDH2</italic></td><td align=\"left\">4.1</td><td align=\"left\">3.9</td></tr><tr><td align=\"left\"><italic>YKL141W</italic></td><td align=\"left\"><italic>SDH3</italic></td><td align=\"left\">3.7</td><td align=\"left\">2.3</td></tr><tr><td align=\"left\"><italic>YDR178W</italic></td><td align=\"left\"><italic>SDH4</italic></td><td align=\"left\">5.3</td><td align=\"left\">3.2</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Glyoxylate cycle</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YIR029W</italic></td><td align=\"left\"><italic>DAL2</italic></td><td align=\"left\">3.0</td><td align=\"left\">3.0</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Respiratory chain</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YPL078C</italic></td><td align=\"left\"><italic>ATP4</italic></td><td align=\"left\">2.1</td><td align=\"left\">2.0</td></tr><tr><td align=\"left\"><italic>YKL016C</italic></td><td align=\"left\"><italic>ATP7</italic></td><td align=\"left\">2.3</td><td align=\"left\">2.4</td></tr><tr><td align=\"left\"><italic>YPR020W</italic></td><td align=\"left\"><italic>ATP20</italic></td><td align=\"left\">2.5</td><td align=\"left\">3.4</td></tr><tr><td align=\"left\"><italic>YGR174C</italic></td><td align=\"left\"><italic>CBP4</italic></td><td align=\"left\">2.8</td><td align=\"left\">2.1</td></tr><tr><td align=\"left\"><italic>YBL045C</italic></td><td align=\"left\"><italic>COR1</italic></td><td align=\"left\">2.0</td><td align=\"left\">2.0</td></tr><tr><td align=\"left\"><italic>YGL187C</italic></td><td align=\"left\"><italic>COX4</italic></td><td align=\"left\">2.0</td><td align=\"left\">2.3</td></tr><tr><td align=\"left\"><italic>YNL052W</italic></td><td align=\"left\"><italic>COX5A</italic></td><td align=\"left\">2.0</td><td align=\"left\">2.1</td></tr><tr><td align=\"left\"><italic>YHR051W</italic></td><td align=\"left\"><italic>COX6</italic></td><td align=\"left\">2.1</td><td align=\"left\">2.2</td></tr><tr><td align=\"left\"><italic>YLL009C</italic></td><td align=\"left\"><italic>COX17</italic></td><td align=\"left\">2.8</td><td align=\"left\">4.7</td></tr><tr><td align=\"left\"><italic>YOR065W</italic></td><td align=\"left\"><italic>CYT1</italic></td><td align=\"left\">2.5</td><td align=\"left\">2.3</td></tr><tr><td align=\"left\"><italic>YKL087C</italic></td><td align=\"left\"><italic>CYT2</italic></td><td align=\"left\">2.2</td><td align=\"left\">2.8</td></tr><tr><td align=\"left\"><italic>YIL098C</italic></td><td align=\"left\"><italic>FMC1</italic></td><td align=\"left\">2.5</td><td align=\"left\">2.3</td></tr><tr><td align=\"left\"><italic>YDL181W</italic></td><td align=\"left\"><italic>INH1</italic></td><td align=\"left\">2.2</td><td align=\"left\">5.5</td></tr><tr><td align=\"left\"><italic>YBR185C</italic></td><td align=\"left\"><italic>MBA1</italic></td><td align=\"left\">2.3</td><td align=\"left\">2.3</td></tr><tr><td align=\"left\"><italic>YML120C</italic></td><td align=\"left\"><italic>NDI1</italic></td><td align=\"left\">4.1</td><td align=\"left\">4.4</td></tr><tr><td align=\"left\"><italic>YEL024W</italic></td><td align=\"left\"><italic>RIP1</italic></td><td align=\"left\">2.3</td><td align=\"left\">2.4</td></tr><tr><td align=\"left\"><italic>YPR151C</italic></td><td align=\"left\"><italic>SUE1</italic></td><td align=\"left\">2.2</td><td align=\"left\">3.0</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Plasma membrane transport</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YLL052C</italic></td><td align=\"left\"><italic>AQY2</italic></td><td align=\"left\">3.0</td><td align=\"left\">3.0</td></tr><tr><td align=\"left\"><italic>YLR411W</italic></td><td align=\"left\"><italic>CTR3</italic></td><td align=\"left\">3.0</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\"><italic>YGL255W</italic></td><td align=\"left\"><italic>ZRT1</italic></td><td align=\"left\">0.5</td><td align=\"left\">0.1</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Mitochondrial transport</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YNR002C</italic></td><td align=\"left\"><italic>ATO2</italic></td><td align=\"left\">3.8</td><td align=\"left\">3.8</td></tr><tr><td align=\"left\"><italic>YKL217W</italic></td><td align=\"left\"><italic>JEN1</italic></td><td align=\"left\">10.3</td><td align=\"left\">2.9</td></tr><tr><td align=\"left\"><italic>YPL134C</italic></td><td align=\"left\"><italic>ODC1</italic></td><td align=\"left\">6.3</td><td align=\"left\">3.3</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Transcriptional regulation</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YOL071W</italic></td><td align=\"left\"><italic>EMI5</italic></td><td align=\"left\">2.1</td><td align=\"left\">2.9</td></tr><tr><td align=\"left\"><italic>YNL333W</italic></td><td align=\"left\"><italic>SNZ2</italic></td><td align=\"left\">3.0</td><td align=\"left\">6.4</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Fatty acid metabolism</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YKL150W</italic></td><td align=\"left\"><italic>MCR1</italic></td><td align=\"left\">3.1</td><td align=\"left\">2.8</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Biosynthesis</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YFL030W</italic></td><td align=\"left\"><italic>AGX1</italic></td><td align=\"left\">3.6</td><td align=\"left\">7.8</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Ribosomal proteins in the mitochondria or cytosol</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YDR296W</italic></td><td align=\"left\"><italic>MHR1</italic></td><td align=\"left\">2.0</td><td align=\"left\">2.5</td></tr><tr><td align=\"left\"><italic>YGL068W</italic></td><td align=\"left\"><italic>MNP1</italic></td><td align=\"left\">2.1</td><td align=\"left\">2.4</td></tr><tr><td align=\"left\"><italic>YDR116C</italic></td><td align=\"left\"><italic>MRPL1</italic></td><td align=\"left\">2.6</td><td align=\"left\">3.5</td></tr><tr><td align=\"left\"><italic>YKL138C</italic></td><td align=\"left\"><italic>MRPL31</italic></td><td align=\"left\">2.2</td><td align=\"left\">2.8</td></tr><tr><td align=\"left\"><italic>YGR165W</italic></td><td align=\"left\"><italic>MRPS35</italic></td><td align=\"left\">2.0</td><td align=\"left\">2.7</td></tr><tr><td align=\"left\"><italic>YHR038W</italic></td><td align=\"left\"><italic>RRF1 (FIL1)</italic></td><td align=\"left\">2.2</td><td align=\"left\">4.8</td></tr><tr><td align=\"left\"><italic>YDR041W</italic></td><td align=\"left\"><italic>RSM10</italic></td><td align=\"left\">2.1</td><td align=\"left\">2.4</td></tr><tr><td align=\"left\"><italic>YDR175C</italic></td><td align=\"left\"><italic>RSM24</italic></td><td align=\"left\">2.4</td><td align=\"left\">2.9</td></tr><tr><td align=\"left\"><italic>YFR049W</italic></td><td align=\"left\"><italic>YMR31</italic></td><td align=\"left\">2.5</td><td align=\"left\">2.6</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Other</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YBR262C</italic></td><td align=\"left\"><italic>AIM5</italic></td><td align=\"left\">2.3</td><td align=\"left\">2.1</td></tr><tr><td align=\"left\"><italic>YFR011C</italic></td><td align=\"left\"><italic>AIM13</italic></td><td align=\"left\">2.1</td><td align=\"left\">3.4</td></tr><tr><td align=\"left\"><italic>YLR168C</italic></td><td align=\"left\"><italic>AIM30</italic></td><td align=\"left\">2.6</td><td align=\"left\">3.3</td></tr><tr><td align=\"left\"><italic>YML087C</italic></td><td align=\"left\"><italic>AIM33</italic></td><td align=\"left\">4.2</td><td align=\"left\">11.6</td></tr><tr><td align=\"left\"><italic>YBR230C</italic></td><td align=\"left\"><italic>OM14</italic></td><td align=\"left\">3.3</td><td align=\"left\">2.4</td></tr><tr><td align=\"left\"><italic>YDL104C</italic></td><td align=\"left\"><italic>QRI7</italic></td><td align=\"left\">2.1</td><td align=\"left\">2.7</td></tr><tr><td align=\"left\"><italic>YOR187W</italic></td><td align=\"left\"><italic>TUF1</italic></td><td align=\"left\">2.3</td><td align=\"left\">2.4</td></tr><tr><td align=\"left\"><italic>YKL067W</italic></td><td align=\"left\"><italic>YNK1</italic></td><td align=\"left\">3.1</td><td align=\"left\">2.1</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Not characterized</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YGR110W</italic></td><td align=\"left\">N/A</td><td align=\"left\">2.3</td><td align=\"left\">2.0</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Biologically-verified data on transcription factors that act on genes also found in Additional file ##SUPPL##0##1##: transcript dependence on Cat8 and Adr1.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>ORF</bold></td><td align=\"left\"><bold>GENE</bold></td><td align=\"right\"><bold>FACTOR CHANGE IN GENE EXPRESSION FROM V5 TO V5.TM6*P^a</bold></td><td align=\"right\"><bold>FACTOR CHANGE IN GENE EXPRESSION USING DATA FROM CHROMATIN IMMUNOPRECIPITATION FOR Cat8^b</bold>[##REF##15743812##25##]<bold>, mRNA EXPRESSION RATIOS FOR </bold><bold>WT</bold><bold>/Δcat8^c</bold>[##REF##11024040##16##]<bold> AND IMMUNOPRECIPITATION DATA FOR Adr1^d</bold>[##REF##15743812##25##].</td></tr></thead><tbody><tr><td align=\"left\" colspan=\"4\"><bold>Gluconeogenesis</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YLR377C</italic></td><td align=\"left\"><italic>FBP1</italic></td><td align=\"right\">5.3</td><td align=\"right\">5.0^b; MAX^c</td></tr><tr><td align=\"left\"><italic>YKR097W</italic></td><td align=\"left\"><italic>PCK1</italic></td><td align=\"right\">6.9</td><td align=\"right\">7.7^b; 151^c</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Alternative carbon source utilization</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YPL061W</italic></td><td align=\"left\"><italic>ALD6</italic></td><td align=\"right\">2.3</td><td align=\"right\">5.2^b; 8^c</td></tr><tr><td align=\"left\"><italic>YML054C</italic></td><td align=\"left\"><italic>CYB2</italic></td><td align=\"right\">10.1</td><td align=\"right\">7.4^d</td></tr><tr><td align=\"left\"><italic>YHL032C</italic></td><td align=\"left\"><italic>GUT1</italic></td><td align=\"right\">6.2</td><td align=\"right\">3.7^d</td></tr><tr><td align=\"left\"><italic>YIL155C</italic></td><td align=\"left\"><italic>GUT2</italic></td><td align=\"right\">3.3</td><td align=\"right\">4.6^d</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>TCA cycle</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YGR244C</italic></td><td align=\"left\"><italic>LSC2</italic></td><td align=\"right\">2.9</td><td align=\"right\">5.2^d</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Glyoxylate cycle</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YCR005C</italic></td><td align=\"left\"><italic>CIT2</italic></td><td align=\"right\">2.2</td><td align=\"right\">1.7^c</td></tr><tr><td align=\"left\"><italic>YLR174W</italic></td><td align=\"left\"><italic>IDP2</italic></td><td align=\"right\">4</td><td align=\"right\">2.9^b; 23^c</td></tr><tr><td align=\"left\"><italic>YOL126C</italic></td><td align=\"left\"><italic>MDH2</italic></td><td align=\"right\">3</td><td align=\"right\">12.7^b; 3.6^c</td></tr><tr><td align=\"left\"><italic>YNL117W</italic></td><td align=\"left\"><italic>MLS1</italic></td><td align=\"right\">3.5</td><td align=\"right\">8.9^b; 79^c</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Respiratory chain</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YPR151C</italic></td><td align=\"left\"><italic>SUE1</italic></td><td align=\"right\">3.7</td><td align=\"right\">4.4^d</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Plasma membrane transport</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YOR348C</italic></td><td align=\"left\"><italic>PUT4</italic></td><td align=\"right\">5.8</td><td align=\"right\">2.7^b; 6.1^c; 2.0^d</td></tr><tr><td align=\"left\"><italic>YDR536W</italic></td><td align=\"left\"><italic>STL1</italic></td><td align=\"right\">21.7</td><td align=\"right\">4.4^c</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Mitochondrial transport</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YNR002C</italic></td><td align=\"left\"><italic>ATO2</italic></td><td align=\"right\">3.8</td><td align=\"right\">0.33^c</td></tr><tr><td align=\"left\"><italic>YML042W</italic></td><td align=\"left\"><italic>CAT2</italic></td><td align=\"right\">2.6</td><td align=\"right\">2.8^b; 2.4^c</td></tr><tr><td align=\"left\"><italic>YOR100C</italic></td><td align=\"left\"><italic>CRC1</italic></td><td align=\"right\">2.2</td><td align=\"right\">8.1^c</td></tr><tr><td align=\"left\"><italic>YKL217W</italic></td><td align=\"left\"><italic>JEN1</italic></td><td align=\"right\">10.3</td><td align=\"right\">3.6^b; 10^c; 10.5^d</td></tr><tr><td align=\"left\"><italic>YPL134C</italic></td><td align=\"left\"><italic>ODC1</italic></td><td align=\"right\">6.3</td><td align=\"right\">3.2^b</td></tr><tr><td align=\"left\"><italic>YJR095W</italic></td><td align=\"left\"><italic>SFC1</italic></td><td align=\"right\">18.3</td><td align=\"right\">6.2^b; MAX^c</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Biologically-verified data on transcription factors that act on genes also found in Additional file ##SUPPL##0##1##: transcript dependence on Mig1.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>ORF</bold></td><td align=\"left\"><bold>GENE</bold></td><td align=\"right\"><bold>FACTOR CHANGE IN GENE EXPRESSION FROM V5 TO V5.TM6*P</bold>^a</td><td align=\"right\"><bold>Mig1 BINDING VERIFIED USING DATA FROM PROTEIN-BINDING MICROARRAY^b</bold>[##REF##15543148##27##]<bold> AND <italic>IN VIVO</italic>^c</bold>[##REF##9832517##26##]<bold> ANALYSES</bold></td></tr></thead><tbody><tr><td align=\"left\" colspan=\"4\"><bold>Glycolysis</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YFR053C</italic></td><td align=\"left\"><italic>HXK1</italic></td><td align=\"right\">9</td><td align=\"right\">b,c</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Gluconeogenesis</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YLR377C</italic></td><td align=\"left\"><italic>FBP1</italic></td><td align=\"right\">5.3</td><td align=\"right\">b</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Alternative carbon source utilization</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YHL032C</italic></td><td align=\"left\"><italic>GUT1</italic></td><td align=\"right\">6.2</td><td align=\"right\">b</td></tr><tr><td align=\"left\"><italic>YIL162W</italic></td><td align=\"left\"><italic>SUC2</italic></td><td align=\"right\">6.8</td><td align=\"right\">c</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Glyoxylate cycle</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YCR005C</italic></td><td align=\"left\"><italic>CIT2</italic></td><td align=\"right\">2.2</td><td align=\"right\">b</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Plasma membrane transport</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YEL069C</italic></td><td align=\"left\"><italic>HXT13</italic></td><td align=\"right\">3.9</td><td align=\"right\">b,c</td></tr><tr><td align=\"left\"><italic>YDL245C</italic></td><td align=\"left\"><italic>HXT15</italic></td><td align=\"right\">3.2</td><td align=\"right\">c</td></tr><tr><td align=\"left\"><italic>YBR298C</italic></td><td align=\"left\"><italic>MAL31</italic></td><td align=\"right\">3.1</td><td align=\"right\">b</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Mitochondrial transport</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YNR002C</italic></td><td align=\"left\"><italic>ATO2</italic></td><td align=\"right\">3.8</td><td align=\"right\">b</td></tr><tr><td align=\"left\"><italic>YKL217W</italic></td><td align=\"left\"><italic>JEN1</italic></td><td align=\"right\">10.3</td><td align=\"right\">b</td></tr><tr><td align=\"left\"><italic>YPL134C</italic></td><td align=\"left\"><italic>ODC1</italic></td><td align=\"right\">6.3</td><td align=\"right\">b</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Transcriptional regulation</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YDR516C</italic></td><td align=\"left\"><italic>EMI2</italic></td><td align=\"right\">3.6</td><td align=\"right\">b,c</td></tr><tr><td align=\"left\"><italic>YEL066W</italic></td><td align=\"left\"><italic>HPA3</italic></td><td align=\"right\">2.2</td><td align=\"right\">b</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\" colspan=\"4\"><bold>Other</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>YEL070W</italic></td><td align=\"left\"><italic>DSF1</italic></td><td align=\"right\">9.4</td><td align=\"right\">b,c</td></tr><tr><td align=\"left\"><italic>YGR243W</italic></td><td align=\"left\"><italic>FMP43</italic></td><td align=\"right\">4.9</td><td align=\"right\">b</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p><bold>Transcriptome analysis of V5.TM6*P compared with the parental V5 strain</bold>. Genes were tabulated if their expression was changed on going from the V5 parental strain to V5.TM6*P, as described in 'Materials and Methods'. All genes were then sorted according to their roles in yeast cellular physiology and alphabetically by gene name under each sub-heading. Note that <italic>HXT1-7 </italic>are deleted in the V5.TM6*P with a part of <italic>HXT1 </italic>and <italic>HXT7 </italic>reincorporated as the <italic>TM6</italic>* chimera. Due to overall high homology within the hexose transporter family we still observed changes on the cDNA array (<italic>HXT1 </italic>(<italic>YHR094C</italic>); factor 0.4; <italic><underline>HXT4 </underline></italic>(<italic>YHR092C</italic>); 0.2; <italic>HXT6 </italic>(<italic>YDR343C</italic>); 2.1; these are not listed in Additional file ##SUPPL##1##2##). As YLL053C is annotated as continuous with YLL052C (AQY2) in some strains, these ORFs are counted once here as YLL052C. Genes that were dubious were not included in Additional file ##SUPPL##0##1## and neither were the Ty-transposable elements, <italic>YGR161C-C </italic>(encoding TyA gag protein) and <italic>YNL054W-B </italic>(encoding TyB gag protein). The top 30 genes with the largest numerical factor change (both up- and down-regulated) are underlined. If no functional information was available, phenotypic data from a deletion mutant was entered under 'Function'. In the 'Factor Change' column, change is expressed as a factor, where that factor is x when a gene expressed with intensity '1' in V5 is expressed with intensity 'x' in V5.TM6*P. All T-tests were jointly adjusted for multiple testing using the false discovery rate method (disregarding correlations between genes): p-values were adjusted so that when selecting all genes with p-values less than a threshold q, a proportion of q false positives would be expected amongst these genes. The genes shown have p-values &lt; 0.05, and thus the expected false discovery rate is 5%. Transcription factors are listed if a binding site was predicted as described in the text. Data on 'Copies/cell' are from the yeast GFP fusion localization database and were collected from wild-type yeast grown on glucose. 'Protein Localisation' was extracted from the <italic>Saccharomyces </italic>Genome Database on 14^thFebruary 2008 and in each case has been manually curated by the site's curators unless the entry is underlined, indicating that the localization has been extracted from a genome wide study.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p>This table provides the information from Additional file ##SUPPL##0##1##, but in Excel format, with base expression values as well as SAGE [##REF##9008165##41##] and other literature data also included. In the 'Factor Change' column, change is expressed as a factor, where that factor is x when a gene expressed with intensity '1' in V5 is expressed with intensity 'x' in V5.TM6*P. In Additional file ##SUPPL##0##1## we have removed dubious genes and Ty-elements, but all genes that were originally spotted on the array are included in Additional file ##SUPPL##1##2##. As for Additional file ##SUPPL##0##1##, <italic>YLL052C </italic>and <italic>YLL053C </italic>are treated as one data point.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p>A further 636 genes changed their expression levels either up by a factor of &lt;2.0 (317 genes) or down by a factor &gt; 0.5 (319 genes) and are presented in Excel format.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S4\"><caption><title>Additional file 4</title><p>Binding sites are listed for the Hap complex, Cat8 and Mig1. The file lists exact position and sequences of the relevant sites.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S5\"><caption><title>Additional file 5</title><p>A computation in which the datasets with the two lowest glucose values have been removed (V5: 10.5, 7.4 g/L glucose and V5.TM6*P: 13.5, 5.6 g/L glucose). Genes are only listed if they are not present in Additional files ##SUPPL##0##1##, ##SUPPL##1##2##, or ##SUPPL##2##3##. Overall, the amplitudes of the factor changes varied by only 9.3% in this new calculation compared with the original one, which indicated that the glucose dependence of the V5 strain does not have a major influence on the way we generated the data presented in Additional file ##SUPPL##0##1##.</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>Yields were calculated as g product (dry weight, ethanol or glucose, respectively)/g of glucose consumed/catabolised for V5 and V5.TM6*P, cultured as described in 'Materials and Methods'. The numbers in parentheses are the standard error of the mean, together with the number of replicates (n) for each experiment. The seemingly low biomass yield for V5 was measured after consumption of all glucose. Yield increases as all ethanol and glycerol are consumed during subsequent respiratory growth.</p></table-wrap-foot>", "<table-wrap-foot><p>Genes were tabulated and sorted according to their roles in yeast cellular physiology, as described in the legend for Additional file ##SUPPL##0##1##. Note that the seven hexose transporter genes, <italic>HXT1</italic>-<italic>7</italic>, are deleted in the V5.TM6*P with the amino-terminal half of <italic>HXT1 </italic>and the carboxy-terminal half of <italic>HXT7 </italic>reincorporated as the <italic>TM6* </italic>chimera. In the 'Factor Change' column, change is expressed as a factor, where that factor is x when a gene expressed with intensity '1' in V5 is expressed with intensity 'x' in V5.TM6*P. All T-tests were jointly adjusted for multiple testing using the false discovery rate method (disregarding correlations between genes): p-values were adjusted so that when selecting all genes with p-values less than a threshold q, a proportion of q false positives would be expected amongst these genes. The genes shown have p-values &lt; 0.05, and thus the expected false discovery rate is 5%. The factor in the final column is taken from Lascaris <italic>et al </italic>[##REF##12537548##19##]. ^†Note that due to overall high homology within the hexose transporter family, changes were still observed on the cDNA array for <italic>HXT4 </italic>(<italic>YHR092C</italic>) even though it is deleted as already mentioned.</p></table-wrap-foot>", "<table-wrap-foot><p>The table is arranged as for Table 2. Factor changes denoted '^a' are from Additional file ##SUPPL##0##1##, '^b' refer to chromatin immunoprecipitation data for Cat8 [##REF##15743812##25##], '^c' refer to mRNA expression ratios for wt/<italic>Δcat8 </italic>[##REF##11024040##16##] and '^d' refer to chromatin immunoprecipitation data for Adr1 [##REF##15743812##25##].</p></table-wrap-foot>", "<table-wrap-foot><p>The table is arranged as for Table 2. Factor changes denoted '^a' are from Additional file ##SUPPL##0##1##. Mig1 binding was verified using data from protein-binding microarray [##REF##15543148##27##] studies denoted '^b'and <italic>in vivo </italic>[##REF##9832517##26##] analyses denoted '^c'although no factor changes were indicated in those studies.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2164-9-365-1\"/>", "<graphic xlink:href=\"1471-2164-9-365-2\"/>", "<graphic xlink:href=\"1471-2164-9-365-3\"/>", "<graphic xlink:href=\"1471-2164-9-365-4\"/>" ]

[ "<media xlink:href=\"1471-2164-9-365-S1.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-365-S2.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-365-S3.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-365-S4.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-365-S5.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Microarrays are most often developed from transcript information in the form of EST (Expressed Sequence Tags) sequences [e.g. [##REF##15453914##1##]]. The annotation of those sequences with information on genetics, homology, functions, metabolic regulations and toxicology, are key to unlocking the biological meaning of the microarray results. Since each single microarray hybridization experiment produces massive amounts of data, handling, processing and analyzing the data become challenging tasks and the application of bioinformatics is absolutely essential. It is desirable to store and organize the results in a database, which needs to be extensible and flexible in order to have the capabilities to compare data from different microarray experiments. The Stanford Microarray Database (SMD) [##REF##17182626##2##] is an example of such a resource, developed primarily with Stanford researchers and their collaborators in mind. Many other communities also develop resource databases, building the tools and functions to suit their organism, for example BarleyBase/PlexDB [##REF##18287702##3##] for Barley genomics, MELOGEN [##REF##17767721##4##] for melon genomics and the Tomato Expression Database (TED) [##REF##16381976##5##].</p>", "<p>A soybean gene expression database has been published, SGMD (the Soybean Genomics and Microarray Database – please see Availability and requirements for more information) [##REF##14681442##6##]. SGMD stores EST and microarray data to explore the interaction of soybean with the major pest soybean cyst nematode (SCN). The SGMD web interface provides on-the-fly statistics analysis to compare cDNA microarray data, which consists of around 4,000 spots and 20,000 EST sequences from the soybean root libraries [##REF##14681442##6##].</p>", "<p>We developed a new database, SoyXpress, with web tools to retrieve and explore the results of Affymetrix microarray experiments, linking also to other soybean genomic information in order to help researchers identify changes in gene expression and determine whether these changes alter biological processes in soybean. We designed SoyXpress for the potential of exploring the entire soybean transcriptome, integrating Affymetrix gene expression data (37,583 soybean probe sets) with 380,095 ESTs from <italic>G. max </italic>and <italic>G. soja</italic>, annotated with metabolic pathways, Gene Ontology terms, with SwissProt identifiers for maximum knowledge extraction. Currently, SoyXpress houses array data from 25 chips, comprising a leaf gene expression profiling experiment including two transgenic and three conventional (non-transgenic) soybean genotypes. SoyXpress is expansible and future gene expression experiments will be integrated.</p>" ]

[]

[]

[ "<title>Utility and Discussion</title>", "<p>We present SoyXpress, where we have integrated 380,095 soybean EST sequences and Affymetrix microarray data with functional annotations such as metabolic pathways and gene ontology. The SoyXpress user web interface was developed to access the database and display the data in a tabular format. Figure ##FIG##0##1## shows the Search Page to retrieve all available IDs and annotations for a soybean transcript or a group of transcripts that share similar protein name or function from our database. SoyXpress can be queried by EST ID, GenBank accession number, Affymetrix probe ID, SwissProt protein ID/name, EC enzyme number or GO term/number. A clickable GO tree that illustrates the hierarchy structure of the ontology is available to select a GO term for searching the associated IDs and annotation for the corresponding soybean sequences from the database. Figure ##FIG##2##3## shows the Search Result page for displaying all available IDs and annotations for the (EST) query IDs from the Search Page. After receiving the query ID, the corresponding EST sequence, Affymetrix probe sequence and DFCI/TIGR TC contig will be retrieved from our database. The BLASTX results for the EST and the Affymetrix probe sequences, such as the SwissProt protein IDs and descriptions, BLAST scores and evalues (represented by the negative exponent of the e-value), are displayed in the EST and AFFY tables. The associated GO numbers, GO terms and EC enzyme number are also displayed. The TC table displays the information for the DFCI/TIGR contig, such as TC ID, the IDs and GenBank accession numbers for the EST members of that contig, the associated GO number/term and EC enzyme number. To facilitate detailed database searches for the soybean search results, all these IDs are hyperlinked to the original public databases, such as GenBank (please see Availability and requirements for more information) for the sequence information, the SwissProt protein database (please see Availability and requirements for more information) for the protein information, the Gene Ontology (please see Availability and requirements for more information) for the functional annotations, the KEGG database (please see Availability and requirements for more information) for the biological pathway maps and enzyme information and the DFCI/TIGR Gene Index for the contig sequence and information (please see Availability and requirements for more information). The information on the cDNA libraries are presented on static html pages adapted from the former Soybean Genomics Initiative website (at the former Center for Computational Genomics and Bioinformatics, University of Minnesota).</p>", "<p>The results for the microarray experiment can be retrieved from the database through a special section of the interface. An overview of the query flow is presented in Figure ##FIG##3##4##. At the top SOY Microarray Analysis webpage, any of the five soybean cultivar samples can be selected for pair-wise comparison (Step 1). Diagrams to assess the quality of the data, such as boxplots of probe intensities, RNA degradation plots and the individual chip images are presented. After selecting two samples, the web page allows a choice of normalization method for pre-processing the raw data (Step 2). Diagrams such as boxplot, PCA analysis and hierarchical clustering are available to visualize the preprocessed data. After selecting the pre-processing method, the webpage allows selecting the cut-off p-value and fold change for differentially expressed genes from the results of the statistical analysis (Step 3). The list of differentially expressed genes from the pairwise comparison is displayed, ordered by probe IDs (Step 4). Statistical scores such as t-score, p-value and fold change are also displayed. A hyperlink is provided to display a plot of the intensities of an individual probe against five soybean cultivars. Check boxes are also available to submit a list of probe IDs to the Soybean Search Page to retrieve all the available IDs and annotations for those probes. It links to the annotation view by clicking on the Annotated Probe List button on the left panel. The annotation view (Step 5) displays the associated SwissProt protein ID/description and the GO number/term with the fold change and p-value for the list of differentially expressed genes. All these IDs are hyperlinked to the original public databases to facilitate detailed database searches. To retrieve results from the gene class analysis based on GO term annotations, similar query pages are developed (Figure ##FIG##3##4##). The list of GO terms (which represents changes of the gene class) is displayed in the result pages with the statistical scores and the number of the genes involved in each gene class. The intensities of the individual genes of each identified GO term can be displayed with log2 fold change, which can allow users to identify whether the genes were regulated in a similar pattern. A special page for BLAST analysis against the sequences from the Affymetrix probe set is also available in order for users to see whether their gene sequence of interest is present in the probe set.</p>", "<p>SoyXpress was developed with two main types of users in mind: researchers and regulators with scientific background. Researchers can explore the annotated sequences in a way that relates to their metabolic pathway or process under study. For instance, a researcher interested in the flavonoid pathway could instantly retrieve all ESTs known from soybean that match the enzymes in this pathway. Regulators are asking for tools to help in assessment of novel crops, be they transgenic or obtained by conventional breeding. With SoyXpress, we have provided such a tool, where differences in global gene expression can be compared between any cultivars or groups of cultivars and where the gene expression is linked to metabolic pathways and to literature resources such as PubMed and TOXLINE. This can help regulators decide on whether a novel cultivar is, at the gene expression level, substantially equivalent to conventional cultivars that are generally recognized as safe (GRAS).</p>", "<p>There is no other web-based database that makes the soybean transcriptome available. The SGMD is limited to 4000 genes expressed in root and has the aim to explore the Soybean Cyst Nematode and soybean interactions [##REF##14681442##6##]; there are only GenBank IDs and BLASTX reports to show the homology of genes and proteins, and no annotations are provided to give information of the biological function and metabolic pathways. The draft of the soybean genome sequence was announced in January 2008 (please see Availability and requirements for more information), and it is envisioned that SoyXpress can be a helpful tool for the soybean genome annotation phase. Predicted genes can be compared with the information in SoyXpress and more reliable annotation can follow.</p>", "<p>Planned future developments of SoyXpress include addition of promoter motif information, UTR features and links to the soybean genome sequence. It is also our hope that other groups will want to house their Affymetrix soybean data in SoyXpress, and an online submission protocol is planned.</p>" ]

[ "<title>Conclusion</title>", "<p>Our scope was to develop a database with a suite of web interfaces to allow users to easily retrieve data and results of microarray experiments with cross-referenced annotations of the expressed sequence tags (EST) and hyperlinks to external public databases. The SoyXpress environment is the most comprehensive bioinformatics tool to date for soybean gene expression analysis and it makes it possible to explore differences in gene expression and to interpret the results based on gene functional annotations to determine any changes that could alter biological processes.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Experiments using whole transcriptome microarrays produce massive amounts of data. To gain a comprehensive understanding of this gene expression data it needs to be integrated with other available information such as gene function and metabolic pathways. Bioinformatics tools are essential to handle, organize and interpret the results. To date, no database provides whole transcriptome analysis capabilities integrated with terms describing biological functions for soybean (<italic>Glycine max </italic>(L) Merr.). To this end we have developed SoyXpress, a relational database with a suite of web interfaces to allow users to easily retrieve data and results of the microarray experiment with cross-referenced annotations of expressed sequence tags (EST) and hyperlinks to external public databases. This environment makes it possible to explore differences in gene expression, if any, between for instance transgenic and non-transgenic soybean cultivars and to interpret the results based on gene functional annotations to determine any changes that could potentially alter biological processes.</p>", "<title>Results</title>", "<p>SoyXpress is a database designed for exploring the soybean transcriptome. Currently SoyXpress houses 380,095 soybean Expressed Sequence Tags (EST), linked with metabolic pathways, Gene Ontology terms, SwissProt identifiers and Affymetrix gene expression data. Array data is presently available from an experiment profiling global gene expression of three conventional and two genetically engineered soybean cultivars. The microarray data is linked with the sequence data, for maximum knowledge extraction. SoyXpress is implemented in MySQL and uses a Perl CGI interface.</p>", "<title>Conclusion</title>", "<p>SoyXpress is designed for the purpose of exploring potential transcriptome differences in different plant genotypes, including genetically modified crops. Soybean EST sequences, microarray and pathway data as well as searchable and browsable gene ontology are integrated and presented. SoyXpress is publicly accessible at <ext-link ext-link-type=\"uri\" xlink:href=\"http://soyxpress.agrenv.mcgill.ca\"/>.</p>" ]

[ "<title>Construction and Content</title>", "<title>Schema and implementation: Sequence core tables</title>", "<p>SoyXpress (Figure ##FIG##0##1##) is implemented in MySQL (version 5.0.18 – please see Availability and requirements for more information). Perl (version 5.8.6) and Perl CGI (please see Availability and requirements for more information) scripts were written for data file parsing, database loading and to create web-interfaces connected to the database using the perl modules DBI and DBD::mysql (please see Availability and requirements for more information). The CGIwithR package [##UREF##0##7##] was used in order to run R within CGI. The sequence core tables are adapted for MySQL (please see Availability and requirements for more information) from the Oracle-based open source ESTIMA (Expressed Sequence Tag Information Management and Annotation project) database [##REF##14706121##8##]. Figure ##FIG##1##2A## shows the tables that organize the sequence information. The table DNA_SEQUENCE specifies the sequence ID, length and location (file path) where the sequence is stored in our file system (adapted from the BioData system at the former Center for Computational Genomics and Bioinformatics, University of Minnesota). The ancillary information about the EST sequences such as the locations of the clone vector, polyA-tail, repeat sequence and the trim site are stored in the tables VECTOR, TAIL, REPEATS and TRIM, respectively. The cDNA library information including the library ID, tissue type, and growing conditions are stored in the table LIBRARY. The library information is linked to the sequence information through table SEQ_ACCESSION, which maps the sequence ID, library ID and GenBank accession number.</p>", "<title>Schema and implementation: Sequence annotation tables</title>", "<p>Figure ##FIG##1##2B## shows the annotation section of the database. The SEQ_ACCESSION table links to the BLAST table by using the sequence ID as the query ID for linking to the BLASTX search results. The GenBank accession number from the SEQ_ACCESSION table links to the DFCI (Dana Farber Cancer Institute) Gene Index (formerly TIGR Gene Index [##REF##10592205##9##]) contig information to obtain the corresponding contig ID for the EST sequences (from table TIGR_GB) and the GO terms associated with each contig (from the table TIGR_GO) (Figure ##FIG##1##2A##). Other information for the additional 8,936 EST sequences downloaded from NCBI websites are stored in the table GB_ACCESSION, which also links to the BLAST table using the GenBank accession number as the query ID. The Gene Ontology databases [##REF##10802651##10##] include the MySQL tables: TERM, TERM_DEFINITION, TERM2TERM, a n d GRAPH_PATH (please see Availability and requirements for more information), which were downloaded and directly reproduced in our database. The BLASTX analysis [##REF##2231712##11##] against SwissProt [##REF##12824418##12##] allowed us to assign protein annotations to 175,910 ESTs (over half of the 318,422 EST sequences). The BLAST table contains the BLASTX search results and links our EST data to their corresponding protein information. Of the 37,637 soybean probe sequences on the Affymetrix GeneChip, we assigned protein annotations to 8,667 sequences. These BLASTX search results are also incorporated into the BLAST table and link to other protein and function annotations. The SwissProt protein names are stored as the hit IDs. Other information about the proteins such as the protein descriptions, hit scores and e-values (negative exponents) are also stored in the BLAST table. The SwissProt protein IDs link to other functional annotations such as gene ontology (GO terms) and KEGG molecular pathways [##REF##16381885##13##] through the GENE_ANNOTATION and EC_SWISS tables. The protein descriptions that describe the enzymes with appropriate EC (enzyme commission) numbers are linked to the KEGG pathways (stored as tables EC_DEF, and EC_MAP) through EC_SWISS table. There are 73,996 EST sequences with assigned EC numbers, around 23% of the EST sequences were enzymes. By linking the transcript sequences data to SwissProt annotations through BLASTX search results in the BLAST table, we can map the transcript sequences to their corresponding functional annotations such as GO terms and KEGG molecular pathways providing a more comprehensive description of the soybean data.</p>", "<title>Schema and implementation: soybean microarray experiment data tables</title>", "<p>The section of the database that organizes the microarray data is shown in Figure ##FIG##1##2C##. Data for the Affymetrix Soybean GeneChip [##UREF##1##14##], for example the probe IDs, the sequences of the probes, and the locations of the probes on the chip are stored in the table CDF_FILE. The whole transcript sequences representing the genes with the corresponding probe IDs and GenBank accession number are stored in the table PROBE_SEQ. The PROBE_SET table contains the probe IDs, GenBank accession number, and the corresponding sequence and clone IDs to map to our soybean EST data, and hence associates the microarray data with corresponding transcript, protein and functional annotations. Also, the microarray data can directly link to the BLAST table by using probe ID as the query ID to provide biological context for our microarray experiment. The raw data for our microarray experiment are stored in the table CEL_DATA, which contains the information for every chip, such as the chip IDs, probe IDs, and probe intensity. The processed data for our microarray experiment using three normalization methods RMA, MAS, dCHIP are stored in three tables RMA_RESULT, MAS_RESULT and LIWONG_RESULT respectively. All the raw and processed microarray data is linked to the PROBE_SET table by the probe IDs. For the analyzed results, the EXPERIMENT table describes which chips are used for the pair-wise comparison. The NORMALIZE table describes which normalization method are used in each pair-wise comparison. The microarray results for each pair-wise comparison analyzed by the LIMMA package are stored in the LIMMA_RESULT table. It includes the scores and p-value from the statistical test for each probe in all pair-wise comparisons. Also, the fold change and average intensity for each probe in all pair-wise comparisons are stored in the table FOLD_CHANGE. All the analyzed microarray results are linked to the PROBE_SET table and hence integrated with the soybean transcript, protein and functional annotations that can provide insight into biological and functional differences between samples.</p>", "<title>Sequence and microarray data sources</title>", "<p>The sequence data annotated and stored in SoyXpress comprises a total of 380,095 public ESTs from <italic>G. max </italic>and <italic>G. soja</italic>. Information on 31,928 tentative consensus (TC) sequences was downloaded from The TIGR (The Institute for Genomic Research) Glycine max Gene Index Project (Release 12.0) [##REF##10592205##9##] (now hosted at the Dana Farber Cancer Institute/Computational Biology and Functional Genomics Laboratory at Harvard University – please see Availability and requirements for more information). The microarray data currently available consists of twenty-five raw data files (CEL files) of an experiment using the Affymetrix Soybean GeneChip [##UREF##1##14##]. These were pre-processed and analyzed by standard methods as previously described [##UREF##2##15##]). The data consists of five biological replicates of leaf gene expression measure of three conventional and two genetically engineered soybean lines. The microarray data is accessible at NCBI (Gene Expression Omnibus) GEO under the accession numbers: GSE9374: GSM238030, GSM238031, GSM238032, GSM238033, GSM238034, GSM238036, GSM238038, GSM238039, GSM238041, GSM238043, GSM238047, GSM238048, GSM238049, GSM238050, GSM238051, GSM238052, GSM238053, GSM238054, GSM238055, GSM238056, GSM238057, GSM238058, GSM238059, GSM238060, GSM238061. Microarray chip information (from Affymetrix), raw data and results are stored in SoyXpress, and each probe is linked to the sequence information and meta-data.</p>", "<title>Informatics of data generation and quality control</title>", "<p>The EST sequences were annotated by command line BLASTX [##REF##2231712##11##] searches against 168,297 SwissProt protein sequences (please see Availability and requirements for more information) [##REF##12824418##12##] to obtain corresponding protein annotations. The SwissProt protein IDs were used to associate the sequences with GO terms, using the file: \"UniProt GO Annotations\" (please see Availability and requirements for more information). Recommended enzyme names and EC numbers were obtained from the Enzyme Nomenclature site (please see Availability and requirements for more information) and also extracted from MeSH (Medical Subject Headings, National Library of Medicine – please see Availability and requirements for more information). Enzyme EC numbers to SwissProt ID associations were obtained from the ExPASy Enzyme nomenclature database (version 36, please see Availability and requirements for more information). Metabolic and regulatory pathways were downloaded from KEGG (Kyoto Encyclopedia of Genes and Genomes – please see Availability and requirements for more information [##REF##16381885##13##]). Enzyme identities within each pathway were obtained by extracting EC numbers from each of the pathways (downloadable XML files from the ftp KGML/map folders, version 0.6 Mar 2005). EC numbers, pathway names and map numbers where extracted and integrated into the database.</p>", "<title>Availability and requirements</title>", "<p>Project name: SoyXpress: a database for the soybean transcriptome</p>", "<p>Project home page: <ext-link ext-link-type=\"uri\" xlink:href=\"http://soyxpress.agrenv.mcgill.ca/\"/>.</p>", "<p>Operating system: Platform independent</p>", "<p>Programming language: Perl</p>", "<p>Other requirements: None</p>", "<p>Licence: None required</p>", "<p>Any restrictions to use by non-academics: None</p>", "<p>SGMD (the Soybean Genomics and Microarray Database): <ext-link ext-link-type=\"uri\" xlink:href=\"http://psi081.ba.ars.usda.gov/SGMD/Default.htm\"/></p>", "<p>MySQL (version 5.0.18): <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.mysql.com\"/></p>", "<p>Perl (version 5.8.6): <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.perl.com\"/></p>", "<p>Perl CGI: <ext-link ext-link-type=\"uri\" xlink:href=\"http://search.cpan.org/dist/CGI.pm/\"/></p>", "<p>DBI and DBD::mysql: <ext-link ext-link-type=\"uri\" xlink:href=\"http://dev.mysql.com/downloads/dbi.html\"/></p>", "<p>MySQL: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.mysql.com\"/></p>", "<p>Gene Ontology databases: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.geneontology.org/GO.downloads.database.shtml\"/></p>", "<p>Glycine max Gene Index Project: <ext-link ext-link-type=\"uri\" xlink:href=\"http://compbio.dfci.harvard.edu/tgi/\"/></p>", "<p>SwissProt: <ext-link ext-link-type=\"uri\" xlink:href=\"http://ca.expasy.org/sprot/\"/></p>", "<p>UniProt GO Annotations: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.geneontology.org/GO.current.annotations.shtml\"/></p>", "<p>Enzyme Nomenclature site: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.chem.qmul.ac.uk/iubmb/enzyme/\"/></p>", "<p>MeSH (Medical Subject Headings, National Library of Medicine): <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.nlm.nih.gov/mesh/filelist.html\"/></p>", "<p>ExPASy Enzyme nomenclature database (version 36): <ext-link ext-link-type=\"uri\" xlink:href=\"http://ca.expasy.org/enzyme/\"/></p>", "<p>KEGG (Kyoto Encyclopedia of Genes and Genomes): <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.genome.jp/kegg/download/ftp.html\"/></p>", "<p>GenBank: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov\"/></p>", "<p>SwissProt protein database: <ext-link ext-link-type=\"uri\" xlink:href=\"http://ca.expasy.org\"/></p>", "<p>Gene Ontology: <ext-link ext-link-type=\"uri\" xlink:href=\"http://amigo.geneontology.org\"/></p>", "<p>KEGG database: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.genome.jp\"/></p>", "<p>DFCI/TIGR Gene Index: <ext-link ext-link-type=\"uri\" xlink:href=\"http://compbio.dfci.harvard.edu\"/></p>", "<p>Draft of the soybean genome sequence: <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.phytozome.org/soybean\"/></p>", "<title>Authors' contributions</title>", "<p>KCCC designed the database, performed the data analyses and web implementation. MVS conceived and designed the overall project, and assisted in the database design. Both authors have participated in writing the manuscript and have read and approved the final submitted version.</p>" ]

[ "<title>Acknowledgements</title>", "<p>The authors wish to thank Lee Zamparo, Julie Livingstone, Ernest Retzel and Frederic Latour for bioinformatics assistance. We are also grateful to Lee Zamparo for critical reading of the manuscript. This project was funded by the Advanced Foods and Materials Network (AFMNet). We also acknowledge the Natural Sciences and Engineering Research Council of Canada (NSERC), le Fonds de recherche sur la nature et les technologies (FQRNT) and the Centre Sève for financial support.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>The main page of the soy database</bold>. Users can submit queries to the database to retrieve all available IDs and annotations for the soybean transcript of interest. Queries can be made using EST ID, GenBank accession number, GenBank GI number, Affymetrix probe Set ID, SwissProt protein ID, name or keyword, GO number or term. A clickable GO tree is available to assist searching for a GO term from the gene ontology hierarchy structure.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Database structure</bold>. A) Gene transcript (sequence) information section. Tables for EST sequences (DNA_SEQUENCE, LIBRARY, REPEATS, SEQ_ACCESSION, TAIL, TRIM, VECTOR), a table for mRNA sequences downloaded from GenBank (GB_ACCESSION) and tables for DFCI/TIGR contigs data (TIGR_GB, TIGR_GO) link through the tables GB_ACCESSION and SEQ_ACCESSION to annotation and microarray data. B) Protein and functional annotations section. A table, BLAST, for BLASTX search results, tables for gene ontology terms information (GENE_ANNOTATION, GRAPH_PATH, TERM, TERM2TERM, TERM_DEFINITION) and tables for KEGG pathways with the enzyme commission numbers (EC_DEF, EC_SWISS, EC_MAP) organize the annotation section. The BLAST table links protein annotation data to transcript sequence information and to microarray experiments data. C) Microarray experiment data section. Tables for chip information (CDF_FILE, PROBE_SET), a table for raw data (CEL_DATA), tables for normalized data (LIWONG_RESULT, MAS_RESULT, RMA_RESULT), and tables for analyzed results (EXPERIMENT, FOLD_CHANGE, LIMA_RESULT, NORMALIZE) organize the microarray data section. The PROBE_SET table links the microarray data to transcript sequence information and protein annotations data.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Database result page, showing information about an EST sequence sag44h11.y1</bold>. Data records are presented in three tables: a) EST table displays the BLASTX result with scores and the negative exponent for the e-value for the EST sequence. b) AFFY table displays the BLASTX result with scores and the negative exponent for the e-value for the Affymetrix probe sequence. c) TC table displays the information for the corresponding DFCI/TIGR contig. The SwissProt protein ID/description, GO number/terms and EC enzyme numbers are displayed with the corresponding EST ID, GenBank accession number or Affymetrix probe ID with hyperlinks to the original public database. If available, links to TOXLINE are also given.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Flowchart of the microarray web interface to access the database for displaying differentially expressed genes or functional gene classes based on GO terms</bold>. A) The query is built in three steps. B) There are four different views for the query results: by probe or functional level analysis either displayed as a list or as more informative annotated list.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2164-9-368-1\"/>", "<graphic xlink:href=\"1471-2164-9-368-2\"/>", "<graphic xlink:href=\"1471-2164-9-368-3\"/>", "<graphic xlink:href=\"1471-2164-9-368-4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Lysophosphatidic acid (LPA; monoacyl-glycerol-3-phoshate) is a lipid mediator that stimulates the proliferation, migration and survival of many cell types [##REF##15273989##1##]. LPA acts through at least five distinct G protein-coupled receptors (GPCRs), termed LPA_1–5, which show overlapping signaling properties and tissue distribution [##REF##15189145##2##,##REF##17459484##3##]. LPA signaling has been implicated in a great variety of biological processes, ranging from embryonic development to wound healing and tumor progression [##REF##15273989##1##, ####REF##15189145##2##, ##REF##17459484##3##, ##REF##16406843##4##, ##REF##12894246##5####12894246##5##]. This multitude of activities is consistent with the broad distribution of LPA receptors and their coupling to multiple G proteins. LPA is produced from more complex lysophospholipids by a secreted lysophospholipase D known as autotaxin (ATX), originally identified as an 'autocrine motility factor' for tumor cells (for review see ref. [##REF##17459484##3##]). ATX is essential for vascular development [##REF##16782887##6##,##REF##16829511##7##] and in addition promotes tumor aggressiveness and angiogenesis [##REF##11559573##8##]. This strongly suggests that LPA is a key pro-angiogenic factor during development and a signifcant effector of cancer progression in the stroma-tumor microenviroment.</p>", "<p>Although LPA-induced signaling pathways and cellular responses have been extensively analyzed over the years [##REF##15273989##1##], it remains unknown how LPA affects global gene expression in its target cells. Gene expression analysis may uncover previously unknown activities of LPA, lead to a better understanding of GPCR signaling in general, and help to predict the behavior of cells in an LPA-enriched environment.</p>", "<p>In this report, we analyze the global transcriptional response to LPA in mouse embryo fibroblasts (MEFs). Fibroblasts are abundant mesenchymal cells in the stroma of many tissues and organs where they regulate epithelial-mesenchymal interactions during development, tissue regeneration and tumor progression [##REF##15549095##9##,##REF##16572188##10##]. LPA has long been known to stimulate the proliferation and migration of fibroblasts [##REF##15273989##1##,##REF##2551506##11##], while excessive LPA signaling in these cells can lead to fibrosis [##REF##18066075##12##]. The present study identifies many novel LPA-regulated genes and shows especially that LPA commits fibroblasts, at the transcriptional level, to create a microenvironment that supports tissue remodeling, leukocyte recruitment, angiogenesis and tumor progression. Since fibroblasts are also responsive to epidermal growth factor (EGF), acting on its cognate receptor tyrosine kinase, we examined in addition to what extent the LPA- and EGF-induced expression patterns overlap.</p>" ]

[ "<title>Methods</title>", "<title>Cell culture</title>", "<p>Mouse embryonic fibroblasts (MEFs) were immortalized at passage 2 by retroviral introduction of the T-box member Tbx2 (LZRS-TBX2-ires-EGFP) to represses p53 function [##REF##11062467##39##]. MEFs were cultured in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 8% fetal calf serum (FCS), penicillin and streptavidin, and seeded every three days at a density of 1.6 × 10⁴/cm²according to the 3T3 protocol.</p>", "<title>RNA isolation and amplification</title>", "<p>Detailed protocols for RNA isolation, amplification, labeling and microarray hybridization can be found at <ext-link ext-link-type=\"uri\" xlink:href=\"http://microarray.nki.nl/download/protocols.html\"/>. MEFs were seeded at density of 2.5 × 10⁴/cm², and 24 h later were starved for 40 h in serum-free medium supplemented with 0.2% bovine serum albumin (BSA; Sigma). Cells were treated with oleoyl LPA (0.3–5 uM) coupled to fatty acid free BSA (Sigma) at a 3:1 molar ratio, or EGF (20 ng/ml) for the indicated time period. Before harvesting, cells were washed twice with ice-cold phosphate buffered saline. Isolation of total RNA was performed with RNAzol Bee (Campro Scientific, Amersfoort, the Netherlands). Isolated total RNA was subsequently Dnase1-treated by using the Qiagen RNase-free DNase kit (Cat. no. 74106) and RNeasy spin columns (Qiagen, West Sussex, UK, Cat. no. 79254) and dissolved in RNase-free H₂O. RNA concentration and purity was measured on a NanoDrop ND-1000 spectrophotometer (Isogen Life-Science), while RNA integrity was determined by agarose gel electrophoresis. Four micrograms of total RNA was used to synthesise cDNA with a T7-(dT)24 primer and RT superscript III (Invitrogen Life Technologies; Cat. no. L1016-01). After second-strand synthesis and removal of contaminating RNA, cRNA was generated by <italic>in vitro </italic>transcription using T7 RNA polymerase. Amplification yields were 1,000- to 2,000-fold.</p>", "<title>Microarray processing and data analysis</title>", "<p>The cRNA of each sample was labeled with Cy5 or Cy3 (Universal Linkage System, Kreatech Biotechnology, Cat. no. EA-006) according the manufacturer's protocol, with minor adjustments. For each labeling, 0.3 μl of the Cy5-ULS or 1 μl Cy3-ULS was mixed with 1 μg of cRNA and 2 μl of labeling solution. The volume of this mix was adjusted to 20 μl and incubated for 30 minutes at 85°C, followed by purification with a KREA <italic>pure </italic>column. Dye incorporation was determined after measuring the labeled cRNA on a NanoDrop ND-1000 spectrophotometer and mixed with the same amount of reverse color Cy-labeled cRNA from the untreated control. Labeled cRNAs were fragmented to pieces of 60–200 nt (Ambion, Cat. no. AM8740). Before hybridization, 20 μg of COT-1 DNA (Invitrogen-Life Technologies, Cat. no. 15279-011), 8 μg of yeast tRNA (Roche Diagnostics B.V., Cat. no. 10109495001) and 20 μg of poly-d(A) (GE Healthcare Biosciences Europe GmbH, Cat. no. 27-7836-03) was added and the volume was adjusted to 60 μl. To this volume 60 μl of a formamide buffer containing 50% Formamide, 50% 20 × SSC (NaCl/Na-citrate) and 0.1% SDS was added. All hybridizations were performed in a hybridization station (Tecan, Cat. no. Hs4800). Before hybridization with the labeled RNA samples, the microarray was pre-hybridized with a bovine serum albumin solution (1% BSA, 5 × SSC and 0.1% SDS) for one hour at 42°C and washed with water and a 5 × SSC, 0,1% SDS solution. The labeled material was denatured at 95°C for 3 minutes and cooled to 42°C before injection in the hybridization chamber. After hybridization for 16 hrs, slides were washed in the hybridization station with a 5 × SSC, 0,1% SDS solution, a 2 × SSC, 0,1% SDS solution and a 1 × SSC solution at 42°C and a 0,2 × SSC solution at 23°C. Finally, the slides were dried with medical grade nitrogen for 3 min. at 30°C. A DNA Microarray scanner (Agilent Technologies, Cat. no. G2505B) was used to scan the slides. To monitor the consistency of the array experiments, \"self-self\" experiments were performed using the same sample as reference. Fluorescent intensities of the images were quantified by using ImaGene v6.0 software (Biodiscovery Inc.). This software has an output of two text files which were uploaded to the CMF database (CMFdb, <ext-link ext-link-type=\"uri\" xlink:href=\"http://cmfdb.nki.nl\"/>) for further analysis. The background-corrected intensities from the Cy5 and Cy3 channel were used to calculate log₂transformed ratios. These ratios were normalized using a lowest fit per subarray [##REF##11842121##40##]. Experiments done in dye-swap fashion were combined to create one dataset on which an outlier analysis was performed. A weighted average ratio and confidence level (P-value) was calculated per gene by a NKI platform adjusted error model [##REF##10929718##41##], which was fine-tuned by self-self hybridizations. Differentially expressed genes between sample and reference were selected based on their P-value (a gene with a P-value &lt; 0.01 is considered an outlier). Genes were selected for further analysis if they had p-values &lt; 0.01 and 2log ratios greater than 0.67 or smaller than -0.67 corresponding to a 1.7-fold change, unless indicated otherwise. To identify structural patterns of gene expression, selected outliers were used for hierarchical clustering using the complete linkage algorithm and for K-means clustering analysis (Euclidian distance) using the GENESIS program [##REF##11836235##42##]. To identify genes that differ between LPA- and EGF induced treatment Anova analysis was applied (two groups) in which the expression values of two consecutive time points were used. \"Ingenuity\" pathway analysis was used for functional analysis on the set of significantly modulated genes to identify affected biological pathways and functional processes. In addition, the web-based platform Gene Ontology Tree Machine (GOTM) was used to identify GO terms with relatively enriched gene numbers [##REF##14975175##43##].</p>", "<title>Microarray slides</title>", "<p>Mouse 32 k Operon v3.0 oligo arrays from the Central Microarray Facility (CMF) at the Netherlands Cancer Institute were used for hybridization. A complete list of genes and controls present on the slides is available on the CMF web site <ext-link ext-link-type=\"uri\" xlink:href=\"http://microarrays.nki.nl/download/geneid.html\"/>. The 70-mer oligo's (Operon, AROS v3.0) were printed on UltraGaps slides (Corning) with a BioRobotics MicroGrid II (Genomic Solutions) print robot. The description of our microarray study follows the MIAME guidelines and the entire microarray data set has been deposited in the EBI/ArrayExpress database and is accessible through accession numbers E-NCMF-16, E-NCMF-17, E-NCMF-18 and E-NCMF-19).</p>", "<title>Quantitative real-time RT-PCR</title>", "<p>Total RNA was extracted using the RNeasy minikit combined with on-column Dnase-I treatment (Qiagen, West Sussex, UK) and resolved in diethyl pyrocarbonate-treated H₂O (DEPC; Sigma). First-strand cDNA synthesis was done using 2 ug of total RNA in presence of 0.5 ug oligo-dT primers (Invitrogen), 40 units RNase inhibitor Rnasin (Promega), 500 nM deoxynucelotide triphosphates (Roche), 10 uM DTT and 10 units Superscript II RT in 1× reverse transcriptase buffer (Invitrogen). Sequences of real-time quantitative PCR primers were designed using Primer express software (PE Biosystems, Foster City, CA). Primer sequences are available upon request. Detection and quantification of each gene was accomplished by SYBRgreen incorporation using the ABI PRISM 7700 sequence detection system (Applied Biosystems). Quantitative RT-PCR was carried out using 40 ng cDNA, 300 nM of each oligo in presence of 1× SYBRgreen mix in 20 ul reactions (Applied Biosystems). Cycling parameters were: 2 min incubation at 50°C, 10 min. incubation at 95°C, followed by 50 PCR cycles consisting of 15 sec at 95°C and 1 min. at 60°C. Product sizes were routinely verified by collecting a melting curve from 55°C to 95°C after final amplification. The relative product levels were quantified using the 2^-ΔΔCTmethod. Data are presented as relative induction of each target gene, normalized to the expression of HPRT, and are representative of two independent experiments.</p>" ]

[ "<title>Results and Discussion</title>", "<title>Characterization of MEFs</title>", "<p>We examined the mitogenic response and LPA receptor expression profile of immortalized mouse embryonic fibroblasts (MEFs). Stimulation of serum-deprived, nearly confluent MEFs with saturating doses of LPA (5 μM) resulted in a significant increase in DNA synthesis to about 40% the level induced by 10% serum. EGF (50 ng/ml) was somewhat less efficacious than LPA in stimulating DNA synthesis (Figure ##FIG##0##1A##). LPA signals via at least five distinct GPCRs (termed LPA_1–5) that couple to multiple G proteins, including G_i/o, G_q/11and G_12/13. Quantitative PCR analysis showed that our MEFs co-expressed LPA₁, LPA₂and LPA₄, with LPA₁and LPA₄being the predominant transcripts (Figure ##FIG##0##1B##). Since MEFs do not express ATX (encoded by <italic>Enpp2</italic>; data not shown), autocrine LPA signaling is not operative in our cell system.</p>", "<title>The global transcriptional response to LPA</title>", "<p>We examined the temporal program of gene expression in serum-starved MEFs treated with LPA (5 μM, i.e. about the normal concentration in serum [##REF##8489494##13##]). Total RNA was isolated at different time points after LPA stimulation (0–24 hrs). Global transcription profiles were determined using oligonucleotide microarrays containing 31,770 mouse transcripts. Amplified RNA of the treated samples was matched with the untreated control and hybridized in duplicate with reversal of the Cy3 and Cy5 dyes; the normalized Cy5/Cy3 ratios were combined and used for further analysis. We selected genes that were significantly regulated (p &lt; 0.01) at two or more consecutive time points or in replicate measurements, which yielded 1508 LPA-regulated genes (see additional file ##SUPPL##0##1##: complete dataset). The entire dataset has been deposited in the EBI/ArrayExpress database (see Methods). We restricted our data set to genes that were induced by &gt;1.7-fold at two or more time points and grouped them according to the temporal profile of gene induction using K-means clustering. This resulted in ten clusters, each containing genes that show similarly shaped waves of transcription (Figure ##FIG##1##2##). Seven clusters contained genes that were upregulated by LPA (424 transcripts), whereas three other clusters mainly comprised the down-regulated genes (209 transcripts) (Figure ##FIG##1##2##; for details see additional files ##SUPPL##1##2## and ##SUPPL##2##3##). The genes that were most strongly regulated at different time points are listed in additional file ##SUPPL##3##4## (Table 1: upregulated genes; Table 2: downregulated genes). The microarray results were validated by examining the expression of 22 representative genes from different clusters using real-time PCR (Figure ##FIG##2##3##). The -fold inductions of expression generally were higher in the qPCR assays than in the corresponding microarray experiments, reflecting the different sensitivities of both methods (see additional file ##SUPPL##4##5##: correlation plot of all data points). Gene ontology analysis revealed that LPA regulated the expression of genes in multiple functional categories that mostly corresponded to the different gene clusters (see additional file ##SUPPL##5##6##).</p>", "<title>Regulation of genes associated with growth regulation and cytoskeletal reorganization</title>", "<title>Upregulated genes</title>", "<p>The immediate transcriptional response to LPA was characterized by &gt;100 upregulated immediate-early genes (mRNA levels peaking at 0.5–1 hr) (Figure ##FIG##1##2##; see additional files ##SUPPL##1##2## and ##SUPPL##1##2##/Table 1). This gene set contained various transcription factors associated with growth stimulation and cell cycle progression, notably genes encoding the AP-1 complex (<italic>Fos, Fra1/FosL1, Jun, Atf3) </italic>and other growth regulatory genes <italic>(Egr1, Egr2</italic>, <italic>Klf6</italic>, <italic>Myc)</italic>. Strong upregulation was also observed for <italic>Ptgs2</italic>, a highly inducible gene that encodes cyclooxygenase-2 (Cox-2) and has important roles in normal tissue homeostasis and inflammation. Upregulation of some of these immediate-early genes, including <italic>Egr1</italic>, <italic>Jun, Myc </italic>and <italic>Ptgs2</italic>, has earlier been found in LPA-treated MEFs and ovarian carcinoma cells [##REF##16027114##14##,##REF##15781636##15##]. LPA also induced the expression of genes encoding growth-regulatory protein kinases, including Sgk1, Dyrk3, Nuak2 and Map2k3 (Mek3). In addition to these 'forward-driving' genes, the early gene clusters contained various 'feedback regulators' known to turn off gene expression and/or attenuate prolonged signaling (see additional file ##SUPPL##3##4##: Table 1). Co-expression of both negative and positive regulators may be critical for the precise control of cell cycle progression. Among the feedback regulators induced by LPA were transcriptional repressors (<italic>Ztbtb16</italic>, <italic>Lfrd1/Tis7</italic>, <italic>Pawr</italic>), genes that promote the degradation of inducible mRNAs (<italic>Zfp36, Nocturnin/Ccrnl2</italic>) and a number of dual-specificity phosphatase (DUSP) genes, notably <italic>Dusp1, Dusp5, Dusp6</italic>, and <italic>Dusp10</italic>, whose products attenuate the activity of MAP kinases [##REF##17473844##16##]. It is further of note that LPA induced a very robust upregulation of <italic>Mig-6 </italic>also known as <italic>Errfi1 </italic>(ErbB receptor feedback inhibitor 1; ~60-fold induction after 2 h). MIG-6 is a scaffold protein that interacts with the EGF receptor to inhibit its catalytic activity and all downstream signaling events [##REF##16648858##17##,##REF##18046415##18##]. By upregulating <italic>Mig-6</italic>, LPA may keep fibroblasts poised to prevent uncontrolled EGF receptor activation.</p>", "<p>In addition to the above growth-regulatory genes, prominent upregulation was observed for genes whose products regulate the cytoskeleton, including various actin isoforms, vinculin and integrin alpha-subunits (<italic>Itga5 </italic>and <italic>Itga6</italic>), consistent with LPA's function as a regulator of cell shape and motility (see additional file ##SUPPL##3##4##: Table 1).</p>", "<title>Downregulated genes</title>", "<p>Strongest downregulation of gene expression was mostly observed at ~4–6 hrs after LPA stimulation, i.e. coincident with the mid-G1 phase of the cell cycle (Figure ##FIG##1##2##; see additional file ##SUPPL##2##3##). The most strongly down-regulated genes are listed in additional file ##SUPPL##3##4## (Table 2); it comprises the transcription factor <italic>Sox4</italic>, recently identified as a mediator of metastasis [##REF##18185580##19##], the transcriptional repressor <italic>Slug </italic>(<italic>Snai2</italic>), an inducer of the mesenchymal phenotype and a marker of malignancy [##REF##17550342##20##], as well as genes that promote growth arrest. The latter set included the cell cycle inhibitors <italic>Ccng2 </italic>(encoding cyclin G2) and <italic>Cdkn1b </italic>(<italic>p27Kip</italic>), and the 'growth arrest-specific' genes <italic>Gas1-3 </italic>and <italic>Ccn5</italic>. The latter gene encodes a growth inhibitory matrix protein (CCN5) whose reduced expression promotes mesenchymal cell motility [##REF##12507905##21##]. Other strongly downregulated genes were <italic>Nedd9/Hef1 </italic>and <italic>Cdh2</italic>. The Nedd9 scaffold is a downstream effector of focal adhesion kinase that may transduce integrin \"inside-out\" signaling to regulate cell-matrix adhesion and invasion [##REF##16288224##22##,##REF##16814714##23##], while <italic>Cdh2 </italic>encodes N-cadherin, a key regulator of intercellular adhesion strength in fibroblasts [##REF##15247242##24##]. Through the coincident loss of N-cadherin, Nedd9 and CCN5, together with the upregulation of secreted metalloproteases (see below), LPA-stimulated fibroblasts may reduce their adhesive contacts and acquire a more motile and invasive phenotype.</p>", "<title>Induction of genes that encode secreted factors</title>", "<p>Aside from genes associated with cell proliferation and motility, the immediate-early and early clusters contained many genes that encode paracrine factors such as chemokines, cytokines, mitogens and pro-angiogenic factors that are involved in inflammation, tissue remodeling and wound healing. Previous studies have shown that LPA induces the expression of the chemokine CXCL1/Gro-alpha and the cytokines IL-6 and IL-8 in ovarian cancer cells [##REF##14670967##25##,##REF##16510595##26##]. Figure ##FIG##3##4## shows a heat map for 34 selected genes that encode secreted factors, with the most strongly induced genes at the top of the map. These include genes encoding CXCL1/Gro-alpha, IL-6, the EGF family members HB-EGF, epiregulin and amphiregulin, PDGF-A, CSF-1, VEGF-A and the 'pro-fibrotic' and vasoactive ligand endothelin-1 (<italic>End1</italic>). LPA also triggered robust upregulation of <italic>Ccn1/Cyr61 </italic>(peaking at 0.5 hr) and <italic>Ccn2/Ctgf </italic>(connective-tissue growth factor; peaking at 2 hrs). The <italic>Ccn </italic>genes encode matrix proteins that control cell attachment and migration, but also are important players in the pathogenesis of fibrosis [##REF##17130294##27##].</p>", "<p>Components of the urokinase-type plasminogen activator (uPA) system, notably PlauR (uPAR; peaking at 2 hrs) and the two major inhibitors of this system, plasminogen activator inhibitor-1 (PAI-1; peaking at 2–4 hrs) and PAI-2 (peaking at 4–6 hrs) were also strongly upregulated by LPA. Through its action on extracellular matrix and cell-surface proteins, the uPA system modulates cell migration and cell-matrix interactions and thereby plays a key role in wound healing, angiogenesis and tumor progression.</p>", "<p>Genes encoding additional extracellular mediators and surface-exposed proteins were induced in a second wave of transcription with expression peaking at 4–6 hr after LPA stimulation (Figure ##FIG##1##2##; see additional files ##SUPPL##2##3## and ##SUPPL##3##4##/Table 1). These included the chemokines CCL-2, CCL-7 and Cx3Cl1, the matrix metalloproteinases Adam19 and MMP3, and the transmembrane glycoprotein CD44, which plays a key role in cell-cell interactions. A very strongly upregulated gene was <italic>ILl1rl1</italic>, which encodes an IL-1 receptor family member (murine ST2; ~40-fold induction at 4 hrs). The secreted form of ST2 inhibits the production of cytokines in inflammatory cells [##REF##17623648##28##]. Thus, ST2 upregulation in LPA-stimulated fibroblasts may serve to temper inflammatory responses. Consistent with our data, ST2 was recently identified as major LPA target gene in osteoblastic cells, where it is thought to play an anti-inflammatory role during bone healing [##REF##17719864##29##]. Strong upregulation was also observed for tissue factor (coagulation factor III, encoded by <italic>F3</italic>), a cell-surface glycoprotein that initiates the clotting cascade and has additional roles in cell migration and angiogenesis [##REF##16479459##30##]. Late upregulation (peak expression at ~6 hrs: see additional file ##SUPPL##2##3##) was observed for <italic>Lgals3 </italic>and <italic>Timp1</italic>. <italic>Lgals3 </italic>encodes galactin-3, a mediator of inflammation, while secreted Timp-1 has a role in wound healing and the creation of a prometastatic niche.</p>", "<p>In conclusion, LPA-stimulated fibroblasts are transcriptionally committed to produce numerous factors known to act on nearby epithelial cells, leukocytes and endothelial cells. LPA stimulation thus enables fibroblasts to promote tissue remodeling, inflammation, angiogenesis, wound healing and, in a tumor context, cancer progression. A schematic representation of the LPA-induced gene expression program over time is shown in additional file ##SUPPL##6##7##.</p>", "<title>LPA dose dependence</title>", "<p>The early cellular responses to LPA, such as cytoskeletal reorganization and migration, usually show their maximal induction in the submicromolar concentration range, while cell cycle progression requires 1–5 μM doses. We determined the dose-efficacy of LPA on gene transcription using three different concentrations (0.3, 1.0 and 5 μM) and analyzed expression profiles over time (0–4 hrs). The regulation of many LPA target genes was preserved at the lowest LPA dose tested (0.3 μM). About 65% of all target genes showed significant regulation by LPA at all three LPA doses (P-values &lt; 0.01; although in many cases the ratios decreased below the threshold of 1.7-fold induction). Increasing the LPA concentration caused increasingly stronger gene expression, often with more prolonged kinetics, as visualized by heat map (Figure ##FIG##4##5##) and quantitated for selected genes by qPCR (Figure ##FIG##5##6##). It is of note that many of the genes encoding secreted factors (<italic>Il1rl1, Pai2, Ccl2, Ccl7, Cx3Cl1</italic>, <italic>Hbegf</italic>, <italic>Vegf</italic>) reached their maximal expression already at 0.3 μM LPA. \"Ingenuity\" pathway analysis indicated that the functional categories modulated by LPA were preserved at all three concentrations, with the notable exception that lowering the LPA dose to 0.3 μM led to a relative enrichment of genes associated with \"cell movement\" (Figure ##FIG##6##7##). This result is consistent with LPA's propensity to act as a motility factor and chemo-attractant rather than a growth factor in the lower concentration range.</p>", "<title>Expression profiles of LPA and EGF show broad overlap, but at least 100 genes are differentially regulated</title>", "<p>Fibroblasts have long been used as a model to study peptide growth factor signaling. When stimulated by distinct peptide growth factors (EGF, FGF, PDGF), fibroblasts show a strongly conserved gene-expression signature [##REF##16737555##31##]. This is not too surprising since the cognate receptor tyrosine kinases (RTKs) all use the same signaling principle. To our knowledge, however, it is unknown to what extent the transcriptional response to GPCR stimulation bears comparison with that to RTK stimulation in the same cell type. We therefore compared the temporal gene expression programs of LPA and EGF in MEFs at five different time points (0.5–6 hrs). We found that EGF (20 ng/ml) induced many of the same genes as LPA (5 μM), although LPA stimulation often led to a higher level of induction and/or more prolonged kinetics (Figure ##FIG##7##8A,B##). For example, LPA caused a much more prolonged upregulation of the immediate early genes <italic>Fos</italic>, <italic>Dusp1 </italic>and <italic>Cxcl1 </italic>than did EGF (Figure ##FIG##7##8B##; see additional file ##SUPPL##7##8##: cluster 1). LPA was also more efficacious in inducing genes that encode paracrine factors (<italic>Ccl2, Ereg, Il1RL1, Ctgf, Vegfa</italic>) and components of the plasminogen activator system (<italic>Plaur, Pai-1</italic>) (Figure ##FIG##7##8B##). Quantitative PCR analysis confirmed the differential regulation of selected genes by LPA and EGF (Figure ##FIG##8##9##). A complete list of the differentially regulated genes is shown in additional file ##SUPPL##8##9##. To what extent these quantitative differences reflect different expression levels of the respective receptors is currently unknown.</p>", "<p>Despite this large overlap of the LPA and EGF expression profiles, approx. 7% of the genes (105 out of 1508 transcripts) was differentially regulated by LPA (Figure ##FIG##7##8##; see additional files ##SUPPL##7##8## and ##SUPPL##8##9##). Immediate-early genes that were upregulated by LPA, but not EGF, include <italic>Edn1 </italic>(endothelin-1), <italic>Fgf16</italic>, <italic>Nfkbia </italic>(NF-kappaB inhibitor alpha) and several protein kinase genes (<italic>Bmp2k</italic>, <italic>Plk2, Tesk2</italic>, <italic>Pim1</italic>), as shown in additional file ##SUPPL##7##8## (cluster 3). Interestingly, LPA also induced the expression of a newly identified GPCR for LPA, termed P2Y5 (encoded by <italic>P2ry5 </italic>[##REF##18297070##32##]), which adds an element of feedback to the fibroblast response to LPA (see additional file ##SUPPL##7##8##: cluster 3). At 2–4 hrs, more LPA-specific transcripts could be identified (Figure ##FIG##7##8B##; see additional files ##SUPPL##7##8## and ##SUPPL##8##9##). Gene ontology analysis revealed that the LPA-specific gene set was enriched for genes associated with cytoskeletal organization and integrin signaling, notably those encoding various actin isoforms (<italic>Actb, Acta1, Actg2</italic>), palladin (<italic>Palld</italic>), vinculin (<italic>Vcl</italic>), an Arp2/3 subunit (<italic>Arpc5</italic>), calponins (<italic>Cnn1, Cnn3</italic>), a Rho GTPase (<italic>Rhoj</italic>), Rho-kinase (<italic>Rock2</italic>), myosin X (<italic>Myo10</italic>) and an integrin subunit (<italic>Itga5</italic>) (see additional file ##SUPPL##8##9##). Specific upregulation of cytoskeleton-associated genes is in line with LPA's role as an efficacious regulator of cell shape and motility.</p>", "<p>It has long been proposed that GPCR ligands such as LPA signal through 'transactivation' of the EGF receptor [##REF##8596637##33##,##REF##14647423##34##]. According to this model, GPCR agonists rapidly activate the EGF receptor to exploit the tyrosine-posphorylated receptor as a signaling intermediate. However, blocking EGF receptor activity by the selective EGF receptor kinase inhibitor AG1478 (250 nM) had no effect on LPA-induced MAP kinase activation, <italic>Ccl2 </italic>expression and DNA synthesis, while the responses to EGF were fully inhibited (additional file ##SUPPL##9##10## and results not shown). This is in agreement with a previous study showing that LPA mitogenic signaling in MEFs does not require EGF receptor tyrosine phosphorylation [##REF##11274221##35##]. Figure ##FIG##9##10## illustrates that the transcriptional response to LPA was only little affected by EGF receptor inhibitor treatment (expression of 528 genes, reproducibly regulated by LPA at three different concentrations at T = 4 hr). About 15% of the LPA-induced genes (81 out of 528 transcripts) was &gt;70% inhibited after drug treatment. Otherwise, EGF receptor inhibition did not affect the induction of key immediate-early and early genes by LPA, such as transcription factors and paracrine mediators. While it remains formally possible that basal EGF receptor activity has a permissive effect on some LPA-induced signaling events, we conclude that LPA and EGF signal independently to regulate broadly overlapping sets of genes in MEFs. It thus appears that the transcriptional program induced by either LPA-GPCR or EGF-RTK stimulation in fibroblasts is more strongly conserved than previously appreciated.</p>" ]

[ "<title>Results and Discussion</title>", "<title>Characterization of MEFs</title>", "<p>We examined the mitogenic response and LPA receptor expression profile of immortalized mouse embryonic fibroblasts (MEFs). Stimulation of serum-deprived, nearly confluent MEFs with saturating doses of LPA (5 μM) resulted in a significant increase in DNA synthesis to about 40% the level induced by 10% serum. EGF (50 ng/ml) was somewhat less efficacious than LPA in stimulating DNA synthesis (Figure ##FIG##0##1A##). LPA signals via at least five distinct GPCRs (termed LPA_1–5) that couple to multiple G proteins, including G_i/o, G_q/11and G_12/13. Quantitative PCR analysis showed that our MEFs co-expressed LPA₁, LPA₂and LPA₄, with LPA₁and LPA₄being the predominant transcripts (Figure ##FIG##0##1B##). Since MEFs do not express ATX (encoded by <italic>Enpp2</italic>; data not shown), autocrine LPA signaling is not operative in our cell system.</p>", "<title>The global transcriptional response to LPA</title>", "<p>We examined the temporal program of gene expression in serum-starved MEFs treated with LPA (5 μM, i.e. about the normal concentration in serum [##REF##8489494##13##]). Total RNA was isolated at different time points after LPA stimulation (0–24 hrs). Global transcription profiles were determined using oligonucleotide microarrays containing 31,770 mouse transcripts. Amplified RNA of the treated samples was matched with the untreated control and hybridized in duplicate with reversal of the Cy3 and Cy5 dyes; the normalized Cy5/Cy3 ratios were combined and used for further analysis. We selected genes that were significantly regulated (p &lt; 0.01) at two or more consecutive time points or in replicate measurements, which yielded 1508 LPA-regulated genes (see additional file ##SUPPL##0##1##: complete dataset). The entire dataset has been deposited in the EBI/ArrayExpress database (see Methods). We restricted our data set to genes that were induced by &gt;1.7-fold at two or more time points and grouped them according to the temporal profile of gene induction using K-means clustering. This resulted in ten clusters, each containing genes that show similarly shaped waves of transcription (Figure ##FIG##1##2##). Seven clusters contained genes that were upregulated by LPA (424 transcripts), whereas three other clusters mainly comprised the down-regulated genes (209 transcripts) (Figure ##FIG##1##2##; for details see additional files ##SUPPL##1##2## and ##SUPPL##2##3##). The genes that were most strongly regulated at different time points are listed in additional file ##SUPPL##3##4## (Table 1: upregulated genes; Table 2: downregulated genes). The microarray results were validated by examining the expression of 22 representative genes from different clusters using real-time PCR (Figure ##FIG##2##3##). The -fold inductions of expression generally were higher in the qPCR assays than in the corresponding microarray experiments, reflecting the different sensitivities of both methods (see additional file ##SUPPL##4##5##: correlation plot of all data points). Gene ontology analysis revealed that LPA regulated the expression of genes in multiple functional categories that mostly corresponded to the different gene clusters (see additional file ##SUPPL##5##6##).</p>", "<title>Regulation of genes associated with growth regulation and cytoskeletal reorganization</title>", "<title>Upregulated genes</title>", "<p>The immediate transcriptional response to LPA was characterized by &gt;100 upregulated immediate-early genes (mRNA levels peaking at 0.5–1 hr) (Figure ##FIG##1##2##; see additional files ##SUPPL##1##2## and ##SUPPL##1##2##/Table 1). This gene set contained various transcription factors associated with growth stimulation and cell cycle progression, notably genes encoding the AP-1 complex (<italic>Fos, Fra1/FosL1, Jun, Atf3) </italic>and other growth regulatory genes <italic>(Egr1, Egr2</italic>, <italic>Klf6</italic>, <italic>Myc)</italic>. Strong upregulation was also observed for <italic>Ptgs2</italic>, a highly inducible gene that encodes cyclooxygenase-2 (Cox-2) and has important roles in normal tissue homeostasis and inflammation. Upregulation of some of these immediate-early genes, including <italic>Egr1</italic>, <italic>Jun, Myc </italic>and <italic>Ptgs2</italic>, has earlier been found in LPA-treated MEFs and ovarian carcinoma cells [##REF##16027114##14##,##REF##15781636##15##]. LPA also induced the expression of genes encoding growth-regulatory protein kinases, including Sgk1, Dyrk3, Nuak2 and Map2k3 (Mek3). In addition to these 'forward-driving' genes, the early gene clusters contained various 'feedback regulators' known to turn off gene expression and/or attenuate prolonged signaling (see additional file ##SUPPL##3##4##: Table 1). Co-expression of both negative and positive regulators may be critical for the precise control of cell cycle progression. Among the feedback regulators induced by LPA were transcriptional repressors (<italic>Ztbtb16</italic>, <italic>Lfrd1/Tis7</italic>, <italic>Pawr</italic>), genes that promote the degradation of inducible mRNAs (<italic>Zfp36, Nocturnin/Ccrnl2</italic>) and a number of dual-specificity phosphatase (DUSP) genes, notably <italic>Dusp1, Dusp5, Dusp6</italic>, and <italic>Dusp10</italic>, whose products attenuate the activity of MAP kinases [##REF##17473844##16##]. It is further of note that LPA induced a very robust upregulation of <italic>Mig-6 </italic>also known as <italic>Errfi1 </italic>(ErbB receptor feedback inhibitor 1; ~60-fold induction after 2 h). MIG-6 is a scaffold protein that interacts with the EGF receptor to inhibit its catalytic activity and all downstream signaling events [##REF##16648858##17##,##REF##18046415##18##]. By upregulating <italic>Mig-6</italic>, LPA may keep fibroblasts poised to prevent uncontrolled EGF receptor activation.</p>", "<p>In addition to the above growth-regulatory genes, prominent upregulation was observed for genes whose products regulate the cytoskeleton, including various actin isoforms, vinculin and integrin alpha-subunits (<italic>Itga5 </italic>and <italic>Itga6</italic>), consistent with LPA's function as a regulator of cell shape and motility (see additional file ##SUPPL##3##4##: Table 1).</p>", "<title>Downregulated genes</title>", "<p>Strongest downregulation of gene expression was mostly observed at ~4–6 hrs after LPA stimulation, i.e. coincident with the mid-G1 phase of the cell cycle (Figure ##FIG##1##2##; see additional file ##SUPPL##2##3##). The most strongly down-regulated genes are listed in additional file ##SUPPL##3##4## (Table 2); it comprises the transcription factor <italic>Sox4</italic>, recently identified as a mediator of metastasis [##REF##18185580##19##], the transcriptional repressor <italic>Slug </italic>(<italic>Snai2</italic>), an inducer of the mesenchymal phenotype and a marker of malignancy [##REF##17550342##20##], as well as genes that promote growth arrest. The latter set included the cell cycle inhibitors <italic>Ccng2 </italic>(encoding cyclin G2) and <italic>Cdkn1b </italic>(<italic>p27Kip</italic>), and the 'growth arrest-specific' genes <italic>Gas1-3 </italic>and <italic>Ccn5</italic>. The latter gene encodes a growth inhibitory matrix protein (CCN5) whose reduced expression promotes mesenchymal cell motility [##REF##12507905##21##]. Other strongly downregulated genes were <italic>Nedd9/Hef1 </italic>and <italic>Cdh2</italic>. The Nedd9 scaffold is a downstream effector of focal adhesion kinase that may transduce integrin \"inside-out\" signaling to regulate cell-matrix adhesion and invasion [##REF##16288224##22##,##REF##16814714##23##], while <italic>Cdh2 </italic>encodes N-cadherin, a key regulator of intercellular adhesion strength in fibroblasts [##REF##15247242##24##]. Through the coincident loss of N-cadherin, Nedd9 and CCN5, together with the upregulation of secreted metalloproteases (see below), LPA-stimulated fibroblasts may reduce their adhesive contacts and acquire a more motile and invasive phenotype.</p>", "<title>Induction of genes that encode secreted factors</title>", "<p>Aside from genes associated with cell proliferation and motility, the immediate-early and early clusters contained many genes that encode paracrine factors such as chemokines, cytokines, mitogens and pro-angiogenic factors that are involved in inflammation, tissue remodeling and wound healing. Previous studies have shown that LPA induces the expression of the chemokine CXCL1/Gro-alpha and the cytokines IL-6 and IL-8 in ovarian cancer cells [##REF##14670967##25##,##REF##16510595##26##]. Figure ##FIG##3##4## shows a heat map for 34 selected genes that encode secreted factors, with the most strongly induced genes at the top of the map. These include genes encoding CXCL1/Gro-alpha, IL-6, the EGF family members HB-EGF, epiregulin and amphiregulin, PDGF-A, CSF-1, VEGF-A and the 'pro-fibrotic' and vasoactive ligand endothelin-1 (<italic>End1</italic>). LPA also triggered robust upregulation of <italic>Ccn1/Cyr61 </italic>(peaking at 0.5 hr) and <italic>Ccn2/Ctgf </italic>(connective-tissue growth factor; peaking at 2 hrs). The <italic>Ccn </italic>genes encode matrix proteins that control cell attachment and migration, but also are important players in the pathogenesis of fibrosis [##REF##17130294##27##].</p>", "<p>Components of the urokinase-type plasminogen activator (uPA) system, notably PlauR (uPAR; peaking at 2 hrs) and the two major inhibitors of this system, plasminogen activator inhibitor-1 (PAI-1; peaking at 2–4 hrs) and PAI-2 (peaking at 4–6 hrs) were also strongly upregulated by LPA. Through its action on extracellular matrix and cell-surface proteins, the uPA system modulates cell migration and cell-matrix interactions and thereby plays a key role in wound healing, angiogenesis and tumor progression.</p>", "<p>Genes encoding additional extracellular mediators and surface-exposed proteins were induced in a second wave of transcription with expression peaking at 4–6 hr after LPA stimulation (Figure ##FIG##1##2##; see additional files ##SUPPL##2##3## and ##SUPPL##3##4##/Table 1). These included the chemokines CCL-2, CCL-7 and Cx3Cl1, the matrix metalloproteinases Adam19 and MMP3, and the transmembrane glycoprotein CD44, which plays a key role in cell-cell interactions. A very strongly upregulated gene was <italic>ILl1rl1</italic>, which encodes an IL-1 receptor family member (murine ST2; ~40-fold induction at 4 hrs). The secreted form of ST2 inhibits the production of cytokines in inflammatory cells [##REF##17623648##28##]. Thus, ST2 upregulation in LPA-stimulated fibroblasts may serve to temper inflammatory responses. Consistent with our data, ST2 was recently identified as major LPA target gene in osteoblastic cells, where it is thought to play an anti-inflammatory role during bone healing [##REF##17719864##29##]. Strong upregulation was also observed for tissue factor (coagulation factor III, encoded by <italic>F3</italic>), a cell-surface glycoprotein that initiates the clotting cascade and has additional roles in cell migration and angiogenesis [##REF##16479459##30##]. Late upregulation (peak expression at ~6 hrs: see additional file ##SUPPL##2##3##) was observed for <italic>Lgals3 </italic>and <italic>Timp1</italic>. <italic>Lgals3 </italic>encodes galactin-3, a mediator of inflammation, while secreted Timp-1 has a role in wound healing and the creation of a prometastatic niche.</p>", "<p>In conclusion, LPA-stimulated fibroblasts are transcriptionally committed to produce numerous factors known to act on nearby epithelial cells, leukocytes and endothelial cells. LPA stimulation thus enables fibroblasts to promote tissue remodeling, inflammation, angiogenesis, wound healing and, in a tumor context, cancer progression. A schematic representation of the LPA-induced gene expression program over time is shown in additional file ##SUPPL##6##7##.</p>", "<title>LPA dose dependence</title>", "<p>The early cellular responses to LPA, such as cytoskeletal reorganization and migration, usually show their maximal induction in the submicromolar concentration range, while cell cycle progression requires 1–5 μM doses. We determined the dose-efficacy of LPA on gene transcription using three different concentrations (0.3, 1.0 and 5 μM) and analyzed expression profiles over time (0–4 hrs). The regulation of many LPA target genes was preserved at the lowest LPA dose tested (0.3 μM). About 65% of all target genes showed significant regulation by LPA at all three LPA doses (P-values &lt; 0.01; although in many cases the ratios decreased below the threshold of 1.7-fold induction). Increasing the LPA concentration caused increasingly stronger gene expression, often with more prolonged kinetics, as visualized by heat map (Figure ##FIG##4##5##) and quantitated for selected genes by qPCR (Figure ##FIG##5##6##). It is of note that many of the genes encoding secreted factors (<italic>Il1rl1, Pai2, Ccl2, Ccl7, Cx3Cl1</italic>, <italic>Hbegf</italic>, <italic>Vegf</italic>) reached their maximal expression already at 0.3 μM LPA. \"Ingenuity\" pathway analysis indicated that the functional categories modulated by LPA were preserved at all three concentrations, with the notable exception that lowering the LPA dose to 0.3 μM led to a relative enrichment of genes associated with \"cell movement\" (Figure ##FIG##6##7##). This result is consistent with LPA's propensity to act as a motility factor and chemo-attractant rather than a growth factor in the lower concentration range.</p>", "<title>Expression profiles of LPA and EGF show broad overlap, but at least 100 genes are differentially regulated</title>", "<p>Fibroblasts have long been used as a model to study peptide growth factor signaling. When stimulated by distinct peptide growth factors (EGF, FGF, PDGF), fibroblasts show a strongly conserved gene-expression signature [##REF##16737555##31##]. This is not too surprising since the cognate receptor tyrosine kinases (RTKs) all use the same signaling principle. To our knowledge, however, it is unknown to what extent the transcriptional response to GPCR stimulation bears comparison with that to RTK stimulation in the same cell type. We therefore compared the temporal gene expression programs of LPA and EGF in MEFs at five different time points (0.5–6 hrs). We found that EGF (20 ng/ml) induced many of the same genes as LPA (5 μM), although LPA stimulation often led to a higher level of induction and/or more prolonged kinetics (Figure ##FIG##7##8A,B##). For example, LPA caused a much more prolonged upregulation of the immediate early genes <italic>Fos</italic>, <italic>Dusp1 </italic>and <italic>Cxcl1 </italic>than did EGF (Figure ##FIG##7##8B##; see additional file ##SUPPL##7##8##: cluster 1). LPA was also more efficacious in inducing genes that encode paracrine factors (<italic>Ccl2, Ereg, Il1RL1, Ctgf, Vegfa</italic>) and components of the plasminogen activator system (<italic>Plaur, Pai-1</italic>) (Figure ##FIG##7##8B##). Quantitative PCR analysis confirmed the differential regulation of selected genes by LPA and EGF (Figure ##FIG##8##9##). A complete list of the differentially regulated genes is shown in additional file ##SUPPL##8##9##. To what extent these quantitative differences reflect different expression levels of the respective receptors is currently unknown.</p>", "<p>Despite this large overlap of the LPA and EGF expression profiles, approx. 7% of the genes (105 out of 1508 transcripts) was differentially regulated by LPA (Figure ##FIG##7##8##; see additional files ##SUPPL##7##8## and ##SUPPL##8##9##). Immediate-early genes that were upregulated by LPA, but not EGF, include <italic>Edn1 </italic>(endothelin-1), <italic>Fgf16</italic>, <italic>Nfkbia </italic>(NF-kappaB inhibitor alpha) and several protein kinase genes (<italic>Bmp2k</italic>, <italic>Plk2, Tesk2</italic>, <italic>Pim1</italic>), as shown in additional file ##SUPPL##7##8## (cluster 3). Interestingly, LPA also induced the expression of a newly identified GPCR for LPA, termed P2Y5 (encoded by <italic>P2ry5 </italic>[##REF##18297070##32##]), which adds an element of feedback to the fibroblast response to LPA (see additional file ##SUPPL##7##8##: cluster 3). At 2–4 hrs, more LPA-specific transcripts could be identified (Figure ##FIG##7##8B##; see additional files ##SUPPL##7##8## and ##SUPPL##8##9##). Gene ontology analysis revealed that the LPA-specific gene set was enriched for genes associated with cytoskeletal organization and integrin signaling, notably those encoding various actin isoforms (<italic>Actb, Acta1, Actg2</italic>), palladin (<italic>Palld</italic>), vinculin (<italic>Vcl</italic>), an Arp2/3 subunit (<italic>Arpc5</italic>), calponins (<italic>Cnn1, Cnn3</italic>), a Rho GTPase (<italic>Rhoj</italic>), Rho-kinase (<italic>Rock2</italic>), myosin X (<italic>Myo10</italic>) and an integrin subunit (<italic>Itga5</italic>) (see additional file ##SUPPL##8##9##). Specific upregulation of cytoskeleton-associated genes is in line with LPA's role as an efficacious regulator of cell shape and motility.</p>", "<p>It has long been proposed that GPCR ligands such as LPA signal through 'transactivation' of the EGF receptor [##REF##8596637##33##,##REF##14647423##34##]. According to this model, GPCR agonists rapidly activate the EGF receptor to exploit the tyrosine-posphorylated receptor as a signaling intermediate. However, blocking EGF receptor activity by the selective EGF receptor kinase inhibitor AG1478 (250 nM) had no effect on LPA-induced MAP kinase activation, <italic>Ccl2 </italic>expression and DNA synthesis, while the responses to EGF were fully inhibited (additional file ##SUPPL##9##10## and results not shown). This is in agreement with a previous study showing that LPA mitogenic signaling in MEFs does not require EGF receptor tyrosine phosphorylation [##REF##11274221##35##]. Figure ##FIG##9##10## illustrates that the transcriptional response to LPA was only little affected by EGF receptor inhibitor treatment (expression of 528 genes, reproducibly regulated by LPA at three different concentrations at T = 4 hr). About 15% of the LPA-induced genes (81 out of 528 transcripts) was &gt;70% inhibited after drug treatment. Otherwise, EGF receptor inhibition did not affect the induction of key immediate-early and early genes by LPA, such as transcription factors and paracrine mediators. While it remains formally possible that basal EGF receptor activity has a permissive effect on some LPA-induced signaling events, we conclude that LPA and EGF signal independently to regulate broadly overlapping sets of genes in MEFs. It thus appears that the transcriptional program induced by either LPA-GPCR or EGF-RTK stimulation in fibroblasts is more strongly conserved than previously appreciated.</p>" ]

[ "<title>Conclusion</title>", "<p>Dissecting the transcriptional response to growth factors in selected cell systems may help to better understand various aspects of embryonic development, adult tissue homeostasis and cancer. The present study characterizes the global transcriptional program of MEFs to LPA and thereby provides new insights into the normal physiological response of quiescent fibroblasts to this multifunctional lipid mediator. In addition to genes associated with cell proliferation, adhesion and migration, LPA induces a host of genes that encode secreted factors known to promote tissue remodeling, wound healing, inflammation, angiogenesis and tumor progression, depending on cellular context. This highlights the importance of LPA signaling in profoundly modifying the fibroblast microenvironment.</p>", "<p>Previous transcriptional profiling of serum-stimulated human skin fibroblasts has identified a 'core serum response' (CSR) that is characterized by cell-cycle-independent genes and reflects various aspects of wound healing, notably the induction of genes involved in matrix remodeling and re-epithelialization [##REF##9872747##36##]. This fibroblast CSR or \"wound-response signature\" is recapitulated in human carcinomas and may help predict tumor progression [##REF##14737219##37##]. Comparative analysis revealed, however, that the expression profile of LPA-stimulated MEFs shows only limited overlap with the canonical CSR of serum-stimulated human fibroblasts. At first sight, this result is somewhat unexpected since LPA is a major serum constituent [##REF##15273989##1##,##REF##8489494##13##]. On the other hand, serum is an ill-defined mixture of numerous bioactive factors and it is likely that the combined action of many different factors obscures the comparison between serum and LPA. Moreover, MEFs differ from human skin fibroblasts not only in their biological and anatomic origin, but also in their LPA receptor expression pattern (data not shown). Yet, it should be emphasized that LPA-stimulated MEFs and serum-stimulated human skin fibroblasts both show a gene expression profile that is strongly associated with tissue remodeling as well as tumor progression.</p>", "<p>Of final note is our finding that the transcriptional response of MEFs to LPA versus EGF shows an overlap of &gt;90%, at least qualitatively, despite the fact that LPA and EGF signal via completely different mechanisms. LPA-induced gene expression was largely independent of EGF receptor activity, which argues against the notion that LPA exploits the EGF receptor as a signaling intermediate. The broad overlap between LPA- and EGF-induced gene expression shows that GPCRs and RTKs have more in common than previously appreciated. One should not conclude, however, that the transcriptional response to receptor stimulation is less dependent on the nature of the receptor than on the cell type. For example, the common expression pattern of LPA and EGF shows hardly any overlap with that induced by Wnt signaling in fibroblasts [##REF##17895986##38##]. The great diversity of fibroblast responses to LPA as reported here is an important area for further study.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Lysophosphatidic acid (LPA) is a lipid mediator that acts through specific G protein-coupled receptors to stimulate the proliferation, migration and survival of many cell types. LPA signaling has been implicated in development, wound healing and cancer. While LPA signaling pathways have been studied extensively, it remains unknown how LPA affects global gene expression in its target cells.</p>", "<title>Results</title>", "<p>We have examined the temporal program of global gene expression in quiescent mouse embryonic fibroblasts stimulated with LPA using 32 k oligonucleotide microarrays. In addition to genes involved in growth stimulation and cytoskeletal reorganization, LPA induced many genes that encode secreted factors, including chemokines, growth factors, cytokines, pro-angiogenic and pro-fibrotic factors, components of the plasminogen activator system and metalloproteases. Strikingly, epidermal growth factor induced a broadly overlapping expression pattern, but some 7% of the genes (105 out of 1508 transcripts) showed differential regulation by LPA. The subset of LPA-specific genes was enriched for those associated with cytoskeletal remodeling, in keeping with LPA's ability to regulate cell shape and motility.</p>", "<title>Conclusion</title>", "<p>This study highlights the importance of LPA in programming fibroblasts not only to proliferate and migrate but also to produce many paracrine mediators of tissue remodeling, angiogenesis, inflammation and tumor progression. Furthermore, our results show that G protein-coupled receptors and receptor tyrosine kinases can signal independently to regulate broadly overlapping sets of genes in the same cell type.</p>" ]

[ "<title>Authors' contributions</title>", "<p>CS performed the experiments, did the data and bioinformatics analysis and drafted parts of the manuscript. RK participated in the microarray experiments and helped with data analysis. WHM conceived and coordinated the study and wrote the manuscript. All authors read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>We thank Marloes Bagijn for PCR and biochemical experiments, Mike Heimerikx and Marja Nieuwland for assistance with the microarray studies, Arno Velds and Daoud Sie for bioinformatics assistance and Lodewijk Wessels for help with statistical analysis. We thank Panthea Taghavi for MEFs and the TBX2 vector. This study was supported by the Dutch Cancer Society and the Centre for Biomedical Genetics. The authors declare no conflict of interest.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Characterization of MEFs</bold>. <bold>(A) </bold>Mitogenic responsiveness of MEFs as measured by [³H]thymidine incorporation. Serum-deprived cells (\"control\") were treated with fetal calf serum (FCS; 10%), LPA (10 μM) or EGF (25 ng/ml). The response to FCS was set at 100%. Bars represent means ± SD (N = 3). <bold>(B) </bold>LPA receptor expression in MEFs. Expression levels were determined by qPCR using the GAPDH gene for normalization. Expression levels of LPA₃and LPA₅are negligible.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Global transcriptional response of MEFs to LPA</bold>. Gene expression patterns induced by LPA (5 μM) clustered into 10 different classes (K-means clustering analysis; N = 10). Clusters were generated based on the time point of maximal induction. Data indicate ²log ratios of transcripts with p ≤ 0.001 in at least two consecutive time points (resulting in 633 selected transcripts). Each column represents one time point of LPA treatment; the last column shows the expression in non-synchronized MEFs. The time point or period of maximal induction (red) or reduction (green) is indicated in each cluster by an arrow and the respective hour. Solid lines (pink) indicate the median temporal pattern of expression; dotted lines indicate the median level for each cluster. See additional files ##SUPPL##1##2## and ##SUPPL##2##3## for details of the individual clusters.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Validation of microarray data</bold>. Comparison of the relative expression levels of genes selected from different clusters, as determined by microarray (closed symbols) and qPCR (open symbols). qPCR data were normalized to HPRT mRNA concentration and plotted relative to the level at time zero. Data are presented as means ± SD of duplicate experiments. Note that the qPCR assays generally yielded higher mRNA values than the microarray analysis. See also the correlation plot in additional file ##SUPPL##5##6##.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Temporal expression pattern of LPA-regulated genes that encode secreted factors</bold>. The heat map shows expression profiles of 34 selected genes as indicated. See also additional file ##SUPPL##3##4## (Table 1).</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>LPA dose dependence</bold>. The global transcriptional response of quiescent MEFs to different LPA concentrations (0.3, 1.0 and 5.0 μM) at the indicated time points (0.5–4 hrs). For each time point, transcripts with p &lt; 0.01 in 3 out of 4 measurements were selected; the resulting 915 genes were subjected to hierarchical clustering analysis.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Temporal expression of selected genes at different LPA concentrations validated by qPCR</bold>. qPCR data were normalized to HPRT mRNA concentration and plotted relative to the level at time zero. Data are presented as means ± SD of duplicate experiments.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p><bold>Ingenuity pathway analysis of gene expression at two different LPA concentrations</bold>. Note relative enrichment of 'cell movement' genes at the lower LPA concentration.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p><bold>Differential expression patterns of LPA and EGF</bold>. <bold>(A) </bold>Global transcriptional responses to LPA (5 μM) versus EGF (50 ng/ml) over time. Hierarchical clustering analysis was done on the 915-gene data set of Figure 5. <bold>(B) </bold>Heat map of 111 reporters (encoding 105 genes) that are differentially regulated by LPA (cyan) and EGF (red), or inversely regulated by either agonist (blue). See also additional file ##SUPPL##7##8##: heat map of selected genes. Numbers (1–9) refer to additional file ##SUPPL##8##9##, which shows a list of all 111 reporters. Genes were identified by Anova analysis (two groups, p &lt; 0.01).</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p><bold>Differential regulation of selected genes by LPA (black) and EGF (red) validated by qPCR</bold>. qPCR data were normalized to HPRT mRNA concentration and plotted relative to the level at time zero. Data are presented as means ± SD of duplicate experiments.</p></caption></fig>", "<fig position=\"float\" id=\"F10\"><label>Figure 10</label><caption><p><bold>Correlation plot of the effect of EGF receptor inhibition</bold>. MEFs were treated with a mixture of two EGF receptor inhibitors (AG1478 and PD168393, 250 nM each) or DMSO (control) prior to stimulation with LPA (5 μM) for 4 hrs. Microarray hybridization was performed using the corresponding time-zero control with the same pretreatment. The ²log expression level of 528 LPA-regulated genes after 4 hrs is shown as a dot plot to correlate the effect of drug treatment on the expression level of individual genes. The expression level of 528 LPA-regulated genes in control cells was set at 100%. The overall reduction of LPA-induced expression by the inhibitors was approx. 35%, as inferred from the correlation coefficient.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p><bold>Complete data set of LPA-regulated target genes in MEFs</bold>. 1508 genes were selected based on the criteria p &lt; 0.01 in at least two samples (i.e. two different time points or the same time point in independent dose-response experiments; Figure ##FIG##4##5##), with a minimal fold-change of &gt;1.5. Indicated are the ²log ratios.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p><bold>Gene expression profiles clustered into different classes: immediate-early and early genes</bold>. See Figure ##FIG##1##2## for details.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p><bold>Gene expression profiles clustered into different classes: delayed and down-regulated genes</bold>. See Figure ##FIG##1##2## for details.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S4\"><caption><title>Additional file 4</title><p><bold>List of selected genes that are most strongly regulated by LPA (5 μM)</bold>. Excel files showing 141 upregulated genes (Table 1) and 38 downregulated genes (Table 2).</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S5\"><caption><title>Additional file 5</title><p><bold>Correlation plot of qPCR versus MA assays</bold>. Comparison of mRNA levels measured by microarray and qPCR assays. Each data point represents a single gene at a single time point (Y = 0.932x1.24; R2 = 0.8862, R = 0.941).</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S6\"><caption><title>Additional file 6</title><p><bold>Gene ontology analysis of the LPA-induced gene expression program in MEFs</bold>. Functional categories of genes showing peak expression at 2–4 hrs.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S7\"><caption><title>Additional file 7</title><p><bold>Schematic representation of the LPA-induced expression program over time</bold>. Ingenuity pathway analysis. Red: upregulated genes. Green: downregulated genes.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S8\"><caption><title>Additional file 8</title><p><bold>Heat map of selected genes that that are differentially regulated by LPA and EGF. </bold>Numbers of the clusters (1–4) refer to those in the heat map of Figure ##FIG##7##8B## and the list of genes in additional file ##SUPPL##8##9##.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S9\"><caption><title>Additional file 9</title><p><bold>List of genes that are differentially regulated by LPA and EGF over time. </bold>Numbers of the clusters (1–9) refer those in Figure ##FIG##7##8B##. Genes were identified using Oneway Anova (two groups, p &lt; 0.05, using at least two time points).</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S10\"><caption><title>Additional file 10</title><p><bold>Effect of AG1487 (250 nM) on LPA- and EGF-induced cellular responses. </bold>Upper panel: MAP kinase activation (pERK) as determined by Western blot; tubulin (tub) served as a loading control. Lower panel: Ccl2 mRNA expression after 1 hr of agonist stimulation (qPCR determination).</p></caption></supplementary-material>" ]

[]

[ "<graphic xlink:href=\"1471-2164-9-387-1\"/>", "<graphic xlink:href=\"1471-2164-9-387-2\"/>", "<graphic xlink:href=\"1471-2164-9-387-3\"/>", "<graphic xlink:href=\"1471-2164-9-387-4\"/>", "<graphic xlink:href=\"1471-2164-9-387-5\"/>", "<graphic xlink:href=\"1471-2164-9-387-6\"/>", "<graphic xlink:href=\"1471-2164-9-387-7\"/>", "<graphic xlink:href=\"1471-2164-9-387-8\"/>", "<graphic xlink:href=\"1471-2164-9-387-9\"/>", "<graphic xlink:href=\"1471-2164-9-387-10\"/>" ]

[ "<media xlink:href=\"1471-2164-9-387-S1.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S2.ppt\" mimetype=\"application\" mime-subtype=\"vnd.ms-powerpoint\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S3.ppt\" mimetype=\"application\" mime-subtype=\"vnd.ms-powerpoint\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S4.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S5.ppt\" mimetype=\"application\" mime-subtype=\"vnd.ms-powerpoint\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S6.ppt\" mimetype=\"application\" mime-subtype=\"vnd.ms-powerpoint\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S7.ppt\" mimetype=\"application\" mime-subtype=\"vnd.ms-powerpoint\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S8.ppt\" mimetype=\"application\" mime-subtype=\"vnd.ms-powerpoint\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S9.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2164-9-387-S10.ppt\" mimetype=\"application\" mime-subtype=\"vnd.ms-powerpoint\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>It has been established that co-regulated genes exhibit similar expression patterns as a norm and not as an exception [##REF##15173566##1##]. Microarray allows sensitive, detection of small differences in transcript abundance [##UREF##0##2##], therefore it is utilized extensively to study co-regulation of genes. The gene expression imprinted in the microarray is the manifestation of the pathway activity undergone by the organism, and every gene performs its obligatory function in various pathways [##REF##12429058##3##]. The genes do not co-express with the same set of genes all the time, but under various conditions will be expressed with different sets of genes termed as conditional coregulation [##REF##12429058##3##,##REF##15575966##4##]. Each gene is estimated to interact with four to eight other genes and associated with 10 biological functions [##REF##11099257##5##]. The DNA microarray therefore assists in measuring the difference in transcriptional activity by comparing their mRNA levels under different experimental conditions like developmental stages, stress, or osmotic shock [##REF##11714516##6##]. Various approaches exist for interpretation of relative gene expression. One of the basic strategies is to set the expression level to three states, i.e. underexpression, baseline and overexpression using a fold change cutoff like two times fold change against the control [##REF##11473000##7##]. Other strategies include setting thresholds representing significant changes between subsequent timepoints and storing in bins [##UREF##1##8##], an adaptive procedure that takes gene-specific variation into consideration to derive the gene expression in different states [##REF##11473000##7##]. Further, microarray data consists of variations generally termed as interesting variations, which are biologically important, and are superimposed by \"obscuring variation\" or systematic variation [##REF##12582260##9##].</p>", "<p>We have developed an alternate algorithm based on graph theory that takes a discretized expression matrix as input and emits output string. Not all genes are expressed all the time and are required at different developmental and maturation phases of plant [##UREF##2##10##]. We therefore categorized each differentially expressed gene as \"ON/OFF\" from every experiment against its control in three states: overexpression (+1), underexpression (-1) and no expression (0). We utilized the regularized t-test to derive differentially expressed genes to overcome low replicates and extract meaningful biological variances [##REF##11395427##11##]. We derived a discretized expression matrix for all the genes for various time series experimental conditions from differentially expressed genes. The similarity between two (or more) discretized vectors can be calculated through various distance measures such as number of positions the vector has similar values excluding 0 [##UREF##0##2##]. Our implementation results in the output string that is the representation of the pattern the genes have undergone during transition from one state to another. Any gene can be queried against all the other genes by matching stored output strings in the database, and a 'score' is generated representing the similarity index between any given set of genes.</p>", "<p>Affymetrix™ provides a calculation of absolute signal values for each gene for a given set of experiments, which can be viewed as points in n-dimensional space (where n is the number of experiments) [##UREF##3##12##]. Similarity between point representations of genes can be calculated using various metrics like Pearson correlation or Euclidean distance using various clustering algorithms [##UREF##4##13##]. Most databases (e.g. Comprehensive Systems-Biology Database (CSB.DB) [##REF##15247097##14##] and the Arabidopsis Microarray Database and Analysis Toolbox (GENEVESTIGATOR) [##REF##15375207##15##] help to seek shared biological roles based on correlation. These databases utilize existing methods like Pearson correlation and Spearman rank correlation to measure coregulation. These clustering methods provide reliable information for performing <italic>internal comparison </italic>of experimental conditions. But the usage of these for <italic>cross comparisons </italic>of various groups/clusters obtained through clustering various experimental conditions tends to obscure information important for identification of coregulated genes [##UREF##1##8##]. This implies that the clustering methods, which have shortcomings in identifying all the relationships existing in microarray expressions and different algorithms, will identify unique relationships thereby limiting them to the constraints of conditional coregulation [##UREF##1##8##,##REF##11099257##5##]. Fuzzy k-means clustering implementation recognizes the concept of conditional coregulation and assigns 'membership' to each gene belonging to various clusters/groups [##REF##12429058##3##].</p>", "<p>We tried to incorporate the conditional coregulation for calculating relationships among genes by mining transcriptional data consisting of experiments in various conditions from AtGenExpress. For each condition in a temporal microarray experiment, the state of the gene at a particular time point is defined by alphabet according to the algorithm. The individual output string comprising of concatenated alphabets is stored on a per gene, per condition basis, meaning the length of the string equal to the number of time points for a particular experimental condition. Therefore, the number of alphabets and the complexity of a generated string for the given temporal experiment increases with an increase in number of timepoints. This implies that the algorithm performs better and produces more reliable results as the number of timepoints increase in the experiment. We tried to further reduce random matching of similar alphabets (at the same timepoint) by introducing an option to award extra increments if the preceding alphabet is matched.</p>" ]

[ "<title>Methods</title>", "<title>Microarray datasets</title>", "<p>The dataset consists of a total of 18 groups of experiments which were already preprocessed in MAS5.0 <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.affymetrix.com\"/> taken from aboveground samples of the abiotic stress series of microarray experiments conducted by AtGenExpress <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.weigelworld.org/resources/microarray/AtGenExpress/\"/>.</p>", "<title>Test data</title>", "<p>The data consists of two sets of enzymes, i.e. nucleotide sugar interconversion enzymes [##REF##15134748##16##] and glycosyltransferases <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.CAZY.org\"/>, hypothesized to be involved in cell wall biosynthesis which consists of 493 genes for analyzing results.</p>", "<p>The overview of the entire analysis and database construction is shown in Figure ##FIG##0##1##.</p>" ]

[ "<title>Results and discussion</title>", "<p>Genes represented in the ATH1-1250 chip can be queried based on Affymetrix™ Probeid or AGICode [##REF##15173566##1##]. We utilised carbohydrate biosynthesis genes as a test case and at present, the database (hosted at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.asidb.com\"/>) refers to 493 genes annotated as glycosyltransferases and nucleotide sugar interconversion enzymes for <italic>Arabidopsis thaliana</italic>. The 'scores' generated by querying the database are retrieved in descending order of 'score' which we refer as 'rank' and used synonymously with 'score' for discussion. We have classified the results into two sections:</p>", "<title>A. Comparison of results with correlation coefficients derived from CSB.DB</title>", "<title>Single gene query</title>", "<p>We queried QUA1 (At3g25140) from our database and listed the top 15 genes (Table ##TAB##0##1##). We merged our results with CSB.DB by utilizing parametric Pearson's linear product moment correlation coefficient and the output using positive co-responding genes with probability &lt; 0.05 by performing single gene query for QUA1 for atge0200 dataset. We found that many genes which generate high scores in ASIDB also have high confidence and high pearson in CSB.DB. Also, our results do not exclude genes unlike CSB.DB for the same dataset giving a more comprehensive picture for better analysis.</p>", "<title>Subgroup (five UGE isoforms) comparison</title>", "<p>Performing co-response analysis for UGE isoforms by comparing fluctuation of transcript abundance between CSB.DB (Spearman correlation) and ASIDB revealed UGE1 and -3 behaved differently than UGE2, -4, -5 [##REF##16644739##19##].</p>", "<title>Random comparison of genes from CSB.DB</title>", "<p>We randomly selected around 10% of the genes (~52 genes) from our database and compared them with Pearson correlation generated from CSB.DB. The correlation coefficient measures the relationship between two variables ranging between +1 and -1 where +0.7 to +1.0 is considered strong positive association, +0.3 to +0.7 as weak positive association and +0.3 to 0 as no association [##UREF##6##20##]. Comparing 1326 correlated pairs generated from CSB.DB and ASIDB, we found that at high pearson correlation, i.e. .8, out of 14 correlated pairs generated by CSB.DB, 10 were identified by ASIDB at rank cutoff of 10 genes (top10 genes) and 13 correlated pairs at rank cutoff of 20 (Figure ##FIG##1##2##). Similarly for pearson correlation cutoff at .7, we were able to identify half (21) of the correlated pairs at rank cutoff of 10 and 30 at rank cutoff of 20. We observe the trend that ASIDB identifies most of the correlated pairs at high pearson value and identification reduces with declining pearson for the same genes. ASIDB also generates a higher number of relationships that are not identified by pearson correlation and might hold biological importance. The comparison file can be downloaded from <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.asidb.com/fDownloads.html\"/>.</p>", "<title>B. Biological validity of the results</title>", "<p>We utilized genes hypothesized to be involved in cell wall biosynthesis (glcosyltranferases and nucleotide interconversion enzyme) as test data. Out of these, around 420 genes that are present in carbohydrate active enzymes <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.cazy.org\"/> are glycosyltransferases while the remaining genes are linked with nucleotide sugar interconversion enzymes. The glycosyltransferases are specific for both the donor sugar nucleotide and the acceptor molecule, which might be another sugar or aglycones. One glycosyltransferase usually catalzses the formation of only one glycosidic linkage. As a result, though many glycosyltransferases catalyze chemically similar reactions, they display remarkable diversity in their donor, acceptor and product specificity and thereby generate a potentially infinite number of glycoconjugates, oligosaccharides and polysaccharides [##REF##12691742##21##]. The high specificity of glycosyltransferases results in difficulty in assessing the biochemical function of the enzymes encoded by these genes [##REF##11312132##22##]. Furthermore, the application of clustering or any other technique has not yielded a precise donor, acceptor or product specificity of glycosyltransferases [##REF##15134749##23##].</p>", "<p>The activated sugars, known as nucleotide sugars, form the substrates for nucleotide sugar interconversion enzymes dedicated to the generation of new sugar species [##UREF##7##24##]. The biochemical and molecular aspects of sugar nucleotide interconversion enzymes are fairly well understood [##REF##11554483##25##]. Also, nucleotide sugars are the substrates for glycosyltransferases that catalyze the polymerization of monosaccharides into glycosides, oligosaccharides, glycolipids, glycogen, starch, cellulose and a large variety of extracellular complex carbohydrates [##UREF##7##24##,##REF##14659703##26##]. An improved description of the link between nucleotide sugar interconversion genes and glycosyltransferases might help in understanding the control of cell wall biosynthesis [##REF##15134748##16##].</p>", "<title>Relationship of UGE with glycosyltransferase</title>", "<p>On querying five UGE isoforms (Table ##TAB##1##2##), we subgrouped them into two groups based on their relationship with each other and glcosyltransferases MUR3 (At2g20370), GOLS2 (At1g56600) and ATGT18 (At5g62220).</p>", "<p>i) UGE1, -3</p>", "<p>UGE1 and UGE3 co-regulated with trehalose 6-phosphate synthases [##REF##9681010##27##] indicating their catabolic role [##REF##16644739##19##,##UREF##8##28##].</p>", "<p>ii) UGE2, -4, -5</p>", "<p>We observed that UGE2,-4 and -5 co-regulated with known galactosyltransferases like MUR3, GolS2, ATGT18 indicating biosynthesis role [##REF##16644739##19##,##UREF##8##28##].</p>", "<title>Relationship among cellulose synthases</title>", "<p>The Arabidopsis genome encodes 10 isoforms of the cellulose synthase catalytic subunit – CESA [##REF##11842152##29##] and we have broken down these groups of genes into two sets based on the rank derived from Table ##TAB##2##3##.</p>", "<p>i) AtCesa1, AtCesa3, AtCesa6, AtCesa5, AtCesa2</p>", "<p>Cesa1,-3,-6 are responsible for cellulose production during primary cell wall development in various tissues [##REF##15932943##30##]. On querying ASIDB, we found that not only CESA2 which have been earlier found to be highly co-regulated with CESA1,-3,-6 came at higher rank [##REF##15932943##30##], but CESA5 also shows co-regulation with this set of genes indicating its role in deposition of cellulose in the primary cell wall.</p>", "<p>ii) AtCesa4, AtCesa 7, AtCesa 8</p>", "<p>The irx1, irx3 and irx5 mutants are the members of CesA gene family and AtCesA4 (IRX5), AtCesA7 (IRX3) and AtCesA8 (IRX1) take part in the synthesis of the complex that is required to synthesize cellulose in the secondary cell wall [##REF##15932943##30##,##REF##15980264##31##]. Other genes, which have shown significant relationship with these genes and have also been mentioned in previous work, are At5g54690, At2g37090 [##REF##15932943##30##].</p>" ]

[ "<title>Results and discussion</title>", "<p>Genes represented in the ATH1-1250 chip can be queried based on Affymetrix™ Probeid or AGICode [##REF##15173566##1##]. We utilised carbohydrate biosynthesis genes as a test case and at present, the database (hosted at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.asidb.com\"/>) refers to 493 genes annotated as glycosyltransferases and nucleotide sugar interconversion enzymes for <italic>Arabidopsis thaliana</italic>. The 'scores' generated by querying the database are retrieved in descending order of 'score' which we refer as 'rank' and used synonymously with 'score' for discussion. We have classified the results into two sections:</p>", "<title>A. Comparison of results with correlation coefficients derived from CSB.DB</title>", "<title>Single gene query</title>", "<p>We queried QUA1 (At3g25140) from our database and listed the top 15 genes (Table ##TAB##0##1##). We merged our results with CSB.DB by utilizing parametric Pearson's linear product moment correlation coefficient and the output using positive co-responding genes with probability &lt; 0.05 by performing single gene query for QUA1 for atge0200 dataset. We found that many genes which generate high scores in ASIDB also have high confidence and high pearson in CSB.DB. Also, our results do not exclude genes unlike CSB.DB for the same dataset giving a more comprehensive picture for better analysis.</p>", "<title>Subgroup (five UGE isoforms) comparison</title>", "<p>Performing co-response analysis for UGE isoforms by comparing fluctuation of transcript abundance between CSB.DB (Spearman correlation) and ASIDB revealed UGE1 and -3 behaved differently than UGE2, -4, -5 [##REF##16644739##19##].</p>", "<title>Random comparison of genes from CSB.DB</title>", "<p>We randomly selected around 10% of the genes (~52 genes) from our database and compared them with Pearson correlation generated from CSB.DB. The correlation coefficient measures the relationship between two variables ranging between +1 and -1 where +0.7 to +1.0 is considered strong positive association, +0.3 to +0.7 as weak positive association and +0.3 to 0 as no association [##UREF##6##20##]. Comparing 1326 correlated pairs generated from CSB.DB and ASIDB, we found that at high pearson correlation, i.e. .8, out of 14 correlated pairs generated by CSB.DB, 10 were identified by ASIDB at rank cutoff of 10 genes (top10 genes) and 13 correlated pairs at rank cutoff of 20 (Figure ##FIG##1##2##). Similarly for pearson correlation cutoff at .7, we were able to identify half (21) of the correlated pairs at rank cutoff of 10 and 30 at rank cutoff of 20. We observe the trend that ASIDB identifies most of the correlated pairs at high pearson value and identification reduces with declining pearson for the same genes. ASIDB also generates a higher number of relationships that are not identified by pearson correlation and might hold biological importance. The comparison file can be downloaded from <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.asidb.com/fDownloads.html\"/>.</p>", "<title>B. Biological validity of the results</title>", "<p>We utilized genes hypothesized to be involved in cell wall biosynthesis (glcosyltranferases and nucleotide interconversion enzyme) as test data. Out of these, around 420 genes that are present in carbohydrate active enzymes <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.cazy.org\"/> are glycosyltransferases while the remaining genes are linked with nucleotide sugar interconversion enzymes. The glycosyltransferases are specific for both the donor sugar nucleotide and the acceptor molecule, which might be another sugar or aglycones. One glycosyltransferase usually catalzses the formation of only one glycosidic linkage. As a result, though many glycosyltransferases catalyze chemically similar reactions, they display remarkable diversity in their donor, acceptor and product specificity and thereby generate a potentially infinite number of glycoconjugates, oligosaccharides and polysaccharides [##REF##12691742##21##]. The high specificity of glycosyltransferases results in difficulty in assessing the biochemical function of the enzymes encoded by these genes [##REF##11312132##22##]. Furthermore, the application of clustering or any other technique has not yielded a precise donor, acceptor or product specificity of glycosyltransferases [##REF##15134749##23##].</p>", "<p>The activated sugars, known as nucleotide sugars, form the substrates for nucleotide sugar interconversion enzymes dedicated to the generation of new sugar species [##UREF##7##24##]. The biochemical and molecular aspects of sugar nucleotide interconversion enzymes are fairly well understood [##REF##11554483##25##]. Also, nucleotide sugars are the substrates for glycosyltransferases that catalyze the polymerization of monosaccharides into glycosides, oligosaccharides, glycolipids, glycogen, starch, cellulose and a large variety of extracellular complex carbohydrates [##UREF##7##24##,##REF##14659703##26##]. An improved description of the link between nucleotide sugar interconversion genes and glycosyltransferases might help in understanding the control of cell wall biosynthesis [##REF##15134748##16##].</p>", "<title>Relationship of UGE with glycosyltransferase</title>", "<p>On querying five UGE isoforms (Table ##TAB##1##2##), we subgrouped them into two groups based on their relationship with each other and glcosyltransferases MUR3 (At2g20370), GOLS2 (At1g56600) and ATGT18 (At5g62220).</p>", "<p>i) UGE1, -3</p>", "<p>UGE1 and UGE3 co-regulated with trehalose 6-phosphate synthases [##REF##9681010##27##] indicating their catabolic role [##REF##16644739##19##,##UREF##8##28##].</p>", "<p>ii) UGE2, -4, -5</p>", "<p>We observed that UGE2,-4 and -5 co-regulated with known galactosyltransferases like MUR3, GolS2, ATGT18 indicating biosynthesis role [##REF##16644739##19##,##UREF##8##28##].</p>", "<title>Relationship among cellulose synthases</title>", "<p>The Arabidopsis genome encodes 10 isoforms of the cellulose synthase catalytic subunit – CESA [##REF##11842152##29##] and we have broken down these groups of genes into two sets based on the rank derived from Table ##TAB##2##3##.</p>", "<p>i) AtCesa1, AtCesa3, AtCesa6, AtCesa5, AtCesa2</p>", "<p>Cesa1,-3,-6 are responsible for cellulose production during primary cell wall development in various tissues [##REF##15932943##30##]. On querying ASIDB, we found that not only CESA2 which have been earlier found to be highly co-regulated with CESA1,-3,-6 came at higher rank [##REF##15932943##30##], but CESA5 also shows co-regulation with this set of genes indicating its role in deposition of cellulose in the primary cell wall.</p>", "<p>ii) AtCesa4, AtCesa 7, AtCesa 8</p>", "<p>The irx1, irx3 and irx5 mutants are the members of CesA gene family and AtCesA4 (IRX5), AtCesA7 (IRX3) and AtCesA8 (IRX1) take part in the synthesis of the complex that is required to synthesize cellulose in the secondary cell wall [##REF##15932943##30##,##REF##15980264##31##]. Other genes, which have shown significant relationship with these genes and have also been mentioned in previous work, are At5g54690, At2g37090 [##REF##15932943##30##].</p>" ]

[ "<title>Conclusion</title>", "<p>We tried to incorporate the conditional co-regulation for calculating the relationship among genes by addressing the limitation of clustering methods in which genes that are expressed in most of the measurements are highlighted while genes that are co-expressed in the subset of conditions are omitted. Our implementation leads to the inclusion of every gene regardless of its expression values thereby highlighting an important relationship absent in other contemporary databases. We found that our approach not only recognizes most of the randomly selected correlated pairs from pearson correlation in CSB.DB, but also generates new relationships. This has resulted in highlighting subtle relationships for example UGE isoforms [##REF##16644739##19##,##UREF##8##28##].</p>", "<p>Taking carbohydrate metabolism as a test case, we observed that those genes known to be involved in similar functions and pathways generate a high 'score' with the queried gene. The 'score' is therefore computed for all the genes present in the database and reflects the magnitude of co-regulation existing with the queried gene. The higher the intersection of expressional patterns under varying conditions, the higher the score generated for the gene calculated by the algorithm (Figure ##FIG##2##3##) and ranked in descending order of 'score'. Interpretation of the results is done by considering genes as nodes linked with each other through the edges [##REF##12540298##32##]. Edges represent interactions between the connected genes, with higher rank/score depicting higher functional similarity (Figure ##FIG##3##4##). We can utilize the interaction of edges and nodes for the construction of the networks.</p>", "<p>After implementing the carbohydrate biosynthesis cycle, we intend to incorporate other cycles like amino acid, nucleotide, and lipid. We have recently added ARACYC <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.arabidopsis.org\"/> annotation to our database.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Most existing transcriptional databases like Comprehensive Systems-Biology Database (CSB.DB) and Arabidopsis Microarray Database and Analysis Toolbox (GENEVESTIGATOR) help to seek a shared biological role (similar pathways and biosynthetic cycles) based on correlation. These utilize conventional methods like Pearson correlation and Spearman rank correlation to calculate correlation among genes. However, not all are genes expressed in all the conditions and this leads to their exclusion in these transcriptional databases that consist of experiments performed in varied conditions. This leads to incomplete studies of co-regulation among groups of genes that might be linked to the same or related biosynthetic pathway.</p>", "<title>Results</title>", "<p>We have implemented an alternate method based on graph theory that takes into consideration the biological assumption – conditional co-regulation is needed to mine a large transcriptional data bank and properties of microarray data. The algorithm calculates relationships among genes by converting discretized signals from the time series microarray data (AtGenExpress) to output strings. A 'score' is generated by using a similarity index against all the other genes by matching stored strings for any gene queried against our database.</p>", "<p>Taking carbohydrate metabolism as a test case, we observed that those genes known to be involved in similar functions and pathways generate a high 'score' with the queried gene. We were also able to recognize most of the randomly selected correlated pairs from Pearson correlation in CSB.DB and generate a higher number of relationships that might be biologically important. One advantage of our method over previously described approaches is that it includes all genes regardless of its expression values thereby highlighting important relationships absent in other contemporary databases.</p>", "<title>Conclusion</title>", "<p>Based on promising results, we understand that incorporating conditional co-regulation to study large expression data helps us identify novel relationships among genes. The other advantage of our approach is that mining expression data from various experiments, the genes that do not express in all the conditions or have low expression values are not excluded, thereby giving a better overall picture. This results in addressing known limitations of clustering methods in which genes that are expressed in only a subset of conditions are omitted.</p>", "<p>Based on further scope to extract information, ASIDB implementing above described approach has been initiated as a model database. ASIDB is available at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.asidb.com\"/>.</p>" ]

[ "<title>Setting up database</title>", "<title>Storage of preprocessed data</title>", "<p>We utilized Cyber-T <ext-link ext-link-type=\"uri\" xlink:href=\"http://CYBERt.microarray.ics.uci.edu\"/> to measure the confidence value associated with fold change for each gene. The Cyber-T analysis (Control versus Experiment) uses Bayesian probabilistic framework to calculate a background variance for each of the genes under analysis. By combining the empirical variance with the local background variance associated with neighboring genes, it calculates the confidence associated with the differential expression [##REF##18629084##17##,##REF##11395427##11##]. This is supposed to compensate for the limited number of replicates by giving proper estimates of variances (which might be biologically relevant).</p>", "<title>Generation of string</title>", "<p>The input matrices f₁f₂f₃f_nfor fold change and p₁p₂p₃p_nfor confidence value are generated from the analysis of Cyber-T, for each Control + Experiment. The fold change is the ratio of expression value between experiment and control, and negative fold change indicates lower expression in the experiment and vice versa. Discretization is utilized to reduce the variables in sample space resembling <italic>\"lossy compression method for the data</italic>\" [##UREF##1##8##]. Relative gene expression of every gene in each column (control plus experiment) is discretized into three distinct levels of differential expression – overexpressed, underexpressed and not expressed. These levels of differential expression are validated by the following rules:</p>", "<p>i. An upregulated gene with a fold change (Expression_Experiment/Expression_Control) greater than +1 and <italic>p </italic>value less than the threshold is deemed as overexpressed and represented as positive state.</p>", "<p>ii. A downregulated gene with a fold change (Expression_Control/Expression_Experiment) less than -1 and <italic>p </italic>value less than the threshold is categorized as underexpressed and represented as negative state.</p>", "<p>iii. A gene with p value above the threshold is categorized as not expressed and represented as neutral state (0).</p>", "<p>The algorithm utilizes an assumption that genes in a time series do not occupy discrete, independent expression devoid of any relationship with their previous experiments in a temporal series. In fact, the present state of each gene is dependent on its immediate preceding <italic>state or vector </italic>for magnitude and therefore generates an output string utilizing the graph theory [##UREF##5##18##] to depict a pattern. The algorithm uses these three levels of discretization to set the present state of each gene based on present and previous states.</p>", "<p>The implementation of the algorithm is discussed as pseudo code below. The initial vector (state) of the gene is set by the above rules of discretization, which states the direction is positive or negative based on the level of expression and magnitude, is equal to the number of timepoint experiments as in line 6. Lines 10 to 21 of the algorithm describe the transition of genes in a timepoint experiment with resulting variable 'out'. The out variable in the algorithm is based on both the direction and magnitude of its previous state and present state as in lines 12, 14, 17 and 19. Each 'out' variable can be considered as an alphabet representing the state of the gene and stored in the matrix on a per gene, per experiment basis.</p>", "<title>Implementation of the algorithm</title>", "<p><bold>a) i </bold>is the number of the gene in the row</p>", "<p><bold>b) j </bold>is the number of the column (experiment)</p>", "<p><bold>c) |len| </bold>is the initial output assigned to the output string equal to number of time points for time series experiments</p>", "<p><bold>d) state </bold>is level of differential expression, i.e. overexpression, underexpression or no expression for a gene at a timepoint</p>", "<p><bold>e) out </bold>is the output alphabet for the transition of each gene i in j^thcolumn</p>", "<p><bold>f) PV(i, j) </bold>is the matrix of confidence value(Bayes P) of i gene in j^thcolumn</p>", "<p><bold>g) FC(i, j) </bold>is the matrix of Fold Change of i gene in j^thcolumn</p>", "<p><bold>h) DM(i, j) </bold>is the matrix storing the output alphabet (out) as string</p>", "<p><bold>i) threshold </bold>is the user defined Bayes P value cutoff which has been taken as 5% arbitrarily</p>", "<p>1. iterate for each gene: 0 to i {</p>", "<p>2. iterate for each time series experiment: 0 to j {</p>", "<p>3. if (PV(i, j) &lt;= threshold) {</p>", "<p>4. if (previous state !exists) { # true for first column</p>", "<p>5. if (present state = (overexpressed | underexpressed)) {</p>", "<p>6. out = ± len; DM(i, j) = out; } #+ overexpression, - underexpression</p>", "<p>7. else if (present state = (not expressed)) {</p>", "<p>8. out = 0; DM(i, j) = out; } }</p>", "<p>9. else if (previous state exists) {</p>", "<p>10. if (present state = previous state) {</p>", "<p>11. if (FC(i, j) &gt; FC(i, j-1) {</p>", "<p>12. increment out; DM(i, j) = ± out; } #direction (±) based on present state</p>", "<p>13. else {</p>", "<p>14. decrement out; DM(i, j) = ± out; } }</p>", "<p>15. else if (present state != previous state) {</p>", "<p>16. if (FC(i, j) &gt; FC(i, j-1) {</p>", "<p>17. increment out; DM(i, j) = +out; }</p>", "<p>18. else {</p>", "<p>19. decrement out; DM(i, j) = -out; } }</p>", "<p>20. } # 9 ends</p>", "<p>21. } # 3 ends</p>", "<p>22. else {</p>", "<p>23. out = 0; DM(i, j) = out; }</p>", "<p>24. } } # 1 ends</p>", "<p>Note: # is the comment</p>", "<p>The total number of alphabets generated for a time series n is 3+4^(n-1), and the number of strings (complexity) resultant from these alphabets for a time series n greater than 1 is 3^(n-1)+2^(2n-1). For each individual experiment, a seperate string is generated with length equal to the timepoint of each experiment. These input values produce sequences of transition of a gene resulting in an output string o₁o₂o₃...o_nrepresenting the expressional changes a gene has undergone that are stored in the database.</p>", "<title>Storage of generated string</title>", "<p>The strings generated from the algorithm, for the time series experiment under different experimental conditions are stored in a RDBMS database. Each individual string is generated for a particular gene in a particular time series experiment and is stored separately.</p>", "<title>Generating a score through querying</title>", "<p>The score is generated by matching strings of query gene with all the other genes in the database. For each experiment, the string of query gene is compared with strings of other genes, per experiment. The comparison is performed for all the experiments and aggregate score is computed. The string matching comprises of two determinants as follows:</p>", "<title>Match</title>", "<p>An alphabet of query string matching with alphabet of other genes at same temporal point of a time series experiment is awarded a unit.</p>", "<title>Weight</title>", "<p>Discretization leads to generalization of entire data. To overcome this aspect, any random similarity of the output alphabet is checked by further awarding an extra unit to any matching output alphabet having a preceding match. Weight provides an additional thrust to seek for genes undergoing similar patterns by separating them from genes with random similarity.</p>", "<p></p>", "<p>T: is the number of time series experiments</p>", "<p>E: is the number of experiments conducted in various conditions</p>", "<p>Gn: is the value calculated in relation to a gene against a query gene using match and weight factors.</p>", "<p>The web interface ASIDB at present defaults match and weight as a unit. The standalone java interface with a local database has a more dynamic interface allowing the user to enter these variables.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>AR conceived the datamining project, wrote the algorithm, created the database, initiated ASIDB and wrote the manuscript. YD coordinated the project and revised the manuscript. All authors read and agree to publish the manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>AR acknowledges thanks to Dr Georg Seifert for involvement in the data mining project at John Innes Centre, UK and his useful inputs in writing the manuscript. AR is thankful to UCOST (grant UCS&amp;T/R&amp;D/LS-09/06) for supporting the datamining project leading to ASIDB. The authors thank Mississippi Functional Genomics Network (DHHS/NIH/NCRR grant #2PORR016476-04) for providing the support.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Overview of the database construction</bold>. E₁, E₂, E₃, E_nis the representation of the time points in temporal experiments. G₁, G₂, G₃, G_nis the representation of the gene. Confidence values and fold change for relative gene expression generated from Cyber-T.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Comparison of pearson correlation with ASIDB score</bold>. Graph depicts the number of correlated pairs at different pearson cutoff and the number of those same correlated pairs identified by ASIDB at different rank cutoff.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Generation of scores from different experimental conditions</bold>. The generation of a score resulting from the overlapping of similar expressional patterns under varying conditions with the query gene.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Depiction of network relationship between genes</bold>. The bi-directional relationship generated between gene A and gene B connected by edges. The edges represent score/rank by performing querying for each gene.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Comparison of ASIDB score with CSB.DB for single gene. Descending order of score from ASIDB compared with the pearson correlation and the p value from CSB.DB for QUA1 for atge0200</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Agicode</td><td align=\"left\">pearson coeff.</td><td align=\"left\">p value</td><td align=\"left\">Rank</td><td align=\"left\">Score</td><td align=\"left\">Name</td><td align=\"left\">Cazy Group</td></tr></thead><tbody><tr><td align=\"left\">At5g60920</td><td align=\"left\">.6424</td><td align=\"left\">3.15e-08</td><td align=\"left\">1</td><td align=\"left\">126</td><td align=\"left\">COB</td><td align=\"left\">-</td></tr><tr><td align=\"left\">At2g20370</td><td align=\"left\">.8077</td><td align=\"left\">6.44E-15</td><td align=\"left\">2</td><td align=\"left\">123</td><td align=\"left\">MUR3</td><td align=\"left\">GT47</td></tr><tr><td align=\"left\">At2g47650</td><td align=\"left\">.3248</td><td align=\"left\">.0113</td><td align=\"left\">3</td><td align=\"left\">116</td><td align=\"left\">AUD2 UXS4</td><td align=\"left\">-</td></tr><tr><td align=\"left\">At3g29360</td><td align=\"left\">.7636</td><td align=\"left\">0.00000000000129</td><td align=\"left\">4</td><td align=\"left\">116</td><td align=\"left\">UGD2</td><td align=\"left\">-</td></tr><tr><td align=\"left\">At5g15490</td><td align=\"left\">.4858</td><td align=\"left\">0.0000833</td><td align=\"left\">5</td><td align=\"left\">112</td><td align=\"left\">UGD3</td><td align=\"left\">-</td></tr><tr><td align=\"left\">At1g19360</td><td align=\"left\">**</td><td align=\"left\">-</td><td align=\"left\">6</td><td align=\"left\">112</td><td align=\"left\">-</td><td align=\"left\">GT74</td></tr><tr><td align=\"left\">At4g22580</td><td align=\"left\">.3035</td><td align=\"left\">0.0184</td><td align=\"left\">7</td><td align=\"left\">111</td><td align=\"left\">-</td><td align=\"left\">GT47</td></tr><tr><td align=\"left\">At1g80290</td><td align=\"left\">**</td><td align=\"left\">-</td><td align=\"left\">8</td><td align=\"left\">110</td><td align=\"left\">-</td><td align=\"left\">GT64</td></tr><tr><td align=\"left\">At5g39320</td><td align=\"left\">.3551</td><td align=\"left\">0.00537</td><td align=\"left\">9</td><td align=\"left\">110</td><td align=\"left\">UGD1</td><td align=\"left\">-</td></tr><tr><td align=\"left\">At1g16900</td><td align=\"left\">.3636</td><td align=\"left\">0.0043</td><td align=\"left\">10</td><td align=\"left\">109</td><td align=\"left\">-</td><td align=\"left\">GT22</td></tr><tr><td align=\"left\">At3g23820</td><td align=\"left\">.7862</td><td align=\"left\">9.88E-14</td><td align=\"left\">11</td><td align=\"left\">109</td><td align=\"left\">GAE6</td><td align=\"left\">-</td></tr><tr><td align=\"left\">At2g22900</td><td align=\"left\">.6505</td><td align=\"left\">0.0000000185</td><td align=\"left\">12</td><td align=\"left\">109</td><td align=\"left\">-</td><td align=\"left\">GT34</td></tr><tr><td align=\"left\">At1g53500</td><td align=\"left\">.6859</td><td align=\"left\">0.00000000147</td><td align=\"left\">13</td><td align=\"left\">109</td><td align=\"left\">RHM2</td><td align=\"left\">-</td></tr><tr><td align=\"left\">At1g08660</td><td align=\"left\">**</td><td align=\"left\">-</td><td align=\"left\">14</td><td align=\"left\">105</td><td align=\"left\">-</td><td align=\"left\">GT29</td></tr><tr><td align=\"left\">At1g06000</td><td align=\"left\">**</td><td align=\"left\">-</td><td align=\"left\">15</td><td align=\"left\">104</td><td align=\"left\">-</td><td align=\"left\">GT1</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Comparison of ASIDB ranking with pearson correlation for UGE homologs. The ranking of genes depicting relationship between UGE homologs and glycosyltransferases with pearson correlation is in parenthesis.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">RANK</td><td align=\"center\">MUR3</td><td align=\"center\">AtGT18</td><td align=\"center\">UGE4</td><td align=\"center\">UGE2</td><td align=\"center\">UGE5</td><td align=\"center\">UGE1</td><td align=\"center\">UGE3</td></tr></thead><tbody><tr><td align=\"left\">UGE1</td><td align=\"center\">477 (.4408)</td><td align=\"center\">460 (.2894)</td><td align=\"center\">478 (.2306)</td><td align=\"center\">473 (.3102)</td><td align=\"center\">374 (NA)</td><td align=\"center\">-</td><td align=\"center\">10 (.6303)</td></tr><tr><td align=\"left\">UGE2</td><td align=\"center\">14 (.5818)</td><td align=\"center\">42 (.2842)</td><td align=\"center\">97 (.1159)</td><td align=\"center\">-</td><td align=\"center\">9 (NA)</td><td align=\"center\">473 (.3102)</td><td align=\"center\">340 (.6475)</td></tr><tr><td align=\"left\">UGE3</td><td align=\"center\">474 (-.424)</td><td align=\"center\">461 (.2785)</td><td align=\"center\">404 (.1084)</td><td align=\"center\">340 (.6475)</td><td align=\"center\">402 (NA)</td><td align=\"center\">4 (.6303)</td><td align=\"center\">-</td></tr><tr><td align=\"left\">UGE4</td><td align=\"center\">1 (.0748)</td><td align=\"center\">291 (.3508)</td><td align=\"center\">-</td><td align=\"center\">131 (.1159)</td><td align=\"center\">43 (NA)</td><td align=\"center\">483 (.2306)</td><td align=\"center\">431 (.1084)</td></tr><tr><td align=\"left\">UGE5</td><td align=\"center\">22 (NA)</td><td align=\"center\">57 (NA)</td><td align=\"center\">52 (NA)</td><td align=\"center\">13 (NA)</td><td align=\"center\">-</td><td align=\"center\">365 (NA)</td><td align=\"center\">409 (NA)</td></tr><tr><td align=\"left\">MUR3</td><td align=\"center\">-</td><td align=\"center\">63 (.3995)</td><td align=\"center\">13 (.0748)</td><td align=\"center\">46 (.5818)</td><td align=\"center\">66 (NA)</td><td align=\"center\">484 (.4408)</td><td align=\"center\">482 (-.424)</td></tr><tr><td align=\"left\">QUA</td><td align=\"center\">2 (.8077)</td><td align=\"center\">159 (.1981)</td><td align=\"center\">35 (.1153)</td><td align=\"center\">118 (.5349)</td><td align=\"center\">226 (NA)</td><td align=\"center\">447 (.3241)</td><td align=\"center\">486 (.3395)</td></tr><tr><td align=\"left\">GOLS2</td><td align=\"center\">34 (NA)</td><td align=\"center\">46 (NA)</td><td align=\"center\">51 (NA)</td><td align=\"center\">1 (NA)</td><td align=\"center\">8 (NA)</td><td align=\"center\">465 (NA)</td><td align=\"center\">324 (NA)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Comparison of ASIDB ranking with pearson correlation for CESA homologs. The ranking of genes depicting relationship between CESA subgroup derived from the score with the pearson correlation in parenthesis.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">RANK</td><td align=\"left\">CESA1</td><td align=\"left\">CESA2</td><td align=\"left\">CESA3</td><td align=\"left\">CESA5</td><td align=\"left\">CESA6</td><td align=\"left\">CESA4</td><td align=\"left\">CESA7</td><td align=\"left\">CESA8</td></tr></thead><tbody><tr><td align=\"left\">CESA1</td><td align=\"left\">-</td><td align=\"left\">53 (.4319)</td><td align=\"left\">1 (.8554)</td><td align=\"left\">25 (.4822)</td><td align=\"left\">2 (.7231)</td><td align=\"left\">155 (.3368)</td><td align=\"left\">152 (.2495)</td><td align=\"left\">439 (.3224)</td></tr><tr><td align=\"left\">CESA3</td><td align=\"left\">2 (.8554)</td><td align=\"left\">9 (.5283)</td><td align=\"left\">-</td><td align=\"left\">7 (.5252)</td><td align=\"left\">5 (.7179)</td><td align=\"left\">312 (.1678)</td><td align=\"left\">219 (.1093)</td><td align=\"left\">350 (.1853)</td></tr><tr><td align=\"left\">CESA6</td><td align=\"left\">3 (.7231)</td><td align=\"left\">14 (.5684)</td><td align=\"left\">17 (.7179)</td><td align=\"left\">1 (.6678)</td><td align=\"left\">-</td><td align=\"left\">240 (.1605)</td><td align=\"left\">125 (-.0458)</td><td align=\"left\">365 (-.033)</td></tr><tr><td align=\"left\">CESA4</td><td align=\"left\">165 (.3368)</td><td align=\"left\">453 (-.2676)</td><td align=\"left\">368 (.1678)</td><td align=\"left\">287 (.022)</td><td align=\"left\">231 (.1605)</td><td align=\"left\">-</td><td align=\"left\">1 (.6776)</td><td align=\"left\">8 (.5777)</td></tr><tr><td align=\"left\">CESA7</td><td align=\"left\">118 (.2495)</td><td align=\"left\">376 (-.4548)</td><td align=\"left\">242 (.1093)</td><td align=\"left\">282 (-.0267)</td><td align=\"left\">73</td><td align=\"left\">1 (.6776)</td><td align=\"left\">-</td><td align=\"left\">2 (.7249)</td></tr><tr><td align=\"left\">CESA8</td><td align=\"left\">448 (.3224)</td><td align=\"left\">193 (-.186)</td><td align=\"left\">397 (.1853)</td><td align=\"left\">382 (.046)</td><td align=\"left\">323 (-.033)</td><td align=\"left\">4 (.5777)</td><td align=\"left\">1 (.7249)</td><td align=\"left\">-</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S7-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtext>Score</mml:mtext>\n <mml:mo>=</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mi>E</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mi>T</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:mi>G</mml:mi>\n <mml:mi>n</mml:mi>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>** Genes not present in CSB.DB (atge0200 matrix)</p></table-wrap-foot>", "<table-wrap-foot><p>*NA represents genes omitted in the CSB.DB (atge0200 matrix)</p></table-wrap-foot>", "<table-wrap-foot><p>*NA is genes omitted in the CSB.DB (atge0200 matrix)</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S7-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S7-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S7-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S7-4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Metastasis (originating from Greek <italic>μετισταναι</italic>, to change) is the single most important event changing the course of cancer from manageable to fatal. For metastasis to occur, tumor cells must acquire the ability to detach from the original tumor, relocate through the blood or lymphatic circulation and start a new colony in a different part of the organism [##REF##12494885##1##]. In spite of intensive research [##REF##12778135##2##, ####REF##887927##3##, ##REF##17229949##4##, ##REF##12673112##5##, ##REF##16531451##6##, ##REF##12469122##7##, ##REF##11823860##8##, ##REF##7563201##9##, ##REF##16491073##10##, ##REF##16889869##11####16889869##11##] there is no consensus regarding the origin of metastases. According to one model, metastatic transformation can be triggered in primary solid tumors by certain conditions, while another model links metastatic potential to a very rare subtype of tumor cells, occurring on the order of one in many millions. Genetic background is also viewed as an important determining factor in metastatic transformation [##REF##16491073##10##,##REF##16889869##11##]. This difference is important for both diagnostic and prognostic purposes.</p>", "<p>Early cancer is clinically heterogeneous, and many patients can have an \"indolent\" disease course that does not significantly affect their survival. Once metastatic disease is documented clinically, the majority of patients die from their tumors as opposed to other causes [##REF##12154061##12##]. This has led some researchers to consider the disease as a series of \"states\" that include clinically localized tumors and those that have metastasized, as a framework to assess the clinical and biological factors associated with specific phenotypes and outcomes [##REF##10699601##13##]. However, there are other plausible concepts. Analysis of relations between different molecular subtypes of cancer and identification of genes specific to such subtypes is important for understanding the biological basis for this clinical heterogeneity and thus is critical in assessing prognosis, selecting therapy, and evaluating treatment effects. Metastatic transformation is a multi-stage process involving complex interactions between tumor cells and the host [##UREF##0##14##]. Cells from primary tumors must detach, invade stromal tissue, and penetrate blood or lymphatic vessels by which they disseminate. They must survive in the circulation to reach a secondary site in which they lodge because of physical size or binding to specific tissues. To form clinically significant tumors, metastatic cells must also adjust their metabolism and signaling systems to proliferate in the new microenvironment. Tumor cells growing at metastatic sites are continually selected for their growth advantage. This is a complex and dynamic process that is expected to involve alterations in many genes and transcriptional programs.</p>", "<p>Considerable amounts of gene expression data have been deposited in public databases and/or are available upon request from other investigators. Analysis of these data is generally limited to one set at a time. However, recent years have seen multiple attempts to conduct meta-analysis across independent data sets. Among the more successful of these is a study by Ramaswamy et al. of molecular signatures of metastasis in primary solid tumors aiming to elucidate the molecular nature of metastasis [##REF##12469122##7##]. The authors analyzed the gene-expression profiles of 12 metastatic adenocarcinoma nodules of diverse origin (lung, breast, prostate, colorectal, uterus, ovary) and compared them with the expression profiles of 64 primary adenocarcinomas representing the same spectrum of tumor types obtained from different individuals. They identified 128 genes differentially expressed between primary and metastatic adenocarcinomas. A similar pattern was found in some primary tumors, which suggests that a gene expression program for metastatic transformation is present in some primary solid tumors. Further refining produced a subset of 17 unique genes that the authors presented as the most significant contributors to the difference between primary and metastatic tumors and thus the most likely diagnostic markers for the metastatic potential.</p>", "<p>In this work, we present an alternative analysis of gene expression data based on a holistic approach integrating fragmented biological evidence and strengthening the unreliable conclusions by bringing more data rather than cutting straight to a few most consistent observations. We start with the analysis of the same meta-set of metastatic and primary tumors utilized by Ramaswamy et al., but supplement the analysis by algorithms, methodology and data not available to the original authors.</p>", "<title>Data</title>", "<p><bold>The Ramaswamy <italic>et al. </italic>meta-set </bold>combines genes represented by different probes across multiple distinct microarray platforms (Affymetrix U95A, Hu6800 and Hu35K oligonucleotide microarrays as well as Rosetta inkjet microarrays) traced through by mapping to UniGene build #147. The data have been scaled to account for different microarray intensities in a given set. Each column (sample) has been multiplied in the data set by 1/slope of a least-squares linear fit of the sample versus a reference (the first sample in the data set) using only genes that had 'Present' calls in both the sample being re-scaled and the reference. A typical sample (that is, one with the closest number of 'Present' calls to the average over all samples in the data set) was chosen as reference.</p>", "<p>The authors performed thresholding using a ceiling of 16,000 units and a floor of 20 units then subjected gene-expression values to a variation filter that excluded genes with minimal variation across the samples being analyzed by testing for a fold-change and absolute variation over samples, comparing max/min and max - min with predefined values and excluding genes not obeying both conditions. The resulting data are available at <ext-link ext-link-type=\"uri\" xlink:href=\"http://wwwgenome.wi.mit.edu/cancer/solid_tumor_metastasis\"/>.</p>", "<title>Colorectal cancer data sets</title>", "<p>The GDS756 dataset provided insight the progression of cancer from primary tumor growth to metastasis by comparison of gene expression in SW480, a primary tumor colon cancer cell line, to that in SW620, an isogenic metastatic colon cancer cell line. Both cell lines were derived from one individual. The GDS1780 set reflects comparison of polysomal RNA from isogenic cell lines established from a colorectal cancer (CRC) patient [##REF##16531451##6##]. The cell lines constitute a cellular model of CRC transition from invasive carcinoma to metastasis. The RNA samples were submitted to microarray analysis using the HG-U133A chip from Affymetrix, (Santa Clara, CA). Three biological replicates were carried out for each cell line and six hybridized arrays obtained. Raw data were analyzed using two microarray analysis software packages, dChip (13) and R-Robust Microarray Analysis (R-RMA) (14). We have downloaded and used these data sets from GEO (GDS756 and GDS1780). Each data set contains 22283 features (probesets) and 6 columns (samples) representing two contrast classes, each with three replicate experiments.</p>", "<title>Breast cancer data</title>", "<p>The data set we used in this study was downloaded from the GEO database (GDS2617); it contains 22283 probe sets. Tumorigenic and non-tumorigenic breast cancer cells were compared. Tumorigenic breast cancer cells were considered those expressing cell-surface proteins CD44 and CD24. Tumorigenic breast cancer cells isolated from 6 individuals were compared with normal breast epithelium derived from 3 individuals. In terms adopted by the authors of the original paper [##REF##17229949##4##], tumorigenic cancer samples are those with invasive potential, resulting in metastatic progression.</p>" ]

[ "<title>Methods</title>", "<title>Overview of the analysis pipeline</title>", "<p>The general overview of the analysis pipeline is given in Supplemental Figure ##FIG##0##1## (Additional File ##SUPPL##0##1##). Our pipeline includes most of the standard analysis steps, but has a few important differences. We extend the analysis to maximize the advantage of pathway analysis. The genes important for understanding the biological processes involved in metastatic transformation are selected not solely by the difference in signal emitted by microarray probes. Instead, we concentrate on the \"group behavior\" of genes, their ability to interact and pre-existing annotation placing the genes into the same biological pathway, linking to the same cellular function. Thus, the inference is done with very liberal selection criteria and not adjusted for multiple testing. We select a large list of potentially differential genes which may contain a large number of false-positives. We then select biological pathways, molecular function and GO terms which are over-represented in the initial intensity-based list. The benefits of the use of pathway and ontological analyses of microarray data have been presented previously [##REF##12620386##15##,##REF##16895928##16##]. More recent GSEA [##REF##16199517##17##] and SAFE [##REF##15647293##18##] methods can be very effective in highlighting the joint effect of a group of genes which may not be significantly differential if considered one by one. However, these methods require additional assumptions that may not be correct in every study. The significance of biological pathways is estimated through a variation of Fisher's exact test as implemented in Metacore or IPA and adjusted for multiple testing using Benjamini-Hochberg FDR analysis (which is a build-in function of GeneGo Metacore software). Single genes that do not map into any statistically significant pathway (i.e. missing all regulators, downstream targets, ligands and other components necessary for a functional molecular mechanism) may be still considered significant if reproducible and independently validated in additional experiments. However, in our analysis pipeline we leave such genes out regardless of their individual difference between primary and metastatic samples in a particular experiment. Our approach is based on collective effects of the groups of genes interlinked by functional relationships, which is inapplicable to some genes lacking information on function, regulation and interaction with other genes.</p>", "<title>Normalization</title>", "<p>The data were normalized using a quantile algorithm similar to one described by Bolstad <italic>et al. </italic>[##REF##12538238##19##]. We applied our own C++ software for normalization, available from A. Ptitsyn upon request. Box-plots for pre-normalized and normalized expression value distributions are shown in Supplemental Figure ##FIG##1##2## (Additional file ##SUPPL##0##1##).</p>", "<title>Preliminary selection of differentially expressed genes</title>", "<p>A set of differentially expressed genes was selected using University of Pittsburgh Gene Expression Data Analysis suite (GEDA, <ext-link ext-link-type=\"uri\" xlink:href=\"http://bioinformatics.upmc.edu/GE2/GEDA.html\"/>). For selection, we applied the standard J5 metric with threshold 4 and optional 4 iteration of Jackknife procedure to reduce the number of false-positive differential genes [##REF##16323966##20##]. Both J5 metric and threshold parameter are standard pre-set values recommended by the developers. We did not attempt to estimate the confidence level of individual genes and used J5 not as a statistical test, but as a selection procedure providing a shortlist of genes deviating from the expected average value and enriched with differential genes. The MA plot showing selected differential genes is presented in Supplemental Figure ##FIG##2##3## (Additional File ##SUPPL##0##1##). Notably, the plot shows a balanced representation of moderately and highly expressed genes, i.e. the categories most appropriate for selection of diagnostic biomarkers. Application of selection procedures biased away from highly expressed genes may reveal truly differential genes, but fewer suitable biomarker candidates. We then applied DAVID web-based tools to perform functional annotation of all potentially differential genes selected by GEDA. The complete annotated lists for analyzed data sets are given in the Supplemental Materials (Supplemental Tables 2, 3, 4 and 5 found in Additional File ##SUPPL##0##1##).</p>", "<title>Functional annotation and pathway analysis</title>", "<p>Analysis of biological pathways was performed using MetaCore software (GeneGo Inc.), Ingenuity Pathways Analysis (Ingenuity Systems Inc.) and free DAVID tools [##REF##17784955##21##].</p>" ]

[ "<title>Results and discussion</title>", "<p>Analysis of the Ramaswamy <italic>et al</italic>. meta-set identified 741 genes differentially expressed between 64 primary solid tumor samples and 12 metastatic tumor samples. The complete list of these genes with functional annotation is given in Supplementary Table 1 (Additional File ##SUPPL##0##1##). As expected (see explanation in Methods section), this list is much larger than the original 128 genes identified by Ramaswamy et al. It is likely that there are some false positive differential genes mixed in, however the exact number is not relevant to the analysis. Instead, we focused on the biological function of the genes on the selected shortlist. This function can be estimated through the analysis of the biological pathways, canonic interaction maps and gene ontology categories found within the shortlist. Analysis of statistically overrepresented pathways in the shortlist of differential genes revealed 19 canonic pathway maps (by GeneGo Metacore version) with confidence level <italic>p </italic>= 0.05 (adjusted for FDR). The chart of overrepresented metabolic maps is given on Figure ##FIG##0##1##. Analysis of the same shortlist of differential genes with the DAVID Functional Classification tool [##REF##17784955##21##] also reveals 6 clusters of gene functions with 3 to 6 members-functional categories (GO terms, PIR keywords, etc.) significantly overrepresented with <italic>p </italic>&lt; 0.05 after FDR (Benjamini-Hochberg) adjustment. These results, presented in Supplemental Table 1 (Additional File ##SUPPL##0##1##), are based on algorithms and knowledge bases different from those of GeneGo Metacore. However, scrutinizing the contents of these results allows reconstruction of the underlying biological processes, which are common, robust and reproducible in experiments.</p>", "<p>The most remarkable among the pathways differentially represented between primary and metastatic tumors are extracellular matrix/cell adhesion/cytoskeleton remodeling and oxidative phosphorylation. The most common pathways break into three classes: a) related to remodeling of internal cellular structure; b) related to alterations in cellular metabolism; and c) alternations in cell surface, antigen presentation and cell adhesion. Pathways related to cell cycle regulation are also found among differential genes.</p>", "<p>Detailed analysis of each overrepresented pathway would be far beyond the scope of this study. However, it is appropriate to comment on key processes reflected by the metabolic maps.</p>", "<p>One of the most strongly and consistently altered pathways in all evaluated datasets involves glucose utilization, specifically down-regulation of major components of oxidative phosphorylation (Figure ##FIG##1##2##) and up-regulation of genes in the glycolytic pathway (Figure ##FIG##2##3##). Down-regulated genes included mitochontrial ATPase pathway members, cytochrome oxidase, and NADH dehydrogenases. The phenomenon of prefential use of glycolysis for ATP generation in tumors was first observed by Otto Warburg in the first half of the 20^thcentury (as early as 1925) [##UREF##1##22##, ####REF##13298683##23##, ##REF##13351639##24####13351639##24##]. However, more recent studies have demonstrated that exaggeration of the Warburg phenomenon through inhibition of mitochondrial function may promote metastasis via enhancement of tumor cell invasion and reduced sensitivity to apoptosis [##REF##12420221##25##, ####REF##10779348##26##, ##REF##10702275##27####10702275##27##]. Furthermore, other groups have recently demonstrated reduction in oxidative phosphorylation-related genes in metastases versus primary tumors using genomic methods [##REF##18097602##28##,##REF##17487416##29##] and correlation of reduction in ATP synthase function with outcome in patients with lung and colorectal cancer [##REF##14963017##30##,##REF##12438266##31##].</p>", "<p>In addition to providing a potential novel marker for metastatic potential, the broad conservation of alterations in bioenergetic pathways in metastatic tumors across tumor types and datasets suggests that interference with glycolytic pathways might be a viable therapeutic strategy for the prevention of metastasis. Glycolytic pathway analogs such as 2-deoxyglucose and 3-bromopyruvate are showing promise as therapeutic agents targeting hypoxic primary tumor cells [##REF##17595539##32##], but have been poorly evaluated as antimetastatic drugs. However, a recent study demonstrated inhibition of pancreatic cancer metastasis in mice treated with 3-bromopyruvate when combined with a heat shock protein 90 inhibitor [##UREF##2##33##]. Furthermore, epigenetic therapies such as histone deacetylase and DNA methyltransferase inhibitors have been shown to reactivate expression of oxidative phosphorylation genes [##REF##18226465##34##], conceivably reducing metastatic potential and suggesting that some alterations in this pathway may be epigenetically regulated.</p>", "<p>Another critical pathway in our analysis that was differentially expressed robustly in primary versus metastatic tumors involves the extracellular matrix, cell adhesion, adhesion-mediated signal transduction and cytoskeletal organization, all of which are cooperatively important in the metastatic cascade.</p>", "<p>Alterations in extracellular matrix proteins included reductions in collagen, fibronectin, and a shift in keratin isoform expression (Figure ##FIG##3##4##). These reductions in cell matrix proteins could theoretically facilitate cell motility and enhance extravasation. The cell adhesion molecules CD63 and CD151 were upregulated in metastatic tumors as well. Experimental and clinical literature demonstrates a role for CD151 in metastasis [##REF##16798740##35##,##REF##15533898##36##]</p>", "<p>Differential expression of some key proteins responsible for adhesion-mediated cell signaling (RhoA, talin, moesin, ezrin, SPARC) was also observed (Figure ##FIG##4##5##). Encouragingly, up-regulation of some well-characterized metastasis-associated genes such as RhoA and ezrin was observed. RhoA plays a key role in regulating the actin cytoskeleton and controlling cell motility, cell-cell interactions and intracellular trafficking [##REF##16900393##37##]. Upregulation of RhoA has been associated with metastasis and/or negative outcome in carcinomas of the liver, kidney, esophagus, and urinary tract [##REF##17914970##38##, ####REF##17597401##39##, ##REF##12855641##40####12855641##40##]. Upregulation of ezrin has been implicated in metastasis of diverse tumor types, such as osteosarcoma, soft-tissue sarcomas, pancreatic carcinoma, and head and neck carcinoma among others [##REF##18246799##41##, ####REF##17874463##42##, ##REF##16633060##43##, ##REF##16365240##44##, ##REF##16144921##45##, ##REF##14704791##46####14704791##46##].</p>", "<p>Significant upregulation of important cytoskeletal components such as actin, tubulin and vimentin was also observed. These proteins play a key role in cell motility, invasion, cell division and intracellular transport, and differential expression of these members has been implicated in human tumor progression as well [##REF##18324357##47##,##REF##18327819##48##]. Increased vimentin is a well-defined phenotypic indicator of epithelial-mesenchymal transition, which has a known association with carcinoma aggressiveness [##REF##18327819##48##].</p>", "<p>Several components of the extracellular matrix – cell adhesion – adhesion-mediated signaling – cytoskeleton pathway have the potential for \"druggability\". For example, small molecule inhibitors of RhoA are in development [##REF##18167222##49##,##REF##17699722##50##], and rapamycin and its analogs have been shown to inhibit the ezrin-associated metastatic phenotype through inhibition of downstream AKT-mTOR signaling [##REF##15781656##51##].</p>", "<p>The antigen presentation pathway in Figure ##FIG##5##6## also reflects, in part, cytoskeleton remodeling: metastatic samples show increased expression of beta-2-microtubulin in the endoplasmic reticulum. All other elements of the antigen presentation pathway found in the differential genes shortlist are down-regulated. Remarkably, the most down-stream elements of the pathway, the final effectors, are the most down-regulated. Immune avoidance is thought to be another key component in successful metastasis; tumor cells must be able to survive in the circulation and avoid immune destruction upon arrest in the end-organ. Furthermore, evidence exists for epigenetic suppression of antigen presentation in tumor cells, and potential reactivation of expression through drugs blocking histone deacetylase and/or DNA methyltransferase, leading to enhanced tumor cell immunogenicity [##REF##18070359##52##, ####REF##18046553##53##, ##REF##17725605##54####17725605##54##].</p>", "<p>How reproducible are the results of computational analysis of an artificial meta-set of primary and metastatic tumors? We cannot possibly repeat the sample collection, RNA extraction and hybridizations. However, since the time Ramaswamy et al. have published their results there have been quite a few publications reporting microarray analysis of primary vs. metastatic tumors, and the data are available from the public databases such as GEO <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/geo\"/>. We have extracted and analyzed a few of these new data sets [##REF##17229949##4##,##REF##16531451##6##] using the same analysis pipeline. The final results of these analyses are lists of statistically significant pathways, molecular functions and GO terms within the shortlist of potentially differential genes. These lists show remarkable agreement in all studies. Comparison of the pathways represented in these lists reveals none unique to any of the 3 data sets, only common and similar (Supplemental Figure ##FIG##3##4## in Additional File ##SUPPL##0##1##). Overall, the Ramaswamy et al. meta-set produces a shorter list of potentially differential genes and further analysis yields fewer significant pathways. This result is not surprising taking into account that many small differences reproducibly observed in each single-tissue experiment have been leveled in composing the meta-set. However, the essential features reflecting the metabolic changes between primary and metastatic tumors are apparent in every analyzed data set. The oxidative phosphorylation pathways with most components down-regulated, cytoskeleton remodeling and cell adhesion-related pathways are always found among the longer lists of significant pathways in the specific colon and breast cancer datasets. Remarkably, the suppressed oxidative phosphorylation pathway is always near the top of the most statistically significant pathways.</p>", "<p>Taken together, there are dramatic changes in gene expression between primary and metastatic tumors; some are quantitative whereas others reflect a new pattern of expression. But how consistently are those changes revealed by a loose non-parametric J5 procedure? This selection procedure gives no estimation of confidence level for the individual genes. In turn, estimation of significance for the biological pathways is very approximate at best: it does not fully account for interdependencies in gene expression. Pathway maps include genes arbitrarily and the database of gene interactions is filled manually by multiple experts scanning the peer-reviewed literature, i.e. prone to errors and contradictions. These databases and associated tools for pathway analysis have improved significantly in recent years, but quantitative estimation of pathway significance still needs additional validation. In order to select only the most reliably over-represented pathways, we performed a bootstrap analysis randomly re-sampling 50% of the short-listed genes. Comparative analysis of over-represented pathways in the randomly re-sampled and original shortlists is given in Supplemental Table 6 and Supplemental Figure ##FIG##4##5## (Additional File ##SUPPL##0##1##). The main pathways are remarkably robust. The genes (putative biomarkers) diagram is dominated by \"similar\" pathways, i.e. belonging to the same pathway map or involved in the same cellular function. There are also some \"common\" genes (i.e. genes representing the same pathway, which is still statistically significant in the randomly selected half-list) and no \"unique\" genes (i.e. representing unique, but statistically significant pathways). This observation leads to important conclusions: a) microarray experiments may yield extensive variation in specific differentially expressed genes, but are robust and reproducible in elucidating differentially expressed pathways; b) random re-sampling of the large list of differentially expressed genes provides no proof of true difference for any single gene, but the list in general has few (if any) false-positive genes. The latter statement is controversial since the common goal of the inference in microarray analysis is to reduce the dimensionality of the feature space and select a small number of truly differential genes. After selection of a shortlist using a <italic>t</italic>-test or one of its variants, the number of differentially expressed genes is further reduced by application of a False Discovery Rate procedure (typically Benjamini-Hochberg) [##UREF##3##55##,##REF##12710672##56##]. Some authors even claim that microarrays are not optimal for pathway analysis because of poor reproducibility of the resulting pathways [##UREF##4##57##]. Our study suggests the opposite. The previously discussed problem of pathway reproducibility is caused by the misconceived methodology, more specifically in the strategy of microarray data analysis. Apparently, applying stricter criteria for selection of differentially expressed genes results in a very small number of candidates that are further reduced by FDR adjustment. The few remaining candidate genes have a much better chance of being successfully reproduced in another microarray experiment and validated by other techniques such as real-time RT-PCR or immunohistochemistry. On the other hand, a shorter list of candidate genes undermines the basis for the pathway analysis, rendering overrepresentation statistics powerless. This may explain the poor reproducibility of pathway analysis in some studies [##UREF##4##57##]. Such a stringent approach to biomarker selection relies entirely on the signal intensity and associated statistics. This approach can be very effective in cases of lethal mutations, congenital disorders and other diseases caused by a single or few factors. However, in complex multifactorial diseases, the most highly expressed genes and most reproducible differences in gene expression often turn out to be non-specific final effectors, downstream of important switches and regulators in biological pathways. Cancer in general and metastasis in particular are the examples of such multifactorial diseases. Application of a systems biology approach, considering not just the effect of single mutated/healthy genes, but entire networks of interlinked and constantly interacting genes is required not only for understanding the mechanism of disease, but also for the selection of diagnostic and prognostic markers, as well as potential therapeutic targets. As we have demonstrated, the pathways are sufficiently reproducible and robust to serve this purpose. The prevailing methodology in microarray analysis has an internal contradiction: it calls for a strict selection of candidate genes that can be independently verified one by one, but systems biology calls for analysis of large numbers of genes. Furthermore, the number of replicates affordable for a typical microarray study is usually insufficient for reliable reproduction of expression in low-expressed genes. However, important biological functions specific to disease are often performed by low-expressed genes. Pathway analysis has the power to identify such signal transducers and key transcription factors only if a large enough number of candidate genes are input. To resolve this contradiction, we propose an extension of the current prevailing methodology.</p>", "<p>First, the analysis pipeline has to be extended to incorporate functional annotation and pathway analysis. Second, selection of the candidate genes cannot be performed based solely on the intensity of signal and its change in the experiment. Instead, we propose to consider this step a pre-selection and relax the criteria for \"differential\" genes. Third, FDR correction should not be applied to a pre-selected \"long list\" of candidate genes. Combined with a relaxed selection threshold, this will inevitably create an influx of false-positive genes, which can be addressed subsequently. Fourth, the \"long list\" is analyzed in order to identify statistically overrepresented biological pathways, GO terms, molecular functions (as implemented in DAVID, IPA and MetaCore software) and gene set enrichment (for example, using GSEA or SAFE methods [##REF##16199517##17##,##REF##15647293##18##]). It is at this stage of analysis that multiple testing adjustments (Bonferroni, or better FDR) should be applied. Most available software, both free (DAVID tools [##REF##17784955##21##]) and commercial (such as IPA and Metacore) have at least one method of false-positive control implemented. However, we still recommend additional techniques, such as the bootstrapping experiment described above, for computational validation of significant pathways. Finally, the discovered statistically significant pathways, gene sets and molecular functions should be used to reverse-engineer the molecular mechanism of disease and select one or more potential biomarkers and drug targets. In our approach, it is important to combine numeric analysis with biological reasoning and deduction.</p>", "<p>The proposed analysis strategy is not yet implemented in a single analysis tool, although all the components have been developed and some of the software packages (such as ArrayTrack [##REF##14630514##58##]) offer partial integration; pathway analysis packages, although independent, can be easily invoked from within the microarray analysis software. In the future, we would like to unite all the tools used for systems biology analysis of biomarkers in a single automated software pipeline.</p>", "<p>Systems biology approaches to analysis of existing public data reveal a large number of new features overlooked in the original analyses. Meta-analysis and cross-examination of a few data sets allows identification of prospective markers and drug targets. The present day databases available for systems biology empower the researchers beyond the dreams of only a few years ago. Now for each identified significant pathway, we may also correlate expression data with known conserved transcription factor binding sites, and employ siRNA-mediated gene knockdown and known pharmacologic inhibitors (pharmacoprobes) to interrogate the phenotypic effects of interference with identified pathways. The systems approach described here allows identification of a number of key pathways that may serve as therapeutic targets for controlling the metastatic transition of primary solid tumors.</p>" ]

[ "<title>Results and discussion</title>", "<p>Analysis of the Ramaswamy <italic>et al</italic>. meta-set identified 741 genes differentially expressed between 64 primary solid tumor samples and 12 metastatic tumor samples. The complete list of these genes with functional annotation is given in Supplementary Table 1 (Additional File ##SUPPL##0##1##). As expected (see explanation in Methods section), this list is much larger than the original 128 genes identified by Ramaswamy et al. It is likely that there are some false positive differential genes mixed in, however the exact number is not relevant to the analysis. Instead, we focused on the biological function of the genes on the selected shortlist. This function can be estimated through the analysis of the biological pathways, canonic interaction maps and gene ontology categories found within the shortlist. Analysis of statistically overrepresented pathways in the shortlist of differential genes revealed 19 canonic pathway maps (by GeneGo Metacore version) with confidence level <italic>p </italic>= 0.05 (adjusted for FDR). The chart of overrepresented metabolic maps is given on Figure ##FIG##0##1##. Analysis of the same shortlist of differential genes with the DAVID Functional Classification tool [##REF##17784955##21##] also reveals 6 clusters of gene functions with 3 to 6 members-functional categories (GO terms, PIR keywords, etc.) significantly overrepresented with <italic>p </italic>&lt; 0.05 after FDR (Benjamini-Hochberg) adjustment. These results, presented in Supplemental Table 1 (Additional File ##SUPPL##0##1##), are based on algorithms and knowledge bases different from those of GeneGo Metacore. However, scrutinizing the contents of these results allows reconstruction of the underlying biological processes, which are common, robust and reproducible in experiments.</p>", "<p>The most remarkable among the pathways differentially represented between primary and metastatic tumors are extracellular matrix/cell adhesion/cytoskeleton remodeling and oxidative phosphorylation. The most common pathways break into three classes: a) related to remodeling of internal cellular structure; b) related to alterations in cellular metabolism; and c) alternations in cell surface, antigen presentation and cell adhesion. Pathways related to cell cycle regulation are also found among differential genes.</p>", "<p>Detailed analysis of each overrepresented pathway would be far beyond the scope of this study. However, it is appropriate to comment on key processes reflected by the metabolic maps.</p>", "<p>One of the most strongly and consistently altered pathways in all evaluated datasets involves glucose utilization, specifically down-regulation of major components of oxidative phosphorylation (Figure ##FIG##1##2##) and up-regulation of genes in the glycolytic pathway (Figure ##FIG##2##3##). Down-regulated genes included mitochontrial ATPase pathway members, cytochrome oxidase, and NADH dehydrogenases. The phenomenon of prefential use of glycolysis for ATP generation in tumors was first observed by Otto Warburg in the first half of the 20^thcentury (as early as 1925) [##UREF##1##22##, ####REF##13298683##23##, ##REF##13351639##24####13351639##24##]. However, more recent studies have demonstrated that exaggeration of the Warburg phenomenon through inhibition of mitochondrial function may promote metastasis via enhancement of tumor cell invasion and reduced sensitivity to apoptosis [##REF##12420221##25##, ####REF##10779348##26##, ##REF##10702275##27####10702275##27##]. Furthermore, other groups have recently demonstrated reduction in oxidative phosphorylation-related genes in metastases versus primary tumors using genomic methods [##REF##18097602##28##,##REF##17487416##29##] and correlation of reduction in ATP synthase function with outcome in patients with lung and colorectal cancer [##REF##14963017##30##,##REF##12438266##31##].</p>", "<p>In addition to providing a potential novel marker for metastatic potential, the broad conservation of alterations in bioenergetic pathways in metastatic tumors across tumor types and datasets suggests that interference with glycolytic pathways might be a viable therapeutic strategy for the prevention of metastasis. Glycolytic pathway analogs such as 2-deoxyglucose and 3-bromopyruvate are showing promise as therapeutic agents targeting hypoxic primary tumor cells [##REF##17595539##32##], but have been poorly evaluated as antimetastatic drugs. However, a recent study demonstrated inhibition of pancreatic cancer metastasis in mice treated with 3-bromopyruvate when combined with a heat shock protein 90 inhibitor [##UREF##2##33##]. Furthermore, epigenetic therapies such as histone deacetylase and DNA methyltransferase inhibitors have been shown to reactivate expression of oxidative phosphorylation genes [##REF##18226465##34##], conceivably reducing metastatic potential and suggesting that some alterations in this pathway may be epigenetically regulated.</p>", "<p>Another critical pathway in our analysis that was differentially expressed robustly in primary versus metastatic tumors involves the extracellular matrix, cell adhesion, adhesion-mediated signal transduction and cytoskeletal organization, all of which are cooperatively important in the metastatic cascade.</p>", "<p>Alterations in extracellular matrix proteins included reductions in collagen, fibronectin, and a shift in keratin isoform expression (Figure ##FIG##3##4##). These reductions in cell matrix proteins could theoretically facilitate cell motility and enhance extravasation. The cell adhesion molecules CD63 and CD151 were upregulated in metastatic tumors as well. Experimental and clinical literature demonstrates a role for CD151 in metastasis [##REF##16798740##35##,##REF##15533898##36##]</p>", "<p>Differential expression of some key proteins responsible for adhesion-mediated cell signaling (RhoA, talin, moesin, ezrin, SPARC) was also observed (Figure ##FIG##4##5##). Encouragingly, up-regulation of some well-characterized metastasis-associated genes such as RhoA and ezrin was observed. RhoA plays a key role in regulating the actin cytoskeleton and controlling cell motility, cell-cell interactions and intracellular trafficking [##REF##16900393##37##]. Upregulation of RhoA has been associated with metastasis and/or negative outcome in carcinomas of the liver, kidney, esophagus, and urinary tract [##REF##17914970##38##, ####REF##17597401##39##, ##REF##12855641##40####12855641##40##]. Upregulation of ezrin has been implicated in metastasis of diverse tumor types, such as osteosarcoma, soft-tissue sarcomas, pancreatic carcinoma, and head and neck carcinoma among others [##REF##18246799##41##, ####REF##17874463##42##, ##REF##16633060##43##, ##REF##16365240##44##, ##REF##16144921##45##, ##REF##14704791##46####14704791##46##].</p>", "<p>Significant upregulation of important cytoskeletal components such as actin, tubulin and vimentin was also observed. These proteins play a key role in cell motility, invasion, cell division and intracellular transport, and differential expression of these members has been implicated in human tumor progression as well [##REF##18324357##47##,##REF##18327819##48##]. Increased vimentin is a well-defined phenotypic indicator of epithelial-mesenchymal transition, which has a known association with carcinoma aggressiveness [##REF##18327819##48##].</p>", "<p>Several components of the extracellular matrix – cell adhesion – adhesion-mediated signaling – cytoskeleton pathway have the potential for \"druggability\". For example, small molecule inhibitors of RhoA are in development [##REF##18167222##49##,##REF##17699722##50##], and rapamycin and its analogs have been shown to inhibit the ezrin-associated metastatic phenotype through inhibition of downstream AKT-mTOR signaling [##REF##15781656##51##].</p>", "<p>The antigen presentation pathway in Figure ##FIG##5##6## also reflects, in part, cytoskeleton remodeling: metastatic samples show increased expression of beta-2-microtubulin in the endoplasmic reticulum. All other elements of the antigen presentation pathway found in the differential genes shortlist are down-regulated. Remarkably, the most down-stream elements of the pathway, the final effectors, are the most down-regulated. Immune avoidance is thought to be another key component in successful metastasis; tumor cells must be able to survive in the circulation and avoid immune destruction upon arrest in the end-organ. Furthermore, evidence exists for epigenetic suppression of antigen presentation in tumor cells, and potential reactivation of expression through drugs blocking histone deacetylase and/or DNA methyltransferase, leading to enhanced tumor cell immunogenicity [##REF##18070359##52##, ####REF##18046553##53##, ##REF##17725605##54####17725605##54##].</p>", "<p>How reproducible are the results of computational analysis of an artificial meta-set of primary and metastatic tumors? We cannot possibly repeat the sample collection, RNA extraction and hybridizations. However, since the time Ramaswamy et al. have published their results there have been quite a few publications reporting microarray analysis of primary vs. metastatic tumors, and the data are available from the public databases such as GEO <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/geo\"/>. We have extracted and analyzed a few of these new data sets [##REF##17229949##4##,##REF##16531451##6##] using the same analysis pipeline. The final results of these analyses are lists of statistically significant pathways, molecular functions and GO terms within the shortlist of potentially differential genes. These lists show remarkable agreement in all studies. Comparison of the pathways represented in these lists reveals none unique to any of the 3 data sets, only common and similar (Supplemental Figure ##FIG##3##4## in Additional File ##SUPPL##0##1##). Overall, the Ramaswamy et al. meta-set produces a shorter list of potentially differential genes and further analysis yields fewer significant pathways. This result is not surprising taking into account that many small differences reproducibly observed in each single-tissue experiment have been leveled in composing the meta-set. However, the essential features reflecting the metabolic changes between primary and metastatic tumors are apparent in every analyzed data set. The oxidative phosphorylation pathways with most components down-regulated, cytoskeleton remodeling and cell adhesion-related pathways are always found among the longer lists of significant pathways in the specific colon and breast cancer datasets. Remarkably, the suppressed oxidative phosphorylation pathway is always near the top of the most statistically significant pathways.</p>", "<p>Taken together, there are dramatic changes in gene expression between primary and metastatic tumors; some are quantitative whereas others reflect a new pattern of expression. But how consistently are those changes revealed by a loose non-parametric J5 procedure? This selection procedure gives no estimation of confidence level for the individual genes. In turn, estimation of significance for the biological pathways is very approximate at best: it does not fully account for interdependencies in gene expression. Pathway maps include genes arbitrarily and the database of gene interactions is filled manually by multiple experts scanning the peer-reviewed literature, i.e. prone to errors and contradictions. These databases and associated tools for pathway analysis have improved significantly in recent years, but quantitative estimation of pathway significance still needs additional validation. In order to select only the most reliably over-represented pathways, we performed a bootstrap analysis randomly re-sampling 50% of the short-listed genes. Comparative analysis of over-represented pathways in the randomly re-sampled and original shortlists is given in Supplemental Table 6 and Supplemental Figure ##FIG##4##5## (Additional File ##SUPPL##0##1##). The main pathways are remarkably robust. The genes (putative biomarkers) diagram is dominated by \"similar\" pathways, i.e. belonging to the same pathway map or involved in the same cellular function. There are also some \"common\" genes (i.e. genes representing the same pathway, which is still statistically significant in the randomly selected half-list) and no \"unique\" genes (i.e. representing unique, but statistically significant pathways). This observation leads to important conclusions: a) microarray experiments may yield extensive variation in specific differentially expressed genes, but are robust and reproducible in elucidating differentially expressed pathways; b) random re-sampling of the large list of differentially expressed genes provides no proof of true difference for any single gene, but the list in general has few (if any) false-positive genes. The latter statement is controversial since the common goal of the inference in microarray analysis is to reduce the dimensionality of the feature space and select a small number of truly differential genes. After selection of a shortlist using a <italic>t</italic>-test or one of its variants, the number of differentially expressed genes is further reduced by application of a False Discovery Rate procedure (typically Benjamini-Hochberg) [##UREF##3##55##,##REF##12710672##56##]. Some authors even claim that microarrays are not optimal for pathway analysis because of poor reproducibility of the resulting pathways [##UREF##4##57##]. Our study suggests the opposite. The previously discussed problem of pathway reproducibility is caused by the misconceived methodology, more specifically in the strategy of microarray data analysis. Apparently, applying stricter criteria for selection of differentially expressed genes results in a very small number of candidates that are further reduced by FDR adjustment. The few remaining candidate genes have a much better chance of being successfully reproduced in another microarray experiment and validated by other techniques such as real-time RT-PCR or immunohistochemistry. On the other hand, a shorter list of candidate genes undermines the basis for the pathway analysis, rendering overrepresentation statistics powerless. This may explain the poor reproducibility of pathway analysis in some studies [##UREF##4##57##]. Such a stringent approach to biomarker selection relies entirely on the signal intensity and associated statistics. This approach can be very effective in cases of lethal mutations, congenital disorders and other diseases caused by a single or few factors. However, in complex multifactorial diseases, the most highly expressed genes and most reproducible differences in gene expression often turn out to be non-specific final effectors, downstream of important switches and regulators in biological pathways. Cancer in general and metastasis in particular are the examples of such multifactorial diseases. Application of a systems biology approach, considering not just the effect of single mutated/healthy genes, but entire networks of interlinked and constantly interacting genes is required not only for understanding the mechanism of disease, but also for the selection of diagnostic and prognostic markers, as well as potential therapeutic targets. As we have demonstrated, the pathways are sufficiently reproducible and robust to serve this purpose. The prevailing methodology in microarray analysis has an internal contradiction: it calls for a strict selection of candidate genes that can be independently verified one by one, but systems biology calls for analysis of large numbers of genes. Furthermore, the number of replicates affordable for a typical microarray study is usually insufficient for reliable reproduction of expression in low-expressed genes. However, important biological functions specific to disease are often performed by low-expressed genes. Pathway analysis has the power to identify such signal transducers and key transcription factors only if a large enough number of candidate genes are input. To resolve this contradiction, we propose an extension of the current prevailing methodology.</p>", "<p>First, the analysis pipeline has to be extended to incorporate functional annotation and pathway analysis. Second, selection of the candidate genes cannot be performed based solely on the intensity of signal and its change in the experiment. Instead, we propose to consider this step a pre-selection and relax the criteria for \"differential\" genes. Third, FDR correction should not be applied to a pre-selected \"long list\" of candidate genes. Combined with a relaxed selection threshold, this will inevitably create an influx of false-positive genes, which can be addressed subsequently. Fourth, the \"long list\" is analyzed in order to identify statistically overrepresented biological pathways, GO terms, molecular functions (as implemented in DAVID, IPA and MetaCore software) and gene set enrichment (for example, using GSEA or SAFE methods [##REF##16199517##17##,##REF##15647293##18##]). It is at this stage of analysis that multiple testing adjustments (Bonferroni, or better FDR) should be applied. Most available software, both free (DAVID tools [##REF##17784955##21##]) and commercial (such as IPA and Metacore) have at least one method of false-positive control implemented. However, we still recommend additional techniques, such as the bootstrapping experiment described above, for computational validation of significant pathways. Finally, the discovered statistically significant pathways, gene sets and molecular functions should be used to reverse-engineer the molecular mechanism of disease and select one or more potential biomarkers and drug targets. In our approach, it is important to combine numeric analysis with biological reasoning and deduction.</p>", "<p>The proposed analysis strategy is not yet implemented in a single analysis tool, although all the components have been developed and some of the software packages (such as ArrayTrack [##REF##14630514##58##]) offer partial integration; pathway analysis packages, although independent, can be easily invoked from within the microarray analysis software. In the future, we would like to unite all the tools used for systems biology analysis of biomarkers in a single automated software pipeline.</p>", "<p>Systems biology approaches to analysis of existing public data reveal a large number of new features overlooked in the original analyses. Meta-analysis and cross-examination of a few data sets allows identification of prospective markers and drug targets. The present day databases available for systems biology empower the researchers beyond the dreams of only a few years ago. Now for each identified significant pathway, we may also correlate expression data with known conserved transcription factor binding sites, and employ siRNA-mediated gene knockdown and known pharmacologic inhibitors (pharmacoprobes) to interrogate the phenotypic effects of interference with identified pathways. The systems approach described here allows identification of a number of key pathways that may serve as therapeutic targets for controlling the metastatic transition of primary solid tumors.</p>" ]

[]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Metastases are responsible for the majority of cancer fatalities. The molecular mechanisms governing metastasis are poorly understood, hindering early diagnosis and treatment. Previous studies of gene expression patterns in metastasis have concentrated on selection of a small number of \"signature\" biomarkers.</p>", "<title>Results</title>", "<p>We propose an alternative approach that puts into focus gene interaction networks and molecular pathways rather than separate genes. We have reanalyzed expression data from a large set of primary solid and metastatic tumors originating from different tissues using the latest available tools for normalization, identification of differentially expressed genes and pathway analysis. Our studies indicate that regardless of the tissue of origin, all metastatic tumors share a number of common features related to changes in basic energy metabolism, cell adhesion/cytoskeleton remodeling, antigen presentation and cell cycle regulation. Analysis of multiple independent datasets indicates significantly reduced oxidative phosphorylation in metastases compared to primary solid tumors.</p>", "<title>Conclusion</title>", "<p>Our methods allow identification of robust, although not necessarily highly expressed biomarkers. A systems approach relying on groups of interacting genes rather than single markers is also essential for understanding the cellular processes leading to metastatic progression. We have identified metabolic pathways associated with metastasis that may serve as novel targets for therapeutic intervention.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>AAP has collected the data, developed the algorithms, the software and performed data analysis. AAP, MMW and DHT interpreted the results and wrote the paper.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>Dr. Thamm is supported by American Cancer Society grant RSG-04-219-01-CCE.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Biological pathways significantly overrepresented in the shortlist of genes differentially expressed between primary solid and metastatic tumors (Ramaswamy et al. data set).</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Genes differentially expressed between primary and metastatic cancers in the oxidative phosphorylation pathway</bold>. Relative change and direction of change in transcript abundance of differentially expressed are marked with color flags. Red color designates higher and blue color designates lower transcript abundance compared to average between primary tumor (1) and metastatic samples (2). The legend for GeneGo pathway maps is given in Supplemental Figure 6 (Additional File ##SUPPL##0##1##).</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Glycolysis pathway</bold>. In spite of the fragmentary nature of the composed meta-set, the Warburg effect is still reflected in the pathway map through increased abundance of lactate dehydrogenase (LDHB). Relative change and direction of change in transcript abundance of differentially expressed are marked with color flags. Red color designates higher and blue color designates lower transcript abundance compared to average between primary tumor (1) and metastatic samples (2). The legend for GeneGo pathway maps is given in Supplemental Figure 6 (Additional File ##SUPPL##0##1##).</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Alterations in extraceullular matrix and secreted proteins associated with metastatic cancer</bold>. Relative change and direction of change in transcript abundance of differentially expressed are marked with color flags. Red color designates higher and blue color designates lower transcript abundance compared to average between primary tumor (1) and metastatic samples (2). The legend for GeneGo pathway maps is given in Supplemental Figure 6 (Additional File ##SUPPL##0##1##).</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>Alterations in adhesion-mediated signaling and cytoskeleton remodeling in metastatic cancer</bold>. Relative change and direction of change in transcript abundance of differentially expressed are marked with color flags. Red color designates higher and blue color designates lower transcript abundance compared to average between primary tumor (1) and metastatic samples (2). The legend for GeneGo pathway maps is given in Supplemental Figure 6 (Additional File ##SUPPL##0##1##).</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Alterations in the antigen presentation pathway observed in metastatic tumors</bold>. Relative change and direction of change in transcript abundance of differentially expressed are marked with color flags. Red color designates higher and blue color designates lower transcript abundance compared to average between primary tumor (1) and metastatic samples (2). The legend for GeneGo pathway maps is given in Supplemental Figure 6 (Additional File ##SUPPL##0##1##).</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p>Supplementary figures 1–6 and Supplementary tables 1–6.</p></caption></supplementary-material>" ]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S8-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S8-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S8-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S8-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S8-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S8-6\"/>" ]

[ "<media xlink:href=\"1471-2105-9-S9-S8-S1.zip\" mimetype=\"application\" mime-subtype=\"x-zip-compressed\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>DNA microarray technology [##REF##9634850##1##,##REF##7569999##2##] has rapidly advanced due to the intrinsic and unprecedented ability to simultaneously measure gene expression on a whole genome basis. Microarray technology continues to develop and is widely cited as offering much utility for translational science, from improved drug discovery, including target discovery, to improved clinical diagnostics and disease stage determination, prognostics and treatment selection, and more. With the prospect of microarray-derived biomarkers being applied in clinical applications, the bar is substantially raised for identification of informative genes enabling accurate classifiers, and efforts to this end are prevalent in the literature [##UREF##0##3##, ####REF##11389458##4##, ##REF##11251224##5##, ##REF##12461517##6##, ##REF##15379650##7##, ##REF##12499298##8##, ##REF##15852500##9##, ##REF##16709585##10##, ##REF##17338815##11####17338815##11##]. More specifically, there is a compelling need to identify a subset of genes from among the more than 20,000 in the entire genome that allow robust classifiers to be developed. The difficulty and challenge is to overcome the intrinsic characteristics of microarray data that contains a substantially small number of samples when compared to the number of genes [##REF##16108087##12##,##REF##17303535##13##]. These characteristics lead to the risk of fitting to noise as genes with high variability unrelated to phenotype masquerade as informative genes. The truly differentiating signals derived from small numbers of experimental replicates are difficult to distinguish in the sea of noise, leading to the appearance of unstable (i.e., non-reproducible) significant gene lists [##REF##16964229##14##, ####REF##16026597##15##, ##REF##18155896##16####18155896##16##].</p>", "<p>Gene selection is synonymous with feature selection or variable selection in machine learning, a process extensively used to mitigate the so called \"curse of dimensionality\" [##UREF##1##17##, ####UREF##2##18##, ##REF##16249260##19##, ##UREF##3##20####3##20##]. Generally, gene selection is done for either hypothesis testing or hypothesis generation. Selecting a subset of genes as molecular signatures or biomarkers that could be used for developing a generalized and accurate classifier for differentiating phenotypes is a hypothesis testing process [##REF##11983868##21##], wherein rigorous validation is needed. On the other hand, identifying a list of putatively relevant genes related to a phenotype or endpoint of interest for subsequent research is a hypothesis generating process [##REF##16398926##22##], wherein validation of the genes is much more relaxed; the genes so identified often shed light on the fundamental molecular mechanisms and biological processes under study.</p>", "<p>Selecting and validating an \"optimal\" set of genes for a molecular signature or biomarker for a robust classifier is a complicated and time-consuming task. An exhaustive search encompassing all possible gene subsets to find the set yielding the smallest error can be an intractable computational task. Worse still, because the number of genes far outnumber samples, the potential for fitting to random noise is high, making stringent testing and validation essential [##REF##15308542##23##,##REF##16670007##24##].</p>", "<p>Most methods to select informative genes for classification model development reported in the literature rely on ranked genes by fold change, correlation coefficient, or p-value from a t-statistic, Wilcoxon statistic, or analysis of variance (ANOVA), or some combination of these [##REF##16398926##22##,##REF##11751221##25##, ####REF##15450124##26##, ##REF##17915022##27##, ##REF##15680584##28##, ##REF##16504159##29##, ##REF##18175770##30####18175770##30##]. To a greater or lesser degree, all of these methods yield an informative gene list varying on the sample size, which has led doubt on microarray reliability [##REF##16964229##14##, ####REF##16026597##15##, ##REF##18155896##16####18155896##16##]. In theory, true phenotype differentiating genes should be expected to express consistently with each other regardless of the sample size. In other words, the list of informative genes as well as the underlying mechanisms inferred by these genes should have nothing to do with the sample size.</p>", "<p>In this study, a bagging [##UREF##4##31##] based new hybrid gene selection approach was investigated to identify informative genes. The rationale of the approach is that informative genes should consistently show significance for different variations of sample size. Accordingly, many re-sampling iterations are conducted to generate different variations of sample size and the frequency of genes exhibiting significance throughout the iterations formed the basis for identification of the informative genes that are considered as a Very Important Pool (VIP) of genes. In reality, the VIP genes can be identified using any existing gene selection approach or their combinations and can be used to derive molecular signatures to build robust classifiers with good generalization capability, or to narrow subsequent research to reveal relevant, fundamental molecular mechanisms in biological processes. In this study, t-statistic and discriminatory analysis are used to evaluate the significance of genes. In the t-statistic, the significant genes are identified based on p-values. In the discriminatory analysis, disjoint Principal Component Analyses (PCAs) are conducted for each class of samples, and those genes with high discrimination power (DP) [##UREF##5##32##] are identified as significant genes. The VIP genes are those having high frequency of showing significance in the re-sampling iterations. The utility of the proposed approach was demonstrated with nine diverse microarray datasets for identifying the informative genes for classifier development and compared with commonly used p-value ranking gene selection approaches.</p>" ]

[ "<title>Materials and methods</title>", "<title>Microarray datasets and software</title>", "<p>Nine publicly available microarray datasets were used to demonstrate the relative prediction performance of the proposed VIP gene selection approach. The datasets are from Alon <italic>et al</italic>. [##REF##10359783##47##], Beer <italic>et al</italic>. [##REF##12118244##48##], Bhattacharjee <italic>et al</italic>. [##REF##11707567##49##], Chen <italic>et al</italic>. [##REF##12058060##50##], Gordon <italic>et al</italic>. [##REF##12208747##51##], Pomeroy <italic>et al</italic>. [##REF##11807556##52##], Resenwald <italic>et al</italic>. [##REF##12075054##53##], Shipp <italic>et al</italic>. [##REF##11786909##54##], Singh <italic>et al</italic>. [##REF##12086878##55##], Yeoh <italic>et al</italic>. [##REF##12086872##56##], and van't Veer <italic>et al</italic>. [##REF##11823860##57##], that for convenience are hereafter respectively referred to as \"Alon\", \"Beer\", \"Bhattacharjee\", \"Chen\", \"Gordon\", \"Pomeroy\", \"Resenwald\", \"Shipp\", \"Singh\", \"Yeoh\", and \"van't Veer\"; information for each dataset is given in Table ##TAB##0##1##.</p>", "<p>The VIP gene selection approach was developed using the programming language Matlab^®7.0, running on a DELL™ Precision 490 workstation equipped with two Intel^®Dual Core Xeon™ 3.0 GHz processors and 2 GB of memory. The Matlab codes are available upon request.</p>", "<p>The biological interpretation of genes was conducted using PathArt <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.jubilantbiosys.com/ppa.htm\"/> through the FDA genomic tool, ArrayTrack <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.fda.gov/nctr/science/centers/toxicoinformatics/ArrayTrack/\"/>.</p>", "<title>Algorithm</title>", "<p>The VIP gene selection approach combines discriminatory powers derived from two independent principal component analyses and p values from t-statistic to filter genes based on a bagging, re-sampling technique. The algorithmic process is depicted in Figure ##FIG##1##2##, where the training dataset is composed of <italic>n</italic>₁samples of class 1 and <italic>n</italic>₂samples of class 2. Samples of class 1 and class 2 are represented by the matrices <bold>X</bold>₁and <bold>X</bold>₂, respectively. The VIP genes are chosen through the following steps:</p>", "<p>1. Randomly select 75% of samples from the training data, <bold>X</bold>₁and <bold>X</bold>_2,using a bagging, re-sampling strategy. The selected samples are represented with <bold>X</bold>_{1<italic>m </italic>}for class 1 and <bold>X</bold>_{2<italic>m </italic>}for class 2.</p>", "<p>2. Rank genes by their p-values and only keep the top 100 genes for next step. P-values are calculated from a two-tailed and unpaired t-statistic with pooled variance estimate (i.e., equal variances or homoscedastic assumption) on <bold>X</bold>_{1<italic>m </italic>}and <bold>X</bold>_{2<italic>m</italic>}. The remaining data are represented by and , respectively.</p>", "<p>3. Rank genes based on their discrimination powers (DPs) and the increment the frequencies of the top 50 genes by one. The calculation of DPs is described in detail in the next section \"calculation of discrimination power\".</p>", "<p>4. Repeat steps one through three 100 times.</p>", "<p>5. Rank genes by frequencies and choose the top 50 genes as VIP genes.</p>", "<title>Calculation of discrimination power</title>", "<p>DPs are calculated from two independent principal component analyses (PCAs). PCA is performed on each p-value-filtered data, and from step 2. The optimum number of components for each PCA is determined using Malinowski's factor indicator function (IND) [##UREF##6##58##] with eqs. (1) – (3):</p>", "<p></p>", "<p></p>", "<p></p>", "<p>where <bold>X </bold>is either and ; <bold>T </bold>and <bold>P </bold>are the score and loading matrices of the PCA; <italic>λ</italic>_{<italic>i </italic>}is the <italic>i</italic>^{<italic>th </italic>}eigenvalue of the total <italic>g </italic>eigenvalues; and <italic>n </italic>and <italic>p </italic>are the number of samples and the number of genes in the matrix <bold>X</bold>, respectively. The optimum number (<italic>k</italic>) of components for the PCA is the one that yields the minimum <italic>IND </italic>value. The discrimination power (DP_{<italic>j</italic>}) for a gene <italic>j </italic>can be calculated with eq. (4):</p>", "<p></p>", "<p>where , , , and are the <italic>j </italic>columns of matrices <bold>E</bold>₁₁, <bold>E</bold>₁₂, <bold>E</bold>₂₂, and <bold>E</bold>₂₁, respectively. <bold>E</bold>₁₁and <bold>E</bold>₁₂are the residue matrices after projecting into the PCA spaces of class 1 and class 2, respectively, while <bold>E</bold>₂₂and <bold>E</bold>₂₁are the residue matrices after projecting into the PCA spaces of class 1 and class 2, respectively. A residue matrix is calculated with eq. (5).</p>", "<p></p>", "<p>where <bold>E </bold>is one of the four residue matrices <bold>E</bold>₁₁, <bold>E</bold>₁₂, <bold>E</bold>₂₂, and <bold>E</bold>₂₁.</p>", "<title>Prediction performance</title>", "<p>The prediction performance of a Nearest-Centroid classifier in this study is characterized with four metrics: accuracy, specificity, sensitivity, and the Matthew's correlation coefficient (MCC). The metrics can be calculated from the prediction confusion matrix shown in Table ##TAB##4##5## as follows:</p>", "<p></p>", "<p></p>", "<p></p>", "<p></p>", "<p>where TP, TN, FP, FN are respectively the numbers of true positive, true negative, false positive, and false negative predictions in the confusion matrix (Table ##TAB##4##5##).</p>" ]

[ "<title>Results</title>", "<p>The VIP gene selection approach for microarray based molecular signatures was applied to the nine publicly available microarray gene expression datasets described in Table ##TAB##0##1##. For the purpose of comparison, the p-value ranking method was also used. For each dataset, an unbiased sample splitting, gene selection, and validation dataset prediction process as depicted in Figure ##FIG##0##1## was carried out. Briefly, a dataset is first randomly split into a training set with two thirds of the samples and a validation set with the remaining samples. With validation samples set aside, gene selection and classifier development are done using the training samples. Two lists of 50 genes are selected, one using the proposed VIP gene selection approach and the other using p-value ranking. The p-value ranking is based on an unpaired, two-tailed t-statistic with pooled variance estimate. In order to exam whether the VIP gene selection approach can identify informative genes or not, three sets of classifiers were generated, one for the VIP genes, one for the p-value genes and another for the genes uniquely identified by the VIP method (called unique genes hereafter). A Nearest-Centroid[##REF##15705458##33##] classification method was used to develop classifiers. These classifiers are applied to predict the validation samples. The prediction performance of classifiers were compared by accuracies, specificities, sensitivities, and the Matthew's correlation coefficients (MCCs). The definitions of these measures are given in the section titled \"materials and methods\". The sample splitting, gene selection, and validation dataset prediction steps were repeated 50 times for adequate statistics.</p>", "<p>We first compared the classifiers based on the VIP genes with those from the p-value ranking. As shown in Table ##TAB##1##2##, the VIP classifiers exhibited somewhat better performance compared to the classifiers from the p-value selected genes. The p-values from t-statistic for accuracy, specificity, sensitivity and MCC between two groups of classifiers (the VIP classifiers versus the p-value ranking classifiers) are 0.0027, 0.32, 0.059, and 0.0092, respectively. Therefore, at the 0.05 confidence level, the improvement of classifier measured in MCC and accuracy is significant, but not for specificity and sensitivity. The results indicate that the VIP genes may convey more, but not less, biologically relevant information than the p-value selected genes.</p>", "<p>Next, to determine whether the unique genes indeed contribute to the sample differentiation and thus biological relevance, we compared prediction performance of the classifiers built from unique genes with those built from the p-value ranked genes across the nine datasets. The average number of unique genes for each dataset is also listed in Table ##TAB##1##2##. It was shown that the average performance metrics (accuracy, specificity, sensitivity, and MCC) for classifiers built from unique genes (number from 14 to 22) are not very different from those built from top 50 p-value ranked genes for all nine datasets. The difference of each pair of average performance metrics is respectively tested across nine datasets with a null hypothesis that the compared performance metrics (accuracy, specificity, sensitivity, or MCC) is not very different from each other by using a paired and two-tailed t-statistic. The p-values given by t-statistic are 0.63, 0.77, 0.95, and 0.81 for accuracy, specificity, sensitivity, and MCC respectively. Apparently, the differences of all prediction performance metrics among classifiers are not significant at the 0.05 confidence level. This suggests that the unique VIP genes are statistically equivalent as those identified by p-value ranking in distinguishing different types of samples. Therefore, these unique genes could be an additional subset of genes which are equally as important as those selected with p-value ranking. The existence of additional subsets of classifying genes may imply that there exist multiple biological processes for studied endpoints or co-factors.</p>", "<p>Lastly, to gain more understanding of the VIP genes in terms of biology related to the investigated dataset, we further examined the unique genes as well as the common genes shared by the p-value method in the van't Veer dataset using PathArt <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.jubilantbiosys.com/ppa.htm\"/> through the FDA genomic tool, ArrayTrack <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.fda.gov/nctr/science/centers/toxicoinformatics/ArrayTrack/\"/>. PathArt is a pathway analysis tool that contains disease related canonical pathways manually created from the literature. The van't Veer dataset contains 24 unique genes and 26 common genes. Of 24 unique genes, ten genes were found in PathArt and were listed in Table ##TAB##2##3##. Most of these ten genes involve biological processes related to various cancers; for example, IGFBP5 and MMP9 are directly related to breast cancer. We also examined the pathways associated with the 26 common genes and found seven unique genes were involved in seven pathways identified by the common genes (Table ##TAB##3##4##). These results demonstrate that the unique genes not identified by the p-value ranking could convey additional important information for biological interpretation.</p>" ]

[ "<title>Discussion</title>", "<p>Quantitatively assessing the effectiveness of gene selection methods can be problematic owing to several limitations among which selection bias caused by information leakage from training phase to validation phase figures prominently [##REF##16670007##24##]. The most severe bias was described by Ambroise <italic>et al</italic>. [##REF##11983868##21##] and Simon <italic>et al</italic>. [##REF##12509396##34##] as occurring when identifying genes from the entire dataset (i.e., training set and validation set) and using them in cross-validation. Wessels <italic>et al</italic>. [##REF##15817694##35##] and Lai <italic>et al</italic>. [##REF##16670007##24##] describe a less severe bias. Typically, the training samples are used to generate a series of gene subsets, while the performance of a classifier trained with the training samples and tested with the validation samples is applied to estimate the informativeness of each gene subset. The bias derives from the fact that the validation samples are used to select the best performing gene subset. Since optimization of the gene subset is part of the training process, selection of the best gene subset should be conducted with the training samples only. This process as shown in Figure ##FIG##0##1## has been carried out in this study to assess the utility of the proposed VIP gene selection method by entirely avoiding bias due to information leakage from validation dataset in training phase.</p>", "<p>Classification method selection is another important aspect of developing predictive models from microarray expression data. Many classifiers are created with one or more adjustable parameters that affect not only the prediction accuracy but also the complexity of the classifiers and the computational expense of their use. The proper adjustment of the tuneable parameters can affect the fairness of comparative predictive performance assessments. For example, the relatively simple k-Nearest Neighbour (KNN)[##REF##11108479##36##] classification method has a tuneable k in the prediction rules. Adjusting k requires some validation process be carried out. Generally, different validation strategies such as leave-one-out cross validation, k-Fold cross-validation, or Monte Carlo validation, will yield different preferred values of k. Other classification approaches, such as Support Vector Machine (SVM) [##REF##8555380##37##], Partial Least Squares Discriminant Analysis (PLS-DA) [##REF##16808466##38##], Random Forest (RF)[##REF##16398926##22##], and Artificial Neural Networks (ANN) [##REF##12221094##39##] are considerably more complex by comparison, causing more work and computational cost. According to Wessels <italic>et al</italic>. [##REF##15817694##35##], Michiels <italic>et al</italic>. [##REF##15705458##33##], and Lai <italic>et al</italic>. [##REF##16670007##24##], choosing a classification method with a limited complexity can help prevent over-training, thus providing a more robust predictor. In this study, the simple classification approach Nearest-Centroid was used to develop and compare classifiers based on unique VIP genes and top 50 p-value ranked genes. Since the method lacks a tuneable parameter, risks of overtraining are lessened compared to other methods, as are the chances that differences in prediction accuracy are due to method rather than selected genes.</p>", "<p>Commonly used gene selection approaches in DNA microarray data analysis, such as p-value ranking or fold change ranking and others, assume that all genes are stochastic variables that are unrelated to each for purposes of calculating significance. This assumption is inconsistent with the actual biological processes where most genes have some interdependency to and are interlinked with other genes through complex mechanisms and pathways. In contrast, the proposed VIP gene selection approach uses both DPs and p-values to assess the discriminatory capability of genes in differentiating sample types. DPs are calculated from two independent PCAs that fuse discriminating information across whole genes. The interdependence and interlinking effects among genes are embedded within the DP calculation, enhancing rather than reducing many aspects of actual biological processes. Furthermore, the bagging re-sampling technique, which has been used to analyze microarray data for clustering [##REF##18218074##40##, ####REF##18204054##41##, ##REF##12801869##42####12801869##42##] and classification [##REF##15466910##43##, ####REF##15978569##44##, ##REF##15691862##45##, ##REF##17624926##46####17624926##46##], is used here to mitigate the chance selection of genes. Compared with p-value ranking-type gene selection approaches, the proposed VIP gene selection has great potential to select additional informative genes that can be useful for either biological insights or to improve the prediction performance of classifiers.</p>" ]

[ "<title>Conclusion</title>", "<p>The new hybrid gene selection approach was investigated for identifying VIP genes from nine diverse gene expression datasets. The VIP gene selection approach quantifies discriminatory capability for differentiating sample classes using both discrimination analysis and p-value ranking through a bagging sampling process. The classifiers built from those unique VIP genes showed comparable prediction capability to those built from the top 50 t-statistic based p-value ranked genes in predicting the types of unknown samples. Therefore, the VIP gene selection approach could provide an additional subset of genes which are of equivalent performance as those selected with the t-statistic based p-value ranking. The subset of VIP genes could convey additional biological information in terms of associated biological pathways and mechanisms during hypothesis generation. Similarly, the VIP genes could be used to improve molecular fingerprints for use in clinical biomarkers.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Advances in DNA microarray technology portend that molecular signatures from which microarray will eventually be used in clinical environments and personalized medicine. Derivation of biomarkers is a large step beyond hypothesis generation and imposes considerably more stringency for accuracy in identifying informative gene subsets to differentiate phenotypes. The inherent nature of microarray data, with fewer samples and replicates compared to the large number of genes, requires identifying informative genes prior to classifier construction. However, improving the ability to identify differentiating genes remains a challenge in bioinformatics.</p>", "<title>Results</title>", "<p>A new hybrid gene selection approach was investigated and tested with nine publicly available microarray datasets. The new method identifies a Very Important Pool (VIP) of genes from the broad patterns of gene expression data. The method uses a bagging sampling principle, where the re-sampled arrays are used to identify the most informative genes. Frequency of selection is used in a repetitive process to identify the VIP genes. The putative informative genes are selected using two methods, t-statistic and discriminatory analysis. In the t-statistic, the informative genes are identified based on p-values. In the discriminatory analysis, disjoint Principal Component Analyses (PCAs) are conducted for each class of samples, and genes with high discrimination power (DP) are identified. The VIP gene selection approach was compared with the p-value ranking approach. The genes identified by the VIP method but not by the p-value ranking approach are also related to the disease investigated. More importantly, these genes are part of the pathways derived from the common genes shared by both the VIP and p-ranking methods. Moreover, the binary classifiers built from these genes are statistically equivalent to those built from the top 50 p-value ranked genes in distinguishing different types of samples.</p>", "<title>Conclusion</title>", "<p>The VIP gene selection approach could identify additional subsets of informative genes that would not always be selected by the p-value ranking method. These genes are likely to be additional true positives since they are a part of pathways identified by the p-value ranking method and expected to be related to the relevant biology. Therefore, these additional genes derived from the VIP method potentially provide valuable biological insights.</p>" ]

[ "<title>Disclaimer</title>", "<p>The views presented in this article do not necessarily reflect those of the US Food and Drug Administration.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>ZS had the original idea, developed the method, did all calculations and data analysis, and wrote the first draft of manuscript. WT had the original idea, discussed on data analysis and presentation of results. HF, HH, LS, and RP involved in discussion on data analysis, verified some of the calculations and assisted with writing the manuscript. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>ZS is grateful to the National Center for Toxicological Research (NCTR) of U.S. Food and Drug Administration (FDA) for postdoctoral support through the Oak Ridge Institute for Science and Education (ORISE).</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>The flowchart for the classifier development and validation using three gene sets: (A) Top 50 p-value ranked genes; (B) Top 50 VIP genes; and (C) the unique VIP genes. Specifically, the data set is first randomly divided into two thirds of samples for training and the remainder for validation. Next, three sets of genes are generated solely based on the training set, and are subsequently used to develop Nearest-Centroid classifiers. Lastly, the classifiers are used to predict the validation samples and their respective prediction performance measured by accuracy (Acc), specificity (Spec), sensitivity (Sens), and Matthew's correlation coefficient (MCC) are calculated. The process is repeated 50 times and the averaged performance metrics are reported in Table 2.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>The detailed process for identifying a very important pool (VIP) of genes. X₁and X₂are, respectively, the gene expression profiles for class 1 samples and class 2 samples in the training set. X_{1<italic>m </italic>}and X_{2<italic>m </italic>}are samples randomly selected from X₁and X₂in the m^{<italic>th </italic>}bagging step. and are the genes remaining after filtering genes from X_{1<italic>m </italic>}and X_{2<italic>m</italic>}, respectively. Malinowski's factor indicator function (IND) is calculated with equations and <italic>IND</italic>_{<italic>k </italic>}= <italic>RE</italic>_{<italic>k</italic>}/(<italic>n </italic>- <italic>k</italic>)², where <italic>λ</italic>_{<italic>i </italic>}is the <italic>i</italic>^theigenvalue of the total <italic>g </italic>eigenvalues; <italic>n </italic>is the number of samples and <italic>p </italic>is the number of genes. The optimum number (<italic>k</italic>) of components corresponds to the IND minimum. E₁₁and E₂₁are the residue matrices after projecting X_{1<italic>m </italic>}and X_{2<italic>m </italic>}into the PCA space for class 1, respectively, while E₂₂and E₁₂are the residue matrices after projecting X_{2<italic>m </italic>}and X_{1<italic>m </italic>}into the PCA space for class 2, respectively. The discrimination power (DP) of a gene <italic>j </italic>is calculated with the equation: , where , , , and are the <italic>j </italic>columns of residue matrices E₁₁, E₁₂, E₂₂, and E₂₁, respectively.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Nine microarray datasets used in the study.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Name</bold></td><td align=\"center\"><bold>Cancer type</bold></td><td align=\"center\"><bold>Prediction task</bold></td><td align=\"center\"><bold>Sample size</bold></td><td align=\"center\"><bold>Number of events</bold></td><td align=\"center\"><bold>Number of genes</bold></td><td align=\"center\"><bold>Reference</bold></td></tr></thead><tbody><tr><td align=\"center\">Beer</td><td align=\"center\">Lung adenocarcinoma</td><td align=\"center\">Survival</td><td align=\"center\">86</td><td align=\"center\">24</td><td align=\"center\">6532</td><td align=\"center\">[##REF##12118244##48##]</td></tr><tr><td align=\"center\">Bhattacharjee</td><td align=\"center\">Lung adenocarcinoma</td><td align=\"center\">4-year survival</td><td align=\"center\">62</td><td align=\"center\">31</td><td align=\"center\">5403</td><td align=\"center\">[##REF##11707567##49##]</td></tr><tr><td align=\"center\">Chen</td><td align=\"center\">Hepatocellular carcinoma</td><td align=\"center\">Tumors</td><td align=\"center\">156</td><td align=\"center\">82</td><td align=\"center\">3964</td><td align=\"center\">[##REF##12058060##50##]</td></tr><tr><td align=\"center\">Pomeroy</td><td align=\"center\">Medulloblastoma</td><td align=\"center\">Medulloblastoma survival</td><td align=\"center\">60</td><td align=\"center\">21</td><td align=\"center\">7129</td><td align=\"center\">[##REF##11807556##52##]</td></tr><tr><td align=\"center\">Rosenwald</td><td align=\"center\">Non-Hodgkin lymphoma</td><td align=\"center\">Survival</td><td align=\"center\">240</td><td align=\"center\">138</td><td align=\"center\">7399</td><td align=\"center\">[##REF##12075054##53##]</td></tr><tr><td align=\"center\">Shipp</td><td align=\"center\">Diffuse large b-cell lymphoma (DLBCL)</td><td align=\"center\">Cured</td><td align=\"center\">58</td><td align=\"center\">32</td><td align=\"center\">6817</td><td align=\"center\">[##REF##11786909##54##]</td></tr><tr><td align=\"center\">Singh</td><td align=\"center\">Prostate cancer</td><td align=\"center\">Tumors</td><td align=\"center\">102</td><td align=\"center\">52</td><td align=\"center\">12600</td><td align=\"center\">[##REF##12086878##55##]</td></tr><tr><td align=\"center\">Yeoh</td><td align=\"center\">Acute lymphocytic leukaemia</td><td align=\"center\">Relapse-free survival</td><td align=\"center\">233</td><td align=\"center\">32</td><td align=\"center\">12236</td><td align=\"center\">[##REF##12086872##56##]</td></tr><tr><td align=\"center\">van't Veer</td><td align=\"center\">Breast cancer</td><td align=\"center\">5-year metastasis-free survival</td><td align=\"center\">97</td><td align=\"center\">46</td><td align=\"center\">4948</td><td align=\"center\">[##REF##11823860##57##]</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Comparison of prediction performance for Nearest-Centroid classifiers built from unique VIP genes, top 50 p-value ranked genes, and 50 VIP genes. The classifier performance metrics, including accuracy (Acc), specificity (Spec), Sensitivity (Sens), and Matthew's correlation coefficient (MCC) were calculated based on averages of 50 repetitions of sample splitting, gene selection, and validation dataset prediction.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"5\"><bold>Unique VIP genes</bold></td><td align=\"center\" colspan=\"4\"><bold>50 p-value ranked genes</bold></td><td align=\"center\" colspan=\"4\"><bold>50 VIP genes</bold></td></tr><tr><td/><td colspan=\"13\"><hr/></td></tr><tr><td align=\"left\"><bold>Data set</bold></td><td align=\"center\"><bold>Number of genes</bold></td><td align=\"center\"><bold>Acc (%)</bold></td><td align=\"center\"><bold>Spec (%)</bold></td><td align=\"center\"><bold>Sens (%)</bold></td><td align=\"center\"><bold>MCC</bold></td><td align=\"center\"><bold>Acc (%)</bold></td><td align=\"center\"><bold>Spec (%)</bold></td><td align=\"center\"><bold>Sens (%)</bold></td><td align=\"center\"><bold>MCC</bold></td><td align=\"center\"><bold>Acc (%)</bold></td><td align=\"center\"><bold>Spec (%)</bold></td><td align=\"center\"><bold>Sens (%)</bold></td><td align=\"center\"><bold>MCC</bold></td></tr></thead><tbody><tr><td align=\"left\">Beer</td><td align=\"center\">15</td><td align=\"center\">64.7</td><td align=\"center\">38.3</td><td align=\"center\">74.0</td><td align=\"center\">0.13</td><td align=\"center\">64.7</td><td align=\"center\">38.3</td><td align=\"center\">74.0</td><td align=\"center\">0.12</td><td align=\"center\">65.2</td><td align=\"center\">35.4</td><td align=\"center\">75.6</td><td align=\"center\">0.11</td></tr><tr><td align=\"left\">Bhattacharjee</td><td align=\"center\">17</td><td align=\"center\">58.7</td><td align=\"center\">57.8</td><td align=\"center\">59.6</td><td align=\"center\">0.18</td><td align=\"center\">58.0</td><td align=\"center\">59.2</td><td align=\"center\">56.8</td><td align=\"center\">0.16</td><td align=\"center\">58.6</td><td align=\"center\">59.4</td><td align=\"center\">57.8</td><td align=\"center\">0.18</td></tr><tr><td align=\"left\">Chen</td><td align=\"center\">14</td><td align=\"center\">96.5</td><td align=\"center\">99.9</td><td align=\"center\">93.6</td><td align=\"center\">0.93</td><td align=\"center\">95.3</td><td align=\"center\">100.0</td><td align=\"center\">91.2</td><td align=\"center\">0.91</td><td align=\"center\">95.8</td><td align=\"center\">100.0</td><td align=\"center\">92.1</td><td align=\"center\">0.92</td></tr><tr><td align=\"left\">Pomeroy</td><td align=\"center\">20</td><td align=\"center\">60.8</td><td align=\"center\">54.0</td><td align=\"center\">64.2</td><td align=\"center\">0.19</td><td align=\"center\">60.8</td><td align=\"center\">51.7</td><td align=\"center\">65.3</td><td align=\"center\">0.18</td><td align=\"center\">62.4</td><td align=\"center\">55.7</td><td align=\"center\">65.8</td><td align=\"center\">0.22</td></tr><tr><td align=\"left\">Rosenwald</td><td align=\"center\">18</td><td align=\"center\">55.5</td><td align=\"center\">58.1</td><td align=\"center\">53.6</td><td align=\"center\">0.12</td><td align=\"center\">56.8</td><td align=\"center\">63.2</td><td align=\"center\">52.2</td><td align=\"center\">0.15</td><td align=\"center\">57.4</td><td align=\"center\">62.3</td><td align=\"center\">53.8</td><td align=\"center\">0.16</td></tr><tr><td align=\"left\">Shipp</td><td align=\"center\">18</td><td align=\"center\">51.6</td><td align=\"center\">50.8</td><td align=\"center\">52.5</td><td align=\"center\">0.03</td><td align=\"center\">47.9</td><td align=\"center\">51.8</td><td align=\"center\">43.0</td><td align=\"center\">-0.05</td><td align=\"center\">49.0</td><td align=\"center\">47.4</td><td align=\"center\">51.0</td><td align=\"center\">-0.02</td></tr><tr><td align=\"left\">Singh</td><td align=\"center\">15</td><td align=\"center\">94.3</td><td align=\"center\">98.3</td><td align=\"center\">91.7</td><td align=\"center\">0.89</td><td align=\"center\">98.1</td><td align=\"center\">100.0</td><td align=\"center\">96.9</td><td align=\"center\">0.96</td><td align=\"center\">97.8</td><td align=\"center\">100.0</td><td align=\"center\">96.4</td><td align=\"center\">0.96</td></tr><tr><td align=\"left\">Yeoh</td><td align=\"center\">22</td><td align=\"center\">74.6</td><td align=\"center\">37.8</td><td align=\"center\">80.2</td><td align=\"center\">0.15</td><td align=\"center\">78.2</td><td align=\"center\">31.0</td><td align=\"center\">85.4</td><td align=\"center\">0.15</td><td align=\"center\">80.2</td><td align=\"center\">35.0</td><td align=\"center\">87.0</td><td align=\"center\">0.21</td></tr><tr><td align=\"left\">van't Veer</td><td align=\"center\">20</td><td align=\"center\">64.8</td><td align=\"center\">64.8</td><td align=\"center\">64.9</td><td align=\"center\">0.30</td><td align=\"center\">65.2</td><td align=\"center\">61.5</td><td align=\"center\">68.6</td><td align=\"center\">0.31</td><td align=\"center\">66.9</td><td align=\"center\">66.1</td><td align=\"center\">67.6</td><td align=\"center\">0.34</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Pathways identified for the unique VIP genes and common genes for the van't Veer dataset.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Accession number (Symbol)</bold></td><td align=\"left\"><bold>Full Name</bold></td><td align=\"left\"><bold>Pathway name</bold></td><td align=\"left\"><bold>Category (e.g. disease)</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>Unique VIP genes</bold></td><td align=\"center\">AF055033 (IGFBP5)</td><td align=\"left\">Insulin-like growth factor binding protein 5</td><td align=\"left\">Estrogen signaling pathway</td><td align=\"left\">Breast cancer</td></tr><tr><td/><td/><td/><td align=\"left\">IGF signaling pathway</td><td align=\"left\">Lung cancer</td></tr><tr><td/><td align=\"center\">NM_000599 (IGFBP5)</td><td/><td align=\"left\">AR mediated pathway; insulin-like growth factor-1 signaling pathway</td><td align=\"left\">Prostate cancer</td></tr><tr><td/><td/><td/><td align=\"left\">Responsive genes</td><td align=\"left\">Ovarian cancer</td></tr><tr><td/><td align=\"center\">NM_000017 (ACADS)</td><td align=\"left\">Acyl-coenzyme A dehydrogenase, C-2 to C-3 short chain</td><td align=\"left\">Responsive genes</td><td align=\"left\">Colon cancer</td></tr><tr><td/><td align=\"center\">NM_004994 (MMP9)</td><td align=\"left\">Matrix metallopeptidase 9 (gelatinase B, 92 kDa gelatinase, 92 kDa type IV collagenase)</td><td align=\"left\">Heregulin, and CXCL12 signaling pathway</td><td align=\"left\">Breast cancer</td></tr><tr><td/><td/><td/><td align=\"left\">Bombesin, IL10, IL8, TGFbeta, and HGF signaling pathway; responsive genes</td><td align=\"left\">Prostate cancer</td></tr><tr><td/><td/><td/><td align=\"left\">Responsive genes; thrombospondin signaling pathway</td><td align=\"left\">Pancreatic cancer</td></tr><tr><td/><td/><td/><td align=\"left\">Gastrin, HGF, and IL4 signaling pathway; integrin, and UPAR mediated pathway</td><td align=\"left\">Colon cancer</td></tr><tr><td/><td/><td/><td align=\"left\">Responsive genes</td><td align=\"left\">Chronic myeloid leukemia</td></tr><tr><td/><td/><td/><td align=\"left\">EGF signaling pathway; VEGF mediated pathway; responsive genes</td><td align=\"left\">Ovarian cancer</td></tr><tr><td/><td/><td/><td align=\"left\">HGF, and IL6 signaling pathway; Responsive genes</td><td align=\"left\">Lung cancer</td></tr><tr><td/><td align=\"center\">NM_001197 (BIK)</td><td align=\"left\">BCL2-interacting killer (apoptosis-inducing)</td><td align=\"left\">p53 mediated pathway</td><td align=\"left\">Colon cancer</td></tr><tr><td/><td align=\"center\">NM_001809 (CENPA)</td><td align=\"left\">Centromere protein A</td><td align=\"left\">Responsive genes</td><td align=\"left\">Lung cancer</td></tr><tr><td/><td/><td/><td align=\"left\">p21 mediated pathway</td><td align=\"left\">Cell-cycle</td></tr><tr><td/><td align=\"center\">NM_002808 (PSMD2)</td><td align=\"left\">Proteasome (prosome, macropain) 26S subunit, non-ATPase, 2</td><td align=\"left\">Tat signaling pathway</td><td align=\"left\">Acquired immuno deficiency syndrome</td></tr><tr><td/><td align=\"center\">NM_004336 (BUB1)</td><td align=\"left\">BUB1 budding uninhibited by benzimidazoles 1 homolog (yeast)</td><td align=\"left\">Spindle Checkpoint Pathway</td><td align=\"left\">Cell-cycle</td></tr><tr><td/><td align=\"center\">NM_004626 (WNT11)</td><td align=\"left\">Wingless-type MMTV integration site family, member 11</td><td align=\"left\">Cell-cell signaling pathway</td><td align=\"left\">Others</td></tr><tr><td/><td/><td/><td align=\"left\">WNT receptor signaling pathway</td><td align=\"left\">Others</td></tr><tr><td/><td align=\"center\">NM_004887 (CXCL14)</td><td align=\"left\">Chemokine (C-X-C motif) ligand 14</td><td align=\"left\">Signal transduction pathway</td><td align=\"left\">Others</td></tr><tr><td colspan=\"5\"><hr/></td></tr><tr><td align=\"center\"><bold>Common genes</bold></td><td align=\"center\">AL050227 (PTGER3)</td><td align=\"left\">Prostaglandin E receptor 3 (subtype EP3)</td><td align=\"left\">Estrogen signaling pathway</td><td align=\"left\">Breast cancer</td></tr><tr><td/><td/><td/><td align=\"left\">PGE2 mediated pathway</td><td align=\"left\">Lung cancer</td></tr><tr><td/><td align=\"center\">NM_006763 (BTG2)</td><td align=\"left\">BTG family, member 2</td><td align=\"left\">Estrogen signaling pathway</td><td align=\"left\">Breast cancer</td></tr><tr><td/><td/><td/><td align=\"left\">Responsive genes</td><td align=\"left\">Prostate cancer</td></tr><tr><td/><td/><td/><td align=\"left\">CEBP alpha mediated pathway</td><td align=\"left\">Chronic myeloid leukemia</td></tr><tr><td/><td/><td/><td align=\"left\">Miscellaneous</td><td align=\"left\">DNA repair</td></tr><tr><td/><td/><td/><td align=\"left\">BTG mediated pathway</td><td align=\"left\">Cell-cycle</td></tr><tr><td/><td align=\"center\">NM_003862 (FGF18)</td><td align=\"left\">Fibroblast growth factor 18</td><td align=\"left\">WNT signaling pathway</td><td align=\"left\">Colon cancer</td></tr><tr><td/><td align=\"center\">NM_006115 (PRAME)</td><td align=\"left\">Preferentially expressed antigen in melanoma</td><td align=\"left\">Responsive genes</td><td align=\"left\">Ovarian cancer</td></tr><tr><td/><td align=\"center\">X05610 (COL4A2)</td><td align=\"left\">Collagen, type IV, alpha 2</td><td align=\"left\">Responsive genes</td><td align=\"left\">Glioblastoma</td></tr><tr><td/><td align=\"center\">NM_003981 (PRC1)</td><td align=\"left\">Protein regulator of cytokinesis 1</td><td align=\"left\">p21 mediated pathway</td><td align=\"left\">Cell-cycle</td></tr><tr><td/><td align=\"center\">NM_006027 (EXO1)</td><td align=\"left\">Exonuclease 1</td><td align=\"left\">p21 mediated pathway</td><td align=\"left\">Cell-cycle</td></tr><tr><td/><td align=\"center\">NM_002811 (PSMD7)</td><td align=\"left\">Proteasome (prosome, macropain) 26S subunit, non-ATPase, 7</td><td align=\"left\">Tat signaling pathway</td><td align=\"left\">Acquired immuno deficiency syndrome</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>The pathways involved with both unique VIP genes and common genes for the van't Veer dataset</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Pathway name</bold></td><td align=\"left\"><bold>Unique gene</bold></td><td align=\"left\"><bold>Common gene</bold></td><td align=\"left\"><bold>Category</bold></td></tr></thead><tbody><tr><td align=\"left\">Estrogen signaling pathway</td><td align=\"left\">IGFBP5 (AF055033, NM_000599)</td><td align=\"left\">BTG2, PTGER3</td><td align=\"left\">Breast cancer</td></tr><tr><td align=\"left\">p21 mediated pathway</td><td align=\"left\">BUB1B, CENPA</td><td align=\"left\">EX01, PRC1</td><td align=\"left\">Cell-cycle</td></tr><tr><td align=\"left\">CEBPalpha mediated pathway</td><td align=\"left\">MMP9</td><td align=\"left\">BTG2</td><td align=\"left\">Chronic myeloid leukemia</td></tr><tr><td align=\"left\">WNT signaling pathway</td><td align=\"left\">WNT11</td><td align=\"left\">FGF18</td><td align=\"left\">Colon Cancer</td></tr><tr><td align=\"left\">Tat signaling pathway</td><td align=\"left\">PSMD2</td><td align=\"left\">PSMD7</td><td align=\"left\">Acquired immuno deficiency syndrome</td></tr><tr><td align=\"left\">Responsive genes</td><td align=\"left\">MMP9</td><td align=\"left\">BTG2</td><td align=\"left\">Prostate cancer</td></tr><tr><td align=\"left\">Responsive genes</td><td align=\"left\">MMP9, IGFBP5 (AF055033, NM_000599)</td><td align=\"left\">PRAME</td><td align=\"left\">Ovarian cancer</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>The prediction confusion matrix</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Observation</bold></td><td align=\"center\" colspan=\"2\"><bold>Prediction</bold></td></tr><tr><td/><td colspan=\"2\"><hr/></td></tr><tr><td/><td align=\"center\">+1</td><td align=\"center\">-1</td></tr></thead><tbody><tr><td align=\"center\">+1</td><td align=\"center\">TP (True positive)</td><td align=\"center\">FN (False negative)</td></tr><tr><td align=\"center\">-1</td><td align=\"center\">FP (False positive)</td><td align=\"center\">TN (True negative)</td></tr></tbody></table></table-wrap>" ]

[ "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S9-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>1</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S9-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>2</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S9-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>R</mml:mi><mml:msub><mml:mi>E</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:msqrt><mml:mrow><mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mi>k</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>g</mml:mi></mml:munderover><mml:mrow><mml:msub><mml:mi>λ</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow><mml:mo>/</mml:mo><mml:mrow><mml:mi>p</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>n</mml:mi><mml:mo>−</mml:mo><mml:mi>k</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mrow></mml:mrow></mml:msqrt></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S9-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>D</mml:mi><mml:msub><mml:mi>P</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mo stretchy=\"false\">[</mml:mo><mml:mrow><mml:mrow><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>12</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mtext>T</mml:mtext></mml:msup><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>12</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>+</mml:mo><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>21</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mtext>T</mml:mtext></mml:msup><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>21</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo></mml:mrow><mml:mo>/</mml:mo><mml:mrow><mml:mo stretchy=\"false\">[</mml:mo><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>11</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mtext>T</mml:mtext></mml:msup><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>11</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>+</mml:mo><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>22</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mtext>T</mml:mtext></mml:msup><mml:mo stretchy=\"false\">(</mml:mo><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>22</mml:mn></mml:mrow></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">]</mml:mo></mml:mrow></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S9-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>11</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1471-2105-9-S9-S9-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>12</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1471-2105-9-S9-S9-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>22</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1471-2105-9-S9-S9-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>21</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1471-2105-9-S9-S9-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>1</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M10\" name=\"1471-2105-9-S9-S9-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>2</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M11\" name=\"1471-2105-9-S9-S9-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>1</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M12\" name=\"1471-2105-9-S9-S9-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>2</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM1\"><label>(1)</label><bold>X </bold>= <bold>TP</bold></disp-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M13\" name=\"1471-2105-9-S9-S9-i9\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>R</mml:mi>\n <mml:msub>\n <mml:mi>E</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:msqrt>\n <mml:mrow>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mi>k</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>g</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>k</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n </mml:msqrt>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM3\"><label>(3)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M14\" name=\"1471-2105-9-S9-S9-i10\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>I</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:msub>\n <mml:mi>D</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>R</mml:mi>\n <mml:msub>\n <mml:mi>E</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n </mml:mrow>\n <mml:mrow>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>k</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mn>2</mml:mn>\n </mml:msup>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M15\" name=\"1471-2105-9-S9-S9-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>1</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M16\" name=\"1471-2105-9-S9-S9-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>2</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM4\"><label>(4)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M17\" name=\"1471-2105-9-S9-S9-i11\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mrow>\n <mml:mtext>DP</mml:mtext>\n </mml:mrow>\n <mml:mi>j</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>12</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mtext>T</mml:mtext>\n </mml:msup>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>12</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>+</mml:mo>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>21</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mtext>T</mml:mtext>\n </mml:msup>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>21</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>11</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mtext>T</mml:mtext>\n </mml:msup>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>11</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>+</mml:mo>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>22</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mtext>T</mml:mtext>\n </mml:msup>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msubsup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>e</mml:mi>\n </mml:mstyle>\n <mml:mi>j</mml:mi>\n <mml:mrow>\n <mml:mn>22</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M18\" name=\"1471-2105-9-S9-S9-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>11</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M19\" name=\"1471-2105-9-S9-S9-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>12</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M20\" name=\"1471-2105-9-S9-S9-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>22</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M21\" name=\"1471-2105-9-S9-S9-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mstyle mathvariant=\"bold\" mathsize=\"normal\"><mml:mi>e</mml:mi></mml:mstyle><mml:mi>j</mml:mi><mml:mrow><mml:mn>21</mml:mn></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M22\" name=\"1471-2105-9-S9-S9-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>1</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M23\" name=\"1471-2105-9-S9-S9-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mtext>X</mml:mtext><mml:mrow><mml:mn>2</mml:mn><mml:mi>m</mml:mi></mml:mrow><mml:mo>'</mml:mo></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM5\"><label>(5)</label><bold>E </bold>= <bold>X </bold>- <bold>XPP</bold>^T,</disp-formula>", "<disp-formula id=\"bmcM6\"><label>(6)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M24\" name=\"1471-2105-9-S9-S9-i12\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>A</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>u</mml:mi>\n <mml:mi>r</mml:mi>\n <mml:mi>a</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>y</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>N</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM7\"><label>(7)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M25\" name=\"1471-2105-9-S9-S9-i13\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>S</mml:mi>\n <mml:mi>p</mml:mi>\n <mml:mi>e</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mi>f</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mi>t</mml:mi>\n <mml:mi>y</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>P</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM8\"><label>(8)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M26\" name=\"1471-2105-9-S9-S9-i14\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>S</mml:mi>\n <mml:mi>e</mml:mi>\n <mml:mi>n</mml:mi>\n <mml:mi>s</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mi>t</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mi>v</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mi>t</mml:mi>\n <mml:mi>y</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>N</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM9\"><label>(9)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M27\" name=\"1471-2105-9-S9-S9-i15\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>M</mml:mi>\n <mml:mi>C</mml:mi>\n <mml:mi>C</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>×</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>×</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>N</mml:mi>\n </mml:mrow>\n <mml:mrow>\n <mml:msqrt>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>×</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>×</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>×</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:msqrt>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>" ]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S9-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S9-2\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>A fundamental step in most microarray experiments is determining one or more short lists of differentially expressed genes (DEGs) that distinguish biological conditions, such as disease from health. Challenges regarding the reliability of microarray results have largely been founded on the inability of researchers to replicate DEG lists across highly similar experiments. For example, Tan <italic>et al</italic>. [##REF##14500831##1##] found only four common DEGs using an identical set of RNA samples across three popular commercial platforms. Independent studies by the groups of Ramalho-Santos [##REF##12228720##2##] and Ivanova [##REF##12228721##3##] of stem cell-specific genes using the same Affymetrix platform and similar study design found a disappointing six common DEGs among about 200 identified in each study [##REF##14563990##4##]. A comparative neurotoxicological study by Miller <italic>et al</italic>. [##REF##15329391##5##] using the same set of RNA samples found only 11 common DEGs among 138 and 425, respectively, from Affymetrix and CodeLink platforms. All these studies ranked genes by <italic>P</italic>-value from simple <italic>t</italic>-tests, used a <italic>P </italic>threshold to identify DEG lists, and applied the concept of the Percentage of Overlapping Genes (POG), or the Venn diagram, between DEG lists as the measure of reproducibility.</p>", "<p>Criticism of and concerns about microarrays continue to appear in some of the most prestigious scientific journals [##REF##15122300##6##, ####REF##15705458##7##, ##REF##15705441##8##, ##REF##15902768##9##, ##REF##15499004##10####15499004##10##], leading to a growing negative perception regarding microarray reproducibility, and hence reliability. However, in reanalyzing the data set of Tan <italic>et al</italic>. [##REF##14500831##1##], Shi <italic>et al</italic>. [##REF##16026597##11##] found that cross-platform concordance was markedly improved when either simple fold change (FC) or Significance Analysis of Microarrays (SAM) [##REF##11309499##12##] methods were used to rank order genes before determining DEG lists. The awareness that microarray reproducibility is sensitive to how DEGs are identified was, in fact, a major motivator for the MicroArray Quality Control (MAQC) project [##REF##16026597##11##,##REF##16964229##13##,##REF##18155896##14##].</p>", "<p>Several plausible explanations and solutions have been proposed to interpret and address the apparent lack of reproducibility and stability of DEG lists from microarray studies. Larger sample sizes [##REF##16585533##15##]; novel, microarray-specific statistical methods [##REF##16369572##16##]; more accurate array annotation information by mapping probe sequences across platforms [##REF##14500831##1##,##REF##15161944##17##]; eliminating absent call genes from data analysis [##REF##16026597##11##,##REF##12805270##18##,##REF##15345031##19##]; improving probe design to minimize cross-hybridization [##REF##15161944##17##]; standardizing manufacturing processes [##REF##14500831##1##]; and improving data quality by fully standardizing sample preparation and hybridization procedures are among the suggestions for improvement [##REF##14970825##20##].</p>", "<p>The MAQC study [##REF##16964229##13##] was specifically designed to address these previously identified sources of variability in DEG lists. Two distinct RNA samples, Stratagene Universal Human Reference RNA (<italic>i.e</italic>., MAQC sample A) and Ambion Human Brain Reference RNA (<italic>i.e</italic>., MAQC sample B), with thousands of differentially expressed genes, were prepared in sufficient quantities and distributed to three different laboratories for each of the five different commercial whole genome microarray platforms participating in the study. For each platform, each sample was analyzed using five technical replicates with standardized procedures for sample processing, hybridization, scanning, data acquisition, data preprocessing, and data normalization at each site. The probe sequence information was used to generate a stringent mapping of genes across the different platforms and 906 genes were further analyzed with TaqMan^®assays using the same RNA samples.</p>", "<p>In addition to assessing the technical performance of different microarray platforms, the MAQC study also discussed the idea of using fold-change ranking along with a non-stringent <italic>P</italic>-value cutoff for selecting DEGs [##REF##16964229##13##,##REF##17061323##21##]. However, a lot of detailed results have not been formally published to support the idea [##UREF##0##22##]. The MAQC project, while positively received by the community [##REF##17110319##23##, ####REF##16943838##24##, ##REF##16973852##25##, ##UREF##1##26##, ##UREF##2##27####2##27##], also stimulated criticism from the statistical community about appropriate ways of identifying DEGs [##UREF##0##22##,##REF##17110319##23##,##UREF##2##27##, ####REF##17211383##28##, ##UREF##3##29##, ##UREF##4##30##, ##REF##16940966##31##, ##REF##17961233##32##, ##UREF##5##33####5##33##].</p>", "<p>To help the microarray community better understand the issue at debate and move forward, in this study, we conducted a careful analysis of these MAQC data sets, along with numerical simulations and mathematical arguments. We demonstrate that the reported lack of reproducibility of DEG lists can be attributed in large part to identifying DEGs from simple <italic>t</italic>-tests without consideration of FC. The finding holds for intra-laboratory, inter-laboratory, and cross-platform comparisons independent of sample pairs and normalization methods, and is increasingly apparent with decreasing number of genes selected.</p>", "<p>As a basic procedure for improving reproducibility while balancing specificity and sensitivity, choosing genes using a combination of FC-ranking and <italic>P </italic>threshold was investigated. This joint criterion results in DEG lists with much higher POG, commensurate with better reproducibility, than lists generated by <italic>t</italic>-test <italic>P </italic>alone, even when selecting a relatively small numbers of genes. An FC criterion explicitly incorporates the measured quantity to enhance reproducibility, whereas a <italic>P </italic>criterion incorporates control of sensitivity and specificity. The results increase our confidence in the reproducibility of microarray studies while supporting a need for caution in the use of inferential statistics when selecting DEGs. While numerous more advanced statistical modeling techniques have been proposed and compared for selecting DEGs [##REF##16369572##16##,##REF##15479783##34##,##UREF##6##35##], the primary objectives here are to explain that the primary reason for microarray reproducibility concerns is failure to include an FC criterion during gene selection, and to recommend a simple and straightforward approach concurrently satisfying statistical and reproducibility requirements. It should be stressed that robust methods are needed to meet stringent clinical requirements for reproducibility, sensitivity and specificity of microarray applications in, for example, clinical diagnostics and prognostics.</p>" ]

[ "<title>Methods</title>", "<title>MAQC data sets</title>", "<p>The MAQC data sets analyzed in this study are available from GEO under series accession number GSE5350. Analyses identified differentially expressed genes between the primary samples A (Stratagene Universal Human Reference RNA, Catalog #740000) and B (Ambion Human Brain Reference RNA, Catalog #6050) of the MAQC study. Analyses are additionally limited to data sets from the following five commercial genome-wide microarray platforms: ABI (Applied Biosystems), AFX (Affymetrix), AG1 (Agilent one-color), GEH (GE Healthcare), and ILM (Illumina), and to the subset of \"12,091\" genes commonly probed by them. TaqMan^®assay data for 906 genes are used to examine gene list comparability between microarrays and TaqMan^®assays. For more information about the MAQC project and the data sets, refer to Shi <italic>et al </italic>[##REF##16964229##13##].</p>", "<title>Normalization methods</title>", "<p>The following manufacturer's preferred normalization methods were used: quantile normalization for ABI and ILM, PLIER for AFX, and median-scaling for AG1 and GEH [##REF##16964229##13##]. For quantile normalization (including PLIER), each test site is independently considered.</p>", "<title>Gene ranking (selection) rules</title>", "<p>Six gene ranking (selection) methods were examined: (1) FC (fold change ranking); (2) FC_<italic>P</italic>0.05 (FC-ranking with <italic>P </italic>cutoff of 0.05); (3) FC_<italic>P</italic>0.01 (FC-ranking with <italic>P </italic>cutoff of 0.01); (4) <italic>P (P</italic>-ranking, simple t-test assuming equal variance); (5) <italic>P</italic>_FC2 (<italic>P</italic>-ranking with FC cutoff of 2); (6) <italic>P</italic>_FC1.4 (<italic>P</italic>-ranking with FC cutoff of 1.4). When a cutoff value (<italic>e.g</italic>., <italic>P </italic>&lt; 0.05) is imposed for a ranking metric (<italic>e.g</italic>., FC), it is likely that the lists of candidate genes that meet the cutoff value may not be the same for the two test sites or two platforms as a result of differences in inter-site or cross-platform variations. Such differences are part of the gene selection process and have been carried over to the gene ranking/selection stage.</p>", "<title>Evaluation criterion – POG (percentage of overlapping genes)</title>", "<p>The POG (percentage of overlapping genes) calculation [##REF##16026597##11##,##REF##16964229##13##] was applied in three types of comparisons: (1) Inter-site comparison using data from the three test sites of each platform; (2) Cross-platform comparison between ABI, AFX, AG1, GEH, and ILM using data from test site 1; for each sample pair, there are ten cross-platform pairs for comparison; (3) Microarray versus TaqMan^®assay comparisons.</p>", "<p>POG is calculated for many different cutoffs that can be considered as arbitrary.</p>", "<p>The number of genes considered as differentially expressed is denoted as 2L, where L is both the number of genes up- and down-regulated. The number of genes available for ranking and selection in one direction, L, varies from 1 to 6000 (with a step of one) or when there are no more genes in one regulation direction, corresponding to 2L varying from 2 to 12,000. Directionality of gene regulation is considered in POG calculations; genes selected by two sites or platforms but with different regulation directionalities are considered as discordant. Therefore, POG can hardly reach 100% in reality.</p>", "<p>The formula for calculating POG is: POG = 100*(DD+UU)/2L, where DD and UU are the number of commonly down- or up-regulated genes, respectively, from the two lists, and L is the number of genes selected from the up- or down-regulation directionality. To overcome the confusion of different numbers for the denominator, in our POG calculations we deliberately selected an equal number of up-regulated and down-regulated genes, L [##REF##16026597##11##]. The POG graphs shown in this study are essentially the same as the CAT (correspondence at the top) plots introduced by Irizarry <italic>et al</italic>. [##REF##15846361##55##] and the POG graphs that we introduced previously [##REF##16026597##11##] except that in the current POG graphs the x-axis is in log-scale to emphasize the details when fewer genes are selected.</p>", "<title>Noise-filtering</title>", "<p>Most of the analyses in this study exclude flagging information; that is, the entire set of \"12,091\" genes is used in the analyses. Some calculations are limited to subsets of genes commonly detectable (\"common present\") by the two test sites or two platforms under comparison. To be denoted as \"commonly present\", the gene is detected (\"present\") in the majority of replicates (<italic>e.g</italic>., three or more when there are five replicates) for each sample in a sample-pair comparison and for each test site or platform.</p>", "<title>Gene selection simulation</title>", "<p>A simulation was created to emulate the characteristics of the MAQC dataset. Fifteen thousand simulated genes were created where 5,000 were undifferentiated in expression between simulated biological samples A and B and 10,000 were differentiated but at various levels (exponential distribution for the log FC, where almost 4,000 are differentiated two-fold or higher, similar to a typical platform in the MAQC study, divided equally into up and down regulated genes). To simulate instances of technical or biological replicates, multiplicative noise (error) was added to the signal for each gene for each of five simulated replicates for each sample using an error distribution that would produce a replicate CV similar to that typically seen in the MAQC data set (<italic>i.e</italic>., the mean replicate CV would be roughly 10%). The CV for any given gene was randomly selected from a trimmed exponential distribution. To address a variety of additional error scenarios but preserving the same distribution of fold change, we also considered three additional mean CV values (2%, 35%, and 100%). To understand the impact of gene list size on the stability of the DEG list, list sizes of 10, 25, 100, and 500 genes were examined for each mean CV scenario. Several gene selection rules were compared: FC-ranking only, <italic>P</italic>-ranking only, and shrunken t-statistic ranking. Note: <italic>P</italic>-ranking is equivalent to t-statistic ranking as well as ranking based on FDR that monotonically transforms the <italic>P</italic>-value. In addition, shrunken t-statistic ranking is equivalent to ranking based on the test statistic used by SAM and related methods. In addition, rules based on FC-ranking with a <italic>P </italic>threshold were also compared (for <italic>P </italic>= 0.1, 0.01, 0.001, and 0.0001). Finally, to simulate differences in the variation patterns of analytes between platforms and even between laboratories, covariance between laboratories/platforms of the variance for each gene was included in the simulations. For a given mean CV, 20 or more simulated instances of five replicates of simulated biological samples A and B were created and DEG lists were prepared for each instance that were rank ordered using the methods described above. To determine reproducibility of a given method for a given mean CV under a given gene list size, the rank-ordered gene lists from these 20 instances were pair-wise compared for consistency and reproducibility. The results presented in the graphs are averages from those pair-wise comparisons.</p>", "<p>The MAQC actual data is characterized by large magnitudes of differential expression among the vast majority of the 12,091 common genes, with some 4000 exhibiting FC &gt; 2 and hundreds with FC &gt; 10. As such, the data may be atypical of commonplace microarray experiments with biological effects. Consequently, two other simulation data sets were created with far fewer genes with large FC, as might be expected in some actual microarray experiments. Specifically, the Small-Delta data set was created with fewer than 50 genes with FC &gt; 2, and a FC &lt; 1.3 for most differentiated genes, and 10,000 undifferentiated genes. In addition, the variances of the genes were correlated similar to that observed in the MAQC data. The third simulated dataset, termed the Medium-Delta set, had a large number of differentiated genes similar to the MAQC simulated dataset, but with small FC similar to the Small-Delta set. Again, gene variances were correlated similar to that observed in the MAQC data.</p>" ]

[ "<title>Results</title>", "<p>The POG for a number of gene selection scenarios employing <italic>P </italic>and/or FC are compared and a numerical example (see side box) is provided that shows how the simple <italic>t</italic>-test, when sample size is small, results in selection of different genes purely by chance. While the data generate from the MAQC samples A and B lack biological variability, the results are supported by the toxicogenomic data of Guo <italic>et al</italic>. [##REF##17061323##21##] While <italic>P </italic>could be computed from many different statistical methods, for simplicity and consistency, throughout this article <italic>P </italic>is calculated with the two-tailed <italic>t</italic>-test that is widely employed in microarray data analysis.</p>", "<title>Inter-site concordance for the same platform</title>", "<p>Figure ##FIG##0##1## gives plots of inter-site POG versus the number of genes selected as differentially expressed. Since there are three possible inter-site comparisons (S1–S2, S1–S3, and S2–S3, where S = Site) and six gene selection methods (see Methods), there are 18 POG lines for each platform. Figure ##FIG##0##1## shows that inter-site reproducibility in terms of POG heavily depends on the number of chosen differential genes and the gene ranking criterion: Gene selection using FC-ranking gives consistently higher POG than <italic>P</italic>-ranking. The POG from FC-ranking is near 90% for as few as 20 genes for most platforms, and remains at this high inter-site concordance level as the number of selected genes increases. In contrast, the POG from <italic>P</italic>-ranking is in the range of 20–40% for as many as 100 genes, and then asymptotically approaches 90% only after several thousand genes are selected.</p>", "<p>The POG is higher when the analyses are limited to the genes commonly detected (\"Present\" in the majority of replicates for each sample) by both test sites under comparison (Figure ##FIG##1##2##). In addition to a slight increase (2–3%), the inter-site POG lines after noise filtering are more stable than those before noise filtering, particularly for ABI, AG1, and GEH. Furthermore, differences between the three ILM test sites are further decreased after noise filtering, as seen from the convergence of the POG for S1–S2, S1–S3, and S2–S3 comparisons. Importantly, noise filtering does not change either the trend or magnitude of the higher POG graphs for FC-ranking compared with <italic>P</italic>-ranking.</p>", "<p>Inter-site concordance for different FC- and <italic>P</italic>-ranking criteria were also calculated for other MAQC sample pairs having much smaller differences than for sample A versus sample B, and correspondingly lower FC. In general, POG is much lower for other sample pairs regardless of ranking method and ranking order varies more greatly, though FC-ranking methods still consistently give a higher POG than <italic>P</italic>-ranking methods. Figure ##FIG##2##3## gives the plots of POG for Sample C versus Sample D, a 3:1 and 1:3 (A:B) mixture, respectively[##REF##16964229##13##,##REF##16964226##36##], for all inter-site comparisons.</p>", "<p>The substantial difference in inter-site POG shown in Figures ##FIG##0##1## and ##FIG##1##2## is a direct result of applying different gene selection methods to the same data sets, and clearly depicts how perceptions of inter-site reproducibility can be affected for any microarray platform. While the emphasis here is on reproducibility in terms of POG, in practice, this criterion must be balanced against other desirable characteristics of gene lists, such as specificity and sensitivity (when the truth is binary) or mean squared error (when the truth is continuous), considerations that that are discussed further in later sections.</p>", "<title>Cross-platform concordance</title>", "<p>Figure ##FIG##3##4## shows the substantial effect that FC- and <italic>P</italic>-ranking based gene selection methods have on cross-platform POG. Similar to inter-site comparisons, <italic>P</italic>-ranking results in much lower cross-platform POG than FC-ranking. When FC is used to rank DEGs from each platform, the cross-platform POG is around 70–85%, depending on the platform pair. The platforms themselves contribute about 15% differences in the cross-platform POG, as seen from the spread of the blue POG lines. Noise filtering improves FC-ranking based cross-platform POG by about 5–10% and results in more stable POG when a smaller number of genes are selected (Figure ##FIG##3##4b##). Importantly, the relative differences between FC- and <italic>P</italic>-ranking methods remain the same after filtering.</p>", "<title>Concordance between microarray and TaqMan^®assays</title>", "<p>TaqMan^®real-time PCR assays are widely used to validate microarray results [##REF##16551369##37##,##REF##16823376##38##]. In the MAQC project, the expression levels of 997 genes randomly selected from available TaqMan^®assays have been quantified in the four MAQC samples [##REF##16964229##13##,##REF##16964225##39##]. Nine hundred and six (906) of the 997 genes are among the \"12,091\" set of genes found on all of the six genome-wide microarray platforms [##REF##16964229##13##]. There are four technical replicates for each sample and the DEGs for TaqMan^®assays were identified using the same six gene selection procedures as those used for microarray data. The DEGs calculated from the microarray data are compared with DEGs calculated from TaqMan^®assay data. With noise filtering (<italic>i.e</italic>., focusing on the genes detected by both the microarray platform and TaqMan^®assays), 80–85% concordance was observed (Figure ##FIG##4##5##). Consistent with inter-site and cross-platform comparisons, POGs comparing microarray with TaqMan^®assays also show that ranking genes by FC results in markedly higher POG than ranking by <italic>P </italic>alone, especially for short gene lists. POG results without noise filtering (Figure ##FIG##5##6##) are some 5% lower but the notable differences in POG between the FC- and <italic>P</italic>-ranking are unchanged.</p>", "<title>Reproducibility of FC and t-statistic: different metrics for identifying differentially expressed genes (DEGs)</title>", "<p>Figure ##FIG##6##7## shows that the inter-site reproducibility of log2 FC (panel a) is much higher than that of log2 t-statistic (panel b). In addition, the relationship between log2 FC and log2 t-statistic from the same test site is non-linear and the correlation appears to be low (panel c). We see similar results when data from different microarray platforms are compared to each other or when microarray data are compared against TaqMan^®assay data (results not shown). The differences between the reproducibility of FC and <italic>t</italic>-statistic observed here are consistent with the differences between POG results in inter-site (Figures ##FIG##0##1##, ##FIG##1##2##, ##FIG##2##3##), cross-platform (Figure ##FIG##3##4##), and microarray versus TaqMan^®assay (Figures ##FIG##4##5## and ##FIG##5##6##) comparisons. The nonlinear relationship between log2 FC and log2 t-statistic (Figure ##FIG##6##7c##) leads to low concordance between the list of DEGs derived from FC-ranking and the list derived from t-statistic (<italic>P</italic>) ranking (Figure ##FIG##7##8##); an expected outcome due to the different emphases of FC and <italic>P</italic>.</p>", "<title>Joint FC and <italic>P </italic>rule illustrated with a volcano plot: ranking by FC, not by <italic>P</italic></title>", "<p>Figure ##FIG##8##9## is a volcano plot depicting how a joint FC and <italic>P </italic>rule works in gene selection. It uses the MAQC Agilent data, and plots negative log10 <italic>P </italic>on the y-axis versus log2 FC on the x-axis. A joint rule chooses genes that fall in the upper left and upper right sections of the plot (sections A and C of Figure ##FIG##8##9##). Other possible cutoff rules for combining FC and <italic>P </italic>are apparent, but are precluded from inclusion due to space. An important conclusion from this study is that genes should be ranked and selected by FC (x-axis) with a non-stringent <italic>P </italic>threshold in order to generate reproducible lists of DEGs. The more stringent the <italic>P</italic>-value threshold, the less reproducible the DEG list is. Our results provide practical guidance to choose the appropriate FC and <italic>P</italic>-value cutoffs in using the \"volcano plots\" to select DEGs.</p>", "<title>Concordance using other statistical tests</title>", "<p>Numerous different statistical tests including rank tests (<italic>e.g</italic>., Wilcoxon rank-sum test) and shrunken t-tests (<italic>e.g</italic>., SAM) have been used for the identification of DEGs. Although this work is not intended to serve as a comprehensive performance survey of different statistical procedures, we set out to briefly examine a few examples due to their popularity. Figure ##FIG##9##10## shows the POG results of several commonly used approaches including FC-ranking, t-test statistic, Wilcoxon rank-sum test, and SAM using AFX site-site comparison as an example[##REF##16964229##13##]. The POG by SAM (pink line), although greatly improved over that of simple t-test statistic (purple line), approached, but did not exceed, the level of POG based on FC-ranking (green line). In addition, the small numbers of replicates in this study rendered many ties in the Wilcoxon rank statistic, resulting in poor inter-site concordance in terms of rank-order of the DEGs between the two AFX test sites. Similar findings (data not shown) were observed using the toxicogenomics data set of Guo <italic>et al</italic>. [##REF##17061323##21##].</p>", "<title>Gene selection in simulated datasets</title>", "<p>The MAQC data, like data from actual experiments, allows evaluation of DEG list reproducibility, but not of truth. Statistics are used to estimate truth, often in terms of sensitivity and specificity, but the estimates are based on assumptions about data variance and error structure that are also unknown. Simulations where truth can be specified <italic>a priori </italic>are useful to conduct parametric evaluations of gene selection methods, and true false positives and negatives are then known. However, results are sensitive to assumptions regarding data structure and error that for microarrays remains poorly characterized.</p>", "<p>Figure ##FIG##10##11## gives POG versus the number of genes for three simulated data sets (MAQC-simulated set, Small-Delta simulated set, and Medium-Delta simulated set, see Methods) that were prepared in order to compare the same gene selection methods as the MAQC data. The MAQC-simulated set was created to emulate the magnitude and structure of differential expression observed between the actual MAQC samples A and B (<italic>i.e</italic>., thousands of genes with FC &gt; 2). By comparison, the Small-Delta simulated data set had only 50 significant genes with FC &gt; 2 and most genes had FC &lt; 1.3. The Medium-Delta data set had FC profiles in between.</p>", "<p>For the MAQC-simulated data, either FC-ranking or FC-ranking combined with any of the <italic>P </italic>threshold resulted in markedly higher POG than any <italic>P</italic>-ranking method, regardless of gene list length and coefficient of variation (CV) of replicates. The POG is ~100%, ~95%, and ~75%, for replicate CV values of 2%, 10%, and 35% CV, respectively, decreasing to about 20–30% with an exceedingly high (100%) CV. In contrast, POG from <italic>P</italic>-ranking alone varies from a few percent to only ~10% when 500 genes are selected.</p>", "<p>For the Medium- and Small-Delta simulated data sets, differences start to emerge between using FC alone and FC with <italic>P </italic>cutoff. From Figure ##FIG##10##11##, when variances in replicates become larger (CV &gt; 10%), the reproducibility is greatly enhanced using FC-ranking with a suitable <italic>P </italic>cutoff versus FC or <italic>P </italic>by themselves. In addition, when variances are small (CV ≤ 10%), the reproducibility is essentially the same for FC with <italic>P </italic>or without. What is clear is that <italic>P </italic>by itself did not produce the most reproducible DEG list under any simulated condition.</p>", "<p>Although <italic>P</italic>-ranking generally resulted in very low POG, a false positive was rarely produced, even for a list size of 500 (data not shown). Thus, the <italic>P </italic>criterion performed as expected, and identified mostly true positives. Unfortunately, the probability of selection of the same true positives with a fixed <italic>P </italic>cutoff in a replicated experiment appears small due to variation in the <italic>P </italic>statistic itself (see inset). FC-ranking by itself resulted in a large number of false positives with a large number of genes for the Medium and Small-Delta sets when genes with small FCs are selected as differentially expressed. These false positives were greatly reduced to the same level as for the <italic>P</italic>-ranking alone when FC-ranking was combined with a <italic>P</italic>-cutoff.</p>", "<title>Variability of the two-sample t-statistic</title>", "<p>In a two-sample t-test comparing the mean of sample A to the mean of sample B, the t-statistic is given as follows:</p>", "<p></p>", "<p>where is the average of the log2 expression levels of sample A with <italic>n</italic>_{<italic>A </italic>}replicates, and is the average of the log2 expression levels of sample B with <italic>n</italic>_{<italic>B </italic>}replicates, and <italic>S</italic>_{<italic>p</italic>}²= (<italic>SS</italic>_{<italic>A</italic>}+<italic>SS</italic>_{<italic>B</italic>})/(<italic>n</italic>_{<italic>A</italic>}+<italic>n</italic>_{<italic>B</italic>}-<italic>2</italic>) is the pooled variance of samples A and B, and <italic>SS </italic>denotes the sum of squared errors. The numerator of the t-statistic is the fold-change (FC) in log2 scale and represents the signal level of the measurements (<italic>i.e</italic>., the magnitude of the difference between the expression levels of sample A and sample B). The denominator represents the noise components from the measurements of samples A and B. Thus, the t-statistic represents a measure of the signal-to-noise ratio. Therefore, the FC and the t-statistic (<italic>P</italic>) are two measures for the differences between the means of sample A and sample B. The t-statistic is intrinsically less reproducible than FC when the variance is small.</p>", "<p>Assume that the data are normally distributed, the variances of samples A and B are equal (<italic>σ</italic>²), the numbers of replicates in samples A and B are equal (<italic>n = n</italic>_{<italic>A </italic>}= <italic>n</italic>_{<italic>B</italic>}), and that there is a real difference in the mean values between samples A and B, <italic>d </italic>(the true FC in log2 scale). Then the t-statistic has a non-central t-distribution with non-central parameter</p>", "<p></p>", "<p>and the mean and variance of the <italic>t</italic>-statistic (Johnson and Kotz, 1970) are</p>", "<p></p>", "<p>where <italic>v </italic>= (<italic>2n-2</italic>) and is the degrees of freedom of the non-central t-distribution. When <italic>d </italic>= 0 (the two means are equal), then the t-statistic has a t-distribution with mean <italic>E</italic>(<italic>t</italic>) = 0 and <italic>Var</italic>(<italic>t</italic>) = <italic>v</italic>/(<italic>v</italic>-2). The variance of the t-statistic depends on the sample size <italic>n</italic>, the magnitude of the difference between the two means <italic>d</italic>, and the variance <italic>σ</italic>². On the other hand, the variance of the mean difference for the FC is (2/<italic>n</italic>)<italic>σ</italic>². That is, the variance of the FC depends only on the sample size <italic>n </italic>and the variance <italic>σ</italic>², regardless of the magnitude of the difference <italic>d </italic>between the two sample means.</p>", "<p>In an MAQC data set, a typical sample variance for the log2 expression levels is approximately <italic>σ</italic>²= 0.15². With <italic>n </italic>= 5, the standard deviation of the FC (in log2 scale) is 0.09. For a differentially expressed gene with a 4-fold change between 5 replicates of sample A and 5 replicates of sample B, <italic>d </italic>= 2 and the t-values have a non-central t-distribution with (<italic>ν </italic>= <italic>n</italic>_A+<italic>n</italic>_B-2) = 8 degrees of freedom and <italic>δ </italic>= 21.08. From the equations above, the mean and the variance of the t-values are E(<italic>t</italic>) = 23.35 and Var(<italic>t</italic>) = 6.96². Within two standard deviations the expected value of the t-value ranges from 9.43 (= 23.35-2 × 6.96) to 37.27 (= 23.35+2 × 6.96), corresponding to <italic>P</italic>s from 1 × 10^-5to 3 × 10^-10, based on the Student's two-sided t-test with 8 degrees of freedom. In contrast, when <italic>n </italic>= 5 the standard deviation of the FC (in log2 scale) is 0.09. The expected value of the FC ranges only from 3.53 (= 2^1.82) to 4.53 (= 2^2.18) within two standard deviations. In this case, this gene would be selected as differentially expressed using either a FC cutoff of 3.5 or a <italic>P </italic>cutoff of 1 × 10^-5. On the other hand, for a gene with a 2-fold change (<italic>d </italic>= 1), the t-statistic has a non-central t-distribution with <italic>δ </italic>= 10.54. The mean and the variance of the t-statistic are E(<italic>t</italic>) = 11.68 and Var(<italic>t</italic>) = 3.62²with a corresponding <italic>P </italic>of 3 × 10^-6at t = 11.68. Using the same <italic>P </italic>cutoff, 1 × 10^-5, this gene is likely to be selected with the probability greater than 0.5. For the FC criterion, the expected value of the FC ranges from 1.76 (= 2^0.82) to 2.26 (= 2^1.18). Using the same FC cutoff of 3.5, this gene is very unlikely to be selected. Thus, the top ranked gene list based on the FC is more reproducible than the top ranked gene list based on the <italic>P</italic>. The top ranked genes selected by a <italic>P </italic>cutoff may not be reproducible between experiments although both lists may contain mostly differentially expressed genes.</p>", "<p>Reference: N. Johnson and S. Kotz (1970). Continuous Univariate Distributions – 2. Houghton Mifflin, Boston.</p>" ]

[ "<title>Discussion</title>", "<p>A fundamental requirement in microarray experiments is that the identification of DEG lists must be reproducible if the data and scientific conclusion from them are to be credible. DEG lists are normally developed by rank-ordering genes in accordance with a suitable surrogate value to represent biological relevance, such as the magnitude of the differential expression (<italic>i.e</italic>., FC) or the measure of statistical significance (<italic>P</italic>) of the expression change, or both. The results show that concurrent use of both FC-ranking and <italic>P</italic>-cutoff as criteria to identify biological relevant genes can be essential to attain reproducible DEG lists across laboratories and platforms.</p>", "<p>A decade since the microarray-generated differential gene expression results of Schena <italic>et al</italic>[##REF##7569999##40##] and Lockhart <italic>et al</italic>[##REF##9634850##41##] were published, microarray usage has become ubiquitous. Over this time, many analytical techniques for identifying DEGs have been introduced and used. Early studies predominantly relied on the magnitude of differential expression change in experiments done with few if any replicates, with an FC cutoff typically of two used to reduce false positives. Mutch <italic>et al</italic>[##REF##11790248##42##] recommended using intensity-dependent FC cutoffs to reduce biased selection of genes with low expression.</p>", "<p>Gene selection using statistical significance estimates became more prevalent during the last few years as studies with replicates became possible. Incorporation of a t-statistic in gene selection was intended to compensate for the heterogeneity of variances of genes [##REF##11395427##43##]. Haslett <italic>et al</italic>. [##REF##12415109##44##] employed stringent values of both FC and <italic>P </italic>to determine DEGs. In recent years, there has been an increasing tendency to use <italic>P</italic>-ranking for gene selection. Kittleson <italic>et al</italic>. [##REF##15982506##45##] selected genes with a FC cutoff of two and a very restrictive Bonferroni corrected <italic>P </italic>of 0.05 in a quest for a short list of true positive genes. Tan <italic>et al</italic>. [##REF##12177426##46##] used <italic>P </italic>to rank genes. Correlation coefficient, which behaves similarly to the t-statistic, has also been widely used as a gene selection method in the identification of signature genes for classification purposes [##REF##16585533##15##,##REF##15308542##47##,##REF##15640445##48##].</p>", "<p>New and widely employed methods have appeared in recent years and implicitly correct for the large variance in the t-statistic that results when gene variance is estimated with a small number of samples. Allison <italic>et al</italic>. [##REF##16369572##16##] collectively described these methods as \"variance shrinkage\" approaches. They include the popular permutation-based \"SAM\" procedure [##REF##15329391##5##,##REF##11309499##12##,##REF##12933549##49##,##REF##15100403##50##], Bayesian-based approaches [##REF##11395427##43##,##REF##11339905##51##] and others [##REF##15618528##52##]. Qin <italic>et al</italic>. [##REF##15479783##34##] compared several variance shrinkage methods with a simple t-statistic and FC for spike-in gene identification on a two-color platform, concluding that all methods except <italic>P </italic>performed well. All these methods have the effect of reducing a gene's variance to be between the average for the samples, and the average over the arrays.</p>", "<p>In some cases, however, the use of FC for gene selection was criticized and entirely abandoned. For example, Callow <italic>et al</italic>. [##REF##11116096##53##], using <italic>P </italic>alone for identifying DEGs, concluded that <italic>P </italic>alone eliminated the need for filtering low intensity spots because the t-statistic is uniformly distributed across the entire intensity range. Reliance on <italic>P </italic>alone to represent a gene's FC and variability in gene selection has become commonplace. Norris and Kahn [##REF##16407153##54##] describe how false discovery rate (FDR) has become so widely used as to constitute a standard to which microarray analyses are held. However, FDR usually employs a shrunken t-statistic and genes are ranked and selected similar to <italic>P </italic>(see Figure ##FIG##10##11##).</p>", "<p>Prior to MAQC, Irizarry <italic>et al</italic>. [##REF##15846361##55##] compared data from five laboratories and three platforms using the CAT plots that are essentially the same as the POG graphs used in our study. Lists of less than 100 genes derived from FC-ranking showed 30 to 80% intra-site, inter-site, and inter-platform concordance. Interestingly, important disagreements were attributable to a small number of genes with large FC that they posit resulted from a laboratory effect due to inexperienced technicians and sequence-specific effects where some genes are not correctly measured.</p>", "<p>Exactly how to best employ FC with <italic>P </italic>to identify genes is a function of both the nature of the data, and the inevitable tradeoff between sensitivity and specificity that is familiar across research, clinical screening and diagnostics, and even drug discovery. But how the tradeoff is made depends on the application. Fewer false negatives at the cost of more false positives may be desirable when the application is identifying a few hundred genes for further study, and FC-ranking with a non-stringent <italic>P </italic>value cutoff from a simple t-test could be used to eliminate some noise. The gene list can be further evaluated in terms of gene function and biological pathway data, as illustrated in Guo <italic>et al</italic>. [##REF##17061323##21##] for toxicogenomic data. Even for relatively short gene lists, FC-ranking together with a non-stringent <italic>P </italic>cutoff should result in reproducible lists. In addition, DEG lists identified by the ranking of FC is much less susceptible to the impact of normalization methods. In fact, global scaling methods (<italic>e.g</italic>., median- or mean-scaling) do not change the relative ranking of genes based on FC; they do, however, impact gene ranking by <italic>P</italic>-value [##REF##17061323##21##].</p>", "<p>The tradeoffs between reproducibility, sensitivity, and specificity become pronounced when genes are selected by <italic>P </italic>alone without consideration of FC, especially when a stringent <italic>P </italic>cutoff is used to reduce false positives. When sample numbers are small, any gene's t-statistic can change considerably in repeated studies within or across laboratories or across platforms. Each study can select different significant genes, purely by chance. It is entirely possible that separately determined lists will have a small proportion of common genes even while each list comprises mostly true positives. This apparent lack of reproducibility of the gene lists is an expected outcome of statistical variation in the t-statistic for small numbers of sample replicates. In other words, each study fails to produce some, but not all, of the correct results. The side box provides a numerical example of how gene list discordance can result from variation in the t-statistic across studies. Decreasing the <italic>P </italic>cutoff will increase the proportion of true positives, but also diminish the number of selected genes, diminish genes common across lists, and increase false negatives. Importantly, selecting genes based on a small <italic>P </italic>cutoff derived from a simple t-test without consideration of FC renders the gene list non-reproducible.</p>", "<p>Additional insight is gained by viewing gene selection from the perspective of the biologist ultimately responsible for interpreting microarray results. Statistically speaking, a microarray experiment tests 10,000 or more null hypotheses where essentially all genes have non-zero differential expression. Statistical tests attempt to account for an unknowable error structure, in order to eliminate the genes with low probability of biological relevance. To the biologist, however, the variance of a gene with a large FC in one microarray study may be irrelevant if a similar experiment again finds the gene to have a large FC; the second experiment would probably be considered a validating reproduction. This conclusion would be reasonable since the gene's <italic>P </italic>depends on a poor estimate of variance across few samples, whereas a repeated FC measurement is tangible reproducibility which tends to increase demonstrably with increasing FC. The biological interpreter can also consider knowledge of gene function and biological pathways before finally assigning biological relevance, and will be well aware that either <italic>P </italic>or FC is only another indicator regarding biological significance.</p>", "<p>This study shows that genes with smaller expression fold changes generated from one platform or laboratory are, in general, less reproducible in another laboratory with the same or different platforms. However, it should be noted that genes with small fold changes may be biologically important [##REF##10929718##56##]. When a fixed FC cutoff is chosen, sensitivity could be sacrificed for reproducibility. Alternatively, when a high <italic>P </italic>cutoff (or no <italic>P </italic>cutoff) is used, specificity could be sacrificed for reproducibility. Ultimately, the acceptable trade-off is based on the specific question being asked or the need being addressed. When searching for a few reliable biomarkers, high FC and low <italic>P </italic>cutoffs can be used to produce a highly specific and reproducible gene list. When identifying the components of genetic networks involved in biological processes, a lower FC and higher <italic>P </italic>cutoff can be used to identify larger, more sensitive but less specific, gene lists. In this case, additional biological information about putative gene functions can be incorporated to identify reliable gene lists that are specific to the biological process of interest.</p>", "<p>Truly differentially expressed genes should be more likely identified as differentially expressed by different platforms and from different laboratories than those genes with no differential expression between sample groups. In the microarray field, we usually do not have the luxury of knowing the \"truth\" in a given study. Therefore, it is not surprising that most microarray studies and data analysis protocols have not been adequately evaluated against the \"truth\". A reasonable surrogate of such \"truth\" could be the consensus of results from different microarray platforms, from different laboratories using the same platform, or from independent methods such as TaqMan^®assays, as we have extensively explored in this study.</p>", "<p>The fundamental scientific requirement of reproducibility is a critical dimension to consider along with balancing specificity and sensitivity when defining a gene list. Irreproducibility would render microarray technology generally, and any research result, specifically, vulnerable to criticism. New methods for the identification of DEGs continue to appear in the scientific literature. These methods are typically promoted in terms of improved sensitivity (power) while retaining nominal rates of specificity. However, reproducibility is seldom emphasized.</p>" ]

[ "<title>Conclusion</title>", "<p>The results show that selecting DEGs based solely on <italic>P </italic>from a simple t-test most often predestines a poor concordance in DEG lists, particularly for small numbers of genes. In contrast, using FC-ranking in conjunction with a non-stringent <italic>P</italic>-cutoff results in more concordant gene lists concomitant with needed reproducibility, even for fairly small numbers of genes. Moreover, enhanced reproducibility holds for inter-site, cross-platform, and between microarray and TaqMan^®assay comparisons, and is independent of platforms, sample pairs, and normalization methods. The results should increase confidence in the reproducibility of data produced by microarray technology and should also expand awareness that gene lists identified solely based on <italic>P </italic>will tend to be discordant. This work demonstrates the need for a shift from the common practice of selecting differentially expressed genes solely on the ranking of a statistical significance measure (<italic>e.g</italic>., t-statistic) to an approach that emphasizes fold-change, a quantity actually measured by microarray technology.</p>", "<title>Conclusions and recommendations</title>", "<p>1. A fundamental step of microarray studies is the identification of a small subset of DEGs from among tens of thousands of genes probed on the microarray. DEG lists must be concordant to satisfy the scientific requirement of reproducibility, and must also be specific and sensitive for scientific relevance. A baseline practice is needed for properly assessing reproducibility/concordance alongside specificity and sensitivity.</p>", "<p>2. Reports of DEG list instability in the literature are often a direct consequence of comparing DEG lists derived from a simple t-statistic when the sample size is small and variability in variance estimation is large. Therefore, the practice of using <italic>P </italic>alone for gene selection should be discouraged.</p>", "<p>3. A DEG list should be chosen in a manner that concurrently satisfies scientific objectives in terms of inherent tradeoffs between reproducibility, specificity, and sensitivity.</p>", "<p>4. Using FC and <italic>P </italic>together balances reproducibility, specificity, and sensitivity. Control of specificity and sensitivity can be accomplished with a <italic>P </italic>criterion, while reproducibility is enhanced with an FC criterion. Sensitivity can also be improved by better platforms with greater dynamic range and lower variability or by increased sample sizes.</p>", "<p>5. FC-ranking should be used in combination with a non-stringent <italic>P </italic>threshold to select a DEG list that is reproducible, specific, and sensitive, and a joint rule is recommended as a baseline practice.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Reproducibility is a fundamental requirement in scientific experiments. Some recent publications have claimed that microarrays are unreliable because lists of differentially expressed genes (DEGs) are not reproducible in similar experiments. Meanwhile, new statistical methods for identifying DEGs continue to appear in the scientific literature. The resultant variety of existing and emerging methods exacerbates confusion and continuing debate in the microarray community on the appropriate choice of methods for identifying reliable DEG lists.</p>", "<title>Results</title>", "<p>Using the data sets generated by the MicroArray Quality Control (MAQC) project, we investigated the impact on the reproducibility of DEG lists of a few widely used gene selection procedures. We present comprehensive results from inter-site comparisons using the same microarray platform, cross-platform comparisons using multiple microarray platforms, and comparisons between microarray results and those from TaqMan – the widely regarded \"standard\" gene expression platform. Our results demonstrate that (1) previously reported discordance between DEG lists could simply result from ranking and selecting DEGs solely by statistical significance (<italic>P</italic>) derived from widely used simple <italic>t</italic>-tests; (2) when fold change (FC) is used as the ranking criterion with a non-stringent <italic>P</italic>-value cutoff filtering, the DEG lists become much more reproducible, especially when fewer genes are selected as differentially expressed, as is the case in most microarray studies; and (3) the instability of short DEG lists solely based on <italic>P</italic>-value ranking is an expected mathematical consequence of the high variability of the <italic>t</italic>-values; the more stringent the <italic>P</italic>-value threshold, the less reproducible the DEG list is. These observations are also consistent with results from extensive simulation calculations.</p>", "<title>Conclusion</title>", "<p>We recommend the use of FC-ranking plus a non-stringent <italic>P </italic>cutoff as a straightforward and baseline practice in order to generate more reproducible DEG lists. Specifically, the <italic>P</italic>-value cutoff should not be stringent (too small) and FC should be as large as possible. Our results provide practical guidance to choose the appropriate FC and <italic>P</italic>-value cutoffs when selecting a given number of DEGs. The FC criterion enhances reproducibility, whereas the <italic>P </italic>criterion balances sensitivity and specificity.</p>" ]

[ "<title>Disclaimer</title>", "<p>This document has been reviewed in accordance with United States Food and Drug Administration (FDA) policy and approved for publication. Approval does not signify that the contents necessarily reflect the position or opinions of the FDA nor does mention of trade names or commercial products constitute endorsement or recommendation for use. The findings and conclusions in this report are those of the author(s) and do not necessarily represent the views of the FDA. James C. Willey is a consultant for and has significant financial interest in Gene Express, Inc.</p>", "<title>List of abbreviations used</title>", "<p><bold>A</bold>: The MAQC sample A (Stratagene Universal Human Reference RNA); <bold>ABI</bold>: Applied Biosystems microarray platform; <bold>AFX</bold>: Affymetrix microarray platform; <bold>AG1</bold>: Agilent one-color microarray platform; <bold>B</bold>: The MAQC sample B (Ambion Human Brain Reference RNA); <bold>C</bold>: The MAQC sample C (75%A+25%B mixture); <bold>CV</bold>: Coefficient of variation; <bold>D</bold>: The MAQC sample D (25%A+75%B mixture); <bold>DEG</bold>: Differentially expressed genes; <bold>FC</bold>: Fold change in expression levels; <bold>GEH</bold>: GE Healthcare microarray platform; <bold>ILM</bold>: Illumina microarray platform; <bold>MAQC</bold>: MicroArray Quality Control project; <bold><italic>P</italic></bold>: The <italic>P</italic>-value calculated from a two-tailed two-sample t-test assuming equal variance; <bold>POG</bold>: Percentage of Overlapping (common) Genes between two lists of differentially expressed genes. It is used as a measure of concordance of microarray results.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>LS conceived of, designed, and coordinated the study. LS, WDJ, RVJ, SCH, and RDW carried out the data analyses. LS drafted the manuscript. All authors contributed to the design of the study, the preparation of the manuscript, and the sometimes-heated discussions on the topic of this paper. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We thank participants of the MicroArray Quality Control (MAQC) project for generating the large data sets that were used in this study. Many MAQC participants contributed to the sometimes-heated discussions on the topic of this paper during MAQC teleconferences and face-to-face project meetings. The common conclusions and recommendations evolved from this extended discourse. Leming Shi and Feng Qian would like to dedicate this work in memory of Prof. Dr. Zhiliang Li of Chongqing University, China.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Concordance for inter-site comparisons. Each panel represents the POG results for a commercial platform of inter-site consistency in terms of DEGs between samples B and A. For each of the six gene selection methods, there are three possible inter-site comparisons: S1–S2, S1–S3, and S2–S3 (S = Site). Therefore, each panel consists of 18 POG lines that are colored based on gene ranking/selection method. Results shown here are based on the entire set of \"12,091\" genes commonly mapped across the microarray platforms without noise (absent call) filtering. POG results are improved when the analyses are performed using the subset of genes that are commonly detectable by the two test sites, as shown in Figure 2. The x-axis represents the number of selected DEGs, and the y-axis is the percentage (%) of genes common to the two gene lists derived from two test sites at a given number of DEGs.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Concordance for inter-site comparisons based on genes commonly detectable by the two test sites compared. Each panel represents the POG results for a commercial platform of inter-site consistency in terms of DEGs between samples B and A. For each of the six gene selection methods, there are three possible inter-site comparisons: S1–S2, S1–S3, and S2–S3. Therefore, each panel consists of 18 POG lines that are colored based on gene ranking/selection method. The x-axis represents the number of selected DEGs, and the y-axis is the percentage (%) of genes common to the two gene lists derived from two test sites at a given number of DEGs.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Concordance for inter-site comparison with samples C and D. The largest fold change between samples C and D is small (three-fold). For each platform, DEG lists from sites 1 and 2 are compared. Analyses are performed using the subset of genes that are commonly detectable by the two test sites.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>Concordance for cross-platform comparisons. Panel a: Based on the data set of \"12,091\" genes (without noise filtering); Panel b: Based on subsets of genes commonly detected (\"Present\") by two platforms. For each platform, the data from test site1 are used for cross-platform comparison. Each POG line corresponds to comparison of the DEGs from two microarray platforms using one of the six gene selection methods. There are ten platform-platform comparison pairs, resulting in 60 POG lines for each panel. The x-axis represents the number of selected DEGs, and the y-axis is the percentage (%) of genes common to the two gene lists derived from two platforms at a given number of DEGs. POG lines circled by the blue oval are from FC based gene selection methods with or without a <italic>P </italic>cutoff, whereas POG lines circled by the teal oval are from <italic>P </italic>based gene selection methods with or without an FC cutoff. Shown here are results for comparing sample B and sample A.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>Concordance between microarray and TaqMan^®assays. Each panel represents the comparison of one microarray platform to TaqMan^®assays. For each microarray platform, the data from test site 1 are used for comparison to TaqMan^®assays. Each POG line corresponds to comparison of the DEGs from one microarray platform and those from the TaqMan^®assays using one of the six gene selection methods. The x-axis represents the number of selected DEGs, and the Y-axis is the percentage (%) of genes common to DEGs derived from a microarray platform and those from TaqMan^®assays. Shown here are results for comparing sample B and sample A using a subset of genes that are detectable by both the microarray platform and TaqMan^®assays. Results based on the entire set of 906 genes are provided in Figure 6.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>Concordance between microarray and TaqMan^®assays without noise-filtering. Each panel represents the comparison of one microarray platform to TaqMan^®assays. The x-axis represents the number of selected DEGs, and the y-axis is the percentage (%) of genes common to DEGs derived from a microarray platform and those from TaqMan^®assays. Shown here are results for comparing sample B and sample A using the entire set of 906 genes for which TaqMan^®assay data are available.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>Inter-site reproducibility of log2 FC and log2 t-statistic. a: log2 FC of site 1 versus log2 FC of site 2; b: log2 t-statistic of test site 1 versus log2 t-statistic of test site 2; and c: log2 FC of test site 1 versus log2 t-statistic of test site 1. Shown here are results for comparing sample B and sample A for all \"12,091\" genes commonly probed by the five microarray platforms. The inter-site reproducibility of log2 FC (a) is much higher than that of log2 t-statistic (b). The relationship between log2 FC and log2 t-statistic from the same test site is non-linear and the correlation appears to be low (c).</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>Concordance between FC and <italic>P </italic>based gene ranking methods (\"12,091 genes\"; test site 1). Each POG line represents a platform using data from its first test site. The x-axis represents the number of selected DEGs, and the y-axis is the percentage (%) of genes common in the DEGs derived from FC- and <italic>P</italic>-ranking. Shown here are results for comparing sample B and sample A for all \"12,091\" genes commonly probed. When a smaller number of genes (up to a few hundreds or thousands) are selected, POG for cross selection method comparison (FC vs. <italic>P</italic>) is low. For example, there are only about 50% genes in common for the top 500 genes selected by FC and <italic>P </italic>separately, indicating that FC and <italic>P </italic>rank order DEGs dramatically differently. When the number of selected DEGs increases, the overlap between the two methods increases, and eventually approach to 100% in common, as expected. The low concordance between FC- and <italic>P</italic>-based gene ranking methods is not unexpected considering their different definitions and low correlation (Figure 7c).</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p>Volcano plot illustration of joint FC and <italic>P </italic>gene selection rule. Genes in sectors A and C are selected as differentially expressed. The colors correspond to the negative log₁₀<italic>P </italic>and log₂fold change values: Red: 20 &lt; -log₁₀<italic>P </italic>&lt; 50 and 3 &lt; log₂fold &lt; 9 or -9 &lt; log₂fold &lt; -3. Blue: 10 &lt; -log₁₀<italic>P </italic>&lt; 50 and 2 &lt; log₂fold &lt; 3 or -3 &lt; log₂fold &lt; -2. Yellow: 4 &lt; -log₁₀<italic>P </italic>&lt; 50 and 1 &lt; log₂fold &lt; 2 or -2 &lt; log₂fold &lt; -1. Pink : 10 &lt; -log₁₀<italic>P </italic>&lt; 20 and 3 &lt; log₂fold or log₂fold &lt; -3. Light blue: 4 &lt; -log₁₀<italic>P </italic>&lt; 10 and 2 &lt; log₂fold or log₂fold &lt; -2. Light green: 2 &lt; -log₁₀<italic>P </italic>&lt; 4 and 1 &lt; log₂fold or log₂fold &lt; -1. Gray)</p></caption></fig>", "<fig position=\"float\" id=\"F10\"><label>Figure 10</label><caption><p>Inter-site concordance based on FC, t-test, Wilcoxon rank-sum test, and SAM. Affymetrix data on samples A and B from site 1 and site 2 for the \"12,091\" commonly mapped genes were used[##REF##16964229##13##]. No flagged (\"Absent\") genes were excluded in the analysis. For the Wilcoxon rank-sum tests, there were many ties, <italic>i.e</italic>., many genes exhibited the same level of statistical significance because of the small sample sizes (five replicates for each group). The tied genes from each test site were broken (ranked) by random ordering. Concordance between genes selected completely by random choice is shown in red and reaches 50% when all candidate genes are declared as differentially expressed; the other 50% genes are in opposite regulation directions. SAM improves inter-site reproducibility compared to t-test, and approaches, but does not exceed that of fold-change.</p></caption></fig>", "<fig position=\"float\" id=\"F11\"><label>Figure 11</label><caption><p>Gene selection and percentage of agreement in gene lists in simulated data sets. Illustrations of the effect of biological context, replicate CV distribution, gene list size, and gene selection rules/methods on the reproducibility of gene lists using simulated microarray data. In some sense, these three graphs represent some extremes as well as typical scenarios in differential expression assays. However, FC sorting with low <italic>P </italic>thresholds (0.001 or 0.0001; light and medium gray boxes) consistently performed better overall than the other rules, even when FC-ranking or <italic>P</italic>-ranking by itself did not perform as well.</p></caption></fig>" ]

[]

[ "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S10-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>t</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mstyle mathsize=\"140%\" displaystyle=\"true\">\n <mml:mrow>\n <mml:mover accent=\"true\">\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"true\">¯</mml:mo>\n </mml:mover>\n </mml:mrow>\n </mml:mstyle>\n <mml:mi>B</mml:mi>\n </mml:msub>\n <mml:mo>−</mml:mo>\n <mml:msub>\n <mml:mstyle mathsize=\"140%\" displaystyle=\"true\">\n <mml:mrow>\n <mml:mover accent=\"true\">\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"true\">¯</mml:mo>\n </mml:mover>\n </mml:mrow>\n </mml:mstyle>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n <mml:mrow>\n <mml:msqrt>\n <mml:mrow>\n <mml:mstyle scriptlevel=\"+1\">\n <mml:mfrac bevelled=\"true\">\n <mml:mrow>\n <mml:msubsup>\n <mml:mi>S</mml:mi>\n <mml:mi>p</mml:mi>\n <mml:mn>2</mml:mn>\n </mml:msubsup>\n </mml:mrow>\n <mml:mrow>\n <mml:msub>\n <mml:mstyle mathsize=\"140%\" displaystyle=\"true\">\n <mml:mi>n</mml:mi>\n </mml:mstyle>\n <mml:mi>B</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mfrac>\n </mml:mstyle>\n <mml:mo>+</mml:mo>\n <mml:mstyle scriptlevel=\"+1\">\n <mml:mfrac bevelled=\"true\">\n <mml:mrow>\n <mml:msubsup>\n <mml:mi>S</mml:mi>\n <mml:mi>p</mml:mi>\n <mml:mn>2</mml:mn>\n </mml:msubsup>\n </mml:mrow>\n <mml:mrow>\n <mml:msub>\n <mml:mstyle mathsize=\"140%\" displaystyle=\"true\">\n <mml:mi>n</mml:mi>\n </mml:mstyle>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mfrac>\n </mml:mstyle>\n </mml:mrow>\n </mml:msqrt>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S10-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mstyle mathsize=\"140%\" displaystyle=\"true\"><mml:mrow><mml:mover accent=\"true\"><mml:mi>X</mml:mi><mml:mo stretchy=\"true\">¯</mml:mo></mml:mover></mml:mrow></mml:mstyle><mml:mi>A</mml:mi></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S10-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mstyle mathsize=\"140%\" displaystyle=\"true\"><mml:mrow><mml:mover accent=\"true\"><mml:mi>X</mml:mi><mml:mo stretchy=\"true\">¯</mml:mo></mml:mover></mml:mrow></mml:mstyle><mml:mi>B</mml:mi></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S10-i4\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>δ</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mrow>\n <mml:mi>d</mml:mi>\n <mml:mo>/</mml:mo>\n <mml:mi>σ</mml:mi>\n </mml:mrow>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:msqrt>\n <mml:mrow>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>/</mml:mo>\n <mml:mn>2</mml:mn>\n </mml:mrow>\n </mml:mrow>\n </mml:msqrt>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S10-i5\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>E</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mi>t</mml:mi>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:msup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n <mml:mi>ν</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n </mml:mrow>\n </mml:msup>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>Γ</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mi>ν</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>Γ</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n <mml:mi>ν</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:mfrac>\n <mml:mi>δ</mml:mi>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>V</mml:mi>\n <mml:mi>a</mml:mi>\n <mml:mi>r</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mi>t</mml:mi>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mi>ν</mml:mi>\n <mml:mrow>\n <mml:mi>ν</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>2</mml:mn>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>+</mml:mo>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mi>ν</mml:mi>\n <mml:mrow>\n <mml:mi>ν</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>2</mml:mn>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>−</mml:mo>\n <mml:msup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:msup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n <mml:mi>ν</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n </mml:mrow>\n </mml:msup>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>Γ</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mi>ν</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>Γ</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mn>2</mml:mn>\n </mml:mfrac>\n <mml:mi>ν</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mn>2</mml:mn>\n </mml:msup>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:msup>\n <mml:mi>δ</mml:mi>\n <mml:mn>2</mml:mn>\n </mml:msup>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>" ]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S10-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-8\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-9\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-10\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S10-11\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Large-scale or systems-wide analysis of relationship networks (e.g. protein-protein) is built from individual links. Since labor is expensive, the domain of knowledge vast, and time short, automated methods of constructing these networks is of paramount importance in bioinformatics research [##REF##17977867##1##]. The simplest method of constructing such a network is by co-occurrence of terms, either in sentences or abstracts [##REF##11326270##2##, ####UREF##0##3##, ##REF##18321884##4####18321884##4##]. Previous research found that entities co-occurring within a sentence have approximately an 80% chance of being related in a non-trivial manner, while entities co-occurring within an abstract have approximately a 50% chance [##UREF##1##5##,##REF##14960466##6##] (exact numbers vary). However, these co-occurrence based approaches, despite their computational efficiency, necessarily remain agnostic about the nature of the relationship between entities. And if we are to have any hope at understanding how control and cause/effect are propagated in these networks, we must establish directionality.</p>", "<title>NLP systems and biological networks</title>", "<p>Establishing relationship directionality usually requires some means of natural language processing (NLP) to extract relationships from biomedical text, although it is not necessarily the only way [##REF##16438716##7##,##REF##18047707##8##]. Most systems that attempt to characterize the nature of the relationship between entities (e.g., directionality, stimulation, inhibition) use grammatical information from sentences, which is provided by NLP software. These may use either shallow parsing [##REF##16095962##9##], which divides the sentence into chunks, or deep parsing [##REF##17052900##10##,##REF##17134476##11##], which provides a complete representation of a sentence's constituent grammatical relations. For example, Chilibot [##REF##15473905##12##] and MedScan [##REF##15033866##13##, ####REF##12967967##14##, ##REF##16563163##15####16563163##15##] attempt to parse sentences to detect the nature of relationships and develop association networks. Chilibot has only three relationship types – positive, negative, and neutral – while MedScan has many more. Both Chilibot and MedScan attempt to determine which terms are subject and object. Systems can also be distinguished by whether they use grammatical information in a rule-based manner, like Chilibot, or whether they use machine learning techniques, such as SVMs [##UREF##2##16##].</p>", "<p>NLP-based approaches have seen heavy use in identifying protein-protein interactions within text [##REF##16873475##17##, ####REF##18315856##18##, ##REF##18207462##19##, ##REF##17238742##20##, ##REF##12799355##21##, ##REF##11262958##22##, ##REF##12689350##23####12689350##23##], gene-disease relationships [##REF##18433469##24##], and have been successful in bolstering and partially reconstructing regulatory networks using text [##REF##17683642##25##]. The goals for these programs range from a simplified two-class method of labeling extracted relationships (e.g., 'stimulates' (positive) or 'inhibits' (negative)), to detection of several dozen semantic types of relationship, such as 'binds', 'cleaves', 'phosphorylates', etc. Our interests in this line of research are to detect paths of directional inference (e.g. A increases B and B decreases C, therefore A decreases C) and their reliability. To accomplish this we must first develop a method that is scalable. Accuracy is desirable, of course, but if an algorithm has a modest accuracy yet many chances to detect directional relations, then it is possible the large sample size may offset a less-than-optimal accuracy. To do this, however, we need to better understand how these methods perform on large unstandardized corpora and the amount of apparently contradictory relationships that are detected as the amount of text is increased and false-positives accrue. Most approaches to date have unfortunately focused on relatively small corpora and it is not clear how they would perform on large datasets (e.g. millions of records).</p>", "<p>We anticipate some contradiction will be factual in nature (e.g. one author reporting A increases B and another reporting A decreases B), but most likely the primary form of contradiction will be context-dependent. That is, the context of the relationship may enable both scenarios to be true and therefore the different facts extracted from text only appear to be contradictory when examined in isolation. We also hypothesize that some objects will better serve as subjects of study, which will bias observations towards one type of directional relationship when, in theory, both might be considered equally logical. For example, the relationship between insulin and glucose will tend more towards reports on the extracellular effects of insulin on glucose (insulin decreases glucose) rather than the intracellular effects (insulin increases glucose) because the extracellular effects are much easier to assay for (e.g. by drawing blood). This is going to introduce an artifact into any NLP analysis that will eventually need to be resolved in the thrust towards a more \"systems-level\" view of biology. For now, however, we hope that such contextual biases will be relatively uniform (e.g. if insulin increases or decreases other entities, the same perspective biases will apply for them) so that making inferences might be possible.</p>", "<p>Here we present a directional relation extraction (DRE) system that uses a support vector machine (SVM) to classify dependency paths, as SVMs have been shown recently to be suited to this type of task [##REF##18003645##26##]. We term it the directional relation extraction and determination with SVM (DREAD SVM). It has the following notable features: First, it has a robust, extensible dependency path convolution kernel capable of extracting and determining directionality for relations when trained on the GENIA event corpus. Second, it has a thesaurus-based named entity recognition algorithm that is not limited to a particular type of biomedical entity (such as gene or protein). We then present the results of cross-validation on the GENIA event corpus and explore the results of large-scale extraction from MEDLINE abstracts. We observe that relationships are often extracted with a degree of ambiguity in their direction and nature – a phenomenon that is often viewed as inconvenient noise that is to be screened out during construction of biological networks. However, it is these instances we are most interested in for this report. The construction of biological networks often is centered upon the arrow and node (or vertices and edges) concept. Yet, how many of these relationships are of such a simple nature that they can be represented in such a manner and how many might be essentially irresolvable without a 3^rdvariable, such as perspective or time? We do not believe we can answer this conclusively here, but through the analysis of several terms expected to have relatively straightforward relationships, we hope to gain a better understanding of how often these simplified relationship constructs are valid.</p>" ]

[ "<title>Materials and methods</title>", "<p>The DREAD algorithm begins with the input of a sentence containing a term of interest, which must appear in a thesaurus of biomedical objects, including genes, chemicals, and clinical phenotypes (described below). The final output of the algorithm is a set of directed edges between the query term and all other terms that co-occurred with the query term within a sentence. These directed edges are also characterized for the presence of stimulatory or inhibitory interactions. In order to achieve this output, the DREAD algorithm goes through several stages of activity: namely Preprocessing, Parsing, and Relationship Resolution (Figure ##FIG##0##1##). Each stage has a corresponding broad function:</p>", "<p>• The <italic>Preprocessing </italic>phase formats the sentence for parsing by tokenizing and Named Entity Recognition (NER).</p>", "<p>• The <italic>Parsing </italic>phase uses Natural Language Processing tools to construct a grammatical representation of the sentence in Stanford Dependency (SD) format. The SD format casts words (tokens) as a network of nodes connected by grammatical <italic>dependencies </italic>(edges).</p>", "<p>• The <italic>Relationship Classification </italic>phase uses a combination of Support Vector Machines and heuristics to deduce the directionality and possible stimulatory or inhibitory character of a relationship based on the grammatical information provided by the parsing phase.</p>", "<title>Preprocessing</title>", "<p>The LingPipe software first breaks each sentence into a sequence of tokens <ext-link ext-link-type=\"uri\" xlink:href=\"http://alias-i.com/lingpipe/\"/>. Next, NER identifies biologically relevant terms within the sentence using a synonym dictionary of genes, chemicals, and clinical phenotypes developed as part of the IRIDESCENT system [##REF##14960466##6##,##REF##15471547##27##,##REF##14734310##28##]. Multiple-token terms are condensed to a single token in order to simplify the subsequent parsing steps. At this point, if the sentence does not contain at least the query term and one other named entity within the synonym dictionary, it is thrown out.</p>", "<title>Parsing</title>", "<title>Deep parsing</title>", "<p>The preprocessed sentences containing entities of interest are sent to the Charniak-Lease parser [##UREF##3##29##], which performs both part-of-speech tagging and deep parsing, returning a representation of the sentence in Penn Treebank format. The Charniak-Lease parser was previously found to be computationally efficient and accurate on biomedical text compared with other deep parsers [##REF##17254351##30##]. The Penn Treebank (PTB) representation of each sentence parsed is then stored in the local relational database awaiting evaluation by the next steps in the system (Figure ##FIG##1##2##).</p>", "<title>Conversion to Stanford Dependency format</title>", "<p>In order to distill the amount of grammatical information used as feature space for machine-learning-based relation extraction to a small, rich subset, we convert the Penn Treebank-format (PTB) output from the Charniak-Lease parser into the Stanford Dependency (SD) format. This SD format represents sentences as a network of nodes (words) and edges (grammatical relationships between words), as described in de Marneffe <italic>et al </italic>[##UREF##4##31##] (See Figure ##FIG##2##3##). In the DREAD system, the PTB-&gt;SD conversion is accomplished using a suite of tools available from Stanford; however, recent work has shown that conversion to typed dependencies from other formats, such as the HPSG output of the GENIA team's Enju parser, is also possible [##UREF##5##32##].</p>", "<title>Extraction of dependency path</title>", "<p>After converting the sentence to SD format, we further reduce the amount of grammatical information used in relation extraction by using only the \"dependency path\" – that is, the shortest sequence of nodes and edges between two entities in the grammatical network of typed dependencies (Figure ##FIG##3##4##). The DREAD system extracts a dependency path between the query term and each other term in the sentence that was identified by the dictionary-based NER step. If there are multiple instances of the query term within the sentence, this process is repeated, omitting dependency paths connecting the query term to another instance of itself.</p>", "<p>The dependency path has previously been shown to provide a good feature space for machine-learning-based relation extraction [##UREF##6##33##,##REF##17142812##34##]. Others, however, have used sub-trees of the syntactic parse tree (i.e., Penn Treebank format) or of the typed dependency format as feature space for relation extraction, to good effect [##UREF##7##35##]. A study of the optimal feature space for relation extraction is provided in Jiang and Zhai [##UREF##8##36##].</p>", "<title>Relationship classification</title>", "<p>In order for the DREAD system to draw a conclusion about the nature of a relationship based on the grammatical information provided by a dependency path, we use two sub-systems. The first, a pipeline of Support Vector Machines (SVMs), is used to establish the presence and directionality of a relationship for each dependency path. If the SVM classification procedure determines that the dependency path does indeed represent a directional relationship, the second system, a simple set of heuristic rules, classifies the relationship as \"stimulatory\" (+), \"inhibitory\" (-), or \"neutral\" (n). Together, these two pieces of information (directionality and stimulatory/inhibitory classification) are a summarized form of the interaction between two entities in the sentence (Figure ##FIG##4##5##).</p>", "<title>SVM-based directionality determination</title>", "<p>The SVM classification pipeline is composed of three one-against-one \"convolution dependency path kernel\" SVMs (described below). Each SVM has a training set derived by parsing and transforming the GENIA event corpus [##REF##12855455##37##,##REF##18182099##38##] into dependency paths. The three SVMs classify the dependency path in the following ways:</p>", "<p>1. Is there a functional relationship between the two entities?</p>", "<p>2. If there is a relationship, is that relationship directional?</p>", "<p>3. If directional, is the direction forward or reverse?</p>", "<p>If SVM #1 or #2 returns a negative answer (i.e., there is no relationship or the relationship is not directed), then the dependency path is not sent to any further steps.</p>", "<title>Convolution dependency path kernel</title>", "<p>Bunescu described the development of a SVM kernel that computes the similarity of two sequences of features by the number and length of common subsequences between them [##UREF##6##33##]. In that work, however, Bunescu used tokens, part-of-speech tags, and entity classes in the order that they appear within the sentence. As a concrete example, in comparing the sentences \"I went to the store\" and \"You went to the game\", the longest common subsequence of tokens would be \"went...the\", and other subsequences would be \"went...to\" and \"to...the\". In another work, Bunescu <italic>et al </italic>restricted feature classes to those within dependency paths, but simply calculated the number of co-occurring features as the kernel score [##UREF##9##39##]. This method had the limitation that paths with different lengths were computed as having zero similarity, which presumably lowered recall.</p>", "<p>We reasoned that the accuracy of the overall SVM procedure could be improved by combining the two methods; in other words, by restricting the set of features to those within the dependency path but by using Bunescu's subsequence kernel to compare two potential relationships with greater flexibility. Indeed, a recent study by Wang, who coined the name \"convolution dependency path kernel\", showed that combining the two methods leads to greater precision and recall [##UREF##10##40##].</p>", "<p>Formally, the generalized subsequence kernel is defined [##UREF##9##39##]:</p>", "<p></p>", "<p>Where <italic>u </italic>is a subsequence from the set of all possible sequences subject to the constraint that <italic>u </italic>must appear as a subsequence in both of the feature sequences <italic>s </italic>and <italic>t </italic>(that is, <italic>u </italic>≺ s[i], u ≺ t[j]); <bold>i </bold>and <bold>j </bold>are the sequences of indices corresponding to <italic>s </italic>and <italic>t</italic>. Each subsequence <italic>u </italic>is penalized according to the decaying factor, λ, with the result that when the feature sequences <italic>s </italic>and <italic>t </italic>are longer, each subsequence is given less weight. This serves as a way of normalizing the kernel scores for feature sequences of different lengths.</p>", "<title>Feature selection</title>", "<p>There are three primary types of feature available in each dependency path for use by machine-learning classification: the token (word), the part-of-speech tag for each token, and the grammatical dependency (SD edge) connecting two tokens. We elect not to use the token as a feature; because MEDLINE abstracts consist of such a large vocabulary, choosing the token would likely lead to overfitting of the training data. On the other hand, both POS tags and grammatical dependencies provide a relatively sparse set of feature classes, and are more relevant to the grammatical structure of the sentence than the token itself.</p>", "<title>GENIA training corpus</title>", "<p>The GENIA event corpus consists of 1,000 abstracts from MEDLINE that have been manually annotated for various biomedical \"events\", largely molecular interactions. Of the events in this corpus involving two entities, some are directed, while others are not. These events have various annotations; we considered all relations of the <italic>Positive_regulation </italic>type as stimulatory, all relations of the <italic>Negative_regulation </italic>type as inhibitory, and all others as neutral. To derive gold standard relationships from the corpus, we first assembled all possible pair-wise combinations of entities within the sentence (because the GENIA event corpus is in a quasi-XML format, there are sometimes entities within entities; the broadest possible entity names were used). These pairs were then classified as follows:</p>", "<p>1. Exists vs. nonexistent – If an event was annotated between a pair of entities, a functional relationship between the two entities was said to exist; otherwise, the pair was classified as non-interacting.</p>", "<p>2. Directional vs. non-directional – Of the pairs of entities that were annotated as interacting, some relations showed one entity acting on the other (in GENIA parlance, this relationship had one \"THEME\" and one \"CAUSE\"), while other relationships did not show directionality (two \"THEME\" annotations). We divided interactions on this basis for the second training set.</p>", "<p>3. Forward vs. reverse – For directional relations, the \"CAUSE\" was considered to be the agent of the interaction, while the \"THEME\" was the target.</p>", "<p>To complete construction of the training corpus, we parsed all sentences and extracted dependency paths between each pair of entities as previously described. Dependency paths from MEDLINE at large could then be compared by the SVM kernel to dependency paths within the training corpus to classify the existence and directionality of relationships through a series of one-against-one classifiers.</p>", "<title>Heuristic determination of stimulatory or inhibitory relationships</title>", "<p>After using the pipeline of SVMs to establish the presence or absence of a relationship and, if applicable, its directionality, we apply a simple set of heuristic rules to determine whether the nature of the relationship described by the dependency path is \"positive\" (stimulation, induction, activation), \"negative\" (repression, down-regulation, deactivation), or \"neutral\" (none-of-the-above). This is done by matching stemmed tokens in the dependency path to a list of words in each category. We hope to soon characterize the stimulatory or inhibitory nature of relationships using a SVM [##UREF##2##16##]; however, initial efforts using tokens as features have resulted in good cross-validation accuracy albeit poor recall in large-scale MEDLINE extraction.</p>", "<title>Relationship resolution on large-scale datasets</title>", "<p>In order to test the performance of the DREAD algorithm on a large scale, and to test the hypothesis that aggregating relationship data from a large corpus will overcome deficiencies in precision on individual sentences, we sought out summary information for different objects in the database. We chose two common terms in the biomedical literature to focus on for analysis of unstructured text: A chemical, caffeine (17,145 papers with this term in the title and/or abstract) and gene, <italic>c-myc </italic>(11,971 papers). Summary information regarding the direction and nature of relationships between caffeine and other objects in the database was compiled from its Wikipedia entry <ext-link ext-link-type=\"uri\" xlink:href=\"http://en.wikipedia.org\"/>. For c-myc, we used NCBI's Gene Reference into Function (GeneRIF)[##REF##14728215##41##] instead because there was more information. Although Wikipedia may not always be appropriate as a source of summary information, we did feel that analysis of system performance should optimally be constrained to one or two standard sources of summary information so as not to bias the analysis (e.g., noticing that the system worked really well for a certain term and then hunting down supporting information related to that term while ignoring other terms that the system performed poorly on).</p>", "<p>After thus compiling a list of expected relationships between the two terms and other biomedical objects, we pulled all abstracts containing caffeine and <italic>c-myc </italic>from MEDLINE, split each abstract into sentences using the LingPipe utility, and preprocessed each sentence as described above. If the sentence was found to contain the target term, the DREAD system was used to predict the relationships between the target term and each other term in the sentence. If a directional relationship was found, it was categorized as either \"forward\" (F), indicating that caffeine or <italic>c-myc </italic>affects the other object, or as \"reverse\" (R), indicating that the other object acts upon caffeine or <italic>c-myc</italic>. \"+\" is defined as a stimulatory (e.g. increases, raises, upregulates, etc) relationship between two terms and \"-\" is an inhibitory relationship (e.g. decreases, lowers, inhibits, downregulates, etc.). For example, the relationship \"caffeine increases cyclic AMP\" would be notated \"F+\" from the perspective of caffeine.</p>", "<p>These relationships were then aggregated and each binary relationship was assigned a strength score based on the amount of contradiction. Extracted relations were considered contradictory when both stimulatory and inhibitory instances were detected for the same directionality. For example, finding A increases B and A decreases B results in a contradiction, but finding A increases B and B decreases A is not necessarily contradictory. Feedback and feedforward loops, for example, can accommodate such trends. Forward scores are calculated as max(F+, F-)/((F+)(F-)) and reverse scores as max(R+, R-)/((R+)(R-)).</p>" ]

[ "<title>Results</title>", "<title>SVM cross validation on GENIA corpus</title>", "<p>To benchmark the performance of the convolution dependency path kernel on standardized biomedical corpora, we subjected the GENIA training data for each of the three SVMs (see <italic>GENIA training corpus </italic>section) to 10-fold cross-validation. For detecting the existence of a relationship and directionality to the relationship, the precision/recall curves are shown in Figure ##FIG##5##6##. Contrasting with Küffner <italic>et al </italic>[##UREF##2##16##], the DREAD SVM's ability to detect the existence of a relationship achieved a maximum F-measure of <bold>0.879 </bold>(Küffner 0.849) at a precision level of <bold>0.820 </bold>(Küffner 0.827) and recall of <bold>0.948 </bold>(Küffner 0.872). The ability of our algorithm in directional relationship extraction (DRE) achieved a maximum F-measure of <bold>0.794 </bold>(Küffner 0.749) at a precision level of <bold>0.704 </bold>(Küffner 0.846) and recall of <bold>0.912 </bold>(Küffner 0.672). For directional relationships, the precision/recall curves are shown in Figure ##FIG##6##7##. F-measure scores were similar between detection of forward (0.861) and reverse (0.791) relationships.</p>", "<p>Note that these scores are based upon the evaluation of dependency paths in which the recognition of named entities is part of the corpus; therefore, they do not take into account the considerable loss in recall and slight loss in precision that results from a dictionary-based named entity approach. Also, in the DREAD system, these SVMs are arranged in a pipeline fashion, so that the error rate propagates in a multiplicative manner. Thus, these scores are a meaningful marker only of the performance of the SVM classifiers and not of the DREAD system as a whole.</p>", "<title>Method scalability</title>", "<p>For comparable tasks, DREAD SVM performance on the GENIA corpus was comparable to levels of precision and recall reported by Küffner <italic>et al </italic>[##UREF##2##16##]. This suggested that it was reasonable to proceed forward and examine how well performance would scale on much larger corpora.</p>", "<title>Caffeine analysis</title>", "<p>Using summary information for each of our target terms to know what relationships should be found, and the nature of those relationships, we ran the DREAD SVM on all MEDLINE abstracts with the target term. The same directional relationship had to be observed at least twice to be counted as a directional relationship prediction. Table ##TAB##0##1## summarizes the DREAD SVM performance on extracting directional relationships between caffeine and other objects in MEDLINE. Caffeine was chosen for three reasons: First, it was expected to have a predominance of stimulatory relationships. Second, it was expected to affect other objects in the body, but not be affected by many other objects (outside of metabolic pathways). And third, the authors have substantial first-hand experience with the physiological effects of this chemical compound. Table ##TAB##0##1## shows that most relationships are indeed forward (253 versus 116 reverse) and stimulatory (118 versus 44 inhibitory), as expected, but clearly not every extracted directional relationship was unambiguous. The system only identified 12/21 (57%) of the expected relationships. It did, however, correctly identify the directionality of 10/12 (83%) of the relationships.</p>", "<p>Drowsiness would naturally seem like a phenotype that should be lowered upon caffeine intake. However, caffeine withdrawal can cause drowsiness [##REF##15448977##42##] and since \"caffeine withdrawal\" was not present in the database, the shorter term was recognized as the subject instead.</p>", "<title>C-myc analysis</title>", "<p>The oncogene, c-myc is well studied and expected to both affect other objects and in turn be affected, as part of a genetic network. The mechanisms which sometimes govern genetic control, especially over a gene that is estimated to possibly regulate up to 15% of all genes[##REF##11443860##43##] are expected to complicate DRE. Table ##TAB##1##2## summarizes the results. The system identified 21/39 (53%) of the expected relationships, and of those, it correctly identified the directionality for 13/21 (62%). Examining some of the individual sentences to better understand the reasons for failure, it became evident that the natures of relationships for c-myc were less straightforward than for caffeine. For example, the role of c-myc as an oncogene is well known and it is upregulated in breast cancer, yet the system identified many more examples of breast cancer \"stimulating\" c-myc. This is because these examples are reported as correlative relationships (e.g., \"in the breast cancer biopsy, c-myc was upregulated\").</p>", "<p>Another example is JUN (aka c-jun). The actual functional relationship between c-myc and JUN is forward, negative: c-myc inhibits c-jun [##REF##18314492##44##]. However, because both c-myc and c-jun are oncogenes, they are usually mentioned in the context of being upregulated together. Yet another difficulty in the analysis of c-myc is that sentences often refer to an increase or decrease in the effects or properties (e.g. \"oncogenic properties\") of c-myc, rather than c-myc itself. Table ##TAB##2##3## shows examples of sentences whose DRE was incorrectly classified. There is an average of 0.11 stimulatory terms per c-myc:c-jun path, but only 0.03 inhibitory terms. So, in this case, at least, there seems to be a literature bias towards describing relationships as stimulatory rather than inhibitory. We examined the different interaction types to see if any were longer on average or contained more stimulatory or inhibitory keywords, finding that gene-gene candidate interaction paths were considerably shorter than gene-process paths, and contained fewer interaction keywords (see Table ##TAB##3##4##).</p>" ]

[ "<title>Discussion and conclusion</title>", "<p>Analysis of unstandardized text comes with many challenges for term recognition, synonym mapping, homonym resolution, etc. And directional relation detection is yet another challenge, not just in a technical sense of algorithmically detecting relations, but also in the sense of resolving apparently contradictory data that may arise. NLP approaches so far isolate facts from sentences and resolve relationships between two entities, which may not be sufficient to accurately reconstruct or model relationship networks. As we report here, extracted relationships range from those with strong support for one type and direction of relationship over another to apparently contradictory relationships, some of which are quickly resolved by a better understanding of context.</p>", "<p>We find that for these tasks involving directional relation extraction, it is important to have terms be as unambiguous as possible. Stress and metabolism, for example, have common meanings (i.e., psychological stress and the rate of energy consumption by an organism, respectively) and more specific physiological meanings (i.e., a state of increased responsiveness to stimulus and the process by which chemicals change form, respectively). One approach to increase DRE accuracy would entail employing methods to break down broad categories into more specific subcategories (e.g. oxidative stress, physiological stress, stress-related illness, etc.), as the recognition of term relationships was not the limiting factor in this analysis but rather the accurate recognition of full terms.</p>", "<p>We also find that noun modifiers frequently complicate thesaurus-based analysis of terms, as in the case of c-myc. To increase or decrease a gene in a biological sense would likely refer to its mRNA levels, protein expression levels or molecular activity (e.g. catalytic) levels. However, study of genetic effects often proceeds via an assessment of how known function is affected by manipulating the system and thus increasing and decreasing statements refer to the modifier rather than the gene (e.g. \"A decreased gene-B-related apoptosis\", \"A increased geneB phosphorylation\", etc). While NLP systems are capable, in theory, of resolving such sentences, this style of study and writing causes more false-positives as corpus size increases.</p>", "<p>Ultimately, the scientific community is going to want to move towards testing of extracted networks for concordance with observed behavior. But to do this, a greater incorporation of contextual or conditional information will become necessary, which may not be possible to represent in a single graph. Temporal effects, for example, would seem to require a graph for short-term and long-term effects. Additionally, some model organisms are manipulated to alter genetic behavior, and relationships that are true within a mutant strain may not be true outside it. Similarly, some interactions are conditional. Worse, some of these conditional interactions may show up as contradictory relationships merely because not enough is known to understand that \"A increases B\" and \"A decreases B\" are both true depending upon the context. All this unfortunately complicates analysis, but at the same time seems unavoidable. Networks may well need several additional parameters per connection to accommodate contextual information. In part, this type of NLP work may guide some of this along – when apparently contradictory information is found, it is possible to initiate a secondary search for contextual clues that could accurately predict when one relationship type would be true over another. In such cases, a computer need not necessarily \"understand\" the context, but simply be able to identify it algorithmically. In the future, we hope to develop more accurate metrics to identify when directional relationships require context resolution prior to their inclusion in a biological network.</p>" ]

[ "<title>Discussion and conclusion</title>", "<p>Analysis of unstandardized text comes with many challenges for term recognition, synonym mapping, homonym resolution, etc. And directional relation detection is yet another challenge, not just in a technical sense of algorithmically detecting relations, but also in the sense of resolving apparently contradictory data that may arise. NLP approaches so far isolate facts from sentences and resolve relationships between two entities, which may not be sufficient to accurately reconstruct or model relationship networks. As we report here, extracted relationships range from those with strong support for one type and direction of relationship over another to apparently contradictory relationships, some of which are quickly resolved by a better understanding of context.</p>", "<p>We find that for these tasks involving directional relation extraction, it is important to have terms be as unambiguous as possible. Stress and metabolism, for example, have common meanings (i.e., psychological stress and the rate of energy consumption by an organism, respectively) and more specific physiological meanings (i.e., a state of increased responsiveness to stimulus and the process by which chemicals change form, respectively). One approach to increase DRE accuracy would entail employing methods to break down broad categories into more specific subcategories (e.g. oxidative stress, physiological stress, stress-related illness, etc.), as the recognition of term relationships was not the limiting factor in this analysis but rather the accurate recognition of full terms.</p>", "<p>We also find that noun modifiers frequently complicate thesaurus-based analysis of terms, as in the case of c-myc. To increase or decrease a gene in a biological sense would likely refer to its mRNA levels, protein expression levels or molecular activity (e.g. catalytic) levels. However, study of genetic effects often proceeds via an assessment of how known function is affected by manipulating the system and thus increasing and decreasing statements refer to the modifier rather than the gene (e.g. \"A decreased gene-B-related apoptosis\", \"A increased geneB phosphorylation\", etc). While NLP systems are capable, in theory, of resolving such sentences, this style of study and writing causes more false-positives as corpus size increases.</p>", "<p>Ultimately, the scientific community is going to want to move towards testing of extracted networks for concordance with observed behavior. But to do this, a greater incorporation of contextual or conditional information will become necessary, which may not be possible to represent in a single graph. Temporal effects, for example, would seem to require a graph for short-term and long-term effects. Additionally, some model organisms are manipulated to alter genetic behavior, and relationships that are true within a mutant strain may not be true outside it. Similarly, some interactions are conditional. Worse, some of these conditional interactions may show up as contradictory relationships merely because not enough is known to understand that \"A increases B\" and \"A decreases B\" are both true depending upon the context. All this unfortunately complicates analysis, but at the same time seems unavoidable. Networks may well need several additional parameters per connection to accommodate contextual information. In part, this type of NLP work may guide some of this along – when apparently contradictory information is found, it is possible to initiate a secondary search for contextual clues that could accurately predict when one relationship type would be true over another. In such cases, a computer need not necessarily \"understand\" the context, but simply be able to identify it algorithmically. In the future, we hope to develop more accurate metrics to identify when directional relationships require context resolution prior to their inclusion in a biological network.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Relationships between entities such as genes, chemicals, metabolites, phenotypes and diseases in MEDLINE are often directional. That is, one may affect the other in a positive or negative manner. Detection of causality and direction is key in piecing pathways together and in examining possible implications of experimental results. Because of the size and growth of biomedical literature, it is increasingly important to be able to automate this process as much as possible.</p>", "<title>Results</title>", "<p>Here we present a method of relation extraction using dependency graph parsing with SVM classification. We tested the SVM classifier first on gold standard corpora from GENIA and find it achieved 82% precision and 94.8% recall (F-measure of 87.9) on these standardized test sets. We then applied the entire system to all available MEDLINE abstracts for two target interactions with known effects. We find that while some directional relations are extracted with low ambiguity, others are apparently contradictory, at least when considered in an isolated context. When examined, it is apparent some are dependent upon the surrounding context (e.g. whether the relationship referred to short-term or long-term effects, or whether the focus was extracellular versus intracellular).</p>", "<title>Conclusion</title>", "<p>Thesaurus-based directional relation extraction can be done with reasonable accuracy, but is prone to false-positives on larger corpora due to noun modifiers. Furthermore, methods of resolving or disambiguating relationship context and contingencies are important for large-scale corpora.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>JDW conceived of and supervised the project, CBG wrote the programs and implemented them. Both authors contributed to writing the manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We would like to acknowledge funding support from NSF-EPSCoR grant EPS-0447262 and NIH/NLM grant 1 R01 LM009758-01.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Flowchart of the relation extraction process.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>An example of sentence structure for the sentence \"<italic>Both isobutylmethylxanthine and theophylline increased the level of cyclic AMP in rat mast cells</italic>.\". Shown are the grammatical relationships diagrammed in the Penn Treebank format.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Grammatical sentence structure in dependency graph format.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>Path of dependency between database terms.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>SVM identification of directional relationships.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>Precision/recall curve for A) Detecting relationships and B) Detecting directional relationships within the GENIA corpus.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>Precision/recall curve for detecting A) Forward relationships and B) Reverse relationships in the GENIA corpus.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Identifying directional relationships in MEDLINE for the chemical compound caffeine on the basis of summarized relationships after analysis of 17,145 abstracts.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>ObjectName</bold></td><td align=\"left\"><bold>No rel.</bold></td><td align=\"left\"><bold>ND Rel.</bold></td><td align=\"left\"><bold>F+</bold></td><td align=\"left\"><bold>F-</bold></td><td align=\"left\"><bold>Fn</bold></td><td align=\"left\"><bold>R+</bold></td><td align=\"left\"><bold>R-</bold></td><td align=\"left\"><bold>Rn</bold></td><td align=\"left\"><bold>Expected</bold></td><td align=\"left\"><bold>Extracted</bold></td></tr></thead><tbody><tr><td align=\"left\">(-)-Adrenaline</td><td align=\"right\">58</td><td align=\"right\">22</td><td align=\"right\">14</td><td align=\"right\">4</td><td align=\"right\">17</td><td align=\"right\">0</td><td align=\"right\">3</td><td align=\"right\">17</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">(-)-Dopa</td><td align=\"right\">5</td><td align=\"right\">2</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">Cyclic AMP</td><td align=\"right\">30</td><td align=\"right\">9</td><td align=\"right\">21</td><td align=\"right\">11</td><td align=\"right\">12</td><td align=\"right\">4</td><td align=\"right\">3</td><td align=\"right\">13</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">diuresis</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\"><bold>\n <italic>Drowsiness</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>3</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"left\"><bold><italic>F</italic>-</bold></td><td align=\"left\"><bold>\n <italic>F+</italic>\n </bold></td></tr><tr><td align=\"left\">Fatigue</td><td align=\"right\">23</td><td align=\"right\">8</td><td align=\"right\">4</td><td align=\"right\">6</td><td align=\"right\">15</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">11</td><td align=\"left\">F-</td><td align=\"left\">F-</td></tr><tr><td align=\"left\">Fatty acid</td><td align=\"right\">10</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">6</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">gastric acid secretion</td><td align=\"right\">3</td><td align=\"right\">3</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\"><bold>\n <italic>Glycerol</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>15</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>6</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>3</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>8</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>3</italic>\n </bold></td><td align=\"left\"><bold>\n <italic>F+</italic>\n </bold></td><td align=\"left\"><bold><italic>F</italic>-</bold></td></tr><tr><td align=\"left\">insomnia</td><td align=\"right\">16</td><td align=\"right\">3</td><td align=\"right\">3</td><td align=\"right\">0</td><td align=\"right\">5</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">Irritability</td><td align=\"right\">2</td><td align=\"right\">2</td><td align=\"right\">2</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">Lethargy</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"left\">F-</td><td align=\"left\">None</td></tr><tr><td align=\"left\">lipid metabolism</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">Nervousness</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">5</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">Norepinephrine</td><td align=\"right\">61</td><td align=\"right\">19</td><td align=\"right\">10</td><td align=\"right\">2</td><td align=\"right\">18</td><td align=\"right\">4</td><td align=\"right\">0</td><td align=\"right\">16</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">Palpitation</td><td align=\"right\">5</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">PKA</td><td align=\"right\">3</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">3</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">Respiratory alkalosis</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">ryanodine-sensitive calcium-release channel activity</td><td align=\"right\">48</td><td align=\"right\">18</td><td align=\"right\">33</td><td align=\"right\">2</td><td align=\"right\">17</td><td align=\"right\">5</td><td align=\"right\">4</td><td align=\"right\">15</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">vasodilation</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">2</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"left\">F-</td><td align=\"left\">F-</td></tr><tr><td align=\"left\">vasoconstriction</td><td align=\"right\">5</td><td align=\"right\">0</td><td align=\"right\">3</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Identifying directional relationships in MEDLINE for the gene c-myc.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>ObjectName</bold></td><td align=\"left\"><bold>No rel.</bold></td><td align=\"left\"><bold>ND Rel.</bold></td><td align=\"left\"><bold>F+</bold></td><td align=\"left\"><bold>F-</bold></td><td align=\"left\"><bold>Fn</bold></td><td align=\"left\"><bold>R+</bold></td><td align=\"left\"><bold>R-</bold></td><td align=\"left\"><bold>Rn</bold></td><td align=\"left\"><bold>Expected</bold></td><td align=\"left\"><bold>Extracted</bold></td></tr></thead><tbody><tr><td align=\"left\"><italic>\n <underline>Process</underline>\n </italic></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Anaplasia</td><td align=\"right\">6</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">4</td><td align=\"left\">F+/R+</td><td align=\"left\">None</td></tr><tr><td align=\"left\"><bold>\n <italic>Angiogenesis</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>11</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>10</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>7</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>3</italic>\n </bold></td><td align=\"left\"><bold>\n <italic>F+</italic>\n </bold></td><td align=\"left\"><bold><italic>F</italic>-</bold></td></tr><tr><td align=\"left\"><bold>\n <italic>Breast cancer</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>92</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>12</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>44</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>9</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>40</italic>\n </bold></td><td align=\"left\"><bold>\n <italic>F+</italic>\n </bold></td><td align=\"left\"><bold>\n <italic>R+</italic>\n </bold></td></tr><tr><td align=\"left\">Cell growth</td><td align=\"right\">131</td><td align=\"right\">34</td><td align=\"right\">19</td><td align=\"right\">19</td><td align=\"right\">76</td><td align=\"right\">6</td><td align=\"right\">8</td><td align=\"right\">34</td><td align=\"left\">F+</td><td align=\"left\">Inc.</td></tr><tr><td align=\"left\">Cervical carcinoma</td><td align=\"right\">22</td><td align=\"right\">4</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">3</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">8</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\"><bold>\n <italic>DNA Damage</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>33</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>7</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>14</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>3</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>11</italic>\n </bold></td><td align=\"left\"><bold>\n <italic>F+</italic>\n </bold></td><td align=\"left\">R+</td></tr><tr><td align=\"left\">DNA repair</td><td align=\"right\">13</td><td align=\"right\">11</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">4</td><td align=\"left\">F-</td><td align=\"left\">None</td></tr><tr><td align=\"left\">Medulloblastoma</td><td align=\"right\">21</td><td align=\"right\">6</td><td align=\"right\">4</td><td align=\"right\">0</td><td align=\"right\">13</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">20</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">Tumorigenesis</td><td align=\"right\">80</td><td align=\"right\">40</td><td align=\"right\">20</td><td align=\"right\">6</td><td align=\"right\">48</td><td align=\"right\">6</td><td align=\"right\">4</td><td align=\"right\">32</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"><italic>\n <underline>Chemical</underline>\n </italic></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Rapamycin</td><td align=\"right\">4</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"left\">R+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">Reactive oxygen species</td><td align=\"right\">9</td><td align=\"right\">5</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">4</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">10</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">Valproic acid</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"left\">R-</td><td align=\"left\">R-</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"><italic>\n <underline>Gene</underline>\n </italic></td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">AICDA</td><td align=\"right\">4</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">R+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">AURKA</td><td align=\"right\">5</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">R+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">Calcineurin</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"left\">R+</td><td align=\"left\">None</td></tr><tr><td align=\"left\"><bold>\n <italic>CDKN2A</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>95</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>37</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>5</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>26</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>7</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>42</italic>\n </bold></td><td align=\"left\"><bold><italic>F</italic>-</bold></td><td align=\"left\"><bold><italic>R</italic>-</bold></td></tr><tr><td align=\"left\">CREBBP</td><td align=\"right\">7</td><td align=\"right\">3</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">5</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">6</td><td align=\"left\">R-</td><td align=\"left\">None</td></tr><tr><td align=\"left\">EPHA2</td><td align=\"right\">8</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F-</td><td align=\"left\">None</td></tr><tr><td align=\"left\"><bold>\n <italic>FBXW7</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>18</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>5</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>4</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>4</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>17</italic>\n </bold></td><td align=\"left\"><bold><italic>R</italic>-</bold></td><td align=\"left\">R+</td></tr><tr><td align=\"left\">HDAC1</td><td align=\"right\">1</td><td align=\"right\">5</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">4</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">HMGCS2</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">F-</td><td align=\"left\">None</td></tr><tr><td align=\"left\">IFN-gamma</td><td align=\"right\">2</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"left\">R-</td><td align=\"left\">None</td></tr><tr><td align=\"left\"><bold>\n <italic>JUN</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>558</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>28</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>12</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>71</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>12</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>6</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>75</italic>\n </bold></td><td align=\"left\"><bold><italic>F</italic>-</bold></td><td align=\"left\">F+/R+</td></tr><tr><td align=\"left\">NDRG2</td><td align=\"right\">2</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">5</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F-</td><td align=\"left\">None</td></tr><tr><td align=\"left\">NFATC1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">R+</td><td align=\"left\">R+</td></tr><tr><td align=\"left\">PCGF2</td><td align=\"right\">4</td><td align=\"right\">4</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">R-</td><td align=\"left\">None</td></tr><tr><td align=\"left\">PPARG</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">PRL</td><td align=\"right\">15</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">5</td><td align=\"right\">6</td><td align=\"right\">1</td><td align=\"right\">5</td><td align=\"left\">R+</td><td align=\"left\">R+</td></tr><tr><td align=\"left\"><bold>\n <italic>RAC1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>5</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>3</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"left\"><bold>\n <italic>R+</italic>\n </bold></td><td align=\"left\">F+</td></tr><tr><td align=\"left\">SP1</td><td align=\"right\">29</td><td align=\"right\">5</td><td align=\"right\">0</td><td align=\"right\">3</td><td align=\"right\">9</td><td align=\"right\">4</td><td align=\"right\">0</td><td align=\"right\">9</td><td align=\"left\">R+</td><td align=\"left\">R+</td></tr><tr><td align=\"left\"><bold>\n <italic>STAT5A</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>6</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>1</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>2</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>0</italic>\n </bold></td><td align=\"right\"><bold>\n <italic>4</italic>\n </bold></td><td align=\"left\"><bold><italic>R</italic>-</bold></td><td align=\"left\">R+</td></tr><tr><td align=\"left\">Telomerase</td><td align=\"right\">27</td><td align=\"right\">5</td><td align=\"right\">13</td><td align=\"right\">0</td><td align=\"right\">9</td><td align=\"right\">5</td><td align=\"right\">1</td><td align=\"right\">8</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">VEGFA</td><td align=\"right\">49</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"right\">18</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">17</td><td align=\"left\">F+</td><td align=\"left\">None</td></tr><tr><td align=\"left\">WRN</td><td align=\"right\">1</td><td align=\"right\">1</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">3</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"left\">F+</td><td align=\"left\">F+</td></tr><tr><td align=\"left\">ZBTB16</td><td align=\"right\">3</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">R-</td><td align=\"left\">None</td></tr><tr><td align=\"left\">ZBTB17</td><td align=\"right\">24</td><td align=\"right\">17</td><td align=\"right\">1</td><td align=\"right\">8</td><td align=\"right\">11</td><td align=\"right\">1</td><td align=\"right\">3</td><td align=\"right\">9</td><td align=\"left\">F-</td><td align=\"left\">F-</td></tr><tr><td align=\"left\">Zfp472</td><td align=\"right\">1</td><td align=\"right\">2</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">0</td><td align=\"right\">1</td><td align=\"left\">R-</td><td align=\"left\">None</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Example sentences from the failed DRE between c-myc and the entities \"Breast Cancer\" and \"c-jun\". The expected relationships for breast cancer was F+ and for c-jun, F-.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Object</td><td align=\"left\">Extracted Rel.</td><td align=\"left\">PubMed ID</td><td align=\"left\">Sentence</td></tr></thead><tbody><tr><td align=\"left\">Breast Cancer</td><td align=\"left\">R+</td><td align=\"right\">12150449</td><td align=\"left\">The c-myc oncogene is frequently activated in invasive breast cancer and has been associated with high nuclear grade, lymph node metastasis and poorer disease outcome</td></tr><tr><td/><td align=\"left\">R+</td><td align=\"right\">7734397</td><td align=\"left\">The proto-oncogene c-myc is involved in the stimulation of cell proliferation, and its expression is known to be stimulated by estradiol (E2) in human breast cancer cell lines and various non – cancerous E2 – dependent tissues</td></tr><tr><td/><td align=\"left\">F+</td><td align=\"right\">1855215</td><td align=\"left\">In search of critical genes in the mechanism of estrogen action in human breast cancer, we previously showed that estrogen stimulates transcription of the c – myc gene in estrogen-dependent (MCF-7) cells</td></tr><tr><td/><td/><td/><td/></tr><tr><td align=\"left\">c-Jun</td><td align=\"left\">R+</td><td align=\"right\">8417822</td><td align=\"left\">17 beta – Estradiol had little effect on expression of c-jun, jun B, jun D, or c-fos mRNA by MCF-7 cells over 12 h, although it stimulated c-myc expression 4-fold within 30 min</td></tr><tr><td/><td align=\"left\">F+</td><td align=\"right\">14523011</td><td align=\"left\">Furthermore, we identify a phylogenetically conserved AP-1-responsive element in the promoter of the c-myc proto-oncogene that recruits in vivo the c-Jun and JunD AP-1 family members and controls the PDGF-dependent transactivation of the c-myc promoter</td></tr><tr><td/><td align=\"left\">R-</td><td align=\"right\">8219202</td><td align=\"left\">In addition, intracellularly, mitoxantrone-induced PCD was associated with a marked induction of c-jun and significant repression of c-myc and BCL-2 oncogenes</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Path lengths and quantity of stimulatory and inhibitory tokens seem to vary with the type of object.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Number of Paths</bold></td><td align=\"center\"><bold>Average Path Length (tokens)</bold></td><td align=\"center\"><bold>Average stimulatory tokens per path</bold></td><td align=\"center\"><bold>Average inhibitory tokens per path</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>Myc:chemical</bold></td><td align=\"center\">42</td><td align=\"center\">7.64</td><td align=\"center\">0.36</td><td align=\"center\">0.21</td></tr><tr><td align=\"center\"><bold>Myc:process</bold></td><td align=\"center\">850</td><td align=\"center\">7.23</td><td align=\"center\">0.16</td><td align=\"center\">0.11</td></tr><tr><td align=\"center\"><bold>Myc:gene</bold></td><td align=\"center\">1378</td><td align=\"center\">4.93</td><td align=\"center\">0.15</td><td align=\"center\">0.06</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S11-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mi>K</mml:mi>\n <mml:mi>n</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>s</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>t</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munder>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>u</mml:mi>\n <mml:mo>∈</mml:mo>\n <mml:msubsup>\n <mml:mi>Σ</mml:mi>\n <mml:mo>∪</mml:mo>\n <mml:mi>n</mml:mi>\n </mml:msubsup>\n </mml:mrow>\n </mml:munder>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:munder>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>i</mml:mi>\n </mml:mstyle>\n <mml:mo>:</mml:mo>\n <mml:mi>u</mml:mi>\n <mml:mo>≺</mml:mo>\n <mml:mi>s</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>i</mml:mi>\n </mml:mstyle>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:munder>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:munder>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>j</mml:mi>\n </mml:mstyle>\n <mml:mo>:</mml:mo>\n <mml:mi>u</mml:mi>\n <mml:mo>≺</mml:mo>\n <mml:mi>t</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>j</mml:mi>\n </mml:mstyle>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:munder>\n <mml:mrow>\n <mml:msup>\n <mml:mi>λ</mml:mi>\n <mml:mrow>\n <mml:mi>l</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>i</mml:mi>\n </mml:mstyle>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>+</mml:mo>\n <mml:mi>l</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>j</mml:mi>\n </mml:mstyle>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:msup>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S11-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>u</mml:mi><mml:mo>∈</mml:mo><mml:msubsup><mml:mi>Σ</mml:mi><mml:mo>∪</mml:mo><mml:mi>n</mml:mi></mml:msubsup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Possible directional relationships are classified as <bold><underline>F</underline></bold>orward (caffeine affects the object) or <bold><underline>R</underline></bold>everse (the object affects caffeine) and as stimulatory (+), inhibitory (-) or neutral (n). No rel = no relationship found, N.D. rel = no directional relationship found, Expected = the directional relation suggested by the summary information, Extracted = the directional relation with the most support, based on the sentences processed. <bold><italic>Bold italic </italic></bold>font indicates errors.</p></table-wrap-foot>", "<table-wrap-foot><p>See Table 1 for header explanations. (\"Inc.\" = inconclusive - tied for the highest relation score).</p></table-wrap-foot>", "<table-wrap-foot><p>See Table 2 for the particular objects tested.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S11-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S11-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S11-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S11-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S11-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S11-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S11-7\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction and background</title>", "<p>We are exploring hybrid methods where Markov-based statistical profiles, in a log likelihood discriminator framework, are used to create a fixed-length feature vector for Support Vector Machine (SVM) based classification. The core idea of the method is that whenever a log likelihood discriminator can be constructed for classification on stochastic sequential data, an alternative discriminator can be constructed by 'lifting' the log likelihood components into a feature vector description for classification by SVM. Thus, the feature vector uses the individual log likelihood components obtained in the standard log likelihood classification effort, the individual-observation log odds ratios, and 'vectorizes' them rather than sums them. The individual-observation log odds ratios are themselves constructed from positionally defined Markov Models (pMM's), so what results is a pMM/SVM sensor method. This method may have utility in a number of areas of stochastic sequential analysis that are being actively researched, including splice-site recognition and other types of gene-structure identification, file recovery in computer forensics ('file carving'), and speech recognition.</p>", "<p>We test our pMM/SVM method on an interesting discrimination problem in gene-structure identification: splice-site recognition. In this situation the pMM/SVM approach leads to evaluation of the log odds ratio of an observed stochastic sequence, for splice-site and not, by Chow expansion decomposition, with vectorization rather than sum of the log odds ratios of the conditional probabilities on individual observations (where the conditional probabilities are pMM's, and the odds are on splice-site probability versus not-splice-site probability). By focusing on a particular application of the pMM/SVM method, this also allows us to demonstrate some of the subtleties that occur in implementation, and how they can be resolved by information theoretic criteria, here via use of Shannon Entropy in particular.</p>", "<p>Our work makes use of Support Vector Machines for several reasons. Firstly, SVM classifiers have a strong generalized application in machine learning making advances in techniques using them in Bioinformatics directly applicable to other fields utilizing SVM based classifiers. Secondly, the techniques introduced here to automatically target relevant data positions based on entropy analysis have direct contributions to expanding the ability to use SVM classifiers in an unsupervised manner. Finally, though there are existing classifiers currently in use for splice site detection the MM/SVM hybridization is presented here as a novel manner of training against stochastic datasets.</p>", "<title>Shannon Entropy</title>", "<p>Shannon Entropy [##UREF##0##1##] or Information Entropy is a measure of uncertainty or randomness for a given variable in a system. One of the original usages [##UREF##1##2##] for Shannon entropy was the measure of information conveyed on average for symbols in a given language, and it has significant applications in cryptography and other fields where information content must be quantified. The entropy is calculated as a product of probability and the logarithm of probability for each possible state of the targeted variable. Suppose we have the discrete probability distribution p(x_i), for the probability of events x_ifor 'i' in [1..N], i.e., p(x_i) is a discrete probability distribution with N states. Then, Shannon entropy is: -∑p(x_i)log(p(x_i)), where the log function in log₂, ln, or log₁₀results in entropy measured in bit, nat, or dit, respectively. The DNA alphabet, in particular, only has four states: Adenine(A), Cytosine (C), Guanine (G), and Thymine(T), so N = 4 in computations involving this primitive alphabet.</p>", "<title>Splice sites</title>", "<p>Coding regions in eukaryotic DNA are typically interrupted by non-coding regions (95% of cases for protein coding). These non-coding regions are removed by splicing after transcription where pre-mRNA intron segments are removed, and the exon segments remaining are joined together to form the final mRNA. The sequences at the splice region are dominated by GT and AG dinucleotide pairs at the intron side of the Exon-Intron (EI) and Intron-Exon (IE) transitions, respectively (see Fig ##FIG##0##1##).</p>", "<title>Markov Model</title>", "<p>Also known as a Markov Chain, a Markov Model (\"MM\") is a stochastic process with \"short-term\" memory. If there were infinite memory, then the probability of observation X₀, given prior observations {X_∞,..., X₁}, would be expressed as P(X₀|X_∞...X₁). In practice, there is neither the data to support such an infinitely detailed conditioning argument, nor the need. (The existence and utility of highly accurate short-term memory representations relate to fundamental aspects of our physical world, such as equations of motion, causality, entropic increase, and equilibration.) For an Nth-order Markov Model (MM) we have: P(X₀|X_N...X₁) [##UREF##2##3##]. When using MMs part of the model selection problem is the choosing the highest order model that is well-represented by the training data available.</p>", "<title>Positionally-defined Markov Model (pMM)</title>", "<p>In the standard Markov analysis of an event X₀, with prior events {X_N...X₁}, i.e. a memory of the past N events, our fundamental mathematical constructs are the conditional probabilities P(X₀|X_N...X₁). For the analysis we describe here we generalize this formalism, further, to also depend on position vis-à-vis some reference point. In the case of splice-site recognition, positionally-defined Markov Models are used to describe event probabilities at various positions on either side of the splice site (also known as a Profile HMM [##UREF##3##4##]). A pMM is defined as the probability of event X₀, with Markov order N, at position I: P(X₀|I; X_N...X₁).</p>", "<title>Support Vector Machines</title>", "<p>SVMs provide a system for supervised learning which is robust against over training and capable of handling non-separable cases. Learning with structural risk minimization is the central idea behind SVMs, and this is elegantly accomplished by obtaining the separating hyperplane between the binary labeled data sets (± 1) that separates the labeled data sets with a maximum possible margin [##UREF##4##5##,##UREF##5##6##]. The power of this approach is greatly extended by the added modeling freedom provided by a choice of kernel. This is related to preprocessing of data to obtain feature vectors, where, for kernels, the features are now mapped to a much higher dimensional space (technically, an infinite-dimensional space in the case of the popular Gaussian Kernel).</p>", "<p>The hyperplane itself is centered at <bold>w</bold>·<bold>x </bold>- <italic>b </italic>= 0 where <bold>w </bold>is the normal vector to the separating hyperplane, <bold>x </bold>is the vector of points satisfying the above equation, and <italic>b </italic>is the offset from the origin. Given this, <bold>w </bold>and <italic>b </italic>are chosen to maximize the distance or gap between parallel hyperplanes <bold>w</bold>·<bold>x </bold>- <italic>b </italic>= -1 and <bold>w</bold>·<bold>x </bold>- <italic>b </italic>= 1 (see [##REF##17118147##7##] for more details on the implementation we use). The separable case for the SVM occurs where there is no crossover from the labeled groups over the hyperplane. Non-separable cases are handled through the use of slack values [##UREF##5##6##] (see Fig. ##FIG##1##2##) to allow for some cross over in order to still obtain the largest possible margin between the bulk of the labeled groups. One of the strengths of SVMs is that the approach to handling non-separable data is almost identical to that for separable data. Further SVM generalizations, even applications in unsupervised learning/clustering, appear to be possible [##REF##17118147##7##].</p>", "<p>Upon introducing Kernels, the SVM equations are solved by eliminating w and b to arrive at the following Lagrangian formulation: max ∑_{(i = 1...n)}α_i- 1/2 ∑_{(i, j = 1...n)}α_iα_jy_iy_jK(x_i, x_j), subject to α_i≥ 0 and ∑_{(i = 1...n)}α_iy_i= 0, where the decision function is computed as f(x) = sign(∑_{(i = 1...n)}α_iy_iK(x_i, x_j) + b), and where K(x_i, x_j) is the kernel generalization to the inner-product term, &lt;x_i, x_j&gt;, that is obtained in the standard [##UREF##5##6##], intuitively geometric, non-kernel based SVM formulation.</p>" ]

[ "<title>Methods</title>", "<title>pMM/SVM method</title>", "<p>In the typical log likelihood discriminator construction, such as for identification of splice sites, binary classification is provided by the sign of the log odds probability of the splice site vs non-splice-site region. The log odds probability, in turn, is obtained from the sum of the log conditional probabilities from the Chow expansion of observing the observed sequence in the splice-site vs non-splice-site models. In the pMM/SVM method, a sum is not produced from the log conditional probabilites, but a vector. The length of the feature vector depends on the number of terms in the Chow expansion, i.e., on the length of sequence used in the splice-site recognition model. For the splice-site recognition problem described here, an SVM-based classifier is explored for a variety of sequence window sizes (4–20 components). The window size is then determined in an automated fashion, that is minimally sized, by use of Shannon entropy analysis of splice-site alignments.</p>", "<title>Shannon entropy data</title>", "<p>In our research we use Shannon entropy analysis to identify locations of lowered entropy within the sequence surrounding a splice-site. With this automated process we can identify areas of the sequence with lower entropy. These segments of the sequence are less random and therefore contain more information than the remainder of the splice. Using the feature transfer function we transfer the positions identified by Shannon entropy analysis into a feature vector for classification by SVM.</p>", "<p>Initial research utilized a small data set of human splice regions originally extracted from GenBank Rel.123 [##UREF##6##8##]. This set contains approximately 2,700 true EI and 2,800 true EI sequences combined with with 300,000 IE false and 270,000 EI false sequences. Splitting the dataset evenly into four (EI test, EI train, IE test, IE train) created a fast turn around for training and testing amongst the various SVM kernel definitions and parameters (results shown in Figs ##FIG##8##9## and ##FIG##9##10##).</p>", "<p>For more in-depth statistical analysis a larger data set was obtained. Given the resistance of SVMs to over training, we elected to train with a more even ratio of true and false sequence instances. For each species approximately 125,000 true and 125,000 false sequences each for IE and EI, giving a total set of 500,000 sequences for each species between the IE train, IE test, EI train, and EI test sets. Species used for testing include: 1. Chicken; 2. Cow; 3. Dog; 4. Human; 5. Mouse; 6. Opossum; 7. Rat; and 8. Rhesus Monkey.</p>" ]

[ "<title>Results</title>", "<title>Shannon entropy analysis</title>", "<p>We analyzed large data sets using a variety of MM based techniques to study the areas of lowered entropy within splice site sample sequences. This analysis was critical to identifying information-rich sequence regions around the splice site locations, and are used in defining the positional range of pMM's needed in the SVM classification that follows. We perform an analysis of the 0^thorder pMM profile of the Shannon entropy delineated splice site regions, then consider the 1^stand 2^ndorder profiles similarly.</p>", "<p>We begin by analyzing the Shannon-entropy of the pMM at various orders for the sample sequences, and search for contiguous regions with lower than average entropy which we refer to as the low Entropy (\"lEnt\") regions. This is the segment of positional data drawn on to generate feature vectors based on pMM data. The initial entropic analysis using the 0^thorder pMM is used to identify base-positions that have low Shannon entropy. Further analysis using higher order pMMs is used to determine if accounting for greater memory further lowers the entropy of a given position in the sequence. It is found that the positions identified in the lEnt regions carry information about the splice site which a trained SVM can classify with high accuracy.</p>", "<title>EI 0^thOrder pMM</title>", "<p>As shown in Fig. ##FIG##2##3##, the majority of the exon (right) and intron (left) positions maintain a high level of entropy around 1.4 nat but there is a marked decrease in entropy around positions 49 and 50 which correspond to the splice site (see earlier background for high degree of GT for EI splice sites), as expected. There is a noticeable lEnt region corresponding to the 4 positions on the intron side of the splice site (SS+4) with no lEnt region identified in the exon portion of the sequence (using 0^th-order pMM's).</p>", "<title>IE 0^thOrder pMM</title>", "<p>As shown in Fig. ##FIG##3##4##, there is a much larger lEnt region in the IE transition, but with a more gradual drop in entropy which is not nearly as pronounced outside of the splice site consensus at positions 49 and 50 (again corresponding to background information). There is also an interesting spike at 2 positions before the splice site (SS-2) at which entropy returns to the normal base line (consistent with what has been noted by biologists).</p>", "<title>EI pMM 1^st&amp; 2^ndOrder Entropy</title>", "<p>With first order pMM on the EI transition we see the entropy on the first splice site residue increase in proportion to surrounding entropy as compared to the MM Profile entropy for EI (see Figs ##FIG##4##5## &amp;##FIG##5##6##). This is indicative of the high entropy for positions near the splice site. Specifically the position preceding the splice site (SS-1) influences the first splice site position and increases entropy. When we extend the EI pMM to 2^ndorder we observe the entropy increases more evenly the further it extends from the splice site. Additionally we see the lowest entropy point shift further into the intron section under the influence of both residues in the splice site. Along the same lines as the EI 2^ndorder pMM, IE shows a more gradual transition than 1^storder or MM Profile, along with a lessening of the entropy spikes seen previously.</p>", "<title>IE pMM 1^st&amp; 2^ndOrder Entropy</title>", "<p>A similar result is achieved when analyzing IE splice site sequences under pMM 1^stOrder (see Figs ##FIG##6##7## &amp;##FIG##7##8##). We note the decrease in entropy from the exon position following the splice site (SS+1) due to the influence of the low entropy splice site residues. Also of note, however, is the entropy spike toward the end of the intron region (SS-2) which becomes lessened when influenced by the surrounding intron residues in the LET Region. Along the same lines as the EI 2^ndorder pMM, IE shows a more gradual transition than 1^storder or MM Profile, along with a lessening of the entropy spikes seen previously.</p>", "<title>Feature extraction, kernel selection, and SVM training</title>", "<p>Through feature extraction we translate the nucleic acids in the sequence, along with the information garnered from the pMM at various orders, into a numeric value which we transfer into a vector. This is accomplished using a variety of functions with differing amounts of success as detailed in our results. Other feature vector extractions are used that involve ratios between event probability and background probability, as well as direct symbol to numeric transliterations. It appears a number of feature vector rules can be successful, as shown in the Tables in Figures ##FIG##8##9## and ##FIG##9##10##, in the sense that they can provide the basis for strong SVM classification of splice sites.</p>", "<p>Once a feature vector has been produced from the data, by pMM preprocessing in particular, the discriminating task is passed to the SVM. The success of an SVM with a given data set can be greatly improved with a tuning over kernels (and kernel parameters). Efforts to automate this tuning on choice of kernels is currently being explored by use of genetic algorithms (further discussion of that effort is not included here). In the work presented here, we explore a variety of kernels, as shown in the Tables in Figures ##FIG##8##9## and ##FIG##9##10##, including the Dot, Polynomial, Radial, and Neural kernels, where each of the kernels is tuned and scored on its best performing kernel parameters.</p>", "<p>In the tables shown in Figs ##FIG##8##9## and ##FIG##9##10##, the SVM performance is shown for various feature extraction methods. The 0^th-order pMM based method elaborated on here, with log likelihood elements log(e_i(x_i)/q(x_i)), is one of the better performing cases, where e_i(x_i) is the pMM for the i^thposition and q(x_i) is the generic background probability for observation x_i(not positionally dependent). For the null case, or negative instances, we select false splice site locations from the true data by choosing positions outside the splice site regions. These feature vectors are split in half, with one set used to train the SVM and the other used to evaluate the SVM's performance (against data it was not trained against). The accuracy is measured in terms of Sensitivity (\"SN\") and Specificity (\"SP\"). By comparing the {SN, SP} of the training data to the {SN, SP} of the testing data we can evaluate the SVM's classification performance, where the generalization, \"real world\", performance is estimated by the scoring with the test data (and an algorithmic probe of the best performance possible is done by testing on the training data).</p>", "<title>Overview of kernels tested</title>", "<p>A variety of methods for feature extraction as well as kernel types and parameters have been tested to see how well the data sets responded to each. The results for these initial tests based on the data sets obtained from [##UREF##6##8##] are presented in Figs ##FIG##8##9## (EI) and ##FIG##9##10## (IE), which show 2 dimensional table comparisons, where the Y-axis represents the feature transfer function used, and the X-axis represents the specific Kernel function and parameter(s) selected. The table entries themselves show results for Sensitivity (\"SN\") and Specificity (\"SP\"). The Radial Gamma function was chosen to test these results more extensively, along with feature extraction using pMM's.</p>", "<p>Results for this are obtained for four species: Cow, Chicken, Human, and Opossum, and are shown in the EI and IE Results that follow.</p>", "<title>EI splice site results</title>", "<p>We use the Radial kernel with gamma set to 0.5, combined with using Log(e(x)/q(x)) where e(x) is the emission probability, and q(x) is the background probability, for a given residue. These results use much larger data sets than initial trials based on data from [##UREF##6##8##], and show comparison across species boundaries.</p>", "<p>Human was chosen as the base line, with Cow selected for evolutionary similarity as a fellow mammal. Chicken was selected for evolutionary distance between itself and human/cow, and Opossum as a marsupial was similarly chosen for its distance from Chicken, and for not being as close to Human as Cow. Figure ##FIG##10##11## shows the results from training and testing. Classification on training data has sensitivity ranges from 80% to 90%, and specificity in the 80–83% range, except for Opossum which drops to 75% on specificity. These results give an idea what the best-case performance should be. Actual classification on the test data, for a true estimate of learning generalization performance, is found to have a 10% reduction in sensitivity, and a 5% reduction in specificity when compared to the 'best-case' training data performance. Interestingly, the Opossum results are stable with almost negligible change in accuracy when testing on the train and test data sets. The low training results in EI are likely due to the much smaller feature vector size due to a smaller lEnt region for the 0^thorder pMM, this is noticeably less in the IE results as we will now examine.</p>", "<title>IE splice site results</title>", "<p>The IE feature vector size is much larger (15 vs 4) than the EI size. As such, there is a much more stable training result due to IE's SVM being in 15 dimensional space vs the 4 dimensional space for EI. Results are detailed in Fig. ##FIG##11##12##, for the same species examined for EI. In comparison to the EI results, both training sensitivity and specificity are close to 100% accuracy. Transitioning to testing gives a drop of approximately 15% for testing sensitivity, but around 40% in specificity (i.e., resulting in 85% SN and 60% SP). Unlike the EI Opossum results, the IE Opossum results on train and test sets are in line with the Cow, Chicken, and Human behavior.</p>" ]

[]

[ "<title>Conclusion</title>", "<p>The main result of this preliminary study shows pMM/SVMs can be trained as splice site classifiers with high accuracy. We believe this approach is applicable to other problem sets, and represents a new approach that combines entropy analysis for feature selection and eventual pMM/SVM classification. From the specific examples shown, we see that the splice-site classification results using the pMM/HMM approach are very promising, for both IE and EI splice sites. By changing from a 0^thOrder pMM to a higher order pMM, it is possible to extend the low entropy (lEnt) region at the cost of adding noise to the low entropy positions. This increase in the lEnt region allows a lift to an SVM with a higher dimensional feature space, which has an impact on initial training results (as shown in the differences between Figs ##FIG##10##11## and ##FIG##11##12## with vector size 4 and 15, respectively). In ongoing efforts we hope to work with pMMs of higher order, and to begin training SVMs using the 1^stand 2^ndOrder pMM's. This effort is meant to eventually contribute to ongoing construction of a new gene finder approach (by SWH) that leverages the power of SVMs and MM variations (such as those involving gap interpolating MMs).</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<p>In this paper we present preliminary results stemming from a novel application of Markov Models and Support Vector Machines to splice site classification of Intron-Exon and Exon-Intron (5' and 3') splice sites. We present the use of Markov based statistical methods, in a log likelihood discriminator framework, to create a non-summed, fixed-length, feature vector for SVM-based classification. We also explore the use of Shannon-entropy based analysis for automated identification of minimal-size models (where smaller models have known information loss according to the specified Shannon entropy representation). We evaluate a variety of kernels and kernel parameters in the classification effort. We present results of the algorithms for splice-site datasets consisting of sequences from a variety of species for comparison.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>SWH conceptualized the project and performed the preliminary pMM/SVM tests. BR performed the extensive Shannon entropy tests, and the pMM/SVM tests with the large multi-species datasets. SWH and BR each contributed to the writing and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We would like to extend special thanks to Dr. Alexander Tchoubanov for preparing the large, multispecies, sequence set for our experiments. SWH would also like to thank the UNO CSCI 6990 Advanced Machine Learning Methods in Bioinformatics Class of 2004 that worked on this topic as a class project and who helped in doing the initial experiments described in the tables shown in Fig.s ##FIG##8##9## and 10.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Examples of GT – AG splice site sequences.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Illustration of a hyperplane separation of two labeled groups in feature space.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Graph of entropy at each position in the sequence using a 0th Order pMM on an EI SS. The SS occurs at positions 49 and 50.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>Graph of entropy at each position in the sequence using a 0th Order pMM on an IE SS. The SS occurs at positions 49 and 50.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>Graph of entropy at each position in the sequence using a 1st Order pMM on an EI SS. The SS occurs at positions 30 and 31.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>Graph of entropy at each position in the sequence using a 2nd Order pMM on an EI SS. The SS occurs at positions 21 and 22.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>Graph of entropy at each position in the sequence using a 1st Order pMM on an IE SS. The SS occurs at positions 40 and 41.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>Graph of entropy at each position in the sequence using a 2nd Order pMM on an IE SS. The SS occurs at positions 30 and 31.</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p>Table overview of results from feature transfer functions (y-axis) and kernel/parameter selections (x-axis) for EI SS samples.</p></caption></fig>", "<fig position=\"float\" id=\"F10\"><label>Figure 10</label><caption><p>Table overview of results from feature transfer functions (y-axis) and kernel/parameter selections (x-axis) for IE SS samples.</p></caption></fig>", "<fig position=\"float\" id=\"F11\"><label>Figure 11</label><caption><p>Overview of selected results from the larger multi-species datasets using radial kernel on EI sequences.</p></caption></fig>", "<fig position=\"float\" id=\"F12\"><label>Figure 12</label><caption><p>Overview of selected results from the larger multi-species datasets using radial kernel on IE sequences.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S12-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-8\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-9\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-10\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-11\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S12-12\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The bacterium <italic>Staphylococcus aureus </italic>secretes <italic>α</italic>-hemolysin monomers that bind to the outer membrane of susceptible cells. Seven monomers can oligomerize to form a very stable water-filled transmembrane channel [##REF##7809129##1##]. The channel can cause death to the target cell by rapidly discharging vital molecules (such as ATP) and disturbing the membrane potential.</p>", "<p>Suspended in lipid bilayer, as shown in Figure ##FIG##0##1##, the <italic>α</italic>-hemolysin channel can be used as a sensor (nanopore-detector) when large molecules interact with the channel environment under an applied potential (where the open channel has 120 picoAmperes of ion flow under normal conditions). When a 9 bp DNA hairpin enters the pore, the loop is caught at the vestibule mouth, leaving the stem terminus perched to readily bind to the amino acid residues near the limiting aperture, resulting in a consistent toggle for thousands of milliseconds as shown in Figure ##FIG##1##2##.</p>", "<p>Many approaches to characterizing of nucleic acid analyte – channel interactions use 2-D scatter plot analysis [##REF##10585944##2##,##REF##17675346##3##]. A recently proposed method of discriminating translocating RNA polynucleotide orientation [##REF##16214857##4##] uses a combination of six sigmoid phenomenological functional forms to approximate possible blockades. A hybrid method of automated analyte classification was used in [##REF##11231558##5##,##REF##12547778##6##] that discriminates among 8GC, 9GC, 9CG, 9TA and 9AT molecules by first obtaining features extracted with Expectation Maximization (EM) learning on a <italic>single </italic>50-state fully connected Hidden Markov Model (HMM). They then construct a feature vector based on the HMM parameters and pass that to a Support Vector Machine (SVM) for classification (with the binary decision tree shown in Figure ##FIG##2##3##). Although the process shown in Fig. ##FIG##2##3## is scalable, and has high classification accuracy, it can also involve high data rejection rates (good for performing solution assays). This motivates effort to have a less scalable, but lower data-rejection rate (such as what is needed during genomic sequencing). Later study of the data examined in [##REF##11231558##5##,##REF##12547778##6##], with PCA reduction on states followed by a simple, uninformed, AdaBoost classification (not SVM, see [##REF##17118136##7##]), led to similar improvement on zero-rejection accuracy, and thus similar improvements (reductions) on the data-rejection needed for high-accuracy classification [##REF##17118136##7##]. That approach, however, didn't begin with the stronger (but non-scalable in class-number) feature extraction method described here. This is the first test of what is expected to be a highly accurate feature extraction method (better than those employed previously), where the critical limitation in general use, however, is in its scalability in number of classes to discriminate.</p>", "<p>In an interview [In Focus, January 2002], one of the pioneers in the development of nanopore technology, Dr. Mark Akeson, states that getting a machine to learn base pair or nucleotide signatures and report the results automatically will be a key feature of a nanopore sequencing instrument. Here we propose a new method of unsupervised learning of ionic flow blockades with Mixture of Hidden Markov Models (MHMM) profiles that has a number of attractive attributes, at the expense of restricting learning to a smaller state space. For genome sequencing, the problem reduces to identifying the classes {<italic>A</italic>, <italic>C</italic>, <italic>G</italic>, <italic>T</italic>}, i.e., there are only four classes to discriminate. Thus, for some important problems the non-scaling constraint is not an issue and this approach may offer the best performance.</p>", "<p>The Maximum a Posteriori (MAP) molecular classification with our model opens the possibility for making distributed decisions in real time. The EM algorithms running on our model are computationally expensive procedures, thus, an important method in this work involves computational speed-up efforts via distributed processing implementations.</p>" ]

[ "<title>Methods</title>", "<p>In our approach we used unsupervised distributed learning of nanopore ionic flow blockade toggles with an MHMM. MHMMs have a long record of successful implementations that started in speech recognition [##UREF##0##8##] and later were used for clustering protein families [##UREF##1##9##], sequences [##UREF##2##10##] and in the search for splicing enhancers [##REF##16584568##11##]. We use the HMM profile shown in Figures ##FIG##1##2## and ##FIG##8##9(a)## to model the channel blockade process using MHMM components as shown in Figure ##FIG##8##9(b)##. Justification for using such profiles is provided in [##REF##18047713##12##], where we have found the duration of ionic flow blockade levels to be distributed with a simple geometric distribution. The noise at a fixed-level blockade level is typically found to be Gaussian, consistent with the overall thermal and shot noise background for the transient-binding fixed-flow-geometry environments formed by channel and blockading elements.</p>", "<p>The ionic flow blockade records were obtained from the previous studies [##REF##12547778##6##]. Two axon binary files (each containing 500 blockade samples of 300 ms) have been used to learn the probabilistic profiles for each hairpin molecule. The first 100 ms of each channel blockade is the basis of the first test set. Four other axon binary files, with uninterrupted recordings (non-sweep data), for each hairpin molecule and recorded on the same day, are then used for testing. The test set was formed by equiprobable sampling of 500 labeled blockade samples from the pool of test files.</p>", "<p>Another test set was constructed from the above data files to measure accuracy of consecutive same-analyte toggle sample classification. In this instance we take all the available blockade signal coming from the test files of a certain molecule (not just the first 100 ms) and use multiple sample draws from the same signal blockade (i.e., consecutive 100 ms segments). With the 100 ms signal samples drawn from the same blockade event, we perform MAP scoring followed by majority vote (with random resolution of ties). (Note: the data rejection employed in [##REF##12547778##6##] could be made roughly equivalent to the signal resampling approach described here by simply collecting consecutive 100 ms samples, as done here, and having classification on a given blockade once the signal isn't rejected, the only difference with the classification post-processing being that in this effort majority vote is employed instead.) Accuracy is calculated as the number of correct classifications matching the known molecule type to the total number of classification events. As in [##REF##12547778##6##], the ionic flow in each record has been normalized to the open channel current 120 pA prior to learning and testing.</p>", "<p>For our distributed MHMM system implementation we have used cluster of five workstations Sun Ultra 40 M2, each equipped with two AMD Dual-Core Opteron processors (2220SE 2.8 GHz), connected through gigabit Ethernet switch.</p>", "<title>HMM definition and EM learning</title>", "<p>The following parameters describe the conventional HMM implementation according to [##UREF##3##13##]:</p>", "<p>• A set of states <italic>S </italic>= {<italic>S</italic>₁,..., <italic>S</italic>_{<italic>N</italic>}} with <italic>q</italic>_{<italic>t </italic>}being the state visited at time <italic>t</italic>,</p>", "<p>• A set of PDFs <italic>B </italic>= {<italic>b</italic>₁(<italic>o</italic>),..., <italic>b</italic>_{<italic>N</italic>}(<italic>o</italic>)}, describing the emission probabilities <italic>b</italic>_{<italic>j</italic>}(<italic>o</italic>_{<italic>t</italic>}) = <italic>p</italic>(<italic>o</italic>_{<italic>t</italic>}|<italic>q</italic>_{<italic>t </italic>}= <italic>S</italic>_{<italic>j</italic>}) for 1 ≤ <italic>j </italic>≤ <italic>N</italic>, where <italic>o</italic>_{<italic>t </italic>}is the observation at time-point <italic>t </italic>from the sequence of observations <italic>O </italic>= {<italic>o</italic>₁,..., <italic>o</italic>_{<italic>T</italic>}},</p>", "<p>• The state-transition probability matrix A = {<italic>a</italic>_{<italic>i</italic>,<italic>j</italic>}} for 1 ≤ <italic>i</italic>, <italic>j </italic>≤ <italic>N</italic>, where <italic>a</italic>_{<italic>i</italic>, <italic>j </italic>}= <italic>p</italic>(<italic>q</italic>_{<italic>t</italic>+1}= <italic>S</italic>_{<italic>j</italic>}|<italic>q</italic>_{<italic>t </italic>}= <italic>S</italic>_{<italic>i</italic>}),</p>", "<p>• The initial state distribution vector ∏ = {<italic>π</italic>₁,..., <italic>π</italic>_{<italic>N</italic>}}.</p>", "<p>A set of parameters <italic>λ </italic>= (∏, <italic>A</italic>, <italic>B</italic>) completely specifies an HMM. Here we describe the HMM parameter update rules for the EM learning algorithm rigorously derived in [##UREF##4##14##]. When training the HMM using the Baum-Welch algorithm (an Expectation Maximization procedure), first we need to find the expected probabilities of being at a certain state at a certain time-point using the forward-backward procedure as shown in Table ##TAB##0##1##.</p>", "<p>Let us define <italic>ξ</italic>_{<italic>t</italic>}(<italic>i</italic>, <italic>j</italic>) as the probability of being in state <italic>i </italic>at time <italic>t</italic>, and state <italic>j </italic>at time <italic>t </italic>+ 1, given the model and the observation sequence</p>", "<p></p>", "<p>and <italic>γ</italic>_{<italic>t</italic>}(<italic>i</italic>) as the probability of being in state <italic>i </italic>at time <italic>t</italic>, given the observation sequence and the model</p>", "<p></p>", "<p>The HMM maximization step using these probabilities is shown in Table ##TAB##1##2##.</p>", "<title>EM learning of HMM mixture</title>", "<p>The objective of mixture learning is to maximize the likelihood function , i.e. we wish to find the locally optimal set of parameters by using the Expectation Maximization (EM) iterative procedure and the set of data points .</p>", "<p>The Expectation step in the mixture fitting algorithm is done by computing the responsibility matrix of the components given the data points:</p>", "<p></p>", "<p>We use Bayes' rule to find the posterior probability (responsibility) of a mixture component with parameters <italic>λ</italic>_{<italic>m </italic>}and emission sequence <italic>O</italic>_{<italic>k</italic>}:</p>", "<p></p>", "<p>The Expectation step is followed by the maximization step where we re-estimate parameters.</p>", "<p>• Mixture proportions</p>", "<p></p>", "<p>• Initial probabilities</p>", "<p></p>", "<p>where is an estimate of initial probabilities for the component <italic>m </italic>given sequence <italic>O</italic>_{<italic>k</italic>},</p>", "<p>• Transitions</p>", "<p></p>", "<p>where is an estimate of transition probabilities for the component <italic>m </italic>given sequence <italic>O</italic>_{<italic>k</italic>},</p>", "<p>• Emissions</p>", "<p></p>", "<p>where is an estimate of emission parameters for the component <italic>m </italic>given sequence <italic>O</italic>_{<italic>k</italic>}.</p>", "<title>Distributed EM implementation</title>", "<p>As discussed in [##UREF##5##15##], the computational gain of a <italic>parallel </italic>implementation can greatly depend on model topology. In the speech recognition community researchers are able to use a highly parallel HMM architectures for phoneme and dictionary word recognition. Typically, when a large number of Processing Elements (PEs) is used, the utilization of each element drops due to communication overheads. Therefore, the communication overhead in any parallel architecture must be strictly managed, ideally reduced to a constellation of PEs with shared memory [##UREF##5##15##]. In recent work [##REF##18447951##16##] we describe the performance of the following HMM EM algorithms (where we studied the last on the list):</p>", "<p>• Conventional EM due to Leonard E. Baum and Lloyd R. [##UREF##6##17##] takes <italic>O</italic>(<italic>T N</italic>) memory and <italic>O</italic>(2<italic>T N Q</italic>_{<italic>max </italic>}+ <italic>T </italic>(<italic>Q </italic>+ <italic>E</italic>)) time, where <italic>T </italic>is the length of the observed sequence, <italic>N </italic>is the number of HMM states, <italic>Q</italic>_{<italic>max </italic>}is the maximum HMM node out-degree, <italic>E </italic>is the number of free emission parameters, <italic>Q </italic>is the number of free transition parameters.</p>", "<p>• Checkpointing EM [##REF##9088708##18##, ####REF##9682053##19##, ##REF##11159327##20####11159327##20##] takes <italic>O</italic>(<italic>N</italic>) memory and <italic>O</italic>(3<italic>T N Q</italic>_{<italic>max </italic>}+ <italic>T </italic>(<italic>Q </italic>+ <italic>E</italic>)) time,</p>", "<p>• Linear memory EM [##REF##18447951##16##,##REF##16171529##21##] takes only <italic>O</italic>(<italic>N</italic>(<italic>Q </italic>+ <italic>E D</italic>)) memory and <italic>O</italic>(<italic>T NQ</italic>_{<italic>max</italic>}(<italic>Q </italic>+ <italic>E D</italic>)) time.</p>", "<p>Similar improvements are also described for the HMM Viterbi implementation in linear memory [##REF##18447951##16##]. In actual usage with the comparatively small durations generally examined, the checkpointing algorithm was found to be the most memory efficient.</p>", "<title>Distributed checkpointing algorithm for learning from large data samples</title>", "<p>The distributed checkpointing EM algorithm is shown in Figure ##FIG##9##10##. Here are the steps in our distributed checkpointing algorithm implementation:</p>", "<p>1. Client machine splits data sequence <italic>O </italic>into subsequences <italic>O</italic>₁,..., <italic>O</italic>_{<italic>t</italic>},..., each of size and distributes them across the servers along with <italic>λ</italic>,</p>", "<p>2. Find Forward and Backward checkpoints in sequential manner at the corresponding servers where emission matrices for <italic>O</italic>_{<italic>t </italic>}were calculated and stored,</p>", "<p>3. Reconstruct dynamic programming tables of size <italic>N</italic> at the servers according to locally stored checkpoints to make local parameter estimate ,</p>", "<p>4. After calculating local parameter estimate, communicate back to the client machine and calculate ,</p>", "<p>5. Redistribute newly found among the server machines for another EM round.</p>", "<title>Distributed MHMM parameter estimate</title>", "<p>An MHMM can easily split the responsibilities calculation between several cluster nodes with minimum communication overhead in the following way:</p>", "<p>1. For each parameter <italic>λ</italic>₁,..., <italic>λ</italic>_{<italic>m</italic>},..., <italic>λ</italic>_{<italic>M </italic>}and sequence <italic>O</italic>₁,..., <italic>O</italic>_{<italic>k</italic>},..., <italic>O</italic>_{<italic>K </italic>}calculate likelihood <italic>p</italic>(<italic>O</italic>_{<italic>k</italic>}| <italic>λ</italic>_{<italic>m</italic>}) on the server nodes and communicate them back to the client,</p>", "<p>2. Client finds responsibilities for each mixture component and a sequence according to formula (4),</p>", "<p>3. Estimated mixture proportions are found on a client node according to (5),</p>", "<p>4. The server nodes find estimates for parameter <italic>λ</italic>_{<italic>m </italic>}and sequence <italic>O</italic>_{<italic>k </italic>}and send them back to the client,</p>", "<p>5. On the client node these newly computed parameters are weighted according to responsibilities (6), (7), (8),</p>", "<p>6. Newly found HMM parameters are disbursed back to the server nodes for the next round of EM training.</p>" ]

[ "<title>Results</title>", "<p>We have learned blockade signal clusters for five different types of molecules: two such profile mixtures, learned in 50 iterations, are shown in Figure ##FIG##3##4##. The classification accuracy is shown in Figure ##FIG##4##5##, where we used 10-fold resampling of 500 labeled toggle sample subsets from our test set [see Section <italic>Methods</italic>] (the 10-fold resampling is needed to perform majority-vote classification stabilization). The resampling offers a similar stabilization on classifications, and at similar computational expense, to what is done via data-rejection in [##REF##11231558##5##,##REF##12547778##6##]. Accuracy here is defined as</p>", "<p></p>", "<p>where True Positives (TP), True Negatives (TN), False Positives (FP) and False Negatives (FN) are among the classified data samples. We have systematically investigated how the model complexity affects accuracy as shown in Figure ##FIG##5##6##, where average accuracy does not improve for the model of more than 12 components and more than 4 blockade levels, although some individual molecules take advantage of increased model complexity as their classification becomes more accurate. We have also investigated the blockade signal duration needed for proper classification, as shown in Figure ##FIG##6##7##, and for the data-sets examined found that samples with more that 100 ms duration yield little in either average classification accuracy or classification time. We tried using ionic flow blockade samples of 200 ms in the MHMM training, for example, with no apparent improvement to classification accuracy over the 100 ms duration samples. This behavior was not observed with the non-MAP, large-state (but scalable), approach used in [##REF##11231558##5##,##REF##12547778##6##], where greater observation times led to improved classification (although there is agreement that there was diminishing returns on learning sets for signal durations greater then 100 ms, and, especially, if greater than 500 ms).</p>", "<p>The accuracy of consecutive same-analyte toggle samples classification is shown in Figure ##FIG##7##8(b)##, where we reach 100% performance within 14 classifications, except for the 9GC molecule, which underperformed when compared with [##REF##11231558##5##,##REF##12547778##6##]. The difficulty with 9GC classification accuracy convergence could be explained by substantial confusion with 9AT toggles, which reaches ~17% at first classification round and reluctantly reduces to ~3% after 21 classification rounds.</p>", "<p>The accuracy improvement is consistent with the accuracy of the previously reported classification process [##REF##12547778##6##] as shown in Figure ##FIG##7##8(a)## (except for the 9GC molecule). The failure to discern 9GC from 9AT in the approach described here, and not in prior efforts [##REF##11231558##5##,##REF##12547778##6##], may simply be the result of better blockade-level resolution 'fine-structure' with the prior model.</p>", "<p>The better resolution between 9GC and 9AT channel blockades obtained with the 50-state <italic>single </italic>HMM (used in [##REF##11231558##5##,##REF##12547778##6##]) may simply be due to the fixed 1pA resolution (the state quantization bin-size) providing a critical resolving capability between very similar blockade signals. If true, a hybrid solution may be to directly incorporate fine-structure into the 4-state <italic>multiple </italic>HMM processing model that is used here, by adding <italic>fine-structure </italic>states at 1pA distances on either side of the 4 states identified by EM. Efforts along these lines are ongoing (see <italic>Discussion</italic>).</p>", "<p>The MHMM analysis framework first has been implemented in a concurrent fashion on a quad-core Sun Ultra 40 M2 machine with speedup factor 3.66 as compared to a conventional implementation, and then distributed to the five machines of the same type with Java RMI with additional speedup of 4.02, which translates to the total speedup of 3.66 × 4.02 = 14.71.</p>" ]

[ "<title>Discussion and conclusion</title>", "<p>There are several advantages in our approach:</p>", "<p>• Classification is highly accurate with no data dropped from consideration,</p>", "<p>• Model parameters may have intuitive physical interpretation (but not in this study),</p>", "<p>• The MHMM implementation is distributed, such that:</p>", "<p>- Learning can take a larger number of samples (for improved accuracy),</p>", "<p>- Enables real-time analyte classification, currently takes only 0.411 sec to classify 100 ms sample,</p>", "<p>- Checkpointing algorithm keeps the memory profile low both on server and client sides without compromising the running time [##REF##18447951##16##].</p>", "<p>The need for using a mixture model beyond a simple HMM comes from the observation that generally no more than half of hairpin blockades come from the same mode of hairpin molecule interacting with nanopore (the modes correspond to principal components in the channel blockade stationary statistics profile). Other mode contributions require different probabilistic profiles for classification which naturally leads to a mixture analysis problem. The method shown in Figure ##FIG##2##3## doesn't introduce such modes at the HMM-processing stage, relying instead on the strengths of the SVM classifier directly.</p>", "<p>Increasing EM-learning model complexity beyond 4 levels and 12 mixture components increases the log-likelihood of fully trained model, but does not lead to better prediction accuracy as shown in Figure ##FIG##5##6##. The likelihood increase is caused by the model overfitting the data. Overfitting with HMM-profile models, however, isn't found to be as detrimental to the generalization performance as with other learning methods – the main penalty is that the learning and classification times increase dramatically, as we need to estimate progressively increasing number of parameters.</p>", "<p>Since we did not computationally exhaust all the possible parameter settings (number of components, number of levels and sample duration), we provide a rationale for the parameter choice we believe is optimal. With preliminary experiments learning on 9CG toggle samples with MHMM of 15 toggle clusters we have consistently exhausted the number of components, many of them converging to the same simple blockade as shown in figure ##FIG##3##4(a)## at the top right. This observation prompted us to use no more than 12 components in the channel blockade signal-mode mixture model.</p>", "<p>The number of four blockade levels corresponds to the physical model of DNA hairpin interacting with nanopore [##REF##11231558##5##]. From the physical perspective the hairpin molecule can undergo different modes of capture blockade, such as Intermediate Level (IL), Upper Level (UL), Lover Level (LL) conductance states and spikes (S) [##REF##12547778##6##]. When a 9 bp DNA hairpin initially enters the pore, the loop is perched in the vestibule mouth and the stem terminus binds to amino acid residues near the limiting aperture. This results in the IL conductance level. When the terminal basepair desorbs from the pore wall, the stem and loop may realign, resulting in a substantial current increase to UL. Interconversion between the IL and UL states may occur numerous times with UL possibly switching to the LL state. This LL state corresponds to binding of the stem terminus to amino acids near the limiting aperture but in a different manner from IL. From the LL bound state, the duplex terminus may fray, resulting in extension and capture of one strand in the pore constriction resulting into short term S state. The allowed transition events between the levels <italic>IL </italic>⇔ <italic>UL </italic>⇔ <italic>LL </italic>⇔ <italic>S </italic>to happen at any time during the analysis procedure. The spikes model, as described in [##REF##18447951##16##], could possibly be used to increase prediction accuracy. However, with the scenario discussed in this manuscript use of such additions did not lead to higher performance since the primary blockade modes shown in Figures ##FIG##3##4(a)## and ##FIG##3##4(b)## are void of spikes.</p>", "<p>A demo program implementing distributed MHMM analysis framework is available free of charge on our web site <ext-link ext-link-type=\"uri\" xlink:href=\"http://logos.cs.uno.edu/~achurban\"/>.</p>" ]

[ "<title>Discussion and conclusion</title>", "<p>There are several advantages in our approach:</p>", "<p>• Classification is highly accurate with no data dropped from consideration,</p>", "<p>• Model parameters may have intuitive physical interpretation (but not in this study),</p>", "<p>• The MHMM implementation is distributed, such that:</p>", "<p>- Learning can take a larger number of samples (for improved accuracy),</p>", "<p>- Enables real-time analyte classification, currently takes only 0.411 sec to classify 100 ms sample,</p>", "<p>- Checkpointing algorithm keeps the memory profile low both on server and client sides without compromising the running time [##REF##18447951##16##].</p>", "<p>The need for using a mixture model beyond a simple HMM comes from the observation that generally no more than half of hairpin blockades come from the same mode of hairpin molecule interacting with nanopore (the modes correspond to principal components in the channel blockade stationary statistics profile). Other mode contributions require different probabilistic profiles for classification which naturally leads to a mixture analysis problem. The method shown in Figure ##FIG##2##3## doesn't introduce such modes at the HMM-processing stage, relying instead on the strengths of the SVM classifier directly.</p>", "<p>Increasing EM-learning model complexity beyond 4 levels and 12 mixture components increases the log-likelihood of fully trained model, but does not lead to better prediction accuracy as shown in Figure ##FIG##5##6##. The likelihood increase is caused by the model overfitting the data. Overfitting with HMM-profile models, however, isn't found to be as detrimental to the generalization performance as with other learning methods – the main penalty is that the learning and classification times increase dramatically, as we need to estimate progressively increasing number of parameters.</p>", "<p>Since we did not computationally exhaust all the possible parameter settings (number of components, number of levels and sample duration), we provide a rationale for the parameter choice we believe is optimal. With preliminary experiments learning on 9CG toggle samples with MHMM of 15 toggle clusters we have consistently exhausted the number of components, many of them converging to the same simple blockade as shown in figure ##FIG##3##4(a)## at the top right. This observation prompted us to use no more than 12 components in the channel blockade signal-mode mixture model.</p>", "<p>The number of four blockade levels corresponds to the physical model of DNA hairpin interacting with nanopore [##REF##11231558##5##]. From the physical perspective the hairpin molecule can undergo different modes of capture blockade, such as Intermediate Level (IL), Upper Level (UL), Lover Level (LL) conductance states and spikes (S) [##REF##12547778##6##]. When a 9 bp DNA hairpin initially enters the pore, the loop is perched in the vestibule mouth and the stem terminus binds to amino acid residues near the limiting aperture. This results in the IL conductance level. When the terminal basepair desorbs from the pore wall, the stem and loop may realign, resulting in a substantial current increase to UL. Interconversion between the IL and UL states may occur numerous times with UL possibly switching to the LL state. This LL state corresponds to binding of the stem terminus to amino acids near the limiting aperture but in a different manner from IL. From the LL bound state, the duplex terminus may fray, resulting in extension and capture of one strand in the pore constriction resulting into short term S state. The allowed transition events between the levels <italic>IL </italic>⇔ <italic>UL </italic>⇔ <italic>LL </italic>⇔ <italic>S </italic>to happen at any time during the analysis procedure. The spikes model, as described in [##REF##18447951##16##], could possibly be used to increase prediction accuracy. However, with the scenario discussed in this manuscript use of such additions did not lead to higher performance since the primary blockade modes shown in Figures ##FIG##3##4(a)## and ##FIG##3##4(b)## are void of spikes.</p>", "<p>A demo program implementing distributed MHMM analysis framework is available free of charge on our web site <ext-link ext-link-type=\"uri\" xlink:href=\"http://logos.cs.uno.edu/~achurban\"/>.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Ionic current blockade signal processing, for use in nanopore detection, offers a promising new way to analyze single molecule properties with potential implications for DNA sequencing. The <italic>α</italic>-Hemolysin transmembrane channel interacts with a translocating molecule in a nontrivial way, frequently evidenced by a complex ionic flow blockade pattern with readily distinguishable modes of toggling. Effective processing of such signals requires developing machine learning methods capable of learning the various blockade modes for classification and knowledge discovery purposes. Here we propose a method aimed to improve our stochastic analysis capabilities to better understand the discriminatory capabilities of the observed the nanopore channel interactions with analyte.</p>", "<title>Results</title>", "<p>We tailored our memory-sparse distributed implementation of a Mixture of Hidden Markov Models (MHMMs) to the problem of channel current blockade clustering and associated analyte classification. By using probabilistic fully connected HMM profiles as mixture components we were able to cluster the various 9 base-pair hairpin channel blockades. We obtained very high Maximum a Posteriori (MAP) classification with a mixture of 12 different channel blockade profiles, each with 4 levels, a configuration that can be computed with sufficient speed for real-time experimental feedback. MAP classification performance depends on several factors such as the number of mixture components, the number of levels in each profile, and the duration of a channel blockade event. We distribute Baum-Welch Expectation Maximization (EM) algorithms running on our model in two ways. A distributed implementation of the MHMM data processing accelerates data clustering efforts. The second, simultanteous, strategy uses an EM checkpointing algorithm to lower the memory use and efficiently distribute the bulk of EM processing in processing large data sequences (such as for the progressive sums used in the HMM parameter estimates).</p>", "<title>Conclusion</title>", "<p>The proposed distributed MHMM method has many appealing properties, such as precise classification of analyte in real-time scenarios, and the ability to incorporate new domain knowledge into a flexible, easily distributable, architecture. The distributed HMM provides a feature extraction that is equivalent to that of the sequential HMM with a speedup factor approximately equal to the number of independent CPUs operating on the data. The MHMM topology learns clusters existing within data samples via distributed HMM EM learning. A Java implementation of the MHMM algorithm is available at <ext-link ext-link-type=\"uri\" xlink:href=\"http://logos.cs.uno.edu/~achurban\"/>.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>AC conceptualized the project, implemented and tested the MHMM EM algorithm for nanopore ionic flow analysis. SWH helped with writing the manuscript and provided many valuable suggestions directing the study. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This research was partly funded from an NIH K-22 award (K22LM008794), an NIH R-21 award (R21GM073617), and an NIH program grant sub-contract (R01HG003703).</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><italic>α</italic>-hemolysin nanopore with captured hairpin.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Upper Level Toggler (ULT) with profile example.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Existing classification process with HMM feature extraction followed by SVM binary tree decision.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>Toggle clusters for 9GC and 9CG molecules. The mixture proportions correspond to the frequency of a certain toggle mode. Sixteen possible transitions corresponding to profile shown in Figure 7 (a) are shown as chessboard, the darker the area of a cell the more probable a transition. Emissions corresponding to each of the four hidden HMM states are shown below the transitions matrix. MAP classified 100 ms toggle sample from the learning set corresponding to a certain profile is also shown.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>MAP classification accuracy with 10-fold resampling on a split-sample data (with 4 levels and 15 components).</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>Increasing model complexity affects accuracy.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>Accuracy of molecular classification depends on sample duration.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>Proposed and existing process classification accuracy.</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p>HMM profile and mixture of profiles.</p></caption></fig>", "<fig position=\"float\" id=\"F10\"><label>Figure 10</label><caption><p>Distributed Checkpointing algorithm.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Forward and backward procedures.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\">Forward procedure</td><td align=\"center\">Backward procedure</td></tr></thead><tbody><tr><td align=\"left\"><italic>α</italic>_{<italic>t</italic>}(<italic>i</italic>) ≡ <italic>p</italic>(<italic>o</italic>₁,..., <italic>o</italic>_{<italic>t</italic>}|<italic>q</italic>_{<italic>t </italic>}= <italic>S</italic>_{<italic>i</italic>}, <italic>λ</italic>)</td><td align=\"left\"><italic>β</italic>_{<italic>t</italic>}(<italic>i</italic>) ≡ <italic>p</italic>(<italic>o</italic>_{<italic>t</italic>+1},..., <italic>o</italic>_{<italic>T</italic>}|<italic>q</italic>_{<italic>t </italic>}= <italic>S</italic>_{<italic>i</italic>}, <italic>λ</italic>)</td></tr><tr><td align=\"left\">• Initially <italic>α</italic>₁(<italic>i</italic>) = <italic>π</italic>_{<italic>i</italic>}<italic>b</italic>_{<italic>i</italic>}(<italic>o</italic>₁) for 1 ≤ <italic>i </italic>≤ <italic>N</italic>,</td><td align=\"left\">• Initially <italic>β</italic>_T(<italic>i</italic>) = 1 for 1 ≤ <italic>i </italic>≤ <italic>N</italic>,</td></tr><tr><td align=\"left\">• for t = 2, 3,..., T and 1 ≤ <italic>j </italic>≤ <italic>N</italic>,</td><td align=\"left\">• for <italic>t </italic>= <italic>T </italic>- 1,...,1 and 1 ≤ <italic>i </italic>≤ <italic>N</italic>,</td></tr><tr><td align=\"left\">• Finally is the sequence <italic>likehood</italic>.</td><td align=\"left\">• Finally .</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Maximization step in HMM learning.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Initial probability estimate</td><td align=\"left\">Transition probability estimate</td><td align=\"left\">Emission parameters estimate</td></tr></thead><tbody><tr><td align=\"left\"> = <italic>γ</italic>₁(<italic>i</italic>), for 1 ≤ <italic>i </italic>≤ <italic>N</italic>.</td><td align=\"left\">, for 1 ≤ <italic>i</italic>, <italic>j </italic>≤ <italic>N</italic>.</td><td align=\"left\">Gaussian emission , , for 1 ≤ <italic>j </italic>≤ <italic>N</italic>.</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S13-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>A</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>u</mml:mi>\n <mml:mi>r</mml:mi>\n <mml:mi>a</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>y</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>T</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>P</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>T</mml:mi>\n <mml:mi>N</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mi>F</mml:mi>\n <mml:mi>N</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S13-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>α</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>j</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>α</mml:mi><mml:mrow><mml:mi>t</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>i</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mo>,</mml:mo><mml:mi>j</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:mstyle></mml:mrow><mml:mo>]</mml:mo></mml:mrow><mml:msub><mml:mi>b</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>o</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S13-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>β</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>i</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>α</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mo>,</mml:mo><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:msub><mml:mi>b</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>o</mml:mi><mml:mrow><mml:mi>t</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:msub><mml:mi>β</mml:mi><mml:mrow><mml:mi>t</mml:mi><mml:mo>+</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>j</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S13-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>O</mml:mi><mml:mo>|</mml:mo><mml:mi>λ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>α</mml:mi><mml:mi>T</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>i</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S13-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>O</mml:mi><mml:mo>|</mml:mo><mml:mi>λ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>π</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:msub><mml:mi>b</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>o</mml:mi><mml:mn>1</mml:mn></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:msub><mml:mi>β</mml:mi><mml:mn>1</mml:mn></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>i</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1471-2105-9-S9-S13-i6\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mi>ξ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>j</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>q</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:msub>\n <mml:mi>S</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:msub>\n <mml:mi>q</mml:mi>\n <mml:mrow>\n <mml:mi>t</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:msub>\n <mml:mi>S</mml:mi>\n <mml:mi>j</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:mi>O</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mi>α</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:msub>\n <mml:mi>a</mml:mi>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>j</mml:mi>\n </mml:mrow>\n </mml:msub>\n <mml:msub>\n <mml:mi>b</mml:mi>\n <mml:mi>j</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>o</mml:mi>\n <mml:mrow>\n <mml:mi>t</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msub>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:msub>\n <mml:mi>β</mml:mi>\n <mml:mrow>\n <mml:mi>t</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>j</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>O</mml:mi>\n <mml:mo>|</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM3\"><label>(3)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1471-2105-9-S9-S13-i7\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mi>γ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>q</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:msub>\n <mml:mi>S</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:mi>O</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mi>α</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:msub>\n <mml:mi>β</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>N</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:msub>\n <mml:mi>α</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:msub>\n <mml:mi>β</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>=</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>j</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>N</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>ξ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>j</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1471-2105-9-S9-S13-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>π</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1471-2105-9-S9-S13-i9\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>a</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mrow><mml:mi>i</mml:mi><mml:mo>,</mml:mo><mml:mi>j</mml:mi></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>t</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:mi>T</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msubsup><mml:mrow><mml:msub><mml:mi>ξ</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>i</mml:mi><mml:mo>,</mml:mo><mml:mi>j</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>t</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mrow><mml:mi>T</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msubsup><mml:mrow><mml:msub><mml:mi>γ</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>i</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M10\" name=\"1471-2105-9-S9-S13-i10\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>b</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>o</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>→</mml:mo><mml:mi>μ</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>t</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>T</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>o</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:msub><mml:mi>γ</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>j</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>t</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>T</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>γ</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>j</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M11\" name=\"1471-2105-9-S9-S13-i11\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>b</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>o</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>→</mml:mo><mml:msup><mml:mi>σ</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>t</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>T</mml:mi></mml:msubsup><mml:mrow><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>o</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>μ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:msub><mml:mi>γ</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>j</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>t</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>T</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>γ</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>j</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M12\" name=\"1471-2105-9-S9-S13-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi mathvariant=\"script\">O</mml:mi><mml:mo>|</mml:mo><mml:mi>Θ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:msubsup><mml:mi>Π</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:msubsup><mml:mi>p</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>o</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>|</mml:mo><mml:mi>Θ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mi>ℒ</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>Θ</mml:mi><mml:mo>|</mml:mo><mml:mi mathvariant=\"script\">O</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M13\" name=\"1471-2105-9-S9-S13-i13\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msup><mml:mi>Θ</mml:mi><mml:mo>∗</mml:mo></mml:msup><mml:mo>=</mml:mo><mml:mi>a</mml:mi><mml:mi>r</mml:mi><mml:mi>g</mml:mi><mml:mi>m</mml:mi><mml:mi>a</mml:mi><mml:mi>x</mml:mi><mml:mi>ℒ</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>Θ</mml:mi><mml:mo>|</mml:mo><mml:mi mathvariant=\"script\">O</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M14\" name=\"1471-2105-9-S9-S13-i14\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">O</mml:mi></mml:semantics></mml:math></inline-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M15\" name=\"1471-2105-9-S9-S13-i15\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mrow>\n <mml:mrow>\n <mml:munder>\n <mml:munder>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mn>1</mml:mn>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mn>1</mml:mn>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mo>⋯</mml:mo>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>M</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mn>1</mml:mn>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mn>1</mml:mn>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mn>2</mml:mn>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mo>⋯</mml:mo>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>M</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mn>2</mml:mn>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mn>1</mml:mn>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mn>3</mml:mn>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mo>⋯</mml:mo>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>M</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mn>3</mml:mn>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋯</mml:mo>\n </mml:mtd>\n <mml:mtd>\n <mml:mo>⋯</mml:mo>\n </mml:mtd>\n <mml:mtd>\n <mml:mo>⋯</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mn>1</mml:mn>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>K</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mo>⋯</mml:mo>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>M</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>K</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>Θ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo stretchy=\"true\">︸</mml:mo>\n </mml:munder>\n <mml:mrow>\n <mml:mi>M</mml:mi>\n <mml:mtext> mixture components</mml:mtext>\n </mml:mrow>\n </mml:munder>\n </mml:mrow>\n <mml:mo>}</mml:mo>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n <mml:mtext> data points</mml:mtext>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM4\"><label>(4)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M16\" name=\"1471-2105-9-S9-S13-i16\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mi>α</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>j</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>M</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:msub>\n <mml:mi>α</mml:mi>\n <mml:mi>j</mml:mi>\n </mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>j</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM5\"><label>(5)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M17\" name=\"1471-2105-9-S9-S13-i17\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mover accent=\"true\">\n <mml:mi>α</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mi>K</mml:mi>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM6\"><label>(6)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M18\" name=\"1471-2105-9-S9-S13-i18\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mover accent=\"true\">\n <mml:mi>Π</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:msubsup>\n <mml:mover accent=\"true\">\n <mml:mi>Π</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>m</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msubsup>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M19\" name=\"1471-2105-9-S9-S13-i19\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mover accent=\"true\"><mml:mi>Π</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>m</mml:mi><mml:mi>k</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM7\"><label>(7)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M20\" name=\"1471-2105-9-S9-S13-i20\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mover accent=\"true\">\n <mml:mi>A</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:msubsup>\n <mml:mover accent=\"true\">\n <mml:mi>A</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>m</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msubsup>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M21\" name=\"1471-2105-9-S9-S13-i21\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mover accent=\"true\"><mml:mi>A</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>m</mml:mi><mml:mi>k</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM8\"><label>(8)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M22\" name=\"1471-2105-9-S9-S13-i22\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mover accent=\"true\">\n <mml:mi>B</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:msubsup>\n <mml:mover accent=\"true\">\n <mml:mi>B</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>m</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msubsup>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>K</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>λ</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>O</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mi>λ</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M23\" name=\"1471-2105-9-S9-S13-i23\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mover accent=\"true\"><mml:mi>B</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>m</mml:mi><mml:mi>k</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M24\" name=\"1471-2105-9-S9-S13-i24\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msqrt><mml:mi>T</mml:mi></mml:msqrt></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M25\" name=\"1471-2105-9-S9-S13-i25\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>O</mml:mi><mml:mrow><mml:msqrt><mml:mi>T</mml:mi></mml:msqrt></mml:mrow></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M26\" name=\"1471-2105-9-S9-S13-i24\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msqrt><mml:mi>T</mml:mi></mml:msqrt></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M27\" name=\"1471-2105-9-S9-S13-i24\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msqrt><mml:mi>T</mml:mi></mml:msqrt></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M28\" name=\"1471-2105-9-S9-S13-i24\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msqrt><mml:mi>T</mml:mi></mml:msqrt></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M29\" name=\"1471-2105-9-S9-S13-i26\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>t</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>Π</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>t</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>A</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>t</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>B</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M30\" name=\"1471-2105-9-S9-S13-i27\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>t</mml:mi></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M31\" name=\"1471-2105-9-S9-S13-i28\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mo>=</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:mover accent=\"true\"><mml:mi>Π</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mo>,</mml:mo><mml:mover accent=\"true\"><mml:mi>A</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mo>,</mml:mo><mml:mover accent=\"true\"><mml:mi>B</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M32\" name=\"1471-2105-9-S9-S13-i29\" overflow=\"scroll\"><mml:semantics><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M33\" name=\"1471-2105-9-S9-S13-i30\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>α</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mn>1</mml:mn></mml:msub><mml:mo>,</mml:mo><mml:mn>...</mml:mn><mml:mo>,</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>α</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>m</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:mn>...</mml:mn><mml:mo>,</mml:mo><mml:msub><mml:mi>α</mml:mi><mml:mi>M</mml:mi></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M34\" name=\"1471-2105-9-S9-S13-i31\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>m</mml:mi><mml:mi>k</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M35\" name=\"1471-2105-9-S9-S13-i32\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mn>1</mml:mn></mml:msub><mml:mo>,</mml:mo><mml:mn>...</mml:mn><mml:mo>,</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>λ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>M</mml:mi></mml:msub></mml:mrow></mml:semantics></mml:math></inline-formula>" ]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S13-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-8\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-9\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S13-10\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Understanding the statistical properties of genomic sequences helps recreate the dynamical processes that led to the current DNA structure, and determine gene-related diseases like cancer and Alzheimer disease. For instance, based on the view that non-coding DNA exhibits long-range correlations [##REF##1301010##1##, ####UREF##0##2##, ##REF##11542924##3##, ##UREF##1##4##, ##UREF##2##5##, ##UREF##3##6####3##6##], Li [##REF##9904836##7##] proposed an expansion-modification model of gene evolution. The model incorporates the two basic features of DNA evolution: (i) sequence elongation due to gene duplication and (ii) mutations. It can be shown that the limiting sequence created by this dynamical process exhibits a long-range correlation structure, as attested by a 1/<italic>f</italic>^{<italic>α </italic>}spectrum, where the exponent <italic>α </italic>is a function of the probability of mutation. To understand the relationship between the DNA correlation structure and possible gene abberations, Dodin et al. [##REF##9015453##8##] designed a simple correlation function intended to visualize the regular patterns encountered in DNA sequences. This function is used to revisit the intriguing question of triplet repeats with the aim of providing a visual estimate of the propensity of genes to be highly expressed and/or to lead to possible aberrant structures formed upon strand slippage.</p>", "<p>Statistical analysis of genomic sequences has, however, relied, for a long time, on signal processing techniques for stationary signals (correlation and power spectrum) [##UREF##0##2##,##UREF##1##4##,##REF##10045801##9##,##UREF##4##10##], and statistical tools for slowly-varying trends within stationary signals (Detrended Fluctuation Analysis or DFA) [##REF##1301010##1##,##UREF##2##5##,##UREF##3##6##]. Stationarity can be argued as a valid assumption for time-series of short duration. However, such an assumption rapidly loses its credibility in the enormous databases maintained by various genome projects. Standard statistical tests (e.g., Priestley's test for stationarity) have been used to verify that genomic sequences are not stationary and the nature of their non-stationarity varies and is often more complex than a simple trend [##UREF##5##11##,##UREF##6##12##]. Subsequently, more recent analysis of genomic data [##REF##1301010##1##] has relied on time-varying power spectral methods (the evolutionary spectrum and periodogram) to capture the statistical characteristics of genomic sequences [##UREF##5##11##,##UREF##6##12##]. The main difficulty in using time-varying spectral methods is that they are extremely unstable and very noisy. Typically, the power spectrum and the evolutionary spectrum are averaged over time in order to obtain smooth and less jittery curves. Moreover, as was pointed out in [##UREF##7##13##], the evolutionary spectrum is restricted to the class of oscillatory processes. A stochastic process, <italic>X</italic>(<italic>t</italic>), is oscillatory if it has a representation of the form</p>", "<p></p>", "<p>Where <italic>Z</italic>(<italic>λ</italic>) is an orthogonal increment process, and the evolutionary power spectrum of the process is defined by <italic>P </italic>(<italic>t</italic>, <italic>λ</italic>) = |<italic>A</italic>(<italic>t</italic>, <italic>λ</italic>)|². This definition of the evolutionary power spectrum has the following disadvantages [##UREF##7##13##]:</p>", "<p><bold>(i) </bold>It is not uniquely defined for a given non-stationary process.</p>", "<p><bold>(ii) </bold>The estimation procedure for the evolutionary spectrum depends greatly on the nature of theamplitude function <italic>A</italic>(<italic>t</italic>, <italic>λ</italic>), which is not known a priori.</p>", "<p><bold>(iii) </bold>An increase in the number of observations does not provide added information on the local behavior of the evolutionary spectrum, and thus does not improve estimation accuracy.</p>", "<p>We propose to model non-stationary genomic sequences by a time-dependent autoregressive moving average (TD-ARMA) process. Cramer [##UREF##8##14##] showed that a non-stationary process still possesses a Wold decomposition in terms of its innovation and its generating system. However, the linear system generating the non-stationary signal, <italic>x</italic>(<italic>t</italic>), when driven by the innovation, <italic>w</italic>(<italic>t</italic>), is no longer shift-invariant; the parameters of the impulse response, <italic>h</italic>_{<italic>u</italic>}, of this system are time-dependent so that</p>", "<p></p>", "<p>The existence of a time-varying ARMA representation of this process is ensured by two theorems due, independently, to Grenier [##UREF##9##15##] and Huang and Aggarwal [##UREF##10##16##]. The uniqueness of the TD-ARMA representation is obtained by constraining the ARMA model to be invertible, but this leads to conditions on the time-varying impulse response {<italic>h</italic>_{<italic>u</italic>}(<italic>t</italic>)} and its inverse (namely to be absolutely summable at any time <italic>t</italic>), which cannot be easily expressed in terms of the time-dependent coefficients of the ARMA model. In this paper, we estimate the time-dependent coefficients of the general TD-ARMA model using mean-squares, least-squares and recursive least-squares algorithms. The mean-squares estimation leads to generalized Yule-Walker type equations [##UREF##9##15##]. Once the non-stationary parameters are estimated (as time series), we use them to provide a basis for statistical inference by defining an index of randomness, which quantitatively assesses how close the non-stationary signal is to white noise. Our simulation results on various gene sequences show that (i) both the coding and non-coding segments of a gene are not random, and (ii) the coding segments are \"closer\" to random sequences than non-coding segments. Our results support the view that both coding and non-coding sequences are not random [##UREF##5##11##,##UREF##6##12##,##REF##10045801##9##,##UREF##11##17##, ####REF##8423849##18##, ##UREF##12##19##, ##UREF##13##20####13##20##], and revokes the stationary study that maintains that non-coding DNA sustains long-range correlations whereas coding DNA behaves like random sequences [##REF##1301010##1##, ####UREF##0##2##, ##REF##11542924##3####11542924##3##,##UREF##2##5##,##UREF##3##6##,##UREF##4##10##].</p>" ]

[ "<title>Methods</title>", "<title>Numerical representation of genomic sequences</title>", "<p>Converting the DNA sequence into a digital signal offers the opportunity to apply powerful signal processing methods for the handling and analysis of genomic information. This is, however, not an easy task as the analysis results might depend on the chosen map. Various numerical mappings have been adopted in the literature. To cite few, Peng et al. [##REF##1301010##1##] construct a one-dimensional map of nucleotide sequences onto a walk, <italic>u</italic>(<italic>i</italic>), which they termed \"DNA walk\". The DNA walk is defined by the rule that the walker steps up (<italic>u</italic>(<italic>i</italic>) = +1) if a pyrimidine resides at position <italic>i</italic>, and steps down (<italic>u</italic>(<italic>i</italic>) = -1) otherwise. Voss [##REF##10045801##9##] represents a DNA sequence by four binary indicator sequences, which indicate the locations of the four nucleotides A, T, C and G. Berthelsen et al. [##REF##9906993##21##] proposed a two-dimensional representation of DNA sequences, characterized by a Hausdorff dimension (also called fractal dimension) that is invariant under (i) complementarity, (ii) reflection symmetry, (iii) compatibility and (iv) substitution symmetry of A͘T and C↔G. The corresponding embedding assignment is given by A = (-1; 0), T = (1; 0), C = (0; -1) and G = (0; 1). In this paper, since we are interested in time-dependent ARMA modeling of one-dimensional non-stationary genomic sequences, we adopt the widely used \"DNA walk\" map proposed by Peng et al [##REF##1301010##1##]. The DNA walk provides a nice graphical representation for each gene. For instance, Figure ##FIG##0##1## shows the structure of the Human gene 276 located in chromosome 1, and its DNA walk is displayed in Fig. ##FIG##1##2##.</p>", "<title>Time-dependent ARMA model</title>", "<p>Grenier [##UREF##14##22##] showed that a discrete non-stationary signal, {<italic>x </italic>[<italic>n</italic>]}, can be represented by finite-order time-varying ARMA processes of the form</p>", "<p></p>", "<p>where <italic>N </italic>is the length of the sequence <italic>x </italic>[<italic>n</italic>], <italic>a</italic>_{<italic>i </italic>}[<italic>n</italic>] and <italic>b</italic>_{<italic>i </italic>}[<italic>n</italic>] are the time-dependent model parameters, <italic>p </italic>and <italic>q </italic>are the model orders and <italic>w </italic>[<italic>n</italic>] is a white noise process. Observe that the coefficients <italic>a</italic>_{<italic>i </italic>}[<italic>n</italic>] and <italic>b</italic>_{<italic>i </italic>}[<italic>n</italic>] appear with an argument <italic>n </italic>- <italic>i </italic>depending on <italic>i</italic>. This is purely arbitrary since any time origin can be chosen, without restraining the generality of the model. We assume that the time-dependent coefficients <italic>a</italic>_{<italic>i </italic>}[<italic>n</italic>] and <italic>b</italic>_{<italic>i </italic>}[<italic>n</italic>] can be expressed as linear combinations of some basis functions ,</p>", "<p></p>", "<p></p>", "<p>The advantage of the basis parametrization is clear from the fact that the identification of the time-dependent coefficients <italic>a</italic>_{<italic>i </italic>}[<italic>n</italic>] and <italic>b</italic>_{<italic>i </italic>}[<italic>n</italic>] reduces to the identification of the constant coefficients and , and therefore the linear non-stationary problem reduces to a linear time-invariant problem. The basis functions do not have to be limited to the standard choices of Legendre, Fourier, or the prolate spheroidal basis but can also take advantage of any prior information, such as the presence of a jump in the coefficients at a known instant [##UREF##14##22##].</p>", "<title>Estimation of the time-dependent ARMA coefficients</title>", "<p>From Eqs. (4) and (5), the unknown parameters of the TD-ARMA model are the weights of the linear combinations defining each time-varying parameter. The linearity is the key to the algorithms proposed in this paper. We will derive mean-squares, least-squares and recursive least-squares solutions to estimate the time-dependent coefficients and .</p>", "<title>Mean-squares estimation</title>", "<p>Define the process</p>", "<p></p>", "<p>and define the vector</p>", "<p></p>", "<p>where the symbol ^{<italic>t </italic>}stands for the transpose of a vector or a matrix. It is possible to derive for this process orthogonality conditions similar to the stationary ARMA model conditions [##UREF##15##23##]. Observe that the process <italic>v </italic>[<italic>n</italic>], defined in Eq. (6), is orthogonal to [<italic>w</italic>[<italic>n </italic>- <italic>q </italic>- 1], <italic>w </italic>[<italic>n </italic>- <italic>q </italic>- 2], ⋯]; hence, it is orthogonal to <italic>x </italic>[<italic>n </italic>- <italic>q </italic>- <italic>i</italic>] for all <italic>i </italic>&gt; 0, and orthogonal to <italic>X </italic>[<italic>n </italic>- <italic>q </italic>- <italic>i</italic>] for all <italic>i </italic>&gt; 0 [##UREF##14##22##]. This orthogonality condition leads to a generalized Yule-Walker equation [##UREF##14##22##]</p>", "<p></p>", "<p>Although the process <italic>x </italic>[<italic>n</italic>] is non-stationary, the stationarity and ergodicity of the process <italic>w </italic>[<italic>n</italic>], together with the linearity of the model, allow us to replace in Eq. (8) the expectation by a summation. However, although consistent, the above estimator is often considered a poor one [##UREF##14##22##].</p>", "<title>Least-squares estimation</title>", "<p>Equations (4) and (5) can be written in vector format as follows</p>", "<p></p>", "<p>where</p>", "<p></p>", "<p>Define</p>", "<p></p>", "<p>Then, we have</p>", "<p></p>", "<p>Using this vector notation, Eq. (3) can be written as</p>", "<p></p>", "<p>Or equivalently</p>", "<p></p>", "<p>where <italic>φ</italic>^{<italic>t </italic>}[<italic>n</italic>] is the row vector</p>", "<p></p>", "<p>and</p>", "<p></p>", "<p>Observe that the vector <italic>θ </italic>contains all the unknowns of the TD-ARMA model. Writing Eq. (10) for <italic>n </italic>= 0, 1, ⋯, <italic>N </italic>- 1 leads to</p>", "<p></p>", "<p>where</p>", "<p></p>", "<p>The least-squares solution of Eq. (11) is given by</p>", "<p></p>", "<p>To overcome the computational complexity associated with the least-squares estimation (involving in particular the inversion of a square (<italic>m </italic>+ 1)(<italic>p </italic>+ <italic>q</italic>) × (<italic>m </italic>+ 1)(<italic>p </italic>+ <italic>q</italic>) matrix), we opted for a recursive least-squares estimation as follows.</p>", "<title>Recursive least-squares estimation</title>", "<p>The recursive least squares algorithm is summarized as [##UREF##16##24##]</p>", "<p></p>", "<p></p>", "<p></p>", "<p>The initial conditions can be chosen arbitrarily.</p>", "<title>Index of randomness</title>", "<p>Over the past decade, there has been a flow of conflicting papers about the long-range power-law correlations detected in eukaryotic DNA [##REF##1301010##1##, ####UREF##0##2##, ##REF##11542924##3####11542924##3##,##UREF##2##5##,##UREF##3##6##,##REF##10045801##9##, ####UREF##4##10##, ##UREF##5##11##, ##UREF##6##12####6##12##,##UREF##11##17##, ####REF##8423849##18##, ##UREF##12##19##, ##UREF##13##20####13##20##]. The controversy is generated by conflicting views that either advocate that non-coding DNA sustains long-range correlations whereas coding DNA behaves like random sequences [##REF##1301010##1##,##UREF##4##10##,##REF##11542924##3##,##UREF##2##5##,##UREF##3##6##], or maintains that both coding and non-coding DNA exhibit long-range power-law correlations [##UREF##5##11##,##UREF##6##12##,##REF##10045801##9##,##UREF##11##17##, ####REF##8423849##18##, ##UREF##12##19##, ##UREF##13##20####13##20##]. Based on the analysis of the time-dependent power spectrum of genomic sequences, Bouaynaya and Schonfeld [##UREF##5##11##,##UREF##6##12##] showed that the statistical differences between coding and non-coding sequences are more subtle than previously concluded using stationary analysis tools. In fact they found that both coding and non-coding sequences are non-random. However, coding sequences are \"whiter\" than non-coding sequences.</p>", "<p>We propose to qualitatively assess the degree of randomness of both coding and non-coding sequences using the time-dependent ARMA coefficients <italic>a</italic>_{<italic>i </italic>}[<italic>n</italic>] and <italic>b</italic>_{<italic>i </italic>}[<italic>n</italic>]. Consider the system function, <italic>H </italic>(<italic>z</italic>), of a stationary ARMA model (whose coefficients <italic>a</italic>_{<italic>i </italic>}and <italic>b</italic>_{<italic>i </italic>}are constant, i.e., independent of time). We know that</p>", "<p></p>", "<p>where (resp. ) are the zeros (resp. poles) of the system function. From the fact that a stationary white noise process has a at spectrum, we observe that the closer (in L₂distance) the zeros and poles are, the flatter is the spectrum of the process. Following the same reasoning locally for non-stationary processes, we define the curve of randomness, <italic>CR </italic>[<italic>n</italic>], of the non-stationary process <italic>x </italic>[<italic>n</italic>] by</p>", "<p></p>", "<p>where the minimum is taken over all pairs (<italic>r</italic>_{<italic>k </italic>}[<italic>n</italic>], <italic>p</italic>_{<italic>k </italic>}[<italic>n</italic>]). Observe that the case <italic>p </italic>&lt;<italic>q </italic>is obtained from the <italic>p </italic>&gt; <italic>q </italic>case by interchanging the roles of <italic>r</italic>_{<italic>k </italic>}and <italic>p</italic>_{<italic>k</italic>}, and the indices <italic>p </italic>and <italic>q</italic>. The curve of randomness defined in Eq. (17) is a measure of how close the zeros and the poles of the system function are, and therefore, is a measure of how flat the system function is, and how close is the process from a white noise. The index of randomness, <italic>IR</italic>(<italic>p</italic>, <italic>q</italic>), of a TD-ARMA(<italic>p</italic>, <italic>q</italic>), is then defined as the average of the curve of randomness, i.e.,</p>", "<p></p>", "<p>In particular, the index of randomness of a TD-ARMA(1,1) (<italic>x </italic>[<italic>n</italic>] + <italic>a</italic>[<italic>n </italic>- 1]<italic>x</italic>[<italic>n </italic>- 1] = <italic>w</italic>[<italic>n</italic>] + <italic>b</italic>[<italic>n</italic>]<italic>w</italic>[<italic>n </italic>- 1]) is given by</p>", "<p></p>", "<p>Observe that the index of randomness of a white noise process is equal to zero. We say that the sequence <italic>x</italic>₁[<italic>n</italic>] with index of randomness <italic>IR</italic>₁is more random than the sequence <italic>x</italic>₂[<italic>n</italic>] with index of randomness <italic>IR</italic>₂if the index of randomness of the former is lower than the index of randomness of the latter, i.e., <italic>IR</italic>₁&lt;<italic>IR</italic>₂.</p>" ]

[ "<title>Results</title>", "<p>All genome sequences considered in this paper have been extracted from the NIH website <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov\"/>. The algorithms were implemented in MATLAB. The DNA sequences were mapped to numerical sequences using the purine-pyrimidine DNA walk proposed in [##REF##1301010##1##]. In our simulations, the recursive least squares algorithm was found to best estimate the time-dependent coefficients of the TD-ARMA model. We used the MATLAB function <italic>rarmax</italic>, which implements the recursive least-squares algorithm for TD-ARMA models. The choice of the orders <italic>p </italic>and <italic>q </italic>of the model were determined experimentally as follows: For each genomic sequence, we computed 100 TD-ARMA models corresponding to the orders (1, 1) up to (10, 10). The best model was chosen to be the one that minimizes the average squared error between the actual and the fitted sequences. Our extensive simulations on various DNA sequences from different organisms show that most of the sequences are best fitted with low-order TD-ARMA models, e.g., TD-ARMA(1,1), TD-ARMA(1,2) and TD-ARMA(2,1). Figure ##FIG##2##3## shows the DNA walk of the Human gene 276 and its TD-ARMA(1,1) fitted sequence. Observe that the TD-ARMA(1,1) model accurately fits this gene sequence. The estimated time-varying coefficients <italic>a </italic>[<italic>n</italic>] and <italic>b </italic>[<italic>n</italic>] are displayed in Fig. ##FIG##3##4## for both the coding and non-coding regions of this gene. Their statistical differences are not clear from the plot of the time-series coefficients. The curves of randomness of the coding and non-coding regions are displayed in Fig. ##FIG##4##5##. Table ##TAB##0##1## shows the index of randomness of various gene sequences. For concise representation, the column titles have been abbreviated as follows: \"C. length\" (resp.\"N.C. length\") denotes the length (in base pairs) of the coding (resp. non-coding) segment of the gene. The total length of the gene is the sum of the lengths of its coding and non-coding regions. \"C. (<italic>p</italic>, <italic>q</italic>)\" (resp. \"N.C. (<italic>p</italic>, <italic>q</italic>)\") denotes the optimal TD-ARMA parameters (<italic>p</italic>, <italic>q</italic>) for the coding (resp. non-coding) region of the gene. Finally, \"C. IR\" (resp. \"N.C. IR\") is the index of randomness of the coding (resp. non-coding) segment of the gene. Observe that, in all considered genes, the index of randomness of both the coding and non-coding segments are strictly positive, and the index of randomness of the coding region is consistently lower than the index of randomness of the non-coding region (recall that the index of randomness of a white noise is zero). These observations bring to bear two important conclusion: (i) Both the coding and non-coding sequences are not random, as attested by an index of randomness greater than zero. (ii) The coding sequences are \"whiter\" than the non-coding sequences. This conclusion revokes previous work on statistical correlation of DNA sequences, which, based on stationary time-series analysis, presumed that coding DNA behaves like random sequences [##REF##1301010##1##, ####UREF##0##2##, ##REF##11542924##3####11542924##3##,##UREF##2##5##,##UREF##3##6##,##UREF##4##10##]; and supports the conflicting view that both coding and non-coding sequences are not random [##UREF##5##11##,##UREF##6##12##,##REF##10045801##9##,##UREF##11##17##, ####REF##8423849##18##, ##UREF##12##19##, ##UREF##13##20####13##20##]. In particular, our conclusion is in accordance with the evolutionary periodogram analysis conducted in [##UREF##5##11##,##UREF##6##12##].</p>" ]

[]

[ "<title>Conclusion</title>", "<p>In this paper, we modelled the non-stationary genomic sequences by a time-dependent autoregressive moving average (TD-ARMA) model. By expressing the time-dependent coefficients as linear combinations of parametric basis functions, we were able to transform a linear non-stationary problem into a linear time-invariant problem. Subsequently, we proposed three methods to estimate the time-dependent coefficients: Mean -square, least-squares, and recursive least-squares algorithms. Based on the estimated TD-ARMA coefficients, we defined an index of randomness to quantify the degree of randomness of both coding and non-coding sequences. We found that both coding and non-coding sequences are not random. However, a higher index of randomness attests that coding sequences are \"whiter\" than non-coding sequences. These results corroborate the evolutionary periodogram analysis of genomic sequences performed in [##UREF##5##11##] and [##UREF##6##12##], and revoke the stationary analysis' conclusion that coding DNA behaves like random sequences.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Over the past decade, many investigators have used sophisticated time series tools for the analysis of genomic sequences. Specifically, the correlation of the nucleotide chain has been studied by examining the properties of the power spectrum. The main limitation of the power spectrum is that it is restricted to stationary time series. However, it has been observed over the past decade that genomic sequences exhibit non-stationary statistical behavior. Standard statistical tests have been used to verify that the genomic sequences are indeed not stationary. More recent analysis of genomic data has relied on time-varying power spectral methods to capture the statistical characteristics of genomic sequences. Techniques such as the evolutionary spectrum and evolutionary periodogram have been successful in extracting the time-varying correlation structure. The main difficulty in using time-varying spectral methods is that they are extremely unstable. Large deviations in the correlation structure results from very minor perturbations in the genomic data and experimental procedure. A fundamental new approach is needed in order to provide a stable platform for the non-stationary statistical analysis of genomic sequences.</p>", "<title>Results</title>", "<p>In this paper, we propose to model non-stationary genomic sequences by a time-dependent autoregressive moving average (TD-ARMA) process. The model is based on a classical ARMA process whose coefficients are allowed to vary with time. A series expansion of the time-varying coefficients is used to form a generalized Yule-Walker-type system of equations. A recursive least-squares algorithm is subsequently used to estimate the time-dependent coefficients of the model. The non-stationary parameters estimated are used as a basis for statistical inference and biophysical interpretation of genomic data. In particular, we rely on the TD-ARMA model of genomic sequences to investigate the statistical properties and differentiate between coding and non-coding regions in the nucleotide chain. Specifically, we define a quantitative measure of randomness to assess how far a process deviates from white noise. Our simulation results on various gene sequences show that both the coding and non-coding regions are non-random. However, coding sequences are \"whiter\" than non-coding sequences as attested by a higher index of randomness.</p>", "<title>Conclusion</title>", "<p>We demonstrate that the proposed TD-ARMA model can be used to provide a stable time series tool for the analysis of non-stationary genomic sequences. The estimated time-varying coefficients are used to define an index of randomness, in order to assess the statistical correlations in coding and non-coding DNA sequences. It turns out that the statistical differences between coding and non-coding sequences are more subtle than previously thought using stationary analysis tools: Both coding and non-coding sequences exhibit statistical correlations, with the coding regions being \"whiter\" than the non-coding regions. These results corroborate the evolutionary periodogram analysis of genomic sequences and revoke the stationary analysis' conclusion that coding DNA behaves like random sequences.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>JSZ derived the different estimation algorithms of the TD-ARMA parameters and performed the simulations. NB proposed the use of the non-stationary analysis and the index of randomness as a basis for statistical inference and biophysical interpretation of genomic data, derived the different estimation algorithms of the TD-ARMA parameters, and drafted the manuscript. DS proposed the use of the non-stationary analysis and the index of randomness as a basis for statistical inference and biophysical interpretation of genomic data and derived the different estimation algorithms of the TD-ARMA parameters. WO proposed the use of TD-ARMA modeling as a non-stationary model of genomic sequences. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>The publication of this paper was partially supported by the Arkansas IDeA Networks of Biomedical Research Excellence (INBRE).</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Gene Structure</bold>. Gene structure of the Human gene 276 located in chromosome 1: The boxes correspond to the exons (coding regions), and the lines between them represent the introns (non-coding regions). The total length of the gene is <italic>N </italic>= 8208 bases, including 1536 coding bases and 6672 non-coding bases.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>DNA Walk</bold>. DNA walk of the Human gene 276.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>TD-ARMA modeling</bold>. TD-ARMA modeling of the Human gene 276: The blue signal is the DNA walk, and the red signal is the TD-ARMA(1,1) fitted signal. The TD-ARMA(1,1) model accurately fits the genomic signal.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>TD-ARMA coefficients estimation</bold>. Estimation of the TD-ARMA(1,1) coefficients of the Human gene 276. The TD-ARMA(1,1) model is given by <italic>x </italic>[<italic>n</italic>] + <italic>a </italic>[<italic>n </italic>- 1] <italic>x </italic>[<italic>n </italic>- 1] = <italic>w </italic>[<italic>n</italic>] + <italic>b </italic>[<italic>n </italic>- 1] <italic>w </italic>[<italic>n </italic>- 1]. The blue and black (resp. red and green) curves depict the time series <italic>a</italic>[<italic>n</italic>] (resp. <italic>b</italic>[<italic>n</italic>]) for the coding and non-coding regions of the gene, respectively.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>Curve of randomness</bold>. The curves of randomness of the coding and non-coding regions of the Human gene 276 are shown in blue and red, respectively. The index of randomness of the coding sequence is equal to 1.0603, whereas its corresponding value for the non-coding sequence is equal to 1.0627.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Index of Randomness of the Coding and Non-Coding segments of Various Gene Sequences</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\">Gene NIH accession number</td><td align=\"center\">C. length</td><td align=\"center\">C. (<italic>p</italic>, <italic>q</italic>)</td><td align=\"center\">C. IR</td><td align=\"center\">N.C. length</td><td align=\"center\">N.C. (<italic>p</italic>, <italic>q</italic>)</td><td align=\"center\">N.C. IR</td></tr></thead><tbody><tr><td align=\"center\">Ashbya gossypii (fungus) AE016815</td><td align=\"center\">180953</td><td align=\"center\">(1,1)</td><td align=\"center\">0.9466</td><td align=\"center\">674919</td><td align=\"center\">(1,1)</td><td align=\"center\">0.9860</td></tr><tr><td align=\"center\">Aspergillus fumigatus (form of fungus) CM000169</td><td align=\"center\">1227993</td><td align=\"center\">(2,1)</td><td align=\"center\">0.9870</td><td align=\"center\">1835394</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0683</td></tr><tr><td align=\"center\">Candida albicans (form of yeast) AP006852</td><td align=\"center\">373390</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0282</td><td align=\"center\">570789</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0429</td></tr><tr><td align=\"center\">Candida albicans AP006852</td><td align=\"center\">373390</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0282</td><td align=\"center\">570789</td><td align=\"center\">(3,1)</td><td align=\"center\">1.0429</td></tr><tr><td align=\"center\">fission yeast GI:157310483</td><td align=\"center\">753661</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0402</td><td align=\"center\">1654671</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0642</td></tr><tr><td align=\"center\">fruit fly AE002620</td><td align=\"center\">21399</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0084</td><td align=\"center\">1222832</td><td align=\"center\">(1,2)</td><td align=\"center\">1.1075</td></tr><tr><td align=\"center\">fruit fly AE002725</td><td align=\"center\">11316</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0145</td><td align=\"center\">659655</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0320</td></tr><tr><td align=\"center\">Homo sapiens hs-gene277 NG-004750</td><td align=\"center\">1639</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0688</td><td align=\"center\">6573</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0808</td></tr><tr><td align=\"center\">Homo sapiens hs-gene276 NG-004750</td><td align=\"center\">1536</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0603</td><td align=\"center\">6672</td><td align=\"center\">(1,1)</td><td align=\"center\">1.0627</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label><italic>X</italic>(<italic>t</italic>) = ∫ <italic>A</italic>(<italic>t</italic>, <italic>λ</italic>)<italic>e</italic>^{2<italic>iπλt </italic>}<italic>dZ</italic>(<italic>λ</italic>),</disp-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S14-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>t</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>u</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mi>∞</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>h</mml:mi>\n <mml:mi>u</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>t</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>t</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>u</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM3\"><label>(3)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S14-i2\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>+</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>a</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>+</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>q</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>b</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n <mml:mo>,</mml:mo>\n <mml:mo>⋯</mml:mo>\n <mml:mo>,</mml:mo>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S14-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>f</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>n</mml:mi><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM4\"><label>(4)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S14-i4\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mi>a</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mi>m</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>c</mml:mi>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>k</mml:mi>\n </mml:mrow>\n </mml:msub>\n <mml:msub>\n <mml:mi>f</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM5\"><label>(5)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S14-i5\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:msub>\n <mml:mi>b</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mi>m</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>d</mml:mi>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>k</mml:mi>\n </mml:mrow>\n </mml:msub>\n <mml:msub>\n <mml:mi>f</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1471-2105-9-S9-S14-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>c</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mo>,</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1471-2105-9-S9-S14-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>d</mml:mi><mml:mrow><mml:mi>i</mml:mi><mml:mo>,</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1471-2105-9-S9-S14-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>f</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>n</mml:mi><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>0</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1471-2105-9-S9-S14-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>n</mml:mi><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>p</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M10\" name=\"1471-2105-9-S9-S14-i9\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>b</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>n</mml:mi><mml:mo stretchy=\"false\">]</mml:mo><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>q</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM6\"><label>(6)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M11\" name=\"1471-2105-9-S9-S14-i10\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>v</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>+</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>a</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>+</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>q</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>b</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n <mml:mo>,</mml:mo>\n <mml:mo>⋯</mml:mo>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM7\"><label>(7)</label><italic>X </italic>[<italic>n</italic>] = [<italic>f</italic>₀[<italic>n</italic>]<italic>x</italic>[<italic>n</italic>], ⋯, <italic>f</italic>_{<italic>m </italic>}[<italic>n</italic>]<italic>x</italic>[<italic>n</italic>]]^{<italic>t</italic>},</disp-formula>", "<disp-formula id=\"bmcM8\"><label>(8)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M12\" name=\"1471-2105-9-S9-S14-i11\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>E</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>X</mml:mi>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo>⋯</mml:mo>\n <mml:mi>X</mml:mi>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n <mml:mi>t</mml:mi>\n </mml:msup>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mi>θ</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mo>−</mml:mo>\n <mml:mi>E</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:mo>⋅</mml:mo>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><italic>a</italic>_{<italic>i </italic>}[<italic>n</italic>] = <bold>f</bold>^{<bold>t </bold>}[<italic>n</italic>] <bold>c</bold>_{<bold>i</bold>},   and   <italic>b</italic>_{<italic>i </italic>}[<italic>n</italic>] = <bold>f</bold>^{<bold>t </bold>}[<italic>n</italic>] <bold>d</bold>_{<bold>i</bold>},</disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M13\" name=\"1471-2105-9-S9-S14-i12\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>f</mml:mi>\n </mml:mstyle>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>f</mml:mi>\n <mml:mn>0</mml:mn>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>f</mml:mi>\n <mml:mi>m</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>c</mml:mi>\n </mml:mstyle>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>i</mml:mi>\n </mml:mstyle>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>c</mml:mi>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>c</mml:mi>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>m</mml:mi>\n </mml:mrow>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>d</mml:mi>\n </mml:mstyle>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>i</mml:mi>\n </mml:mstyle>\n </mml:msub>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>d</mml:mi>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>d</mml:mi>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>m</mml:mi>\n </mml:mrow>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><bold>u</bold>^{<bold>t </bold>}[<italic>n</italic>] = <italic>x </italic>[<italic>n</italic>] <bold>f</bold>^{<bold>t </bold>}[<italic>n</italic>],   and   <bold>v</bold>^{<bold>t </bold>}[<italic>n</italic>] = <italic>w </italic>[<italic>n</italic>] <bold>f</bold>^{<bold>t </bold>}[<italic>n</italic>].</disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M14\" name=\"1471-2105-9-S9-S14-i13\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>a</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:msup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>u</mml:mi>\n </mml:mstyle>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:msub>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>c</mml:mi>\n </mml:mstyle>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>i</mml:mi>\n </mml:mstyle>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>b</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:msup>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>v</mml:mi>\n </mml:mstyle>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:msub>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>d</mml:mi>\n </mml:mstyle>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>i</mml:mi>\n </mml:mstyle>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM9\"><label>(9)</label><italic>x </italic>[<italic>n</italic>] + <bold>u</bold>^{<italic>t </italic>}[<italic>n </italic>- 1] <bold>c</bold>_{<bold>1 </bold>}+ ⋯ + <bold>u</bold>^{<italic>t </italic>}[<italic>n </italic>- <italic>p</italic>] <bold>c</bold>_{<bold>p </bold>}= <italic>w </italic>[<italic>n</italic>] + <bold>v</bold>^{<italic>t </italic>}[<italic>n </italic>- 1] <bold>d</bold>_{<bold>1 </bold>}+ ⋯ + <bold>v</bold>^{<italic>t </italic>}[<italic>n </italic>- <italic>q</italic>] <bold>d</bold>_{<bold>q</bold>}</disp-formula>", "<disp-formula id=\"bmcM10\"><label>(10)</label><italic>x </italic>[<italic>n</italic>] + <italic>φ</italic>^{<italic>t </italic>}[<italic>n</italic>] <italic>θ </italic>= <italic>w </italic>[<italic>n</italic>],</disp-formula>", "<disp-formula><italic>φ</italic>^{<italic>t </italic>}[<italic>n</italic>] = [<bold>u</bold>^{<italic>t </italic>}[<italic>n </italic>- 1], ⋯, <bold>u</bold>^{<italic>t </italic>}[<italic>n </italic>-<italic>p</italic>], - <bold>v</bold>^{<italic>t </italic>}[<italic>n </italic>- 1], ⋯, <bold>v</bold>^{<italic>t </italic>}[<italic>n </italic>-<italic>q</italic>]],</disp-formula>", "<disp-formula><bold><italic>θ </italic></bold>= [<bold>c</bold>_{<bold>1</bold>}, ⋯,<bold>c</bold>_{<bold>p</bold>}, <bold>d</bold>_{<bold>1</bold>}, ⋯, <bold>d</bold>_{<bold>q</bold>}]^{<italic>t</italic>}.</disp-formula>", "<disp-formula id=\"bmcM11\"><label>(11)</label><bold>x = Φ <italic>θ </italic>+ w</bold>,</disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M15\" name=\"1471-2105-9-S9-S14-i14\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>Φ</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:msup>\n <mml:mi>φ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mn>0</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:msup>\n <mml:mi>φ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>x</mml:mi>\n </mml:mstyle>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mn>0</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:mstyle mathvariant=\"bold\" mathsize=\"normal\">\n <mml:mi>w</mml:mi>\n </mml:mstyle>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mn>0</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mo>⋮</mml:mo>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM12\"><label>(12)</label><bold><italic>θ </italic>= (Φ^{<italic>t </italic>}Φ)^-1Φ^{<italic>t </italic>}x</bold></disp-formula>", "<disp-formula id=\"bmcM13\"><label>(13)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M16\" name=\"1471-2105-9-S9-S14-i15\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mover accent=\"true\">\n <mml:mi>θ</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mover accent=\"true\">\n <mml:mi>θ</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>+</mml:mo>\n <mml:mi>L</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mo>{</mml:mo>\n <mml:mi>x</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>+</mml:mo>\n <mml:msup>\n <mml:mi>φ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mover accent=\"true\">\n <mml:mi>θ</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>}</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM14\"><label>(14)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M17\" name=\"1471-2105-9-S9-S14-i16\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>L</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mo>−</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>φ</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mn>1</mml:mn>\n <mml:mo>+</mml:mo>\n <mml:msup>\n <mml:mi>φ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>φ</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM15\"><label>(15)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M18\" name=\"1471-2105-9-S9-S14-i17\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>−</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>φ</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:msup>\n <mml:mi>φ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mn>1</mml:mn>\n <mml:mo>+</mml:mo>\n <mml:msup>\n <mml:mi>φ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msup>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>P</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mtext> </mml:mtext>\n <mml:mi>φ</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM16\"><label>(16)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M19\" name=\"1471-2105-9-S9-S14-i18\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>H</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>z</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mi>q</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:msub>\n <mml:mi>b</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:msup>\n <mml:mi>z</mml:mi>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mi>k</mml:mi>\n </mml:mrow>\n </mml:msup>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:msub>\n <mml:mi>a</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:msup>\n <mml:mi>z</mml:mi>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mi>k</mml:mi>\n </mml:mrow>\n </mml:msup>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∏</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>q</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo>−</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:msup>\n <mml:mi>z</mml:mi>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∏</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo>−</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:msup>\n <mml:mi>z</mml:mi>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M20\" name=\"1471-2105-9-S9-S14-i19\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>r</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>q</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M21\" name=\"1471-2105-9-S9-S14-i20\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mrow><mml:mo>{</mml:mo><mml:msub><mml:mi>p</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo>}</mml:mo></mml:mrow><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>p</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM17\"><label>(17)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M22\" name=\"1471-2105-9-S9-S14-i21\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>{</mml:mo>\n <mml:mrow>\n <mml:mtable columnalign=\"left\">\n <mml:mtr columnalign=\"left\">\n <mml:mtd columnalign=\"left\">\n <mml:mrow>\n <mml:mi>C</mml:mi>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:msub>\n <mml:mrow>\n <mml:mi>min</mml:mi>\n <mml:mo>⁡</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>,</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:msub>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mi>q</mml:mi>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>q</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>−</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>+</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>q</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd columnalign=\"left\">\n <mml:mrow>\n <mml:mtext>if </mml:mtext>\n <mml:mi>p</mml:mi>\n <mml:mo>&gt;</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo>;</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr columnalign=\"left\">\n <mml:mtd columnalign=\"left\">\n <mml:mrow>\n <mml:mi>C</mml:mi>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:msub>\n <mml:mrow>\n <mml:mi>min</mml:mi>\n <mml:mo>⁡</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>,</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:msub>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mi>p</mml:mi>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>−</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>+</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>q</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo>+</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>q</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd columnalign=\"left\">\n <mml:mrow>\n <mml:mtext>if </mml:mtext>\n <mml:mi>q</mml:mi>\n <mml:mo>&gt;</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo>;</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr columnalign=\"left\">\n <mml:mtd columnalign=\"left\">\n <mml:mrow>\n <mml:mi>C</mml:mi>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:msub>\n <mml:mrow>\n <mml:mi>min</mml:mi>\n <mml:mo>⁡</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>,</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:msub>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mi>p</mml:mi>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:msubsup>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>−</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd columnalign=\"left\">\n <mml:mrow>\n <mml:mtext>if </mml:mtext>\n <mml:mi>p</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n </mml:mrow>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM18\"><label>(18)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M23\" name=\"1471-2105-9-S9-S14-i22\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>I</mml:mi>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>q</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mi>N</mml:mi>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:munderover>\n <mml:mrow>\n <mml:mi>C</mml:mi>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM19\"><label>(19)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M24\" name=\"1471-2105-9-S9-S14-i23\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>I</mml:mi>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo>,</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mi>N</mml:mi>\n </mml:mfrac>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:munderover>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:mi>a</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>−</mml:mo>\n <mml:mi>b</mml:mi>\n <mml:mo stretchy=\"false\">[</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo stretchy=\"false\">]</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>" ]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S14-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S14-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S14-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S14-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S14-5\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Originally employed to analyze single or a small collection of targeted molecules, gas chromatography-mass spectrometry (GC-MS) and other chromatography-spectrometry technologies have emerged as viable tools for the wholesale fingerprinting of complex chemical mixtures. This has been made possible by the advent of computer-aided chemometrics, which in principle can lead to identification and quantification of most or all component chemicals. This advancement continues to profoundly benefit scientific disciplines as diverse as petroleum, diesel and biodiesel chemistry [##REF##17727864##1##,##REF##12830915##2##]; biomarker discovery [##REF##17054416##3##]; basic metabolic chemistry; drug metabolite identification; receptor-ligand and enzyme-substrate biochemistry; environmental toxicology [##REF##17417068##4##]; pharmacokinetics; functional genomics [##REF##18251859##5##] and metabolomics [##REF##16921475##6##,##REF##15389842##7##].</p>", "<p>Separations with mass detection yield <italic>mass chromatograms</italic>; with the intensity of the mass detector's response indexed both to the ion mass-to-charge ratio (<italic>mz</italic>) channel being monitored, and to the time elapsed since injection of the biochemical mixture onto the chromatographic column positioned upstream of the detector, <italic>i.e</italic>., to its retention time (<italic>RT</italic>). With modern mass spectrometers, the variation in <italic>mz </italic>of a chemical is usually modest and often can be ignored during data processing. <italic>RT </italic>variation can be appreciable, however, as illustrated in Figure ##FIG##0##1##, and nonlinear over the extent of a chromatogram as dramatically illustrated in Smith et al. [##REF##16448051##8##] and elsewhere [##REF##17466315##9##,##REF##17963529##10##]. Retention-time differences are caused by uncontrolled experimental variables such as column aging and instabilities in flow rates of mobile phases and the shape of thermal or mobile-phase gradients [##REF##17466315##9##,##REF##12184621##11##,##REF##15144199##12##]. Misalignment was a minor issue as long as multidimensional separation technologies were used to quantify a few molecular targets, but manual curation proves arduous if not impossible when each mass chromatogram displays hundreds of potentially significant features and an experiment contains hundreds of such analytical runs.</p>", "<p>The data produced from chromatography coupled with mass spectrometry can be viewed as <italic>three-way</italic>: along <italic>RT </italic>space, <italic>mz </italic>space, and analytical run space. Some of the most attractive and powerful three-way techniques, such as parallel factor analysis (PARAFAC) to further resolve peaks, assume <italic>trilinearity </italic>of data, a mathematical constraint such that multiple instances of a data feature align with each other along all three dimensions, which rarely if ever is achieved in real mass-chromatographic data, primarily due to <italic>RT </italic>misalignment. Techniques making no trilinearity assumption (<italic>e.g.</italic>, PARAFAC2 and MCR-ALS) still usually require alignment to facilitate parsing of the large matrix representing an entire, typical chemical profiling experiment into submatrices of computationally feasible size. This is particularly true since parsing must occur at locations along the chromatogram lacking peaks. Finally, two-way, one-run-at-a-time approaches such as AMDIS [##UREF##0##13##], when applied serially to multiple runs, <italic>e.g.</italic>, with the help of MET-IDEA [##REF##16808440##14##] or SpectConnect [##REF##17263323##15##], have been observed to produce an artifact where single chemical eluates are identified as multiple mass-chromatographic features [##UREF##1##16##], again largely the result of misalignment. Thus, a complete comparative analysis of data acquired in a non-targeted, profile-type experiment, involving many analytical runs, needs to include a robust alignment operation as an obligatory preprocessing step.</p>", "<p>Alignment algorithms have been described as falling within two categories based on whether they use feature detection or not [##REF##17963529##10##]. The best-known method that does not detect features is correlation optimized warping (COW), proposed by Nielson et al. [##UREF##2##17##]; it warps, i.e., linearly interpolates, one chromatogram to another by selecting input parameters such as section length and slack size that maximize the similarity between the two chromatograms using dynamic programming. Optimization of the input parameters is difficult, however, and performance is often questionable [##REF##14719890##18##,##REF##16643929##19##]. Variant warping algorithms, such as parametric and semi-parametric time warping, have been proposed to address these deficiencies (reviewed in [##REF##16643929##19##]). Feature-detection algorithms, in contrast, attempt to identify and match peaks throughout an entire set of runs. Although this approach requires one additional step for alignment, it generally produces superior results and adds the ability to integrate peak areas during the process. Recent examples of such methods include metAlign, MZmine, and XCMS [##REF##16448051##8##,##REF##16403790##20##,##UREF##3##21##]. These methods differ with regard to which features are used for matching, some employing only features evident in <italic>RT </italic>space [##REF##12830915##2##,##REF##14630657##22##], while others also use spectral information [##REF##16448051##8##,##REF##17963529##10##,##REF##12184621##11##,##REF##16403790##20##].</p>", "<p>We have developed and tested an improved <italic>RT </italic>alignment method that relies on feature detection and utilizes matching criteria based on both peak retention time and peak spectral data. Peaks in sample mass chromatograms are detected and matched to peaks in an arbitrarily selected reference chromatogram. Mass spectra provide information required to determine whether peaks from different samples are chemically identical components. In addition to retention data and mass spectra, our method utilizes an inherent property of chromatograms: peaks eluting near to each other tend to show similar deviations in their retention times, and thus can be initially processed as blocks of peaks. Through trial, or simulated, shifts of blocks along the <italic>RT </italic>axis relative to the reference chromatogram, and through reorganization of peaks into new blocks as needed, an optimal shift strategy is discovered. This shift information is applied to both the TIC and the full, two-dimensional matrix of raw data while warping only non-peak regions, in an effort to exactly preserve the shapes and integrated areas of key peaks. Thus, the result matrix can be used as a direct input to subsequent multivariate analysis.</p>" ]

[]

[ "<title>Results</title>", "<title>Algorithm</title>", "<p>Our alignment method operates in a pairwise fashion: one mass chromatogram, a sample <italic>S</italic>, is aligned with a reference chromatogram, <italic>R</italic>. <italic>R </italic>can be any run from the set of all runs but, once selected, must be used for the entire set. Chromatographic peaks with acceptable signal-to-noise ratio (<italic>SN</italic>) and width are detected in <italic>R</italic>, and for each <italic>S </italic>as its processing is begun, by analyzing chromatograms with a published wavelet-based method [##REF##16820428##23##]. A <italic>peak set </italic>contains only those peaks actually used in the alignment process, accompanied by further information about them. An <italic>S </italic>peak set <italic>SP </italic>always includes all detected peaks that satisfy <italic>signal-to-noise </italic>and width criteria. An <italic>R </italic>peak set <italic>RP </italic>will typically contain only a subset of all peaks; it is a dynamic set, in that it is selected anew for every <italic>S</italic>, using only peaks most compatible with the <italic>S </italic>being processed. Accurate alignment is possible through the matching of peaks detected in select <italic>mz </italic>channels, <italic>i.e.</italic>, in select extracted ion chromatograms (EIC), instead of in the TIC, where coelution and higher baselines can muddy the picture. Overall, the alignment process for a pair of chromatograms involves (a) finding EIC peaks for the two, (b) iteratively matching them, which also yields retention discrepancy data, and (c) aligning, <italic>i.e.</italic>, warping and shifting, the sample chromatogram based on peak-match data. This pairwise alignment is repeated for every sample, matching to the same reference. Only the first two steps will be explained in detail in this paper.</p>", "<title>Peak detection</title>", "<p>The purpose of this step is to assemble two sets of peaks, for <italic>S </italic>and <italic>R</italic>, such that they closely resemble each other in size and in the mass-spectral characteristics of their elements. Let the set of peaks from a sample TIC be <italic>SP </italic>= {P₁, P₂,..., P_n} and from a reference \"inferred TIC\" (details below) be <italic>RP </italic>= {Q₁, Q₂,..., Q_m}. Their elements are ordered by elution time and each element can be envisioned as a group of EIC peaks, one resulting from each ion produced upon ionization, with possible fragmentation, of an eluted chemical component. Up to five EIC peaks per TIC peak are recorded in descending order by their signal-to-noise ratio (SN), thus, for instance, element <italic>x </italic>of <italic>SP </italic>is the set of EIC peaks, P_{<italic>x </italic>}= {p_{<italic>x</italic>1}, p_{<italic>x</italic>2},... p_{<italic>xj</italic>}} where 1 ≤ <italic>x </italic>≤ <italic>n </italic>and 1 ≤ <italic>j </italic>≤ 5. Similarly for <italic>RP</italic>, Q_{<italic>x </italic>}= {q_{<italic>x</italic>1}, q_{<italic>x</italic>2},..., q_{<italic>xk</italic>}}, 1 ≤ <italic>x </italic>≤ <italic>m</italic>, 1 ≤ <italic>k </italic>≤ 5, as illustrated in Figure ##FIG##1##2##. In this peak detection stage, all necessary EIC peaks are found for the alignment, accompanied by the requisite peak information: <italic>mz</italic>; width; retention time of apices; and <italic>SN</italic>. Since values for the time axis are in units of MS scan number and are discrete values, perhaps as few as eight across a peak, the peak maxima found by a peak detection algorithm will often deviate from their true apices. Thus, for more precise alignment, fractional top positions are determined for use in the actual alignment. It should be noted, however, that subsequent alignment involves only integral shifts in scan number, in order to preserve the matrix-like structure of mass-chromatographic data. Considering the three points acquired nearest the apex of a peak, each an ordered pair (x, y) where x is scan and y is intensity, we can solve the quadratic equation <italic>y </italic>= <italic>Ax</italic>²+ <italic>Bx </italic>+ <italic>C </italic>describing the unique downward-opening parabola defined by those points, using simple linear algebra. The true apex occurs at the position where the first derivative is zero, and will equal (-<italic>B/2A</italic>, <italic>C-B</italic>²/<italic>4A</italic>).</p>", "<p>For actual detection of both TIC and EIC peaks, we use the continuous wavelet transform algorithm of Du <italic>et al</italic>. [##REF##16820428##23##] since it is robust to noise and readily available in the authors' MassSpecWavelet package for the R statistical language [##UREF##4##24##]. After NetCDF [##UREF##5##25##] files of <italic>S </italic>and <italic>R </italic>are read into matrices of intensities, the peak detection method proceeds as follows (symbol conventions are summarized in Table ##TAB##0##1##):</p>", "<p>1. Detect peaks in a sample TIC <italic>S </italic>whose <italic>SN </italic>ratios are greater than <italic>SNtic</italic>.</p>", "<p>2. Form peak set <italic>SP </italic>by finding, for each detected TIC peak, as many component EIC peaks as possible, not to exceed five, for which the distance between its apex and that of the TIC is less than <italic>pClose</italic>, the <italic>SN </italic>is greater than <italic>SNeic</italic>, the peak width is less than <italic>pWidth</italic>; and the <italic>SN </italic>is among the five highest <italic>SN </italic>of all EIC peaks passing these criteria. Retention times are expressed in fractions of scans by a quadratic interpolation of their apex positions, as described above.</p>", "<p>3. For each peak in <italic>SP</italic>, find EIC peaks in the reference run <italic>R</italic>, which have corresponding <italic>mz </italic>values.</p>", "<p>4. Group the found EIC peaks into a peak set <italic>RP</italic>, an \"inferred TIC\" peak set, by requiring that their apices fall within <italic>sDist </italic>of the corresponding EIC peak in <italic>S </italic>and within <italic>pClose </italic>of the inferred TIC peak in <italic>R</italic>. Additionally, the Pearson correlation of two numeric vectors of <italic>mz</italic>-ordered ion intensities, <italic>i.e.</italic>, spectra for the <italic>S </italic>TIC peak and the <italic>R </italic>inferred TIC peak, whose location is taken as the median <italic>RT </italic>of the grouped EIC peaks, must be greater than <italic>corMass</italic>.</p>", "<title>Iterative peak matching</title>", "<p>Once the peak sets for <italic>S </italic>and <italic>R </italic>are determined, peak matching can be initiated. Figure ##FIG##2##3## illustrates the overall process by which all <italic>S </italic>peaks are <italic>solved</italic>, <italic>i.e.</italic>, matched to an <italic>R </italic>peak or determined to have no match. The basic unit for iterative peak matching is a <italic>block </italic>which is composed of adjacent, unsolved peaks in <italic>S </italic>(or a single one); blocks are bound by either solved peaks or ends of the chromatogram. Initially, in the first iteration, the entire <italic>S </italic>peak set is one block. Each iteration identifies those peaks which remain unmatched, organizes them into new blocks, identifies new <italic>S-to-R </italic>peak matches within blocks, and discovers an <italic>RT </italic>shift value for each match that optimizes the alignment of subsequent peaks in its block. Besides recording the match, the method records the iteration number when a match was made, EIC <italic>mz </italic>information, and, importantly, the retention discrepancy, <italic>i.e.</italic>, the nearest integral number of MS scans by which the sample peak will need to be shifted in the final warp-and-shift alignment step to align it with the corresponding reference peak. After the final iteration, no peak in <italic>SP </italic>remains unsolved, <italic>i.e.</italic>, all are either matched with a reference peak, or evaluated as being unmatchable. In any iteration, if a peak is solved, its final location is fixed and no further adjustment will be made to it in later iterations.</p>", "<p>Retrieval of the best candidate peak from <italic>R </italic>against which to test the current <italic>S </italic>peak (Figure ##FIG##2##3##, upper right) is illustrated also in Figure ##FIG##1##2## (squares and triangles). The first step, comparing TIC peaks, ideally results in pairings of chemically identical chromatographic eluates. For this, one must exploit their underlying mass spectra. Spectra are treated as vectors of mass intensities and tested against each other by requiring that their Pearson correlation exceed a certain value. Once this criterion is met, the prominent EIC components of their spectra are tested to find the <italic>mz</italic>-matched EIC pair with the strongest <italic>SN</italic>. These are the \"model\" EIC peaks for that TIC peak pair, and their peak retention times are used instead of TIC retentions for more precise shifting.</p>", "<p>The peak matching method produces a set of matched results. A peak match is represented by a list containing an <italic>S </italic>EIC peak, the matching <italic>R </italic>EIC peak, the <italic>mz</italic>, the shift amount, and the final iteration number, <italic>e.g.</italic>,</p>", "<p></p>", "<p>where ϕ means there is no matching peak in <italic>R</italic>. Peaks are processed one-by-one according to their elution times. The current peak is matched only when (i) a candidate peak in <italic>R </italic>is within <italic>sDist </italic>of it, (ii) the Pearson correlation of the two peaks' mass spectra is greater than <italic>corMass</italic>, and (iii) the profile value of remaining peak deviations is greater than <italic>prof</italic>.</p>", "<p>A profile value is determined as follows: for a block of peaks <italic>SB </italic>= {b₁, b₂,..., b_{<italic>l</italic>}}, where b₁is the current working peak and <italic>l </italic>does not exceed the initial size of <italic>SP</italic>, initial deviations of peaks from their candidate matches can be represented by the vector <italic>ID </italic>= {id₁, id₂,..., id_{<italic>m</italic>}}, where <italic>m </italic>≤ <italic>l</italic>. The alignment that would perfectly align b₁is simulated by shifting all the peaks in <italic>SB </italic>by the integer-rounded value of id₁, resulting in a vector of deviations after the simulation, <italic>SD </italic>= {sd₁, sd₂,...}; note that sd₁is less than |0.5|. Next, we will have an evaluation vector E = {|id₁| - |sd₁|, |id₂| - |sd₂|,...} where absolute values of the simulated deviations are subtracted from absolute values of the initial deviations. A positive value within <italic>E </italic>means that its corresponding peak in <italic>S </italic>is brought closer as a result of the simulation. The, <italic>profile value </italic>is defined as the ratio of positive values to the total number of values in <italic>E</italic>. A profile value of 0.5 would mean that, if all peaks in a block were shifted by the initially recorded deviation of the current peak from its candidate peak in <italic>R</italic>, then half of the remaining traceable peaks, including the current one, are also improved in alignment. Only if the above three conditions (i, ii and iii) are met will the current peak be recorded as a match. Otherwise, it remains unsolved so that it can be processed again in later iterations with smaller block sizes. For the last peak in <italic>SB</italic>, however, there is only one element in the <italic>E </italic>vector, the current peak itself, so the profile value is always 1 and thus uninformative. In such a case, we model the deviations of already matched peaks by loess regression and use the model to predict the deviation of the current peak. If the actual deviation falls within <italic>lpBound </italic>of the prediction, then the candidate matching peak in <italic>R </italic>is accepted as a match.</p>", "<p>In such a case of a single or the last peak in a block, block processing will always solve the peak, either as a match or as unmatchable, signified by ϕ. After the last peak in <italic>SB </italic>is solved and no more blocks remain to be processed, the iteration number is incremented, unsolved peaks are grouped into new blocks, and the match process continues until there are no unsolved peaks. The final size of the time axis is actually determined by the result of the first iteration during which all peaks are in one block. Peak matching simulations in subsequent iterations can affect the time domain only within the boundary of the peaks within blocks. Iterative peak matching is described in Figure ##FIG##3##4## and Table ##TAB##1##2##. Figure ##FIG##3##4## illustrates peak matching in a 700-scan region containing seven peaks (numbered 1–7) used for alignment testing. After three iterations, all peaks were solved, i.e., matched or unmatched. Four boxes show peak blocks created at the start of an iteration. The <italic>y </italic>values 1, 2 and 3 shown on an axis on the right side of the figure indicate within which iteration corresponding peak blocks were processed. Table ##TAB##1##2## shows shift amounts applied to the peaks in a block at the end of each iteration in the example of Figure ##FIG##3##4##. The processing of peaks in a block proceeds from left to right. A parenthesized shift number implies matching and ϕ means that a peak was determined to have no match. When a match occurs, that shift amount is propagated to the subsequent peaks in the same block. For instance, peak #5 was shifted 11 scans when peak 1 was matched, and an additional 4 scans for a match of peak #3. Since it was not matched in the first iteration, peak matching continues in the next iteration. When peak 4 was matched, a shift of -2 was propagated to peak 5 and, with the resulting total shift of 13, peak 5 was determined by criteria described in Figure ##FIG##2##3## to be matched, thus ending its processing.</p>", "<title>Testing</title>", "<p>Data in 45 files were used to test the alignment algorithm. They were acquired by a quadrupole GC-EI-MS system during a month-long study of the effect of life-span-altering mutations on metabolite levels in the soil nematode <italic>C. elegans</italic>. Unless specified otherwise, run #8 was selected as the reference and the rest of the runs were aligned to it in succession. Figure ##FIG##4##5## shows TIC for all samples, viewed from above with total ion intensities color-encoded, before and after alignment.</p>", "<p>As shown in Figure ##FIG##5##6##, iterative block-shifting identified peak deviations for all runs and aligned them appropriately, thus, drastically decreasing peak deviations to no more that 1 scan, or an average deviation of 0.25 scans. Since a scan lasts 0.78 seconds, this is a mean deviation of 0.2 sec and a maximum deviation of less than one second for runs lasting over an hour. This is a great improvement over the initial deviations (as much as 22 scans or 17.2 seconds) and was achieved with conservation of the shapes and areas of key peaks, because only non-peak regions are warped.</p>", "<p>The small remaining deviation results from the discrete nature of the chromatography time dimension.</p>", "<p>When the alignment was repeated on the same data but with different references, results were similar. Not only were similarly aligned chromatograms produced (see Additional Files ##SUPPL##0##1##, ##SUPPL##1##2## and ##SUPPL##2##3##), but similar progress was made in correcting deviations and solving unsolved peaks as iterations progressed (Figure ##FIG##6##7##). No matter which reference was used, most deviations were corrected in early iterations.</p>", "<title>Comparisons</title>", "<p>Two well known algorithms, COW and XCMS, mentioned earlier in this paper, were selected to further evaluate the performance of our block-shift method with respect to the correctness of the alignment and the preservation of peak areas, using the test data set. A full evaluation of the performance of the three methods under more diverse conditions could be the subject of a separate study, however, to our knowledge, even the limited comparison reported here between COW and XCMS is unprecedented. COW is available as a set of MATLAB scripts [##UREF##6##26##]; XCMS as an R package [##REF##16448051##8##]. As in block shifting, analytical run #8 was chosen as the reference for COW. XCMS does not require the choice of a reference, relying instead on median positions identified, well-behaved <italic>peak-groups </italic>[##REF##16448051##8##]. Four major TIC peaks were selected for these comparisons: one in the beginning; one near the end; and two from the middle of the time interval of chromatography. For each, the most prominent spectral <italic>mz </italic>value was identified, and its EIC chromatogram along the full extent of the chromatogram was used as the input for COW alignment. Both XMCS and block-shifting used as their input the entire set of EIC mass chromatograms for every run.</p>", "<p>The quality of alignment by these three approaches is compared in Figure ##FIG##7##8##. Table ##TAB##2##3## summarized their effects, if any, on peak integrated areas, this calculated by a method that considers area between the apex and a horizontal line drawn at 1/5^ththe height of the apex. Looking at Figure ##FIG##7##8##, Peak #1 appears to have been least precisely aligned by XCMS, peak #4 by COW. For COW and XCMS, the less symmetric peaks #2 and #3 appear to show some dependence of apical position on the height of peaks, a phenomenon not evident with block-shifting. Table ##TAB##2##3## illustrates that COW, and to a lesser extent, XCMS alignments are accompanied by artifactual distortions in peak area. We also observed peak <italic>shape </italic>differences (data not shown). As for the block-shift method, areas of two of the four peaks were perfectly preserved. Two and 13 of 45 analytical runs did show area distortion for peaks #2 and #3, respectively. This can be attributed to the inclusion of a peak tail region during integration which was excluded from the peak region during block-shift alignment, thus, was liable to be warped.</p>" ]

[ "<title>Discussion</title>", "<p>Robust alignment is an important step as it affects not only the quality of comparative post-data analysis but also which type of data analysis can be used [##REF##17466315##9##]. Our iterative block-shifting approach is well suited to subsequent data analysis methods that operate on matrices, because the discrete nature of the time axis is preserved, and should allow approaches that require trilinearity because resulting alignments are precise to within one scan unit. Additionally, it preserves areas and shapes of detected peaks.</p>", "<p>Precise alignment is possible through the recurrent use of mass spectral information in both peak detection and peak matching steps. Some alignment errors may not be prevented by spectral considerations, however, for instance, errors that might occur when multiple isobaric compounds are retained differently during chromatography. There is an additional requirement for peak matching that the match not adversely affect the alignment of too many of the remaining peaks in its block (as set by the <italic>prof </italic>parameter). The effect of the <italic>prof </italic>criterion is to delay the matching of potentially troublesome peaks such as isobaric compounds, ultimately until they exist alone in a block, at which time, the desireability of using them for alignment is evaluated by a loess-based smoothing criterion. This method potentially can calibrate even heavily misaligned peaks since peaks are found in an adjustable search range; we know of no other alignment algorithm for which the deviation in retention time from sample to sample can exceed the time between a peak and its neighbors [##REF##16448051##8##,##REF##14719890##18##].</p>", "<p>One drawback of iterative block shifting is that, while its final step of warping and shifting conserves detected peaks, undetected peaks are liable to be deformed since nonpeak regions are warped. For this reason, the stringency during detection of sample TIC peaks is kept very low to try to detect, and thus preserve, most or all peaks of experimental interest. In cases where much of the retention artifact occurs at the beginning of the chromatogram, warping artifacts will be minimal, since the leftmost correction is a simple block-shift. Finally, if an undetected, and thus, potentially distorted peak is detected by some other means subsequent to alignment and proves important in the experiment, an investigator can always recover true peak area and shape by referring to the original raw data file using adjacent matched and aligned peaks to help locate the feature of interest. Because most peaks are area-preserved by this method, it is expected that fewer instances will occur than with methods that generally distort area that will require a return to the raw data for quantification purposes.</p>", "<p>One disadvantage of typical pairwise alignment approaches is that the selection of the reference chromatogram can affect performance [##REF##17466315##9##,##UREF##6##26##]. No sample chromatogram is likely to include every peak from all the other chromatograms in a series. Our proposed method, while not free from this disadvantage, lessens the difficulty of selecting a good reference by using subsets of available peaks in the reference for the alignment of every other sample.</p>" ]

[]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Metabolomics, petroleum and biodiesel chemistry, biomarker discovery, and other fields which rely on high-resolution profiling of complex chemical mixtures generate datasets which contain millions of detector intensity readings, each uniquely addressed along dimensions of <italic>time </italic>(<italic>e.g.</italic>, <italic>retention time </italic>of chemicals on a chromatographic column), a <italic>spectral value </italic>(<italic>e.g., mass-to-charge ratio </italic>of ions derived from chemicals), and the <italic>analytical run number</italic>. They also must rely on data preprocessing techniques. In particular, inter-run variance in the retention time of chemical species poses a significant hurdle that must be cleared before feature extraction, data reduction, and knowledge discovery can ensue. <italic>Alignment methods</italic>, for calibrating retention reportedly (and in our experience) can misalign matching chemicals, falsely align distinct ones, be unduly sensitive to chosen values of input parameters, and result in distortions of peak shape and area.</p>", "<title>Results</title>", "<p>We present an iterative block-shifting approach for retention-time calibration that detects chromatographic features and qualifies them by retention time, spectrum, and the effect of their inclusion on the quality of alignment itself. Mass chromatograms are aligned pairwise to one selected as a reference. In tests using a 45-run GC-MS experiment, block-shifting reduced the absolute deviation of retention by greater than 30-fold. It compared favourably to COW and XCMS with respect to alignment, and was markedly superior in preservation of peak area.</p>", "<title>Conclusion</title>", "<p>Iterative block-shifting is an attractive method to align GC-MS mass chromatograms that is also generalizable to other two-dimensional techniques such as HPLC-MS.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>MC conceived, developed the algorithm, and drafted the manuscript. RJSR and JT coordinated the project and revised the manuscript. All authors read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>We wish to acknowledge Lulu Xu for conducting the GC-MS experiment and Stephen Jennings for helpful discussions on this study. This project was supported by NIH Grant Number P20 RR-16460 from the IDeA Networks of Biomedical Research Excellence (INBRE) program of the National Center for Research Resources and NIA program project grant #AG020641.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Unaligned chromatograms</bold>. A 500-scan region is shown for each of two total-ion-current chromatograms, <italic>S </italic>and <italic>R</italic>, in a GC-MS experiment.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Diagram of two peak sets</bold>. Peaks in a sample peak set <italic>SP </italic>match with peaks in a reference peak set <italic>RP</italic>. <italic>SP </italic>is composed of detected TIC peaks, <italic>i.e</italic>, capital <italic>P</italic>'s, in which individual EIC peaks, <italic>p</italic>'s, up to five, are arranged in descending order of their signal-to-noise ratios. Note that <italic>RP </italic>is a peak set composed of \"inferred TIC\" peaks since individual EIC peaks, <italic>q</italic>'s, are first identified by using the <italic>mz </italic>values of <italic>SP </italic>and then are grouped into a TIC peak. This also illustrates the matching of TIC peak <italic>P</italic>₁to <italic>Q</italic>₁, but of P₂to <italic>Q</italic>₃because, in the latter case, either peak <italic>Q</italic>₂had no EIC component (<italic>q</italic>₂₁<italic>and q</italic>₂₂) with a matching <italic>mz </italic>value, or the spectra of <italic>P</italic>₂and <italic>Q</italic>₂were insufficiently correlated, whereas Q₃met both of these conditions, with <italic>q</italic>₃₂having matching <italic>mz</italic>.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>A flowchart of iterative peak matching</bold>. The left flowchart shows the overall iterative peak-matching flow, whereas, the right flowchart shows the flow within the subroutine for processing a single block.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>An example of iterative peak matching</bold>. A 700-scan region of <italic>S </italic>TIC is shown which has 7 detected peaks. Peaks are assigned to blocks at the start of each iteration, with blocks shown as boxes of height matching the iteration number. Intensities are log transformed for a better display of weak signals.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>Top plots showing all 45 runs, before and after alignment</bold>. These two heat-map-encoded top plots display the total ion current (TIC) for mass chromatograms of all 45 runs in a <italic>C</italic>. <italic>elegans </italic>experiment (see text), before and after alignment. Run #8 was used as the alignment reference. The brightness is proportional to the logarithm of intensity, so peaks are displayed as bright vertical bars. Initially, as the run number increases, the peaks are skewed to left, meaning the same peaks eluted earlier in higher-numbered (later) runs. The pattern also exhibits serious breaks and other nonlinearities. These imperfections were corrected by the alignment method and are not evident in the bottom image.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Peak deviations before and after alignment</bold>. Retention-time deviations of matched peaks are color-coded in these two panels, before (top) and after (bottom) application of the described alignment method. The heat-map code is displayed to the right (Note the narrower range of deviations represented by colors in the lower panel). White cells represent instances where a peak in a run either was not detected or did not pass signal-to-ratio and peak-width criteria. Twenty-one peaks were omitted from the display because they met criteria in fewer than ten sample runs; their inclusion does not alter the result. Run #8 again was selected as the alignment reference. Several weeks, and a GC-MS re-tuning operation, occurred between runs 17 and 18.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p><bold>Robust iterative peak matching with different references</bold>. (a) Sums of absolute values of all sample-<italic>vs.</italic>-reference peak discrepancies are normalized to the pre-alignment (iter = 0) value and plotted as a function of the number of iterations completed. Data are compared for four independent alignments, with different runs selected as the reference. Regardless of which run was used as the reference, most peak retentions were corrected in early iterations (in the Ref #8 case, all peaks were solved by iteration 5 and no 6^thiteration was done). Comparing the same references, the number remaining unsolved after each iteration is shown in (b).</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p><bold>A comparison of retention-time alignment by three methods</bold>. Top panel: Unabridged total-ion-current (TIC) chromatograms for 45 analytical runs in a GC-MS metabolomics experiment, prior to alignment. Remaining panels: columns 1–4 show details for peaks labelled 1–4 in the top panel, both unaligned (top row), and aligned using COW [##UREF##2##17##] with automated parameter selection [##UREF##6##26##], using XCMS with three iterations [##REF##16448051##8##], and using iterative block-shifting with its default parameters, as described in the text (rows 2–4, respectively).</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Algorithm input variables</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Variable</bold></td><td align=\"center\"><bold>Formal Definition</bold></td><td align=\"center\"><bold>Default</bold></td></tr></thead><tbody><tr><td align=\"left\"><italic>Sntic</italic></td><td align=\"left\"><italic>SN </italic>threshold when detecting peaks in <italic>S </italic>TIC. Low values are used in order to include weak signals.</td><td align=\"center\">1</td></tr><tr><td align=\"left\"><italic>Sneic</italic></td><td align=\"left\"><italic>SN </italic>threshold when detecting peaks in EIC chromatograms for <italic>S </italic>and <italic>R</italic>. Values higher than the <italic>SNtic </italic>are used to reduce the risk of matching noisy EIC peaks.</td><td align=\"center\">5</td></tr><tr><td align=\"left\"><italic>PWidth</italic></td><td align=\"left\">Peak width threshold for every peak detection, in units of scans.</td><td align=\"center\">12</td></tr><tr><td align=\"left\"><italic>PClose</italic></td><td align=\"left\">EIC-to-TIC peak apex distance threshold, in units of scans.</td><td align=\"center\">2</td></tr><tr><td align=\"left\"><italic>SDist</italic></td><td align=\"left\">Search distance when finding candidate peaks in <italic>R</italic>, in units of scan number measuring from the apex of an <italic>S </italic>peak.</td><td align=\"center\">15</td></tr><tr><td align=\"left\"><italic>CorMass</italic></td><td align=\"left\">Correlation coefficient threshold between two peaks.</td><td align=\"center\">0.95</td></tr><tr><td align=\"left\"><italic>Prof</italic></td><td align=\"left\">Profile threshold for peak deviations.</td><td align=\"center\">0.5</td></tr><tr><td align=\"left\"><italic>LpBound</italic></td><td align=\"left\">Lone peak boundary, in units of scan number.</td><td align=\"center\">5</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Record of shifts during the peak matching stage in Figure 4.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\">peak</td><td align=\"center\" colspan=\"4\">shift (in scan)</td></tr><tr><td/><td align=\"center\"><bold>i = 1</bold></td><td align=\"center\"><bold>i = 2</bold></td><td align=\"center\"><bold>i = 3</bold></td><td align=\"center\"><bold>Total</bold></td></tr></thead><tbody><tr><td align=\"center\">1</td><td align=\"center\">(11)</td><td align=\"center\">-</td><td align=\"center\">-</td><td align=\"center\">11</td></tr><tr><td align=\"center\">2</td><td align=\"center\">11</td><td align=\"center\">ϕ</td><td align=\"center\">-</td><td align=\"center\">11</td></tr><tr><td align=\"center\">3</td><td align=\"center\">(15)</td><td align=\"center\">-</td><td align=\"center\">-</td><td align=\"center\">15</td></tr><tr><td align=\"center\">4</td><td align=\"center\">15</td><td align=\"center\">(-2)</td><td align=\"center\">-</td><td align=\"center\">13</td></tr><tr><td align=\"center\">5</td><td align=\"center\">15</td><td align=\"center\">(-2)</td><td align=\"center\">-</td><td align=\"center\">13</td></tr><tr><td align=\"center\">6</td><td align=\"center\">15</td><td align=\"center\">-2</td><td align=\"center\">(-1)</td><td align=\"center\">12</td></tr><tr><td align=\"center\">7</td><td align=\"center\">15</td><td align=\"center\">-2</td><td align=\"center\">(-1)</td><td align=\"center\">12</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Peak integration errors* caused by three alignment methods</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>1</bold></td><td align=\"center\"><bold>2</bold></td><td align=\"center\"><bold>3</bold></td><td align=\"center\"><bold>4</bold></td></tr></thead><tbody><tr><td align=\"left\"><bold>COW area %error ± SD</bold></td><td align=\"center\">8.7 ± 5.2</td><td align=\"center\">4.7 ± 3.8</td><td align=\"center\">3.0 ± 2.4</td><td align=\"center\">4.5 ± 3.2</td></tr><tr><td align=\"left\"><bold>XCMS area %error ± SD</bold></td><td align=\"center\">0.17 ± 00.14</td><td align=\"center\">1.29 ± 0.91</td><td align=\"center\">0.50 ± 0.89</td><td align=\"center\">0.11 ± 0.10</td></tr><tr><td align=\"left\"><bold>Block-shift area %error ± SD</bold></td><td align=\"center\">0.000 ± 0.00</td><td align=\"center\">0.002 ± 0.01</td><td align=\"center\">0.18 ± 0.80</td><td align=\"center\">0.000 ± 0.00</td></tr><tr><td colspan=\"5\"><hr/></td></tr><tr><td align=\"left\"><bold>Block <italic>vs</italic>. COW (<italic>t</italic>-test P val.)</bold></td><td align=\"center\">&lt;10^-10</td><td align=\"center\">&lt;10^-10</td><td align=\"center\">&lt;10^-10</td><td align=\"center\">&lt;10^-10</td></tr><tr><td align=\"left\"><bold>Block <italic>vs</italic>. XCMS (<italic>t</italic>-test P val.</bold><bold>)</bold></td><td align=\"center\">&lt;10^-10</td><td align=\"center\">&lt;10^-10</td><td align=\"center\">0.08</td><td align=\"center\">&lt;10^-10</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label>{(p₁₁, q₁₁, 30, 5, 1), (p₂₁, q₃₂, 40, 3, 1), (p₃₁, ϕ, 40, 3, 2),...}</disp-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p>Alignment result if run 1 is selected as the Reference.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p>Alignment result if run 22 is selected as the Reference.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p>Alignment result if run 36 is selected as the Reference.</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>*area %error = 100% × (area_aligned- area_raw)/area_raw</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S15-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S15-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S15-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S15-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S15-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S15-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S15-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S15-8\"/>" ]

[ "<media xlink:href=\"1471-2105-9-S9-S15-S1.jpg\" mimetype=\"image\" mime-subtype=\"jpeg\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-S9-S15-S2.jpg\" mimetype=\"image\" mime-subtype=\"jpeg\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-S9-S15-S3.jpg\" mimetype=\"image\" mime-subtype=\"jpeg\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Often accompanying the macromolecules deposited in the Protein Data Bank (PDB) [##REF##10592235##1##] are smaller molecules of biological importance. Some of these are energy-carrying cofactors, such as ATP, coenzyme A, and nicotinamide-adenine dinucleotide (NAD). Some analogs of these molecules are either drugs or can be used in drug design [##REF##11587640##2##,##REF##11405646##3##].</p>", "<p>Like other biologically relevant molecules, many of these small molecules contain chiral or prochiral centers. An atom is a chiral center if four different chemical groups are attached to it. A chiral configuration can be designated R or S, depending on the arrangement of the attached groups (Figure ##FIG##0##1##). If, however, two of these groups are identical, then the center atom is prochiral, meaning that it would become chiral if either of the identical groups were substituted for a unique group. These two groups are called diastereotopic, i.e., if either were replaced with a unique group, the molecule would become one or another diastereomer. Within a pair of diastereotopic atoms, one is designated <italic>pro</italic>-R and the other <italic>pro</italic>-S, indicating the configuration of the chiral atom would result from replacing the diastereotopic atom with a group that has higher priority than the other groups. Many ligands contain diphosphate groups that contain at least one prochiral phosphorus atom (Figure ##FIG##1##2##).</p>", "<p>The <italic>pro</italic>-S and <italic>pro</italic>-R oxygen atoms of nucleic acid strands are named \"OP1\" and \"OP2\", respectively [##UREF##0##4##]. Many enzymes treat the <italic>pro</italic>-R and <italic>pro</italic>-S oxygen atoms of DNA and RNA differently [##REF##2411211##5##]. These diastereotopic oxygen atoms are also treated differently in RNA-intron splicing [##REF##1381347##6##,##REF##7527587##7##]. Small diphosphate-containing molecules also participate in enzymatic reactions in which the distinction between diastereotopic atoms or groups is important [##REF##2411211##5##,##REF##2988606##8##,##REF##34427##9##]. Unfortunately, many of these diastereotopic atoms do not have standardized names, an issue that has not been investigated to our knowledge. Consistent naming of diastereotopic atoms is needful when performing all-atom superpositioning or all-atom root mean square deviation (RMSD) calculation [##REF##9192890##10##]. It is also needful for data mining in the PDB, e.g., structure-based virtual screening for drug candidates [##REF##18342797##11##,##REF##11560062##12##]. In this paper, we will conduct a systematic PDB-wide analysis on the diastereotopic atom names of small molecules containing diphosphate.</p>" ]

[ "<title>Methods</title>", "<title>Selection of small molecules for analysis</title>", "<p>PDB files were selected from the January 7, 2008 \"snapshot\" of the Protein Data Bank. The search feature of the Protein Data Bank website <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.pdb.org/pdb/search/advSearch.do\"/> was used to select PDB codes for files containing ligands that had substructures matching the SMILES pattern \"C~O~P(~O)(~O)~O~P(~O)(~O)~O\". Here, \"C\" represents a carbon atom, \"~\" represents any bond, \"O\" represents oxygen, \"P\" represents phosphorus, and the parentheses indicate that the oxygen atoms inside them are bonded to the preceding phosphorus atom in the list, not to subsequent atoms in the list. This matches any ligand containing a (PO₄)₂moiety, such as NAD, ATP, and Coenzyme A, resulting in a list of 4435 PDB codes.</p>", "<p>Since the PDB files corresponding to these codes also included other ligands not meeting our criteria, we analyzed each of the small molecules within each PDB file and selected each one that met the following criteria: (1) It did not have the same residue name as an amino acid or nucleic acid, including names mapped to standard residue names via the \"MODRES\" record. (2) It had an entry in the Chemical Component Dictionary <ext-link ext-link-type=\"uri\" xlink:href=\"http://deposit.rcsb.org/het_dictionary.txt\"/> from the PDB. (3) It had complete coordinates for the non-hydrogen atoms specified in the Chemical Component Dictionary. And (4), it had a diphosphate group attached to carbon, with the diphosphate group consisting of two phosphorus atoms, each covalently bonded to four oxygen atoms. We chose to analyze the prochiral phosphate centers adjacent to carbon atoms because of their abundance and because it allowed a simple and direct application of the CIP algorithm.</p>", "<p>Atoms were considered to be covalently bonded if the distance between their centers was less than the sum of their covalent radii plus a cushion of 0.4 Å, following the custom of the Cambridge Structural Database (CSD) [##REF##12037359##19##]. Covalent radii were obtained from the CSD website <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ccdc.cam.ac.uk/products/csd/radii/\"/>.</p>", "<p>Also excluded were molecules that had alternate conformations that shared the same residue number. This guaranteed that any modeled alternate conformations would contain complete molecules. Those files containing diphosphates were further checked for phosphorus atoms having a prochiral configuration (see Determination of Prochiral Centers below). For those that did, the names of all four atoms attached to the prochiral center were recorded along with their relative stereochemical positions. Of the 4435 files originally selected, 4184 were found to have at least one ligand with a prochiral phosphate atom.</p>", "<title>Determination of prochiral centers</title>", "<p>The CIP algorithm [##UREF##1##20##,##UREF##2##21##] for assigning priorities to atoms within a molecule was implemented using in-house software. CIP priorities were calculated for all four atoms connected to a phosphorus atom. Following the CIP-algorithm, the oxygen atom attached to two phosphorus atoms always had the highest priority and the oxygen atom attached to carbon always had the second highest priority. The two remaining oxygen atoms were not bonded to any other atom besides the phosphorus atom.</p>", "<p>If each atom had a distinct priority, then the phosphorus is chiral and the determinant algorithm of Cieplak and Wisniewski[##UREF##3##22##] could be used to calculate whether the configuration is R or S as shown in Equation (1):</p>", "<p></p>", "<p>X_N, Y_N, and Z_Nare the x, y, and z components of the coordinates for group N. The subscripted letters A, B, C, and D represent the highest, second highest, third highest, and lowest priority atoms, respectively (see Figure ##FIG##0##1##). <italic>m </italic>is the result of calculating the determinant. It is negative for the R configuration and positive for the S configuration. If it is evaluated to be zero, then the atoms are all in the same plane [##UREF##3##22##], which should never be the case for tetrahedrally arranged molecules such as phosphates. For understanding the mathematics behind this equation and how it captures the handedness of four three-dimensional coordinates, we refer the reader to the work of Cieplak and Wisniewski [##UREF##3##22##].</p>", "<p>If two of the atoms attached to the phosphorus atom have identical priorities, then they are diastereotopic and the phosphorus is prochiral. In the case of diphosphate-containing molecules, the diastereotopic atoms are only bonded to phosphorus and therefore have the lowest priority (see Figure ##FIG##1##2##). We will call the atoms attached to the phosphorus atom A, B, C, and C', where A and B have the highest and second highest priority, respectively, while C and C' tie for the lowest priority. In this case, Equation (1) can be adapted to determine whether C is the <italic>pro</italic>-S or <italic>pro</italic>-R atom and, concomitantly, whether C' is the <italic>pro</italic>-R or <italic>pro</italic>-S atom. By definition, a diastereotopic atom being <italic>pro</italic>-S (or <italic>pro</italic>-R) means that, if it were replaced by a group with higher priority than any other substituent, then the prochiral center would become chiral with an S (or R) configuration. Therefore, we treat C as if it had the highest priority and then calculate the resulting configuration. If the calculated configuration is S, then C is <italic>pro</italic>-S; if it is R, then C is <italic>pro</italic>-R. To do this computationally, we artificially raise the priority of C to be the highest (i.e. higher than A) changing Equation (1) to the following:</p>", "<p></p>", "<p>If m is positive, then C is the <italic>pro</italic>-S atom and, concomitantly, C' is the <italic>pro</italic>-R atom (Figure ##FIG##1##2##). If m is negative, then C is the <italic>pro</italic>-R atom and C' is the <italic>pro</italic>-S atom.</p>", "<title>Third-party software used</title>", "<p>COOT [##REF##15572765##23##] was used for visualizing PDB files, which was especially useful during the development of our software. As needed, the SSM [##REF##15572779##24##] module of COOT was also used for superposition of molecules. Pymol was used for viewing NMR models as well as generating depictions of molecular structures for figures [##UREF##4##25##].</p>" ]

[ "<title>Results</title>", "<title>Inconsistencies in PDB files</title>", "<p>There were 4167 PDB files containing a total of 295 distinct ligands having prochiral centers that met our strict criteria. Over half of these ligands (175) had two prochiral phosphate centers that were adjacent to carbon, and one had three (OXT from [PDB:<ext-link ext-link-type=\"pdb\" xlink:href=\"2JI7\">2JI7</ext-link>] [##REF##17637344##13##]), for a total of 472 distinct prochiral centers adjacent to carbon. For example, NAD contains two because it has a diphosphate sandwiched between two ribose moieties. Each distinct prochiral center contains a pair of disastereotopic atoms. We analyzed the names of the atoms at each prochiral center. Of these distinct centers, 354 had a single naming convention but 241 of these also only occurred in a single PDB file. There were 118 distinct prochirality centers that had more than one naming convention.</p>", "<p>We defined a case of swapped names to occur when all of the following were true between two molecules with the same type of prochiral center: (1) the highest and second highest priority names were consistent, (2) the <italic>pro</italic>-R atom of one prochiral center had the same name as the <italic>pro</italic>-S atom of a second center, and (3) the <italic>pro</italic>-S atom of the first center had the same name as the <italic>pro</italic>-R atom of the second center (Figure ##FIG##1##2##). 117 of the 118 centers had swapped naming conventions as defined above. The remaining center, which had two naming conventions, actually had a naming error. Nine of the 117 centers with swapped names had additional naming conventions. In every case, we found that the extra naming conventions were caused by errors rather than mere inconsistencies. For example, in a structure of a surfactin synthetase-activating enzyme [PDB:<ext-link ext-link-type=\"pdb\" xlink:href=\"1QR0\">1QR0</ext-link>] [##REF##10581256##14##], the diastereotopic atoms attached to phosphorus atom P1A are labeled \"O5A\" and \"O4A\" instead of the names \"O2A\" and \"O1A\" defined in the Chemical Component Dictionary <ext-link ext-link-type=\"uri\" xlink:href=\"http://deposit.rcsb.org/het_dictionary.txt\"/> from the PDB. In a similar manner, the diastereotopic atoms attached to P2A are named \"O2A\" and \"O1A\", instead of the names \"O5A\" and \"O4A\" defined in the dictionary file. In another example, in a structure of <italic>E. coli </italic>carbamoyl phosphate synthetase [PDB:<ext-link ext-link-type=\"pdb\" xlink:href=\"1CE8\">1CE8</ext-link>] [##REF##10428826##15##] the O5' oxygen atom is mislabeled as O4' for 8 different ADP molecules. Interestingly, in four of these molecules, the <italic>pro</italic>-S and <italic>pro</italic>-R atoms are labeled \"O1A\" and \"O2A\", respectively, while in the other four molecules they are labeled \"O2A\" and \"O1A\", respectively.</p>", "<p>In Table ##TAB##0##1##, we present statistics for sample cases in which there were at least two nonredundant examples of each naming convention. For additional selected examples, see Supplement Table 1 in Additional File ##SUPPL##0##1##. For our full results, including cases that had no inconsistencies, see Supplemental Table 2 in Additional File ##SUPPL##1##2## (explanation in Additional file ##SUPPL##2##3##). All results, including those resulting from errors, are included in Supplemental Table 2. However, we emphasize that the bulk of the results are due to inconsistencies, not errors.</p>", "<title>Examples of naming inconsistencies</title>", "<p>Most of the atom naming inconsistencies mentioned in this paper relate to differences found between different files. However, there are a few cases in which naming inconsistencies can be found within a single file. One example is an X-ray crystal structure of alcohol dehydrogenase [PDB:<ext-link ext-link-type=\"pdb\" xlink:href=\"2OHX\">2OHX</ext-link>] [##REF##15299346##16##]. This structure contains two NAD molecules (see Figure ##FIG##1##2##). The prochiral center around phosphorus atom PN has consistent naming between the two molecules, however the prochiral center around phosphorus atom PA does not. In one case the <italic>pro</italic>-S and <italic>pro</italic>-R atoms are named \"O1A\" and \"O2A\", respectively, and in the other case, the names are \"O2A\" and \"O1A\", respectively.</p>", "<p>Another example is an NMR structure of bovine acyl-coenzyme A binding protein (Figure ##FIG##2##3##) [PDB:<ext-link ext-link-type=\"pdb\" xlink:href=\"1NVL\">1NVL</ext-link>]. This structure contained 20 NMR models, in which one phosphorus prochiral center was consistently named and the other was not. For the P1A center, models 1, 2, 5 and 18 have <italic>pro</italic>-S and <italic>pro</italic>-R atoms named \"O1A\" and \"O2A\", while the remaining 15 models have them named \"O2A\" and \"O1A\", respectively. Meanwhile, the <italic>pro</italic>-S and <italic>pro</italic>-R atoms at the P2A center are consistently named \"O5A\" and \"O4A\", respectively.</p>" ]

[ "<title>Discussion</title>", "<p>The inconsistent naming of atoms discussed in our paper is due largely to a lack of standardized names, <italic>not </italic>due to errors on the part of crystallographers or NMR researchers. There can be no errors where there are no rules.</p>", "<p>A study of NAD(P) molecules by Carugo and Argos ignored the diastereotopic oxygen atoms for purposes of superimposing molecules because of naming inconsistencies [##REF##9144787##17##]. Despite their use of atom-specific names for other atoms in the molecules, they only generally referred to diastereotopic oxygen atoms as \"terminal oxygen atoms\". That was eleven years ago and only involved a study of 32 protein structures. This was long before the recent remediation project of the PDB [##REF##18073189##18##]. This project has done well to bring molecular and atomic naming conventions for PDB files into conformity with standards established by the International Union of Pure and Applied Chemistry (IUPAC) and the International Union of Biochemistry and Molecular Biology (IUBMB). However, IUPAC and IUBMB do not have standards for most diastereotopic atoms of small molecules.</p>", "<p>There were no obvious overall trends in naming conventions with respect to the <italic>pro</italic>-R and <italic>pro</italic>-S atoms. This is likely due to the lack of naming standardization. However, trends are commonly seen among specific ligands (Table ##TAB##0##1##). One interesting observation is that the P prochiral center of FAD is highly biased in its naming convention (87% for one convention); however, the second center, PA has little bias (54% for one convention). Another observation is that NAD-like ligands tend to have naming conventions such that similar names (e.g. O1A and O1N) are seen on the same \"side\" of the molecule.</p>", "<p>We suggest a general rule that names for <italic>pro</italic>-S atoms come alphanumerically before names for <italic>pro</italic>-R atoms. This is similar to the standard of using \"OP1\" for <italic>pro</italic>-S and \"OP2\" for <italic>pro</italic>-R in nucleic acids. The data indicates that there is no strong bias for this nor for its opposite convention among diphosphate containing ligands.</p>", "<p>Regardless of what rules may become adopted, it is important to know to which atom a particular name refers. Establishing standard names and topologies that take prochirality into consideration will result in less confusion and more accuracy in studies involving small molecules. Until standards are adopted, individuals mining the data need to do their own standardization of the names. This naming can be enforced upfront, prior to the official release of data, or it can be enforced by individuals mining the data.</p>" ]

[ "<title>Conclusion</title>", "<p>Current naming conventions do not completely map unique names to unique diastereotopic atoms, resulting in possible confusion or error, or at least the need for researchers to impose their own naming standardization. We herein describe many cases of naming inconsistencies for small molecules containing diphosphate moieties. A future study will assess naming conventions of all atoms in the PDB, addressing more general issues of chirality and prochirality. The in-house software used in this study is available upon request.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Biological chemistry is very stereospecific. Nonetheless, the diastereotopic oxygen atoms of diphosphate-containing molecules in the Protein Data Bank (PDB) are often given names that do not uniquely distinguish them from each other due to the lack of standardization. This issue has largely not been addressed by the protein structure community.</p>", "<title>Results</title>", "<p>Of 472 diastereotopic atom pairs studied from the PDB, 118 were found to have names that are not uniquely assigned. Among the molecules identified with these inconsistencies were many cofactors of enzymatic processes such as mononucleotides (e.g. ADP, ATP, GTP), dinucleotide cofactors (e.g. FAD, NAD), and coenzyme A. There were no overall trends in naming conventions, though ligand-specific trends were prominent.</p>", "<title>Conclusion</title>", "<p>The lack of standardized naming conventions for diastereotopic atoms of small molecules has left the <italic>ad hoc </italic>names assigned to many of these atoms non-unique, which may create problems in data-mining of the PDB. We suggest a naming convention to resolve this issue. The in-house software used in this study is available upon request.</p>", "<p>A version of the software used for the analyses described in this paper is available at our web site: <ext-link ext-link-type=\"uri\" xlink:href=\"http://digbio.missouri.edu/ddan/DDAN.htm\"/>.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>CAB participated in the design of the study, developed the in-house software, carried out the atom name analysis, and drafted the manuscript. DX participated in the design and coordination of the study, and helped draft the manuscript. Both authors read and approved the final manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>CAB was supported by NIH Grant Number 2-T15-LM07089-16 from the National Library of Medicine. DX was supported by an NIH Grant (1R21GM078601-01).</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>S and R configurations for chiral centers</bold>. (a) S configuration and (b) R configuration, for atoms A, B, C, and D when they have the highest, second, third, and lowest priorities, respectively. Notice that when the three highest priority groups (A, B, and C) are facing the viewer, they have a counter-clockwise arrangement in the S configuration and a clockwise arrangement in the R configuration.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>NAD molecules from X-ray crystal structure </bold><ext-link ext-link-type=\"pdb\" xlink:href=\"2OHX\">2OHX</ext-link>. For comparison purposes, one molecule is superimposed on the other and then offset slightly. The atom names are similarly offset for the diastereotopic oxygen atoms of the adenine-side phosphate group. Note the differences in names for the <italic>pro</italic>-R and <italic>pro</italic>-S atoms. Both molecules of NAD shown are from an alcohol dehydrogenase structure [PDB:<ext-link ext-link-type=\"pdb\" xlink:href=\"2OHX\">2OHX</ext-link>][##REF##15299346##16##]. Following the CIP-algorithm, since all four oxygen atoms have the same atomic number, their priority is determined by subsequent bonded groups. The O3 oxygen atom is bonded to the next phosphorus atom and the O5' oxygen atom is bonded to the preceding C5' carbon atom, while the remaining two oxygen atoms are unbonded except to the original phosphorus atom. Therefore, the O3 oxygen atom has the highest priority, the O5' oxygen atom has the second highest priority, and the remaining two oxygen atoms tie for the lowest priority. The <italic>pro</italic>-S atom is the one that, if it were replaced with an atom of highest priority, would make the phosphorus atom chiral with an S configuration. Both molecules are drawn using red for oxygen, blue for nitrogen, and orange for phosphorus. One is drawn using light blue for carbon and the other is drawn using white for carbon.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Diphosphate of Coenzyme A from NMR structure </bold><ext-link ext-link-type=\"pdb\" xlink:href=\"1NVL\">1NVL</ext-link>. The diphosphate region of coenzyme A of two models from <ext-link ext-link-type=\"pdb\" xlink:href=\"1NVL\">1NVL</ext-link> is shown. The diphosphate region of model 4 (light coloring) is superimposed on the same diphosphate region of model 2 (standard coloring). The diastereotopic names of prochiral center P2A have consistent names (O4A and O5A), but the <italic>pro</italic>-S and <italic>pro</italic>-R names for prochiral center P1A are not (O1A and O2A, respectively, for model 2, and O2A and O1A, respectively, for model 4).</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Naming convention statistics for selected ligands</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>ligand code</bold></td><td align=\"left\"><bold>ligand name</bold></td><td align=\"left\"><bold>center atom</bold></td><td align=\"left\"><bold><italic>pro</italic>-S</bold></td><td align=\"left\"><bold><italic>pro</italic>-R</bold></td><td align=\"left\"><bold>#</bold></td><td align=\"left\"><bold>bias (%)</bold></td><td align=\"left\"><bold>example PDB</bold></td></tr></thead><tbody><tr><td align=\"left\">ACO</td><td align=\"left\">acetyl-coenzyme A</td><td align=\"left\">P1A</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">22</td><td align=\"left\">42%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1DM3\">1DM3</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">30</td><td align=\"left\">58%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1B87\">1B87</ext-link></td></tr><tr><td/><td/><td align=\"left\">P2A</td><td align=\"left\">O4A</td><td align=\"left\">O5A</td><td align=\"left\">25</td><td align=\"left\">48%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1B87\">1B87</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O5A</td><td align=\"left\">O4A</td><td align=\"left\">27</td><td align=\"left\">52%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1DM3\">1DM3</ext-link></td></tr><tr><td align=\"left\">ADP</td><td align=\"left\">adenosine-5'-diphosphate</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">211</td><td align=\"left\">33%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A6E\">1A6E</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">419</td><td align=\"left\">67%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"13PK\">13PK</ext-link></td></tr><tr><td align=\"left\">ATP</td><td align=\"left\">adenosine-5'-triphosphate</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">103</td><td align=\"left\">30%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1B0U\">1B0U</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">240</td><td align=\"left\">70%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A0I\">1A0I</ext-link></td></tr><tr><td align=\"left\">COA</td><td align=\"left\">coenzyme A</td><td align=\"left\">P1A</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">67</td><td align=\"left\">45%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1ACA\">1ACA</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">81</td><td align=\"left\">55%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1CM0\">1CM0</ext-link></td></tr><tr><td/><td/><td align=\"left\">P2A</td><td align=\"left\">O4A</td><td align=\"left\">O5A</td><td align=\"left\">67</td><td align=\"left\">46%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1ESM\">1ESM</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O5A</td><td align=\"left\">O4A</td><td align=\"left\">78</td><td align=\"left\">54%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1ACA\">1ACA</ext-link></td></tr><tr><td align=\"left\">CTP</td><td align=\"left\">cytidine-5'-triphosphate</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">20</td><td align=\"left\">49%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1GQ9\">1GQ9</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">21</td><td align=\"left\">51%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1COZ\">1COZ</ext-link></td></tr><tr><td align=\"left\">FAD</td><td align=\"left\">flavin-adenine dinucleotide</td><td align=\"left\">P</td><td align=\"left\">O1P</td><td align=\"left\">O2P</td><td align=\"left\">554</td><td align=\"left\">87%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A8P\">1A8P</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2P</td><td align=\"left\">O1P</td><td align=\"left\">81</td><td align=\"left\">13%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1B2R\">1B2R</ext-link></td></tr><tr><td/><td/><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">290</td><td align=\"left\">46%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1AHV\">1AHV</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">345</td><td align=\"left\">54%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A8P\">1A8P</ext-link></td></tr><tr><td align=\"left\">GTP</td><td align=\"left\">guanosine-5'-triphosphate</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">35</td><td align=\"left\">36%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1CKM\">1CKM</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">62</td><td align=\"left\">64%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A8R\">1A8R</ext-link></td></tr><tr><td align=\"left\">NAD</td><td align=\"left\">nicotinamide-adenine-dinucleotide</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">144</td><td align=\"left\">27%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A5Z\">1A5Z</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">388</td><td align=\"left\">73%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A4Z\">1A4Z</ext-link></td></tr><tr><td/><td/><td align=\"left\">PN</td><td align=\"left\">O1N</td><td align=\"left\">O2N</td><td align=\"left\">394</td><td align=\"left\">74%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A4Z\">1A4Z</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2N</td><td align=\"left\">O1N</td><td align=\"left\">135</td><td align=\"left\">26%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A7A\">1A7A</ext-link></td></tr><tr><td align=\"left\">NAP</td><td align=\"left\">nadp nicotinamide-adenine-dinucleotide phosphate</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">87</td><td align=\"left\">26%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1CIV\">1CIV</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">247</td><td align=\"left\">74%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A27\">1A27</ext-link></td></tr><tr><td/><td/><td align=\"left\">PN</td><td align=\"left\">O1N</td><td align=\"left\">O2N</td><td align=\"left\">280</td><td align=\"left\">83%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A27\">1A27</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2N</td><td align=\"left\">O1N</td><td align=\"left\">58</td><td align=\"left\">17%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1A80\">1A80</ext-link></td></tr><tr><td align=\"left\">TPP</td><td align=\"left\">thiamine diphosphate (i.e. vitamin B₁)</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">25</td><td align=\"left\">56%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1AY0\">1AY0</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">20</td><td align=\"left\">44%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1B0P\">1B0P</ext-link></td></tr><tr><td align=\"left\">UDP</td><td align=\"left\">uridine-5'-diphosphate</td><td align=\"left\">PA</td><td align=\"left\">O1A</td><td align=\"left\">O2A</td><td align=\"left\">80</td><td align=\"left\">79%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1BGU\">1BGU</ext-link></td></tr><tr><td/><td/><td/><td align=\"left\">O2A</td><td align=\"left\">O1A</td><td align=\"left\">21</td><td align=\"left\">21%</td><td align=\"left\"><ext-link ext-link-type=\"pdb\" xlink:href=\"1C3J\">1C3J</ext-link></td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S16-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:mi>B</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:mi>B</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:mi>B</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:mi>C</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:mi>C</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:mi>C</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:mi>D</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:mi>D</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:mi>D</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mi>m</mml:mi>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S16-i2\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:mi>C</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:mi>C</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:mi>C</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:mi>A</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:mi>B</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:mi>B</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:mi>B</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>X</mml:mi>\n <mml:msup>\n <mml:mi>C</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Y</mml:mi>\n <mml:msup>\n <mml:mi>C</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mrow>\n <mml:msub>\n <mml:mi>Z</mml:mi>\n <mml:msup>\n <mml:mi>C</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:msub>\n </mml:mrow>\n </mml:mtd>\n <mml:mtd>\n <mml:mn>1</mml:mn>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mi>m</mml:mi>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p><bold>Supplemental Table 1</bold>. Contains Table ##TAB##0##1## from this document with about four additional pages of examples of naming convention statistics for selected ligands.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional file 2</title><p><bold>Supplemental Table 2</bold>. Contains all of the calculated results, including those for prochiral centers that appear only once in the PDB.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional file 3</title><p><bold>Explanation of Supplemental Table 2</bold>. Contains an explanation of the columns in Supplemental Table 2.</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p># = number of PDB files in which the given naming convention was observed.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S16-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S16-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S16-3\"/>" ]

[ "<media xlink:href=\"1471-2105-9-S9-S16-S1.doc\" mimetype=\"application\" mime-subtype=\"msword\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-S9-S16-S2.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-S9-S16-S3.doc\" mimetype=\"application\" mime-subtype=\"msword\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Genome-wide association studies (GWAS) aim to identify genetic variants of single nucleotide polymorphisms (SNPs) across the entire human genome that are associated with phenotypic traits, such as disease status and drug response. The International HapMap project determined genotypes of over 3.1 million common SNPs in human populations and computationally assembled them into a genome-wide map of SNP-tagged haplotypes [##REF##16255080##1##,##REF##17943122##2##]. Concurrently, high-throughput SNP genotyping technology advanced to enable simultaneous genotyping of hundreds of thousands of SNPs. These advances combine to make GWAS a feasible and a promising research field for associating genotypes with various disease susceptibilities and health outcomes. Recently, GWAS was successfully applied to identify common genetic variants associated with a variety of phenotypes [##REF##15761122##3##, ####REF##17068223##4##, ##REF##17434869##5##, ##REF##17463246##6##, ##REF##17463249##7##, ##REF##17463248##8##, ##REF##17293876##9##, ##REF##17529967##10##, ##REF##17554300##11##, ##REF##17804789##12##, ##REF##18245381##13##, ##REF##16699517##14##, ##REF##17200669##15##, ##REF##17435756##16##, ##REF##17401366##17##, ##REF##17401363##18##, ##REF##17558408##19##, ##REF##17554260##20##, ##REF##17529973##21##, ##REF##17618284##22##, ##REF##17618283##23##, ##REF##17632509##24##, ##REF##17637780##25##, ##REF##17317784##26##, ##REF##17236132##27##, ##REF##18179894##28##, ##REF##18252221##29##, ##REF##18453082##30##, ##REF##18067574##31####18067574##31##]. Many of these studies used the Affymetrix GeneChip Human Mapping 500 K array set [##REF##17434869##5##,##REF##17463246##6##,##REF##17554300##11##]. The genomic DNA for one of the arrays is cleaved with the <italic>Nsp I </italic>restriction enzyme and ~262,000 SNPs are interrogated. The second chip uses <italic>Sty I </italic>– cleaved genomic DNA and ~238,000 SNPs are analyzed. Genotypes from Affymetrix GeneChip Human Mapping 500 K array set data are usually determined by the calling algorithm BRLMM [##UREF##0##32##] embedded in Affymetrix software packages. Algorithms developed by other laboratories such as PLASQ [##REF##16322765##33##], GEL [##REF##16809396##34##], CRLMM [##REF##17189563##35##], SNiPer-HD [##REF##17062589##36##], MAMS [##REF##17459966##37##], and CHIAMO [##REF##17554300##11##] are also utilized.</p>", "<p>The MPAM algorithm was developed for analysis of raw data (i.e., the CEL files) from the first generation of Affymetrix Mapping 10 K array and is based on clustering of chips for each SNP by modified partitioning around medoids [##REF##14668223##38##]. MPAM was error prone for SNPs with missing genotype groups or low minor allele frequency, a problem more pronounced on the second generation of Affymetrix Mapping 100 K array. This prompted Affymetrix to develop a new dynamic model based calling algorithm called DM for Mapping 100 K array data [##REF##15657097##39##]. DM is a single-chip calling algorithm and usually calls genotypes with high overall call rate and accuracy. However, the algorithm exhibited a higher misclassification rate for heterozygous genotypes than for homozygous genotypes. To improve data analyses for genotyping arrays, the multi-chip genotype calling algorithm RLMM was developed. RLMM is based on a robustly fitted, <underline>l</underline>inear <underline>m</underline>odel that employs <underline>M</underline>ahalanobis distance for classification [##REF##16267090##40##]. RLMM achieved a higher call rate than DM. With the release of the Mapping 500 K SNP array set, Affymetrix extended the RLMM model to BRLMM by adding a Bayesian step that provided improved estimates of cluster centers and variances. The DM and GEL algorithms operate on a single chip, while all others use multiple chips to call genotypes.</p>", "<p>High call rate and accuracy of genotype calling are important and essential issues for success of GWAS, since errors introduced in the genotypes by calling algorithms can inflate false associations and may lose true associations between genotype and phenotype. Each of the algorithms was reported to have a high successful call rate and accuracy, or more precisely, high concordance with genotypes determined by the International HapMap Consortium on the HapMap samples. With the exception of DM and GEL, the algorithms require data from multiple chips (i.e., a batch) to make genotype calls. A GWAS usually involves analyses of thousands of samples that generate thousands of raw data files (i.e., CEL files). The raw data file for one sample (two CEL files for Affymetrix Mapping 500 K array set: one from Nsp-digested genomic DNA and one from Sty-digested DNA) is about 130 MB in size. Computer memory (RAM) limits make it unfeasible to analyze all CEL files in a GWAS in one single batch on a single computer. The samples are, therefore, divided into many batches for genotype calling. Affymetrix suggests 40 to 96 CEL files for a batch for the BRLMM method. To date, the effects on genotype calls caused (potentially) by changing the number and specific combinations of CEL files in batches and propagation of the effects to the downstream association analysis have not been investigated.</p>", "<p>Since BRLMM is recommended by Affymetrix, we analyzed the effect of batch size and composition on the ability of the BRLMM algorithm to consistently call the 270 samples from the International HapMap project.</p>" ]

[ "<title>Methods</title>", "<title>Raw data</title>", "<p>The raw data (CEL files) from the Affymetrix GeneChip Human Mapping 500 K array set of the 270 HapMap samples were downloaded from the International HapMap project website <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.hapmap.org/downloads/raw_data/affy500k/\"/>. The CEL file format was described on Affymetrix's developer pages <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.affymetrix.com/Auth/support/developer/fusion/file_formats.zip\"/>. The file name indicated the population code (CEU/YRI/CHB+JPT), the sample identifier (e.g., NA12345), followed by the Affymetrix array type (based on restriction enzyme name: Nsp or Sty). Three population groups composed the data sets and each group contained 90 samples: CEU had 90 samples from Utah residents with ancestry from northern and western Europe (termed as European in this paper); CHB+JPT had 45 samples from Han Chinese in Beijing, China, and 45 samples from Japanese in Tokyo, Japan (termed as Asian in this paper); YRI had 90 samples from Yoruba in Ibadan, Nigeria (termed as African in this paper).</p>", "<title>Quality of the raw data</title>", "<p>The quality of the raw data from the Affymetrix Human Mapping 500 K array set was assessed using DM [##REF##15657097##39##] before genotype calling by BRLMM. DM is a single array based algorithm; it processes one CEL file at a time in a multiple CEL file batch and statistically assesses experimental qualities with a numerical score between 0 and 100. A high QC (quality control) number means high quality of the experiment (CEL file).</p>", "<title>Genotype calling by BRLMM</title>", "<p>All experiments of genotype calling by BRLMM reported in this paper were conducted using apt-probeset-genotype of Affymetrix Power Tools 1.8.5. Affymetrix Power Tools (APT) contains a set of cross-platform command line programs that implement algorithms for analyzing and working with Affymetrix GeneChip^®arrays. These programs are available on the Affymetrix website <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.affymetrix.com/support/developer/powertools/index.affx\"/>. APT programs are intended for \"power users\" who prefer programs that can be utilized in scripting environments and are sophisticated enough to handle the complexity of extra features and functionality. The function of apt-probeset-genotype in APT is an application for making genotype calls using SNP Arrays (100 K, 500 K, Genome-Wide SNP Arrays 5.0 and 6.0). BRLMM is one of the genotype calling algorithms implemented in this function, and enables many parameters to be changed by a user. For the studies reported here, all the parameters, except as noted in the narrative were set to the default values recommended by Affymetrix. The chip description files (cdf) for both Nsp and Sty chips of the Mapping 500 K array set, as well as files for defining SNPs on chromosome X, were also used before genotype calling. They were downloaded from Affymetrix website. Nsp and Sty CEL files were genotype-called separately.</p>", "<title>Batch size experiments</title>", "<p>Three experiments were designed and conducted in order to assess the effect of batch size. In the first experiment (BS1), the 270 HapMap samples were divided into three batches based on their population groups: 90 Europeans, 90 Asians, and 90 Africans. The genotypes were called separately by BRLMM using the default parameter setting suggested by Affymetrix (CEL files from Nsp and CEL files from Sty were analyzed separately). Genotype calling results on Nsp files and on Sty files of the three batches in this experiment were then merged for comparison with results of other experiments with different batch sizes. The second experiment (BS2) used a batch size of 45 samples. Genotypes were called from the CEL files from 90 European samples in two batches, each with 45 CEL files using BRLMM with the same parameter settings as in the first experiment. The procedure was repeated for the Asian and African samples. In the third experiment (BS3), the batch size was 30 samples from each population groups.</p>", "<title>Batch composition experiments</title>", "<p>The selection of samples (CEL files) to place in each batch can also be anticipated to alter genotyping call rates. The term batch composition effect is used here to denote the selected arrays within batches. BRLMM was used with default parameter settings and the CEL files of 270 HapMap samples to test batch composition effects. In the first experiment (BC1), the 270 samples were placed in three batches. One batch contained 90 samples from the same population group, Europeans, Asians, or Africans. In the second experiment (BC2), the 90 samples in each of the three population groups were evenly divided into two subgroups with each subgroup having 45 unique samples. Genotype calling was then conducted in three batches with composition of: (i) subgroup 1 of Europeans + subgroup 1 of Asians, (ii) subgroup 2 of Europeans + subgroup 1 of Africans, and (iii) subgroup 2 of Africans + subgroup 2 of Asians. In the third experiment (BC3), the 90 samples in each of the three population groups were evenly divided into three subgroups with each subgroup having 30 unique samples. Genotype calling was then conducted in three batches with composition of: (i) subgroup 1 of Europeans + subgroup 1 of Asians + subgroup 1 of Africans, (ii) subgroup 2 of Europeans + subgroup 2 of Asians + subgroup 2 of Africans, and (iii) subgroup 3 of Europeans + subgroup 3 of Asians + subgroup 3 of Africans. In each of the three experiments, genotype calling results of the three batches were merged together before conducting the comparisons.</p>", "<title>Comparing genotype calling results</title>", "<p>In each of the experiments reported here, the genotype calling results by BRLMM from different calling batches were first merged using a set of in-house programs written in C++. When merging the calling results, genotypes of SNPs in Nsp and Sty chips of the same samples were merged followed by assembling together all genotypes of all of the 270 HapMap samples. Thereafter, overall call rates for each of the experiments, call rates of individual samples and SNPs in each of the experiments, and concordant calls between experiments were calculated and exported as tab-delimited text files using the in-house programs written in C++. Comparison of calling results was done using the R package.</p>", "<p>Paired two samples t-test in R package (t.test) was used to statistically test the alternative hypothesis that call rates on samples or SNPs between two calling experiments are different.</p>", "<p>To quantify batch effect, average absolute differences in call rates were calculated for the comparisons using formula (1).</p>", "<p></p>", "<p>where and are call rates of experiments 1 and 2 of sample i or SNP i, respectively; N is the total number of samples (in this case, 270) or SNPs (in this case, 500,668 which includes 50 QC probe sets in both Nsp and Sty chips).</p>", "<title>Association analysis</title>", "<p>In order to study the propagation of batch effect to the significantly associated SNPs, all genotype calling results of the raw data of 270 HapMap samples using BRLMM with different batch sizes and compositions were analyzed using Chi²statistics test for associations between the SNPs and the case-control settings.</p>", "<p>Prior to association analysis, quality control (QC) of the calling results was conducted to remove markers and samples with low quality. For each of the calling results, call rate of 90% was used to remove SNPs and samples. Minor allele frequency was used to filter SNPs and its cut-off was set to 0.01. Departure from Hardy-Weinberg equilibrium (HWE) was check for all SNPs. The p-value of Chi²test for Hardy-Weinberg equilibrium was calculated for all SNPs at first and then the p-values were adjusted for multiple tests using Benjamini and Hochberg false discovery rate (FDR) [##UREF##1##41##]. FDR of 0.01 was set as the cut-off for HWE test. There were no samples removed because of low quality. 54942 (10.97%) to 55496 (11.084%) SNPs were removed in the QC, mainly because of departure from HWE.</p>", "<p>To mimic \"case-control\" in GWAS, for each of the genotype calling results, each of the three population groups (European, African, and Asian) was assigned as \"case\" while the other two as \"control\" to form a data set for association analysis for identifying the SNPs significantly associated with the \"case\" population group.</p>", "<p>In the association analyses, a 2 × 3 contingency table was generated for each SNP and a case-control setting. Then Chi²statistics test was applied on the contingency table to calculate a p-value for measuring the statistical significance of the association between the testing SNP and the corresponding case-control setting. After raw p-values for all SNPs in a data set were calculated, Bonferroni correction was applied to adjust the raw p-values. Lastly, a criterion of Bonferroni-corrected p-value less than 0.01 was used to identify the significantly associated SNPs.</p>" ]

[ "<title>Results</title>", "<title>Batch size effect</title>", "<p>Batch size effect was assessed by comparing the genotypes called from BS1, BS2, and BS3 (see <italic>Methods</italic>) for call rate and concordance. The overall call rates, defined as the proportion of successful calls to the total number of calls (successful calls plus missing calls) for BS1, BS2, and BS3 were 99.48%, 99.50%, and 99.49%, respectively. However, overall call rates are not informative enough to assess the distribution of missed calls on the chip. Batch size effect on genotype calling rates are best compared using one-against-one comparisons of distributions of call rates on individual samples and SNPs. These distributions were calculated from data of samples and SNPs generated from the calling results of the experiments with three batch sizes (BS1, BS2, and BS3).</p>", "<p>The comparison of call rates of samples using MA-like plots is shown in Figure ##FIG##0##1##. The average call rate of two genotype calling results (x-axis) from experiments with two different batch sizes were plotted against the difference of call rates between the two experiments (large batch size – small batch size; y-axis). The horizontal dotted lines at y = 0 represent the expected locations of samples if the missing calls on each sample were exactly the same in the two experiments. Data points above this line are the samples having fewer missing calls (i.e., higher call rate) in the experiment with the larger batch size than in the experiment with the smaller batch size. Data points beneath this line indicate samples having fewer missing calls in the experiment with smaller batch size than in the experiment with the larger batch size. The perpendicular distance from a data point to this line is the difference in call rate of a sample between the two experiments. Figure ##FIG##0##1A## compares the results of BS1 with BS2; ##FIG##0##1B## compares the results of BS1 with BS3; and ##FIG##0##1C## compares the results of BS2 with BS3. Data points at lower average call rates are more distant from the calculated equivalent call rate (dotted line) than the data points at higher average call rates. Thus, batch size affected lower call rates more severely than higher call rates. Furthermore, data points in Figure ##FIG##0##1B## (BS1 versus BS3) are farther away from the dotted line when compared with the data points in Figure ##FIG##0##1A## (BS1 versus BS2), which, in turn, were farther away from the dotted line when compared with Figure ##FIG##0##1C## (BS2 versus BS3). The values of (see <italic>Methods</italic>) were 0.0304, 0.0416, and 0.0257 for comparisons shown in Figure ##FIG##0##1A, B##, and ##FIG##0##1C##, respectively, that are related to the corresponding differences of batch sizes of the compared experiments, 45 (90 – 45), 60 (90 – 30), and 15 (45 – 30). The p-values for comparisons in Figure ##FIG##0##1A, B##, and ##FIG##0##1C## are 1.736 × 10^-6, 0.0296, and 0.0116, respectively, indicating that call rates on samples between calling batch sizes are statistically different.</p>", "<p>The comparisons of the call rates for individual SNPs are depicted by MA-like plots in Figure ##FIG##1##2##. Figure ##FIG##1##2A## compares the results of BS1 with BS2; ##FIG##1##2B## compares the results of BS1 with BS3; and ##FIG##1##2C## compares the results of BS2 with BS3. The trend is similar to that observed in Figure ##FIG##0##1## that batch size affected lower call rates more severely than higher call rates for individual SNPs. The values were calculated to be 0.1563, 0.1982, and 0.1467 for the comparisons shown in Figure ##FIG##1##2A, B##, and ##FIG##1##2C##, respectively. They were positively correlated with the differences of batch sizes of the compared experiments, 45, 60, and 15, respectively. The p-values for comparisons in Figure ##FIG##1##2A, B##, and ##FIG##1##2C## are 2.2 × 10^-16, indicating that the difference of call rates on SNPs between calling batch sizes are statistically significant.</p>", "<p>Comparing call rates in experiments with different batch sizes can only assess the batch size effect on missing calls. Since three genotypes (homozygote, heterozygote, and variant homozygote) are possible for a genotype call, we determined the effect of batch size on the ability to consistently call the genotype. To evaluate the batch size effect on successful calls, concordance of successful genotype calls between experiments with different batch sizes was analyzed (Table ##TAB##0##1##). Batch size affected successful genotype calls since the concordances were not 100% and heterozygous genotype concordances were more affected than homozygous genotype concordances. The largest difference in batch size (60, BS1 versus BS3) led to the lowest concordances (99.986% overall concordance). However, the concordances for BS2 versus BS3 were slightly lower than for BS1 versus BS2, even though the difference of batch sizes for BS2 versus BS3 (45 – 30 = 15) is smaller than that for BS1 versus BS2 (90 – 45 = 45). This result is likely due to the relatively large difference in the number of arrays in the batch (BS1 = 90 arrays and BS3 = 30 arrays). High concordance of genotype calls depends on the difference between batch sizes as well as the actual batch sizes themselves.</p>", "<title>Batch composition effect</title>", "<p>The overall call rate based on all CEL files of the 270 HapMap samples for BC1, BC2, and BC3 (see <italic>Methods</italic>) were 99.48%, 99.43%, and 99.41%, respectively. The genetic homogeneity of the batches in BC1 (samples from 1 population group) is higher than that of BC2 (samples from 2 population groups) which, in turn, is higher than that of BC3 (samples from 3 population groups). The batch sizes were the same for all of the three experiments. Thus, higher call rates were obtained when genotype calling was conducted with samples of higher genetic homogeneity. The effect of batch homogeneity was relatively minor by this measure. Because the distribution of missing calls on samples and SNPs was more informative for assessing batch effect in our first experiments (BS studies), we examined the distribution of call rates in the BC experiments.</p>", "<p>The comparisons of call rates on samples are depicted by MA-like plots (Figure ##FIG##2##3##). Figure ##FIG##2##3A## compares the results of BC1 with BC2; ##FIG##2##3B## compares the results of BC1 with BC3; and ##FIG##2##3C## compares the results of BC2 with BC3. It can be seen that most of the data points are above the dotted lines, indicating fewer missing genotypes (i.e., higher call rate) when samples in batches are of higher genetic homogeneity. Batch composition had a larger effect when the call rate was lower. Moreover, the level of batch composition effects was related to differences in the genetic homogeneity of samples in the compared batch compositions. We quantified genetic homogeneity as , where n is number of population groups of samples in a batch composition. The values of <italic>GH </italic>are 1, 0.5 and 0.33 for BC1, BC2, and BC3, respectively. The values of the comparisons in Figure ##FIG##2##3A, B##, and ##FIG##2##3C## are 0.0552, 0.0774, and 0.0373, respectively. These values are positively correlated with the corresponding <italic>GH </italic>differences between the compared experiments, (1 – 0.5 = 0.5), (1 – 0.33 = 0.67), and (0.5 – 0.33 = 0.17). The p-values for all comparisons are 2.2 × 10^-16. Therefore, the call rates on samples between calling batch compositions are statistically different.</p>", "<p>The comparisons of call rates on SNPs for BC1 versus BC2, BC1 versus BC3, and BC2 versus BC3 are shown in Figure ##FIG##3##4A, B##, and ##FIG##3##4C##, respectively. Data points at lower average call rate were farther away from the dotted line than the data points at higher average call rate; that is, batch composition affected SNPs with lower call rates more severely than SNPs with higher call rates. Furthermore, more SNPs are above rather than below the calculated equivalent call rates (dotted line) indicating fewer missing genotypes per SNP (i.e., higher call rate) when samples in calling batches are of higher genetic homogeneity. Moreover, it was further confirmed that the level of batch composition effects was related to differences in genetic homogeneity of samples in the compared batch compositions. The values are 0.2046, 0.2384, and 0.1749 for comparisons shown in Figure ##FIG##3##4A, B##, and ##FIG##3##4C##, respectively, that are related to the corresponding <italic>GH </italic>differences between the compared experiments: 0.5, 0.67, and 0.17. The p-values for all comparisons are 2.2 × 10^-16, confirming that the call rates on SNPs between calling batch compositions are statistically different.</p>", "<p>To evaluate batch composition effect on successful genotype calls, concordance of successful genotype calls between experiments with different batch compositions was analyzed (Table ##TAB##1##2##). Batch composition not only affected the genotype calls but was more pronounced at heterozygous genotypes compared with homozygous genotypes, since the concordance for heterozygous genotype calls were lower than the corresponding concordance for homozygous genotype calls. Moreover, the concordance of successful genotype calls between the compared batch compositions were negatively related to genetic homogeneity differences between the batch compositions. For example, overall concordances were 99.986%, 99.980%, and 99.991% for BC1 versus BC2, BC1 versus BC3, and BC2 versus BC3, respectively. These are in opposite order of the <italic>GH </italic>differences of the compared experiments, that is, 0.5, 0.67, and 0.17 for BC1 versus BC2, BC1 versus BC3, and BC2 versus BC3, respectively.</p>", "<title>Quality of the raw data</title>", "<p>The quality of the raw data is important for comparative analyses and interpretation. The QC scores of the 270 Nsp CEL files and of the 270 Sty chip CEL files of the 270 HapMap samples were calculated using DM (Figure ##FIG##4##5A## and ##FIG##4##5B##, respectively). The average QC scores for Nsp and Sty CEL files are 97.58 and 98.26, respectively. The lowest QC scores for Nsp and Sty CEL files are 93.49 and 93.18, respectively. The Affymetrix default QC cut-off score is 93. Therefore, we confirmed high QC of the raw data and used all CEL files of 270 HapMap samples in our study.</p>", "<title>Propagation of batch effect to significantly associated SNPs</title>", "<p>The objective of a GWAS is to identify the genetic markers associated with a specific phenotypic trait. It is critical to assess whether and how the batch effect propagates to the significant SNPs identified in the downstream association analysis. Three case-control based association analyses were conducted for each of the calling results with different batch sizes and compositions to assess the propagation of batch effect in genotype calling to the significantly associated SNPs (see <italic>Methods</italic>).</p>", "<p>After removal of low quality SNPs by quality control assessment, each of the three population groups (European, Asian, and African) was set as \"case\" while the other two groups were set as \"control\". Association analyses were conducted to identify SNPs that can differentiate the \"case\" group from the \"control\" group. Different lists of SNPs significantly associated with a same population group, identified using the genotype calling results with different batch sizes and compositions, were compared using Venn diagram.</p>", "<p>The comparisons of the significantly associated SNPs obtained from calling results with different batch sizes are given in Figure ##FIG##5##6##. The significantly associated SNPs from BS1 are in black circles, from BS2 in blue circles, and from BS3 in red circles. Number of significantly associated SNPs common in all three batch sizes is in brown, shared only by two batch sizes in green. The association analyses results for European versus others are depicted in Figure ##FIG##5##6A##, for African versus others in ##FIG##5##6B##, and for Asian versus others in ##FIG##5##6C##.</p>", "<p>It is clear that the batch size effect on genotype calling propagated into the downstream association analyses. Moreover, it was observed that the larger the differences between two batch sizes, the fewer the significantly associated SNPs shared by the two batch sizes. For example, there were 471, 370, and 217 significantly associated SNPs shared only by BS2 and BS3, by BS1 and BS2, and by BS1 and BS3 for the association analyses with European as \"case\", respectively, that are negatively related to the corresponding differences of batch sizes: 15, 45, and 60. Same trends were observed for the association analyses with African as \"case\" and with Asian as \"case\".</p>", "<p>Figure ##FIG##6##7## compares the lists of significantly associated SNPs obtained using the genotypes called by the three batch compositions. The significantly associated SNPs from BC1 are in black circles, from BC2 in blue circles, and from BC3 in red circles. Number of significantly associated SNPs common in all three compositions is in brown, shared only by two compositions in green. Association analyses results for European versus others are depicted in Figure ##FIG##6##7A##, for African versus others in ##FIG##6##7B##, and for Asian versus others in ##FIG##6##7C##.</p>", "<p>The Venn diagrams demonstrated that for a same \"case-control\" setting different lists of significantly associated SNPs were identified by the same statistical test (Chi²test) using the genotype calling results from different batch compositions. Therefore, the batch composition effect on genotype calling propagated to the significantly associated SNPs. Moreover, it was observed that the larger the difference of genetic homogeneity between two batch compositions, the fewer the significantly associated SNPs shared by the two batch compositions. For example, there were 555, 512, and 229 significantly associated SNPs shared only by BC2 and BC3, by BC1 and BC2, and by BC1 and BC3, respectively, for the association analyses with European as \"case\". The numbers are negatively related to the corresponding differences of genetic homogeneity in the batch compositions: 0.17, 0.5, and 0.67. Same trends were observed for the association analyses with African as \"case\" and with Asian as \"case\".</p>" ]

[ "<title>Discussion</title>", "<p>GWAS is increasingly used to identify loci containing genetic variants associated with common diseases and drug responses. The number of SNPs interrogated in a GWAS has grown from thousands to millions; for example, the newest Affymetrix SNPs array 6.0 contains ~2 million probe sets. At the same time, the allele frequency difference of disease-associated or drug-associated SNPs is usually very small. Therefore, a very small error introduced in genotypes by genotype calling algorithms may result in inflated false associations between genotype and phenotype in the downstream association analysis. Reproducibility and robustness are as important to genotype calling as is the accuracy and call rate that are usually used to evaluate performance of genotype calling algorithms. As most genotype calling algorithms are based on multiple chips, and genotype calling for a GWAS is usually conducted in many batches, reproducibility and robustness of multi-chip calling algorithms under different batch sizes and compositions are important variables. Statistical tests of these parameters would increase the confidence for associated SNPs identified in downstream association analysis.</p>", "<p>A heterozygous genotype carries a rare allele. Therefore, the robustness of calling heterozygous reduces false positive associations and the chance of missing true associations. Our studies revealed that both batch size and composition affected genotype calling results, especially for heterozygous genotype calling. It was also demonstrated that batch effect propagates to the downstream association analysis. Genotype calling algorithms that eliminate or reduce batch effects but maintain high call rates and accuracy are preferred for GWAS.</p>", "<p>BRLMM first derives an initial guess for each SNP's genotype using the DM algorithm and then analyzes across SNPs to identify cases of non-monomorphism. This subset of non-monomorphism SNPs is then used to estimate a prior distribution on cluster centers and variance-covariance matrices. This subset of SNP genotypes is revisited and the clusters and variances of the initial genotype guesses are combined with the prior information of the SNP in an ad-hoc Bayesian procedure to derive a posterior estimate of cluster centers and variances. All SNPs in a chip are called according to their Mahalanobis distances from the three cluster centers and confidence scores are assigned to the calls. With default settings, BRLMM randomly picks 10,000 SNPs to estimate cluster centers and variances. But the number of non-monomorphism SNPs used to estimate the prior distribution on cluster centers and variance-covariance matrices varies with changing number of CEL files and changing composition of CEL files in the calling batches. Batch size effect and batch composition effect alter these estimates of prior distribution and variance-covariance matrices. The effect of altering the number of non-monomorphism SNPs was confirmed when using the BRLMM calling algorithm by varying the batch size and composition. The average number of non-monomorphism SNPs used to estimate the prior distributions are 5468 (Nsp) and 5422 (Sty), 4356 (Nsp) and 4358 (Sty), and 3612 (Nsp) and 3618 (Sty) for calling batches in BS1, BS2, and BS3, respectively. The difference of batch sizes is related to the difference of numbers of non-monomorphism SNPs used to estimate the prior distribution which is, in turn, related to the difference of genotype calling results. The average number of non-monomorphism SNPs used to estimate the prior distribution are 5468 (Nsp) and 5422 (Sty), 6399 (Nsp) and 6308 (Sty), and 6788 (Nsp) and 6688 (Sty) for calling batches in BC1, BC2, and BC3, respectively. Differences in genetic homogeneity of samples are related to differences in the numbers of non-monomorphism SNPs used to estimate the prior which, in turn, is related to the difference of genotype calling results.</p>" ]

[ "<title>Conclusion</title>", "<p>As demonstrated above, both batch size and batch composition affect genotype calling results of GWAS using the BRLMM algorithm. The larger the difference of batch sizes, the larger the effect. When the samples in the calling batches are more homogenous, more concordant genotypes are called. Batch effect propagates to the downstream association analysis and makes the significantly associated SNPs identified inconsistent. Therefore, we suggest from our studies that the same or larger batch sizes should be used to make genotype calls for GWAS and homogenous samples should be put into the same batches.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Genome-wide association studies (GWAS) aim to identify genetic variants (usually single nucleotide polymorphisms [SNPs]) across the entire human genome that are associated with phenotypic traits such as disease status and drug response. Highly accurate and reproducible genotype calling are paramount since errors introduced by calling algorithms can lead to inflation of false associations between genotype and phenotype. Most genotype calling algorithms currently used for GWAS are based on multiple arrays. Because hundreds of gigabytes (GB) of raw data are generated from a GWAS, the samples are typically partitioned into batches containing subsets of the entire dataset for genotype calling. High call rates and accuracies have been achieved. However, the effects of batch size (i.e., number of chips analyzed together) and of batch composition (i.e., the choice of chips in a batch) on call rate and accuracy as well as the propagation of the effects into significantly associated SNPs identified have not been investigated. In this paper, we analyzed both the batch size and batch composition for effects on the genotype calling algorithm BRLMM using raw data of 270 HapMap samples analyzed with the Affymetrix Human Mapping 500 K array set.</p>", "<title>Results</title>", "<p>Using data from 270 HapMap samples interrogated with the Affymetrix Human Mapping 500 K array set, three different batch sizes and three different batch compositions were used for genotyping using the BRLMM algorithm. Comparative analysis of the calling results and the corresponding lists of significant SNPs identified through association analysis revealed that both batch size and composition affected genotype calling results and significantly associated SNPs. Batch size and batch composition effects were more severe on samples and SNPs with lower call rates than ones with higher call rates, and on heterozygous genotype calls compared to homozygous genotype calls.</p>", "<title>Conclusion</title>", "<p>Batch size and composition affect the genotype calling results in GWAS using BRLMM. The larger the differences in batch sizes, the larger the effect. The more homogenous the samples in the batches, the more consistent the genotype calls. The inconsistency propagates to the lists of significantly associated SNPs identified in downstream association analysis. Thus, uniform and large batch sizes should be used to make genotype calls for GWAS. In addition, samples of high homogeneity should be placed into the same batch.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>HH coordinated the project, designed the experiments, conducted the genotype calling and association analysis, compared the calling results using R package, and wrote the manuscript. ZS wrote all of the in-house C++ programs, and involved discussions on the experiments and analysis of the calling results. WG calculated all of the call rates and concordant calls and involved discussions on the experiments and analysis of the calling results. LS, RP, JK, JCF, and WT involved discussions on designing the experiments and analysis and assisted the writing manuscript. HF, JX, JC, and TH involved discussions on experimental design and data analysis. All authors read and approved the manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We thank Drs. Federico Goodsaid, Sue Jane Wang, and Li Zhang of CDER/FDA, Ansar Jawaid of AstraZeneca, David Craig of The Translational Genomics Research Institute, Uwe Scherf, Lakshmi Vishnuvajjala, Arkendra De, and Lakshman Ramamurthy of CDRH/FDA, Nick Xiao of Core Genotyping Facility/NCI, and Keith Nangle, Meg E. Ehm, and Gbenga R. Kazeem of GlaxoSmithKline for fruitful discussions. We are grateful to the reviewers for their comments and suggestions for revising and improving the paper. We also thank Dr. Tao Chen and Dr. Lei Guo for reading through the paper and their comments. The views presented in this article do not necessarily reflect those of the US Food and Drug Administration.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>MA-like plots for comparing call rates of samples between two experiments with different batch sizes</bold>. The empty circles depict the 270 samples. The x-axes represent average call rates of individual samples in two experiments with different batch sizes. The horizontal dotted lines indicate where values of the expected call rates are the same in the two compared experiments. <bold>A</bold>: Comparison between BS1 and BS2. The y-axis represents call rate in BS1 – call rate in BS2. <bold>B</bold>: Comparison between BS1 and BS3. The y-axis represents call rate in BS1 – call rate in BS3. <bold>C</bold>: Comparison between BS2 and BS3. The y-axis represents call rate in BS2 – call rate in BS3.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>MA-like plots for comparing call rates of SNPs between two experiments with different batch sizes</bold>. The empty circles depict 500,568 SNPs. The x-axes represent average call rates of individual SNPs in two experiments with different batch sizes. The horizontal dotted lines indicate the expected locations of SNPs where the call rates in the two compared experiments were exactly same. <bold>A</bold>: Comparison between BS1 and BS2. The y-axis represents call rate in BS1 – call rate in BS2. <bold>B</bold>: Comparison between BS1 and BS3. The y-axis represents call rate in BS1 – call rate in BS3. <bold>C</bold>: Comparison between BS2 and BS3. The y-axis represents call rate in BS2 – call rate in BS3.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>MA-like plots for comparing call rates of samples between two experiments with different batch compositions</bold>. The empty circles depict the 270 samples. The x-axes represent average call rates of individual samples in two experiments with different batch compositions. The horizontal dotted lines indicate the expected locations of samples where the call rates in the two compared experiments were exact same. <bold>A</bold>: Comparison between BC1 and BC2. The y-axis represents call rate in BC1 – call rate in BC2. <bold>B</bold>: Comparison between BC1 and BC3. The y-axis represents call rate in BC1 – call rate in BC3. <bold>C</bold>: Comparison between BC2 and BC3. The y-axis represents call rate in BC2 – call rate in BC3.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>MA-like plots for comparing call rates of SNPs between two experiments with different batch compositions</bold>. The empty circles depict 500,568 SNPs. The x-axes represent average call rates of individual SNPs in two experiments with different batch compositions. The horizontal dotted lines indicate the expected locations of SNPs where the call rates in the two compared experiments were exactly same. <bold>A</bold>: Comparison between BC1 and BC2. The y-axis represents call rate in BC1 – call rate in BC2. <bold>B</bold>: Comparison between BC1 and BC3. The y-axis represents call rate in BC1 – call rate in BC3. <bold>C</bold>: Comparison between BC2 and BC3. The y-axis represents call rate in BC2 – call rate in BC3.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>Histograms of QC confidence scores of Affymetrix Human Mapping 500 K Array Set CEL files of 270 HapMap samples</bold>. The x-axes indicate the QC confidence scores range from 0 to 100. The y-axes represent number of CEL files with QC confidence scores within a window depicted at the x-axes. <bold>A</bold>: Nsp chip CEL files of the 270 HapMap samples. <bold>B</bold>: Sty chip CEL files of the 270 HapMap samples.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Venn diagrams for comparisons of the significantly associated SNPs identified using the genotype calling results with different calling batch sizes</bold>. The numbers in circles are the significantly associated SNPs identified in association analyses using calling results from different batch sizes: black circles for BS1, blue circles for BS2, and red circles for BS3. Numbers in brown represent the associated SNPs shared by all three batch sizes, numbers in green represent the associated SNPs shared only by two batch sizes, and the numbers in other colors are the associated SNPs identified only by the corresponding batch sizes. <bold>A</bold>: The association analyses results for European versus others. <bold>B</bold>: The association analyses results for African versus others. <bold>C</bold>: The association analyses results for Asian versus others.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p><bold>Venn diagrams for comparisons of the significantly associated SNPs identified using the genotype calling results with different calling batch compositions</bold>. The numbers in circles are the significantly associated SNPs identified in association analyses using calling results from different batch compositions: black circles for BC1, blue circles for BC2, and red circles for BC3. Numbers in brown represent the associated SNPs shared by all three batch compositions, numbers in green represent the associated SNPs shared only by two batch compositions, and the numbers in other colors are the associated SNPs identified only by the corresponding batch compositions. <bold>A</bold>: The association analyses results for European versus others. <bold>B</bold>: The association analyses results for African versus others. <bold>C</bold>: The association analyses results for Asian versus others.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Concordance of calls between batch sizes</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\" colspan=\"2\">Comparison</td><td align=\"center\">BS1 <italic>vs </italic>BS2</td><td align=\"center\">BS1 <italic>vs </italic>BS3</td><td align=\"center\">BS2 <italic>vs </italic>BS3</td></tr></thead><tbody><tr><td align=\"center\">Successful Calls for Both</td><td align=\"center\">SNPs</td><td align=\"center\">134258764</td><td align=\"center\">134187584</td><td align=\"center\">134265847</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.338</td><td align=\"center\">99.285</td><td align=\"center\">99.343</td></tr><tr><td align=\"center\">Concordant Calls (All)</td><td align=\"center\">SNPs</td><td align=\"center\">134248899</td><td align=\"center\">134187584</td><td align=\"center\">134253973</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.993</td><td align=\"center\">99.986</td><td align=\"center\">99.991</td></tr><tr><td align=\"center\">Concordant Calls (Hom)</td><td align=\"center\">SNPs</td><td align=\"center\">98179772</td><td align=\"center\">98136394</td><td align=\"center\">98204063</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.997</td><td align=\"center\">99.993</td><td align=\"center\">99.995</td></tr><tr><td align=\"center\">Concordant Calls (Het)</td><td align=\"center\">SNPs</td><td align=\"center\">36069127</td><td align=\"center\">36031744</td><td align=\"center\">36049910</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.981</td><td align=\"center\">99.964</td><td align=\"center\">99.980</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Concordance of calls between batch compositions</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\" colspan=\"2\">Comparison</td><td align=\"center\">BC1 <italic>vs </italic>BC2</td><td align=\"center\">BC1 <italic>vs </italic>BC3</td><td align=\"center\">BC2 <italic>vs </italic>BC3</td></tr></thead><tbody><tr><td align=\"center\">Successful Calls for Both</td><td align=\"center\">SNPs</td><td align=\"center\">134128046</td><td align=\"center\">134063768</td><td align=\"center\">134107787</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.241</td><td align=\"center\">99.194</td><td align=\"center\">99.226</td></tr><tr><td align=\"center\">Concordant Calls (All)</td><td align=\"center\">SNPs</td><td align=\"center\">134109060</td><td align=\"center\">134036623</td><td align=\"center\">134095792</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.986</td><td align=\"center\">99.980</td><td align=\"center\">99.991</td></tr><tr><td align=\"center\">Concordant Calls (Hom)</td><td align=\"center\">SNPs</td><td align=\"center\">98050788</td><td align=\"center\">97992008</td><td align=\"center\">98016851</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.989</td><td align=\"center\">99.983</td><td align=\"center\">99.993</td></tr><tr><td align=\"center\">Concordant Calls (Het)</td><td align=\"center\">SNPs</td><td align=\"center\">36058272</td><td align=\"center\">36044165</td><td align=\"center\">36078941</td></tr><tr><td/><td align=\"center\">%</td><td align=\"center\">99.977</td><td align=\"center\">99.970</td><td align=\"center\">99.985</td></tr></tbody></table></table-wrap>" ]

[ "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S17-i1\" overflow=\"scroll\"><mml:semantics><mml:mover accent=\"true\"><mml:mi>D</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S17-i1\" overflow=\"scroll\"><mml:semantics><mml:mover accent=\"true\"><mml:mi>D</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S17-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>G</mml:mi><mml:mi>H</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mi>n</mml:mi></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S17-i1\" overflow=\"scroll\"><mml:semantics><mml:mover accent=\"true\"><mml:mi>D</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S17-i1\" overflow=\"scroll\"><mml:semantics><mml:mover accent=\"true\"><mml:mi>D</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM1\"><label>(1)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1471-2105-9-S9-S17-i3\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mover accent=\"true\">\n <mml:mi>D</mml:mi>\n <mml:mo>¯</mml:mo>\n </mml:mover>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mi>N</mml:mi>\n </mml:munderover>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:mrow>\n <mml:mi>C</mml:mi>\n <mml:msubsup>\n <mml:mi>R</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mn>1</mml:mn>\n </mml:msubsup>\n <mml:mo>−</mml:mo>\n <mml:mi>C</mml:mi>\n <mml:msubsup>\n <mml:mi>R</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mn>2</mml:mn>\n </mml:msubsup>\n </mml:mrow>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mi>N</mml:mi>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1471-2105-9-S9-S17-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>C</mml:mi><mml:msubsup><mml:mi>R</mml:mi><mml:mi>i</mml:mi><mml:mn>1</mml:mn></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1471-2105-9-S9-S17-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>C</mml:mi><mml:msubsup><mml:mi>R</mml:mi><mml:mn>1</mml:mn><mml:mn>2</mml:mn></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Successful calls for both: SNP genotypes successfully called in both of the compared experiments; Concordant calls (All): same genotype called in both of the compared experiments; Concordant calls (Hom): homozygous genotype called in both of the compared experiments; Concordant calls (Het): heterozygous genotype called in both of the compared experiments.</p></table-wrap-foot>", "<table-wrap-foot><p>Successful calls for both: genotype successfully called in both of the compared experiments; Concordant calls (All): same genotype called in both of the compared experiments; Concordant calls (Hom): homozygous genotype called in both of the compared experiments; Concordant calls (Het): heterozygous genotype called in both of the compared experiments.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S17-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S17-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S17-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S17-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S17-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S17-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S17-7\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>A timely response and preparedness in response to the changing environmental cues are essential for life in plants and animals alike. Since plants are dependent on light for photosynthesis, a natural assumption is that circadian (i.e., approximately daily) oscillation should be an even more prominent feature of the plant gene expression than in animals. Multiple studies have reported the existence and detailed mechanism of a circadian molecular clock in plants [##REF##17683202##1##, ####REF##11118138##2##, ##REF##11158533##3##, ##REF##16473970##4##, ##REF##9467557##5####9467557##5##]. Based on the studies of the model plant, <italic>Arabidopsis thaliana</italic>, researchers have determined that a \"substantial\" part of plant transcriptome cycles follows a circadian rhythm. This estimation is based on microarray experiments that search for the genes following the circadian rhythm among the entire set of transcripts that is examined by the microarray. Early attempts to identify these genes employed two-color spotted arrays, resulting in a cumbersome experimental design, or tried to minimize expenses by increasing the time span between the sample collections. The latter produced data with a very low sampling rate, which obscured the oscillation pattern in all but a few of the least noisy genes. More recent studies [##REF##17683202##1##,##REF##16473970##4##] used Affymetrix (Affymetrix Inc., Santa Clara) <italic>Arabidopsis thaliana </italic>expression arrays. These studies focused on the role of circadian oscillation in specific regulatory and signaling systems, but the entire set of data with expression profiles of more than 22,000 transcripts over two days was made available for downloading from the public databases. Although independent, both of these sets share almost identical experimental conditions and samples at the same rate of once every four hours. These features make the data easy to compare not only with one another but also to the large body of murine circadian expression data, which is also sampled every four hours over a period of two days.</p>", "<p>In recent years, we have published a number of studies on circadian oscillation in metabolically active peripheral tissues in mice [##REF##16567517##6##,##REF##17144790##7##]. We have also reanalyzed and reported the discovery of circadian oscillation in a number of independent murine data sets from public sources [##REF##17118131##8##]. The results of our analysis of murine circadian data are in sharp contradiction with previous reports. We were able to demonstrate circadian oscillation in not just a small number of the genes that are presumably linked to the circadian molecular clock but in all transcribed genes. Our most recent studies [##REF##18365000##9##] show that circadian oscillation is traceable not only in expressed but also in genes that were previously considered silent or unexpressed. The prevailing theory reflected in the molecular biology textbooks states that 10–15% of genes cycle within a daily period and are presumably regulated by the circadian molecular clock. Molecular clocks vary significantly in details, and the genes that form the clock may be evolutionarily unrelated, but the molecular clocks of plants, mammals, and insects share the same negative feedback principle that makes oscillation self-sustaining and adjustable. We have previously established that this theory does not reflect the reality, at least in the murine model. While the basic circadian clock is active in all central and peripheral tissues, other genes show robust, noise-free oscillation, particularly those involved in supporting basic energy metabolism and not directly linked to the circadian molecular clock. Moreover, the key elements of the cell transcription machinery itself exhibit a pronounced circadian pattern in the modulation of expression of practically every gene. We now know that the entire animal transcriptome, not just a specially regulated portion, experiences circadian oscillation. However, a reasonable expectation is to find the same observation in the plant transcriptome. Plants are even more dependent on the daily change in lighting conditions. Nevertheless, the most recent studies reported only 10.4 [##REF##17683202##1##] and 16% [##REF##16473970##4##] of \"circadially regulated\" genes in the plant transcriptome. This obvious contradiction demands a uniform re-analysis of the data using advanced methodology that has been tested in multiple previous studies.</p>" ]

[ "<title>Materials and methods</title>", "<title>Circadian expression data</title>", "<title>UC Davis data set</title>", "<p>Col-0 ecotype seeds were stratified at 4°C for 4 days before transfer to a growth chamber (22°C). Seedlings were entrained in 12-h white light (light source was cool white fluorescence tubes)/12-h dark cycles for 7 days before being released into free-running conditions of continuous white light at 22°C. Starting at subjective dawn of day 9, tissue was harvested every 4 h over the course of the next 44 h. Following standard protocols labelled cRNA targets were prepared from total RNA and hybridized to Affymetrix Arabidopsis expression GeneChips. Expression values were estimated at CSU from the original CEL files provided by the authors using dChip-derived Model-Based Expression Index.</p>", "<title>University of Warwick data set</title>", "<p>Wild-type Col-0 seedlings were used for the microarray circadian time-course experiment. Seedlings were placed immediately into LD 12:12 and grown for 7 days at 22°C. At dawn on the 8th day, they were placed into constant cool white fluorescent light. Samples were taken over two circadian cycles at 4-h intervals starting from ZT26. Samples were assayed on the Affymetrix GeneChip oligonucleotide ATH1 array (Affymetrix) according to the manufacturer's instructions. Background correction and normalization and gene expression analysis of the array data were performed using the GC-RMA routine [##REF##16410320##21##] in GeneSpring version 7.2 (Silicon Genetics). The resulting table of gene expression values was downloaded from the GEO database.</p>", "<title>Algorithms</title>", "<title>Data pre-processing</title>", "<p>Profiles have been smoothened by a 3^rddegree polynomial procedure and median-subtracted. For smoothing we use seven-point Savitzky-Golay algorithm [##UREF##0##24##]. To take advantage of all points in the time series a single-pass smoothing has been applied in a circular manner, with the last points contributing to smoothing the starting points. For better compatibility, the same smoothing and median subtraction procedure has been applied to all data sets.</p>", "<title>Spectral analysis</title>", "<p>For purposes of spectral analysis, consider a series of microarray expression values for gene <italic>x </italic>with <italic>N </italic>samples of the form</p>", "<p></p>", "<p>This series can be converted from time-domain, where each variable represents a measurement in time to a frequency domain using Discrete Fourier Transform (DFT) algorithm. Frequency domain representation of the series of experiments is also known as periodogram, which can be denoted by <italic>I</italic>(<italic>ω</italic>) :</p>", "<p></p>", "<p>If a time series has a significant sinusoidal component with frequency <italic>ω </italic>∈ [0, <italic>π</italic>], then the periodogram exhibits a peak at that frequency with a high probability. Conversely, if the time series is a purely random process (a.k.a \"white noise\"), then the plot of the periodogram against the Fourier frequencies approaches a straight line [##UREF##1##25##].</p>", "<title>Fisher's g-test</title>", "<p>The significance of the observed periodicity can be estimated by Fisher <italic>g</italic>-statistics, as recently recommended in [##REF##14693803##14##]. Fisher derived an exact test of the maximum periodogram coordinate by introducing the <italic>g</italic>-statistic</p>", "<p></p>", "<p>where <italic>I</italic>(<italic>ω</italic>_{<italic>k</italic>}) is a <italic>k-</italic>th peak of the periodogram. Large values of g indicate a non-random periodicity. We calculate the <italic>p</italic>-value of the test under the null hypothesis with the exact distribution of <italic>g </italic>using the following formula:</p>", "<p></p>", "<p>where <italic>n </italic>= [<italic>N/</italic>2] and <italic>p </italic>is the largest integer less than 1/<italic>x</italic>.</p>", "<p>This algorithm closely follows the guidelines recommended for analysis of periodicities in time-series microarray data [##REF##14693803##14##] with the exception that we applied a locally developed C++ code instead of R scripts.</p>", "<title>Autocorrelation</title>", "<p>For a given a discrete time series <italic>Y </italic>= <italic>x</italic>₀, <italic>x</italic>₁, <italic>x</italic>₂, <italic>x</italic>_{<italic>N</italic>-1}the autocorrelation is simply the correlation of the expression profile against itself with a frame shift of <italic>k </italic>data points (where 0 ≤ <italic>k </italic>≤ <italic>N </italic>- 1, often referred as the lag). For the time shift <italic>f</italic>, defined as <italic>f </italic>= <italic>i </italic>+ <italic>k </italic>if <italic>i+k&lt;N </italic>and <italic>f </italic>= <italic>i </italic>+ <italic>k </italic>- <italic>N </italic>otherwise</p>", "<p></p>", "<p>For each time series we calculate the maximum positive <italic>R(f) </italic>among all possible phase shifts <italic>f </italic>and use tabulated 0.05 significance cutoff values for correlation coefficient. Time series that shows significant autocorrelation <italic>R(f) </italic>with the lag <italic>f </italic>corresponding to one day (6 time points) are considered circadially expressed.</p>", "<title>Pt-test</title>", "<p>Consider a time series <italic>Y </italic>= <italic>x</italic>₀, <italic>x</italic>₁, <italic>x</italic>₂, ... <italic>x</italic>_{<italic>N</italic>-1}in which technical variation approaches or even exceeds the amplitude of periodic expression. In a very short time series stochastic noise often obscures periodicity. However, the periodic change of the base expression level can still be identified in spite of the high noise level. If the periodogram of the original time series <italic>IY(ω) </italic>contains a significant peak corresponding to a particular frequency (for example, circadian) this peak results from observation is the <italic>Y</italic>. A random permutation would preserve the same noise level, but not the periodicity. Let <italic>YR </italic>be a random permutation of the time series <italic>Y</italic>. Its corresponding periodogram is <italic>IR(ω)</italic>. After DFT a periodogram <italic>IR(ω) </italic>would represent only the peaks occurring by chance. However it will miss the true periodic frequencies unless permutations happen to preserve the period, for example if the rank of each point <italic>x </italic>in permutated series <italic>YR </italic>is equal <italic>x</italic>_{<italic>Y </italic>}± <italic>n </italic>* <italic>p </italic>where <italic>n </italic>is a natural number and <italic>p </italic>is a period corresponding to a significant peak in <italic>IY(ω)</italic>. To avoid random re-institution of periodicity we generate <italic>YR </italic>by multiple shuffling of randomly selected time points <italic>x</italic>_{<italic>n </italic>}⇔ <italic>x</italic>_{<italic>m</italic>}, where |<italic>n </italic>- <italic>m</italic>| ≠ <italic>p</italic>, i.e. each shuffle is swaps time points from different phase. Comparing permutations with deliberately wiped out periodicity to the original time series we can estimate whether a particular order of observations (i.e. time series) is important. For each gene expression profile we generate two series of <italic>min(n!,100) </italic>random permutations. Each permutated series <italic>YR </italic>is transformed to the frequency domain and a single peak of the periodogram <italic>IR(ω) </italic>is stored. The p-value for the null-hypothesis of random nature of a particular peak of periodogram can be estimated by comparing the stored <italic>IR(ω) </italic>values to the observed <italic>I(ω)</italic>:</p>", "<p></p>", "<p>High <italic>p</italic>-value exceeding the threshold, for example 0.05, means that at least 5 out of 100 random permutations of time series produce a periodogram with the same or higher peak, corresponding to a given periodicity. Low <italic>p</italic>-values indicate a significant difference between periodogram <italic>IR(ω) </italic>preserving circadian periodicity and randomly permutated periodogram <italic>IY(ω) </italic>with the same level of technical variation. This difference leads to rejection of the null-hypothesis of purely random nature of variation in the original time series <italic>Y</italic>.</p>", "<title>Phase continuum</title>", "<p>We start with phase classification, assigning each gene a phase based on maximal correlation to an ideal cosine curve. This method is superior to assigning a phase by position of peaks only because it takes into account more data. Each profile is subjected to z-score transformation equalizing the variation between time points. For each profile autocorrelation with circadian lag (<italic>R</italic>_{<italic>c</italic>}) is calculated and all profiles are sorted first by phase then by descending order of <italic>R</italic>_{<italic>c</italic>}. Concatenating all profiles of the same phase with equalized range of variation (amplitude) we generate a continuous stream <italic>C</italic>_{<italic>ph </italic>}of measurements containing a clear signal on one end and stochastic noise on the other. This continuum is treated with low-pass frequency filter and polynomial smoothing. We analyze each phase fraction separately to detect the point at which circadian signal deteriorates beyond p = 0.05 significance cutoff. A window W moving along the stream is transformed to frequency domain using Discrete Fourier Transform (DFT). The resulting periodogram <italic>I</italic>_{<italic>w </italic>}is compared a periodogram of a randomly permutated <italic>W</italic>_{<italic>r </italic>}using Kolmogorov-Smirnov goodness of fit test. Once the point at which <italic>I</italic>_{<italic>w </italic>}does not differ significantly from a random periodogram <italic>I</italic>_{<italic>wr </italic>}is detected, we count all original gene expression profiles that have circadian signal above the established cutoff [##REF##17571920##17##].</p>", "<title>False Discovery Rate analysis</title>", "<p>this methodology often applied to reduce the number of false-positive tests is based in the assumption of independent or mildly dependent [##REF##12883005##26##] hypothesis testing. However, in case of testing timeline expression profiles for periodicity independence could not be assumes for a number of reasons. First, the pattern of circadian oscillation is obvious in the great majority of expression profiles as seen on heatmaps (Figure ##FIG##0##1##, for example). Second, analysis of correlation with phase shift (also used to identify phase groups) confirms high correlation of nearly all profiles to common cosine curves. Third, living cells are known to have more than one oscillator, but these oscillators are normally synchronized to the rhythm of the circadian molecular clock, active in peripheral tissues. Testing individual expression profiles for periodicity we are looking for manifestation of the same factor, hence not independent hypothesis. For these reasons FDR correction has not been applied to reduce the number of detected oscillating genes.</p>" ]

[ "<title>Results and discussion</title>", "<title>Overview of the analysis strategy</title>", "<p>Independent circadian studies in plants or animals rarely use exactly the same analysis pipeline. However, comparing a representative set of studies [##REF##14712921##10##, ####REF##16166217##11##, ##REF##14734811##12##, ##REF##11967526##13##, ##REF##14693803##14##, ##REF##16254148##15####16254148##15##] reveals a commonality in strategy. The most typical approach starts with the normalization and scaling of microarray experiments; then the data is filtered, and only the genes that are present at least <italic>n </italic>times throughout the complete timeline (i.e., the exact number varies) are selected for further analysis. Some studies [##REF##12015981##16##] introduced an additional filter that selects only \"actively expressed\" genes, i.e., genes with an expression level that is estimated above some arbitrary threshold. This much reduced subset of transcripts is subjected to the periodicity test and, in rare cases, a panel of more than one test, followed by a false discovery rate (FDR) correction. The few genes that pass the test are further analyzed to determine the phase and the amplitude of oscillation and to visualize using profile plots and heat maps. This approach produces consistent results across a number of circadian data sets from diverse origins but also shares a common set of problems. First of all, the formulation of the null hypothesis for the statistical tests is based on the assumption of an absence of periodicity, i.e., a steady line rate of transcription for the majority of the genes. This assumption is intuitive but has no foundation in biology. Second, each gene (transcript) is tested independently. On the other hand, the authors realize that that the researchers are looking for a manifestation of the same rhythm that modulates expression of different genes and that this expression is expected to be highly correlated. Another problem common for all circadian studies is that microarray expression profiles are very expensive to generate. Additionally, even the best data sets count two consecutive circadian periods at most with samples collected every four hours. Such a low sampling rate combined with a high level of stochastic noise, which is also typical for microarray estimation of gene expression, makes testing for periodicity particularly challenging.</p>", "<p>A series of papers that we have published since 2006 have reported new algorithms for the analysis of periodicity in gene expression, including a new statistical test for periodicity, a phase classification, an application of digital signal processing, and an analysis of same-phase groups of genes as a continuous signal [##REF##17118131##8##,##REF##18365000##9##,##REF##17571920##17##,##REF##16532060##18##]. These algorithms were instrumental in the characterization of circadian expression in peripheral tissues [##REF##16567517##6##,##REF##17144790##7##], the discovery of a baseline oscillation in the transcript abundance of all genes [##REF##17571920##17##], the discovery of alternative transcripts oscillating with a phase shift [##REF##18047714##19##], and the discovery of the extra-low expression of eukaryotic genes [##REF##18365000##9##]. However, all of our studies have been conducted with murine (circadian) and yeast (metabolic oscillation) models, which obviously does not include plants in the scope of our investigations. This study aims to rectify this shortcoming and determine if our previous findings of pervasive and persistent circadian oscillation in the murine transcriptome are also true for plants.</p>", "<p>Our analysis starts with the same preprocessing normalizing and scaling microarray experiment in a time series. Then a provisional phase of oscillation is assigned to each gene. This is done before any selection or testing for periodicity; thus, a provisional or \"most likely\" phase can be potentially assigned to a non-oscillating noisy expression profile. However, assigning a phase does not introduce change in the data and thus does not preclude non-oscillating profiles from being filtered out in a later step. For further analysis, expression profiles are grouped into classes based on the provisionally assigned phase. In each group, profiles are joined into the phase continuum, which maximizes the statistical power in testing for periodicity and allows the application of digital filters [##REF##17571920##17##]. Our methods also do not attempt the impossible, i.e., aligning all noisy profiles by peaks at particular time point. We use only as many phase classes as possible by generating an artificial cosine curve with the given length of observation, which is typically two complete daily periods, and given sampling rate, which is typically one sample taken every four hours. This strategy of analysis is applicable to a wide variety of data and has been tested on multiple data sets that were produced by collaborators at the Pennington Biomedical Research Center as well as independent data obtained from the public databases and kindly provided by the respective authors. The detailed description of the algorithms that were used in each step of the analysis is given in the Materials and Methods section.</p>", "<p>Among the circadian gene expression data that is available from the GEO (Gene Expression Omnibus, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/geo\"/>), only two sets have a sufficient sampling rate and use a contemporary microarray platform (i.e., Affymetrix <italic>Arabidopsis </italic>expression array). For convenience in this paper, these data sets will be named by the academic affiliation of the majority of the authors, i.e., Davis [##REF##17683202##1##] and Warwick [##REF##16473970##4##] data sets. Unfortunately, no single experiment measures gene expression in the natural, undisturbed form. Both examine which data we reanalyze to collect samples in a constant light. The idea behind such an experimental design was to isolate the genes that are regulated by the molecular clock by presuming that all other genes will experience no oscillation without environmental cues. A brief description of the experimental conditions producing these data sets is given in the Materials and Methods section.</p>", "<p>Our results show that the efforts on the part of the original authors to isolate a small number of genes did not result in the intended outcome. In both data sets, the baseline circadian oscillation is statistically significant and visually detectable in practically 100% of all genes. The overview of the patterns dominating gene expression in plants is given in Figure ##FIG##0##1##. A straightforward application of a Pt-test [##REF##17118131##8##] to one gene at a time identifies 8,639 transcripts (i.e., ~39% of all genes that were examined by the microarray) as oscillating in the Davis data set and 10,001 transcripts (i.e., ~44%) as oscillating in the Warwick data set with the p-value cutoff at 0.05. A less noise-tolerant autocorrelation method identifies circadian oscillation in 3,351 transcripts (i.e., ~15%) in the Davis data set and 4,324 (i.e., ~19%) in the Warwick data set. An application of the Fisher's g-test in the same setting with the same significance cutoff identifies 3,497 (i.e., ~15%) in the Davis data set and 3,918 (i.e., ~17%) in the Warwick data set. Not surprisingly, the Pt-test, which was specifically developed for a short time series with a low sampling rate, outperforms the other algorithms. However, the older algorithms report the numbers of rhythmic transcripts that are generally in agreement with those published by Edwards et al. [##REF##16473970##4##], i.e., 3,505 or approximately 15% of all transcripts that are represented on the microarray. Using the same COSOPT approach [##REF##15063650##20##], Covington and Harmer [##REF##17683202##1##] have identified only 1,610 rhythmic transcripts (i.e., ~7%) in the Davis data set with the same significance cutoff. On one hand, this is consistent with the general observation that all methods tend to identify more rhythmic transcripts in the Warwick data set, which is probably because it has better general signal to noise ratio across all probe sets. On the other hand, such a sharp drop in the number of identified rhythmic transcripts as compared to the algorithms that are applied in this paper may indicate low robustness in the cosine curve fitting in the time domain that is employed by COSOPT; i.e., a little extra noise causes a large decrease in performance. This was one of the motivations for the development of the permutation test for periodicity, which considers only one peak in frequency domain (i.e., a periodogram), thereby making the algorithm more tolerable to noise that contributes to all other peaks [##REF##17118131##8##].</p>", "<p>In spite of the similarities in the experimental conditions, relatively few transcripts are identified as oscillating in the same phase between the two data sets. The diagram of the overlapping phase groups is presented in Figure ##FIG##1##2## and Table ##TAB##0##1##. Roughly two thirds of all transcripts are out of sync between the Warwick and Davis experiments, probably reflecting some minor differences in the environment. These genes are also possibly less important for the plant response to environmental cues and are not directly linked to the circadian clock while still modulated by oscillation in other genes. However, this hypothesis should be corroborated by further studies. This observation is consistent with previous observations in mouse gene expression: phase of expression is volatile and often varies between tissues and experimental conditions [##REF##16567517##6##,##REF##17118131##8##,##REF##16532060##18##]. On the other hand, low sampling rate of microarray experiments also curbs the precision of phase assignment.</p>", "<p>Results of the analysis in a classic \"one gene at a time\" approach are generally in agreement with each other and seem to reflect some of the patterns in the data. However, all of these methods are in acute contradiction with the results that are presented in Figure ##FIG##0##1##. The pattern in Figure ##FIG##0##1## shows exactly two red areas of elevated transcript abundance interspaced with two green areas of lower transcript abundance, and this pattern does not stop on a fraction of 7%, 15%, or even 44%; it involves all or nearly all of the genes. The phase continuum approach [##REF##17571920##17##] applies statistical testing to separate rhythmic transcripts from stochastic ones. This analysis shows remarkable agreement with the intuitively detected pattern but relies on quantitative methodology. In both the Davis and Warwick data sets, this method reported a detectable baseline circadian oscillation in 100% of all transcripts. This number does not exclude any transcript, not even those never considered present or expressed by GC-RMA (i.e., Warwick) [##REF##16410320##21##] or Affymetrix MAS5.0 algorithms. Such rhythmic behavior of the \"non-present\" ghost transcripts has been recently reported based on studies of animal and yeast data [##REF##18365000##9##] and is also supported by experimental studies [##REF##17632057##22##]. Genes that are expressed below the resolution ability (i.e., presence call) for the current microarray technology are not silent; they are expressed at a low level but respond to the changing cellular and external factors. Additionally, they interact with the other related genes in biological pathways. This study confirms that the same is true for plant genes. A separate view of the oscillating pattern in low-expressed (i.e., not called present) genes is depicted in Figure ##FIG##2##3##.</p>", "<p>Because oscillation is so pervasive, it affects not some but all biological pathways. The nitrogen reduction metabolic pathway, which is shown in Figure ##FIG##3##4##, provides an example. The known components of the pathway are traced to the probe sets in the Davis data, and their expression profiles are overlapped with the KEGG pathway map [##REF##18077471##23##]. Even though the Davis data is noisier, a circadian oscillation pattern with two humps over two days of observation is apparent in most expression profiles. Remarkably, a few components of the nitrogen reduction pathway have alternative probe sets that oscillate with a phase shift or directly in counter-phase to each other. This phenomenon has already been reported with animal data [##REF##18047714##19##] as a possible molecular mechanism compensating for constant oscillation and creating a steady transcript abundance over time, thus providing a steady translation rate and stabilizing the volume of signal transduction at any time of the day. A similar pattern of expression in alternative probe sets for the same gene may have a similar explanation as in <italic>Arabidopsis</italic>. A steady abundance of the transcripts that is created by alternating transcripts with a different survival time creates a steady production of enzymes that are required for basic cellular metabolic function at all times.</p>", "<p>While all genes oscillate, all genes are not necessarily specifically regulated to create oscillation. Also, oscillation is not likely determined by the function of each particular gene. In the dynamic cellular environment, all components experience a baseline oscillation expression rate, and the transcript abundance of each gene at any given time is relative to some other genes. These genes are also oscillating. The presence of a fraction of constantly expressed non-oscillating genes is unlikely [##REF##16532060##18##]. Oscillation is simply imposed on all genes, modulating every cellular process. The illustration of this point is presented in Figure ##FIG##4##5##. The expression profiles for the major components of the basal transcription machinery (picture template taken from KEGG) are also traced to their respective probe sets in the Davis data set. Since the microarray data carries a lot of stochastic technical variation, the profiles may deviate from the ideal cosine curve. However, the circadian pattern with two peaks over two days is clearly visible in the majority of plots and is at least consistent with the others. Notably, even the TATA-binding protein expression is explicitly circadian, which is bound to affect many other transcripts.</p>", "<p>The data obtained in two independent studies that were conducted at different times and in opposite hemispheres of the globe are very similar in the general pattern but exhibit some differences in the observed phase and the amplitude of some genes. The experimental design, though described in different words, is almost identical. Differences may be possible outside the methods described in the published papers, but the only significant differences seem to be in the time selected for the starting point (ZT, zeitgeber time) at subjective dawn, although not large, and the technique used to quantify expression values for microarrays. The Davis data set also has lower overall intensity and twice as many genes that are deemed absent as compared to the Warwick data. The latter can possibly explain the difference in signal to noise ratio between almost identically designed experiments. In both studies, the attempt to stop oscillation in the entire transcriptome by removing environmental oscillation (i.e., light) proved futile. In animal studies, changing the lighting regime from oscillating to constant darkness or dim light creates asynchrony between feeding, sleeping, and other activity patterns. As a result, a significant number of transcripts loose synchronization and identifying the baseline oscillation becomes more challenging. From a glance, the plant transcriptome data that was collected in constant light looks like the mouse transcriptome data that was collected under normal conditions with no alternation in the environment. Unfortunately, we do not have plant data that was collected under normal lighting for comparison. However, the robust oscillation under constant light in both the Davis and Warwick data leaves little space for a change. Leveling a single rhythmic environmental factor makes little impression on the pattern of gene expression in plants.</p>", "<p>The authors of the publications that contribute to the Davis and Warwick data sets are referencing one another's works and are aware of some discrepancies, particularly in the number of rhythmic or \"circadially regulated\" genes. Covington and Harmer find a prevalence of higher than 10% of circadially-regulated genes intriguing and thus possible. However, neither of these research teams allowed for the possibility of all genes being expressed in circadian rhythm. This finding undermines the results of the studies that follow the separation of a small portion of rhythmic transcripts through both the analyses of over-represented pathways and the role of the molecular clock in specific pathways. Much of the results and discussion presented in these papers are based on the intuitive but unfounded assumption that all genes are expressed in a steady line pattern. Unfortunately, in the light of knowing that all genes are oscillating in a circadian pattern, these findings will have to be revised. On the other hand, the circadian timeline data that were collected for these studies are invaluable. These data could be an endless source of discovery. However, the analysis should be considered from a different angle, i.e., not whether a particular gene, co-regulated genes, or pathways are circadially-regulated but how changing experimental conditions affects oscillating properties, such as the phase and amplitude of the genes.</p>" ]

[ "<title>Results and discussion</title>", "<title>Overview of the analysis strategy</title>", "<p>Independent circadian studies in plants or animals rarely use exactly the same analysis pipeline. However, comparing a representative set of studies [##REF##14712921##10##, ####REF##16166217##11##, ##REF##14734811##12##, ##REF##11967526##13##, ##REF##14693803##14##, ##REF##16254148##15####16254148##15##] reveals a commonality in strategy. The most typical approach starts with the normalization and scaling of microarray experiments; then the data is filtered, and only the genes that are present at least <italic>n </italic>times throughout the complete timeline (i.e., the exact number varies) are selected for further analysis. Some studies [##REF##12015981##16##] introduced an additional filter that selects only \"actively expressed\" genes, i.e., genes with an expression level that is estimated above some arbitrary threshold. This much reduced subset of transcripts is subjected to the periodicity test and, in rare cases, a panel of more than one test, followed by a false discovery rate (FDR) correction. The few genes that pass the test are further analyzed to determine the phase and the amplitude of oscillation and to visualize using profile plots and heat maps. This approach produces consistent results across a number of circadian data sets from diverse origins but also shares a common set of problems. First of all, the formulation of the null hypothesis for the statistical tests is based on the assumption of an absence of periodicity, i.e., a steady line rate of transcription for the majority of the genes. This assumption is intuitive but has no foundation in biology. Second, each gene (transcript) is tested independently. On the other hand, the authors realize that that the researchers are looking for a manifestation of the same rhythm that modulates expression of different genes and that this expression is expected to be highly correlated. Another problem common for all circadian studies is that microarray expression profiles are very expensive to generate. Additionally, even the best data sets count two consecutive circadian periods at most with samples collected every four hours. Such a low sampling rate combined with a high level of stochastic noise, which is also typical for microarray estimation of gene expression, makes testing for periodicity particularly challenging.</p>", "<p>A series of papers that we have published since 2006 have reported new algorithms for the analysis of periodicity in gene expression, including a new statistical test for periodicity, a phase classification, an application of digital signal processing, and an analysis of same-phase groups of genes as a continuous signal [##REF##17118131##8##,##REF##18365000##9##,##REF##17571920##17##,##REF##16532060##18##]. These algorithms were instrumental in the characterization of circadian expression in peripheral tissues [##REF##16567517##6##,##REF##17144790##7##], the discovery of a baseline oscillation in the transcript abundance of all genes [##REF##17571920##17##], the discovery of alternative transcripts oscillating with a phase shift [##REF##18047714##19##], and the discovery of the extra-low expression of eukaryotic genes [##REF##18365000##9##]. However, all of our studies have been conducted with murine (circadian) and yeast (metabolic oscillation) models, which obviously does not include plants in the scope of our investigations. This study aims to rectify this shortcoming and determine if our previous findings of pervasive and persistent circadian oscillation in the murine transcriptome are also true for plants.</p>", "<p>Our analysis starts with the same preprocessing normalizing and scaling microarray experiment in a time series. Then a provisional phase of oscillation is assigned to each gene. This is done before any selection or testing for periodicity; thus, a provisional or \"most likely\" phase can be potentially assigned to a non-oscillating noisy expression profile. However, assigning a phase does not introduce change in the data and thus does not preclude non-oscillating profiles from being filtered out in a later step. For further analysis, expression profiles are grouped into classes based on the provisionally assigned phase. In each group, profiles are joined into the phase continuum, which maximizes the statistical power in testing for periodicity and allows the application of digital filters [##REF##17571920##17##]. Our methods also do not attempt the impossible, i.e., aligning all noisy profiles by peaks at particular time point. We use only as many phase classes as possible by generating an artificial cosine curve with the given length of observation, which is typically two complete daily periods, and given sampling rate, which is typically one sample taken every four hours. This strategy of analysis is applicable to a wide variety of data and has been tested on multiple data sets that were produced by collaborators at the Pennington Biomedical Research Center as well as independent data obtained from the public databases and kindly provided by the respective authors. The detailed description of the algorithms that were used in each step of the analysis is given in the Materials and Methods section.</p>", "<p>Among the circadian gene expression data that is available from the GEO (Gene Expression Omnibus, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/geo\"/>), only two sets have a sufficient sampling rate and use a contemporary microarray platform (i.e., Affymetrix <italic>Arabidopsis </italic>expression array). For convenience in this paper, these data sets will be named by the academic affiliation of the majority of the authors, i.e., Davis [##REF##17683202##1##] and Warwick [##REF##16473970##4##] data sets. Unfortunately, no single experiment measures gene expression in the natural, undisturbed form. Both examine which data we reanalyze to collect samples in a constant light. The idea behind such an experimental design was to isolate the genes that are regulated by the molecular clock by presuming that all other genes will experience no oscillation without environmental cues. A brief description of the experimental conditions producing these data sets is given in the Materials and Methods section.</p>", "<p>Our results show that the efforts on the part of the original authors to isolate a small number of genes did not result in the intended outcome. In both data sets, the baseline circadian oscillation is statistically significant and visually detectable in practically 100% of all genes. The overview of the patterns dominating gene expression in plants is given in Figure ##FIG##0##1##. A straightforward application of a Pt-test [##REF##17118131##8##] to one gene at a time identifies 8,639 transcripts (i.e., ~39% of all genes that were examined by the microarray) as oscillating in the Davis data set and 10,001 transcripts (i.e., ~44%) as oscillating in the Warwick data set with the p-value cutoff at 0.05. A less noise-tolerant autocorrelation method identifies circadian oscillation in 3,351 transcripts (i.e., ~15%) in the Davis data set and 4,324 (i.e., ~19%) in the Warwick data set. An application of the Fisher's g-test in the same setting with the same significance cutoff identifies 3,497 (i.e., ~15%) in the Davis data set and 3,918 (i.e., ~17%) in the Warwick data set. Not surprisingly, the Pt-test, which was specifically developed for a short time series with a low sampling rate, outperforms the other algorithms. However, the older algorithms report the numbers of rhythmic transcripts that are generally in agreement with those published by Edwards et al. [##REF##16473970##4##], i.e., 3,505 or approximately 15% of all transcripts that are represented on the microarray. Using the same COSOPT approach [##REF##15063650##20##], Covington and Harmer [##REF##17683202##1##] have identified only 1,610 rhythmic transcripts (i.e., ~7%) in the Davis data set with the same significance cutoff. On one hand, this is consistent with the general observation that all methods tend to identify more rhythmic transcripts in the Warwick data set, which is probably because it has better general signal to noise ratio across all probe sets. On the other hand, such a sharp drop in the number of identified rhythmic transcripts as compared to the algorithms that are applied in this paper may indicate low robustness in the cosine curve fitting in the time domain that is employed by COSOPT; i.e., a little extra noise causes a large decrease in performance. This was one of the motivations for the development of the permutation test for periodicity, which considers only one peak in frequency domain (i.e., a periodogram), thereby making the algorithm more tolerable to noise that contributes to all other peaks [##REF##17118131##8##].</p>", "<p>In spite of the similarities in the experimental conditions, relatively few transcripts are identified as oscillating in the same phase between the two data sets. The diagram of the overlapping phase groups is presented in Figure ##FIG##1##2## and Table ##TAB##0##1##. Roughly two thirds of all transcripts are out of sync between the Warwick and Davis experiments, probably reflecting some minor differences in the environment. These genes are also possibly less important for the plant response to environmental cues and are not directly linked to the circadian clock while still modulated by oscillation in other genes. However, this hypothesis should be corroborated by further studies. This observation is consistent with previous observations in mouse gene expression: phase of expression is volatile and often varies between tissues and experimental conditions [##REF##16567517##6##,##REF##17118131##8##,##REF##16532060##18##]. On the other hand, low sampling rate of microarray experiments also curbs the precision of phase assignment.</p>", "<p>Results of the analysis in a classic \"one gene at a time\" approach are generally in agreement with each other and seem to reflect some of the patterns in the data. However, all of these methods are in acute contradiction with the results that are presented in Figure ##FIG##0##1##. The pattern in Figure ##FIG##0##1## shows exactly two red areas of elevated transcript abundance interspaced with two green areas of lower transcript abundance, and this pattern does not stop on a fraction of 7%, 15%, or even 44%; it involves all or nearly all of the genes. The phase continuum approach [##REF##17571920##17##] applies statistical testing to separate rhythmic transcripts from stochastic ones. This analysis shows remarkable agreement with the intuitively detected pattern but relies on quantitative methodology. In both the Davis and Warwick data sets, this method reported a detectable baseline circadian oscillation in 100% of all transcripts. This number does not exclude any transcript, not even those never considered present or expressed by GC-RMA (i.e., Warwick) [##REF##16410320##21##] or Affymetrix MAS5.0 algorithms. Such rhythmic behavior of the \"non-present\" ghost transcripts has been recently reported based on studies of animal and yeast data [##REF##18365000##9##] and is also supported by experimental studies [##REF##17632057##22##]. Genes that are expressed below the resolution ability (i.e., presence call) for the current microarray technology are not silent; they are expressed at a low level but respond to the changing cellular and external factors. Additionally, they interact with the other related genes in biological pathways. This study confirms that the same is true for plant genes. A separate view of the oscillating pattern in low-expressed (i.e., not called present) genes is depicted in Figure ##FIG##2##3##.</p>", "<p>Because oscillation is so pervasive, it affects not some but all biological pathways. The nitrogen reduction metabolic pathway, which is shown in Figure ##FIG##3##4##, provides an example. The known components of the pathway are traced to the probe sets in the Davis data, and their expression profiles are overlapped with the KEGG pathway map [##REF##18077471##23##]. Even though the Davis data is noisier, a circadian oscillation pattern with two humps over two days of observation is apparent in most expression profiles. Remarkably, a few components of the nitrogen reduction pathway have alternative probe sets that oscillate with a phase shift or directly in counter-phase to each other. This phenomenon has already been reported with animal data [##REF##18047714##19##] as a possible molecular mechanism compensating for constant oscillation and creating a steady transcript abundance over time, thus providing a steady translation rate and stabilizing the volume of signal transduction at any time of the day. A similar pattern of expression in alternative probe sets for the same gene may have a similar explanation as in <italic>Arabidopsis</italic>. A steady abundance of the transcripts that is created by alternating transcripts with a different survival time creates a steady production of enzymes that are required for basic cellular metabolic function at all times.</p>", "<p>While all genes oscillate, all genes are not necessarily specifically regulated to create oscillation. Also, oscillation is not likely determined by the function of each particular gene. In the dynamic cellular environment, all components experience a baseline oscillation expression rate, and the transcript abundance of each gene at any given time is relative to some other genes. These genes are also oscillating. The presence of a fraction of constantly expressed non-oscillating genes is unlikely [##REF##16532060##18##]. Oscillation is simply imposed on all genes, modulating every cellular process. The illustration of this point is presented in Figure ##FIG##4##5##. The expression profiles for the major components of the basal transcription machinery (picture template taken from KEGG) are also traced to their respective probe sets in the Davis data set. Since the microarray data carries a lot of stochastic technical variation, the profiles may deviate from the ideal cosine curve. However, the circadian pattern with two peaks over two days is clearly visible in the majority of plots and is at least consistent with the others. Notably, even the TATA-binding protein expression is explicitly circadian, which is bound to affect many other transcripts.</p>", "<p>The data obtained in two independent studies that were conducted at different times and in opposite hemispheres of the globe are very similar in the general pattern but exhibit some differences in the observed phase and the amplitude of some genes. The experimental design, though described in different words, is almost identical. Differences may be possible outside the methods described in the published papers, but the only significant differences seem to be in the time selected for the starting point (ZT, zeitgeber time) at subjective dawn, although not large, and the technique used to quantify expression values for microarrays. The Davis data set also has lower overall intensity and twice as many genes that are deemed absent as compared to the Warwick data. The latter can possibly explain the difference in signal to noise ratio between almost identically designed experiments. In both studies, the attempt to stop oscillation in the entire transcriptome by removing environmental oscillation (i.e., light) proved futile. In animal studies, changing the lighting regime from oscillating to constant darkness or dim light creates asynchrony between feeding, sleeping, and other activity patterns. As a result, a significant number of transcripts loose synchronization and identifying the baseline oscillation becomes more challenging. From a glance, the plant transcriptome data that was collected in constant light looks like the mouse transcriptome data that was collected under normal conditions with no alternation in the environment. Unfortunately, we do not have plant data that was collected under normal lighting for comparison. However, the robust oscillation under constant light in both the Davis and Warwick data leaves little space for a change. Leveling a single rhythmic environmental factor makes little impression on the pattern of gene expression in plants.</p>", "<p>The authors of the publications that contribute to the Davis and Warwick data sets are referencing one another's works and are aware of some discrepancies, particularly in the number of rhythmic or \"circadially regulated\" genes. Covington and Harmer find a prevalence of higher than 10% of circadially-regulated genes intriguing and thus possible. However, neither of these research teams allowed for the possibility of all genes being expressed in circadian rhythm. This finding undermines the results of the studies that follow the separation of a small portion of rhythmic transcripts through both the analyses of over-represented pathways and the role of the molecular clock in specific pathways. Much of the results and discussion presented in these papers are based on the intuitive but unfounded assumption that all genes are expressed in a steady line pattern. Unfortunately, in the light of knowing that all genes are oscillating in a circadian pattern, these findings will have to be revised. On the other hand, the circadian timeline data that were collected for these studies are invaluable. These data could be an endless source of discovery. However, the analysis should be considered from a different angle, i.e., not whether a particular gene, co-regulated genes, or pathways are circadially-regulated but how changing experimental conditions affects oscillating properties, such as the phase and amplitude of the genes.</p>" ]

[]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Circadian rhythm is a crucial factor in orchestration of plant physiology, keeping it in synchrony with the daylight cycle. Previous studies have reported that up to 16% of plant transcriptome are circadially expressed.</p>", "<title>Results</title>", "<p>Our studies of mammalian gene expression revealed circadian baseline oscillation in nearly 100% of genes. Here we present a comprehensive analysis of periodicity in two independent data sets. Application of the advanced algorithms and analytic approached already tested on animal data reveals oscillation in almost every gene of <italic>Arabidopsis thaliana</italic>.</p>", "<title>Conclusion</title>", "<p>This study indicates an even more pervasive role of oscillation in molecular physiology of plants than previously believed. Earlier studies have dramatically underestimated the prevalence of circadian oscillation in plant gene expression.</p>" ]

[ "<title>Competing interests</title>", "<p>The author declares that they have no competing interests.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Overview of the circadian pattern of gene expression in <italic>Arabidopsis thaliana</italic>. All expression profiles are grouped into four separate phase classes by correlation to a cosine curve with the same sampling rate (12 points over 2 complete periods). Profiles are then sorted by decreasing autocorrelation with circadian shift. The pattern of two areas of elevated gene expression (red) interspaced by the areas of lowered gene expression (green areas) reflecting two complete cycles over two days of observation is clearly visible throughout entire data. The heatmaps show all transcripts represented on microarray (over 22,000 probesets).</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Overlap in phase of oscillation between two plant circadian data sets. While practically all genes have baseline oscillation, relatively few genes (34%) are found oscillating in the same phase between Warwick and Davis data sets. The diagram shows the absolute number (above) and percentage (rounded up, below) of transcripts oscillating in the same (overlapping area) and in different phases in each phase of four phase groups.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Overview of circadian oscillation pattern in genes with \"absent\" call. The heatmaps are produced using the same algorithm as for Figure 1. However, only the probesets called \"absent\" at all 12 time points are considered. In spite of being expressed below the estimated noise level in each microarray experiment, these transcripts show the same circadian expression pattern (two elevated expression periods spaced by two lower expression periods over two days) as conventionally detectable transcripts.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>Circadian oscillation in nitrogen metabolism pathway. Expression profiles of microarray probesets are overlapped with the Kyoto Encyclopedia of Genes and Genomes (KEGG) map of the nitrogen metabolism in <italic>Arabidopsis thaliana</italic>. Some genes are represented by more than one set of alternative probes. Expression of the NIA2 (Nitrate Reductase 2, EC:1.7.1.1) at the start of the cascade shows the evidence of alternative probes (and thus most likely alternative transcripts) oscillating in counter-phase. Green color marks the elements of the pathway for which KEGG has additional information accessible by a clickable link and for which Affymetrix probesets are traced in the Davis data set.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>Circadian oscillation of the elements of basal transcription complex of <italic>Arabidopsis thaliana </italic>(Davis data set). All elements of transcription are expressed in oscillating pattern including TATA-binding protein (TBP).</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Difference between phase groups in Davis and Warwick data sets.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td/><td align=\"center\" colspan=\"4\"><bold>Warwick data set</bold></td></tr><tr><td/><td/><td colspan=\"4\"><hr/></td></tr><tr><td/><td/><td align=\"center\"><bold>Phase I</bold></td><td align=\"center\"><bold>Phase II</bold></td><td align=\"center\"><bold>Phase III</bold></td><td align=\"center\"><bold>Phase IV</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>Davis data set</bold></td><td align=\"center\"><bold>Phase I</bold></td><td align=\"center\">733 3%</td><td align=\"center\">518 2%</td><td align=\"center\">607 3%</td><td align=\"center\">1624 7%</td></tr><tr><td/><td align=\"center\"><bold>Phase II</bold></td><td align=\"center\">1450 6%</td><td align=\"center\">2174 10%</td><td align=\"center\">1242 5%</td><td align=\"center\">1480 7%</td></tr><tr><td/><td align=\"center\"><bold>Phase III</bold></td><td align=\"center\">904 4%</td><td align=\"center\">1732 8%</td><td align=\"center\">2268 10%</td><td align=\"center\">1867 8%</td></tr><tr><td/><td align=\"center\"><bold>Phase IV</bold></td><td align=\"center\">817 4%</td><td align=\"center\">854 4%</td><td align=\"center\">1849 8%</td><td align=\"center\">2625 12%</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula><italic>Y </italic>= <italic>x</italic>₀, <italic>x</italic>₁, <italic>x</italic>₂, <italic>x</italic>_{<italic>N</italic>-1}</disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S18-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>I</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>ω</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mi>N</mml:mi>\n </mml:mfrac>\n <mml:msup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>t</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>0</mml:mn>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:munderover>\n <mml:mrow>\n <mml:msub>\n <mml:mi>x</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:msup>\n <mml:mi>e</mml:mi>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n <mml:mi>ω</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:msup>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mn>2</mml:mn>\n </mml:msup>\n <mml:mo>,</mml:mo>\n <mml:mi>ω</mml:mi>\n <mml:mo>∈</mml:mo>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:mn>0</mml:mn>\n <mml:mo>,</mml:mo>\n <mml:mi>π</mml:mi>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S18-i2\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>g</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mrow>\n <mml:mi>max</mml:mi>\n <mml:mo>⁡</mml:mo>\n </mml:mrow>\n <mml:mi>k</mml:mi>\n </mml:msub>\n <mml:mi>I</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:msub>\n <mml:mi>ω</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mrow>\n <mml:mrow>\n <mml:mi>N</mml:mi>\n <mml:mo>/</mml:mo>\n <mml:mn>2</mml:mn>\n </mml:mrow>\n </mml:mrow>\n </mml:msubsup>\n <mml:mrow>\n <mml:mi>I</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:msub>\n <mml:mi>ω</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msub>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S18-i3\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>P</mml:mi>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mi>g</mml:mi>\n <mml:mo>&gt;</mml:mo>\n <mml:mi>x</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mrow>\n <mml:mfrac bevelled=\"true\">\n <mml:mn>1</mml:mn>\n <mml:mi>x</mml:mi>\n </mml:mfrac>\n </mml:mrow>\n </mml:munderover>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>[</mml:mo>\n <mml:mrow>\n <mml:msup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mi>p</mml:mi>\n </mml:msup>\n <mml:mfrac>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>!</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo>!</mml:mo>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>p</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mo>!</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n <mml:msup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mn>1</mml:mn>\n <mml:mo>−</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mi>x</mml:mi>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msup>\n </mml:mrow>\n <mml:mo>]</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S18-i4\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>f</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mn>0</mml:mn>\n <mml:mrow>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:msub>\n <mml:mi>x</mml:mi>\n <mml:mtext>i</mml:mtext>\n </mml:msub>\n <mml:mo>−</mml:mo>\n <mml:mover accent=\"true\">\n <mml:mi>x</mml:mi>\n <mml:mo>¯</mml:mo>\n </mml:mover>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:msub>\n <mml:mi>x</mml:mi>\n <mml:mi>f</mml:mi>\n </mml:msub>\n <mml:mo>−</mml:mo>\n <mml:mover accent=\"true\">\n <mml:mi>x</mml:mi>\n <mml:mo>¯</mml:mo>\n </mml:mover>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:msubsup>\n <mml:mo>∑</mml:mo>\n <mml:mn>0</mml:mn>\n <mml:mrow>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n </mml:msubsup>\n <mml:mrow>\n <mml:msup>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:msub>\n <mml:mi>x</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo>−</mml:mo>\n <mml:mover accent=\"true\">\n <mml:mi>x</mml:mi>\n <mml:mo>¯</mml:mo>\n </mml:mover>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mn>2</mml:mn>\n </mml:msup>\n </mml:mrow>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S18-i5\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mi>N</mml:mi>\n <mml:mrow>\n <mml:mi>I</mml:mi>\n <mml:mi>R</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>ω</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>≥</mml:mo>\n <mml:mi>I</mml:mi>\n <mml:mi>Y</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>ω</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:msub>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>min</mml:mi>\n <mml:mo>⁡</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>n</mml:mi>\n <mml:mo>!</mml:mo>\n <mml:mo>,</mml:mo>\n <mml:mn>100</mml:mn>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>.</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Diagonal cells show the number and percentage (rounded up) of transcripts found in the same phase in both data sets. Cells off the diagonal show the number and percentage of transcripts oscillating in different phases.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S18-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S18-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S18-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S18-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S18-5\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Biological systems can be modeled as complex network systems with many interactions between the components. These interactions give essential information about the function and behavior of the analyzed system. For example, the protein-protein interaction network is the physical basis of multiple cellular functions. The first large-scale protein interaction studies were conducted on yeast [##REF##10688190##1##,##REF##11283351##2##], followed by more recent studies on the fly [##REF##14605208##3##] and the worm [##REF##14704431##4##]. The main goal of current network research in many diverse areas such as biology, social sciences and statistical physics is to understand and characterize underlying network structures. Such an understanding is vital for many systems, simply because <italic>structure always affects function </italic>[##REF##11258382##5##].</p>", "<p>As noted by Barabasi [##REF##14735121##6##], the most important discovery of network research in recent years is that many different complex network systems including biological networks demonstrate some significant common principles that govern their architecture, topology and behavior such as small-world property [##REF##11034217##7##,##UREF##0##8##], power-law degree distribution [##REF##11034217##7##] and highly modular structures [##REF##9623998##9##].</p>", "<p>Among other network analysis tasks network clustering plays a particular role as it helps us to detect modules or communities which are usually good indicators of structural or functional units of the underlying network.</p>", "<p>Various methods have been developed to find partitions in networks. These methods tend to partition networks such that there are a dense set of edges within every partition and few edges between partitions. Modularity-based algorithms [##REF##15729348##10##, ####UREF##1##11##, ##UREF##2##12####2##12##] and normalized cut [##UREF##3##13##,##UREF##4##14##] are widely used examples. However, most of the state-of-the-art algorithms have quadratic running time thus limited usage for large real networks.</p>", "<p>Recently we proposed a new network clustering algorithm, SCAN (Structural Clustering Algorithm for Networks) which runs linearly with the size of given network [##UREF##5##15##]. Despite the common methodology of current methods where maximization or minimization of edges within/across clusters is essential, it defines clusters based on structural similarity of vertices. Number of shared neighbors for two vertices basically defines their similarity, and vertices with similarity values above certain threshold are assigned to the same partition. Similarity definition that takes into account the neighborhood of a vertex becomes more reasonable when social networks are considered: people who share more friends are more likely to be friends each other and more likely to be members of same community.</p>", "<p>Additionally its capability to detect hubs and isolated nodes in the given network makes it a unique approach as no other available methods offer such a function.</p>", "<p>In this study, we show the effectiveness of our methodology using the budding yeast (Saccharomyces cerevisiae) protein-protein interaction network. To validate our clustering results, we compare our clusters with the known functions of each protein. Our predicted functional modules achieve very high clustering scores as compared to other state-of-the-art approaches.</p>" ]

[ "<title>Methods</title>", "<title>The notion of structure-connected clusters</title>", "<p>Our goal is to achieve an optimal clustering of the PPI network, as well as to identify hubs and outliers. Therefore, both connectivity and local structure are used in our definition of optimal clustering. In this section, we formulize the notion of a structure-connected cluster, which extends that of a density-based cluster [##UREF##8##23##] and can distinguish good clusters, hubs, and outliers in networks. In the next section, we present, SCAN, an efficient algorithm to find the optimal clustering of networks [##UREF##5##15##].</p>", "<title>Structure-connected clusters</title>", "<p>The existing network clustering methods are designed to find optimal clustering of networks based on the number of edges that run within or across clusters. Direct connections are important, but they represent only one aspect of the network structure. We believe that the neighborhood around two connected vertices is also important. The neighborhood of a protein includes all the vertices connected to it by an edge. When you consider a pair of connected vertices, their combined neighborhood reveals neighbors common to both vertices.</p>", "<p>Our method is based on common neighbors. Two vertices are assigned to a cluster according to how they share neighbors. This makes sense when you consider social communities. People who share many friends create a community, and the more friends they have in common, the more intimate the community. But in social networks there are different kinds of actors. There are also people who are outsiders (like hermits), and there are people who are friendly with many communities but belong to none (like politicians). The latter plays a special role in small-world networks as <italic>hubs </italic>[##REF##9623998##24##]. Such a hub is illustrated by protein <italic>G </italic>in Figure ##FIG##4##5##.</p>", "<p>More formally, we focus on simple, undirected and unweighted graphs. Let <italic>G </italic>= {<italic>V</italic>, <italic>E</italic>} be a graph, where <italic>V </italic>is a set of vertices -proteins in PPI; and <italic>E </italic>is set of unordered pairs of distinct vertices, called edges. Before formal presentation of SCAN, it is worth to give fundamental definitions exploited by the algorithm. We refer the reader to our previous work [##UREF##5##15##] for extensive formal discussions.</p>", "<p>The structure of a vertex is described by its neighborhood. A formal definition of vertex structure is given as follows.</p>", "<title>Definition 1 (Vertex Structure)</title>", "<p>Let <italic>v ∈ V</italic>, the structure of v is defined by its neighborhood, denoted by</p>", "<p></p>", "<p>Please note that neighborhood of protein <italic>v, Γ (v)</italic>, also includes v in addition to all neighbors of v. For instance, considering Figure ##FIG##4##5##, <italic>Γ (A) </italic>would be <italic>{A, B, E, F, G}</italic>. Having Definition 1, now we can formulize similarity function, which is run for every edge, <italic>{v, w} ∈ E</italic>, in the network. We call the similarity function structural similarity because it is solely derived from vertex structure <italic>Γ (v)</italic>. The structural similarity between two vertices is measured by normalized common neighbors, which is also called cosine similarity measure commonly used in information retrieval. If we only use the number of shared neighbors, hub vertices, such as <italic>G </italic>in Figure ##FIG##4##5##, will be clustered into either of the clusters or two clusters will be mistakenly merged. Therefore, we normalize number of common neighbors by the geometric mean of the two neighborhoods' size. Note that In Figure ##FIG##4##5##, protein <italic>G </italic>should be identified as a hub, shared in neighborhood of both clusters.</p>", "<title>Definition 2 (Structural Similarity)</title>", "<p></p>", "<p>When a member of cluster shares a similar structure with one of its neighbors, their computed structural similarity will be large. Obviously structural similarity is symmetric, <italic>σ (v, w) = σ (w, v)</italic>. Structural similarity between <italic>v </italic>and <italic>w, σ (v, w)</italic>, would be greater than zero if and only if v and w are vertices of an edge <italic>e ∈ E</italic>. Under this circumstance, structural similarity attains values between (0, 1]. However, structural similarity should be restricted to control expansion of the cluster. Therefore, we apply a threshold <italic>ε </italic>to the computed structural similarity when assigning cluster membership, formulized in the following <italic>ε</italic>-neighborhood definition.</p>", "<title>Definition 3 (<italic>ε</italic>-Neighborhood)</title>", "<p></p>", "<p>When a vertex accumulates enough neighbors in its <italic>ε</italic>-neighborhood, it becomes a nucleus or <italic>seed </italic>for a cluster. Such a vertex is called a core vertex. Core vertices are a special class of vertices that have a minimum of <italic>μ </italic>neighbors with a structural similarity that greater than or equals to the threshold <italic>ε</italic>. From core vertices we grow the clusters. In this way only the parameters <italic>μ </italic>and <italic>ε </italic>determine the clustering of networks. For a given <italic>ε</italic>, the minimal size of a cluster is determined by <italic>μ</italic>. If a vertex w is in <italic>ε</italic>-neighborhood of a core vertex <italic>v</italic>, vertex <italic>w </italic>should be included into the same cluster with vertex <italic>v</italic>. Because, they are connected and share a similar structure. This concept is known as <italic>direct structural reachability</italic>.</p>", "<p>Direct structural reachability is symmetric for any pair of cores. However, it is asymmetric if one of the vertices is not core. Also the property of direct structural reachability is basis for the cluster expansion. A newly formed cluster C consists of a core vertex v and v's <italic>ε</italic>-neighborhood. Then we try to expand cluster C through any vertex w in v's <italic>ε</italic>-neighborhood. This approach guarantees that vertex w is directly structure-reachable from vertex v. Iterative queries for direct structural reachability usually add more and more vertices into the current cluster. This procedure mimics a chain effect for core vertices.</p>", "<p>A simple setting is shown in Figure ##FIG##5##6##. Under the conditions of <italic>μ </italic>= <italic>2 </italic>and <italic>ε </italic>= <italic>0.6</italic>, the possible scenario is as follows: We find <italic>E </italic>as the first core vertex since structural similarity between <italic>E </italic>and <italic>B </italic>is greater than <italic>ε, 0.6</italic>. Now a cluster of <italic>{E, B} </italic>is formed, and it should be expanded if possible. At the second step we look for any vertex that is similar to <italic>B</italic>. Among neighbors of <italic>B</italic>, vertex <italic>D </italic>is selected and inserted into the current cluster <italic>{E, B} </italic>due to similarity value of <italic>0.77 </italic>between <italic>B </italic>and <italic>D</italic>. After the insertion, the cluster has now three vertices <italic>{E, B, D}</italic>. At this stage of algorithm, it is noticeable that vertex <italic>E </italic>and <italic>B </italic>are core vertices; <italic>B </italic>is directly structure-reachable from <italic>E</italic>; and <italic>D </italic>is directly structure-reachable from <italic>B</italic>.</p>", "<p>After given example, we introduce another property of SCAN algorithm: structural reachability, which can be considered as chained form of direct structural reachability. The structural reachability is transitive, but it is asymmetric. It is only symmetric for a pair of cores, as appears in previous example. More specifically, the structural reachability is a transitive closure of direct structural reachability.</p>", "<p>Two non-core vertices in the same cluster may not be structure-reachable because the core condition may not hold for them. But they still belong to the same cluster because they both are structure- reachable from the same core. This idea is known as structural connectivity, and explained more formally as follows. A vertex <italic>v ∈ V </italic>is structure-connected to a vertex <italic>w ∈ V </italic>w.r.t <italic>ε </italic>and <italic>μ</italic>, if there is a vertex <italic>u ∈ V </italic>such that both v and w are structure-reachable from <italic>u</italic>. The structural connectivity is a symmetric relation. For the structure-reachable vertices, it is also reflective. Now we are ready to define a cluster as structure-connected vertices, which is maximal w.r.t. structural reachability.</p>", "<p>A non-empty subset <italic>C </italic>⊆ <italic>V </italic>is called a structure-connected cluster w.r.t <italic>ε </italic>and <italic>μ</italic>, if all vertices in <italic>C </italic>are structure-connected and <italic>C </italic>is maximal w.r.t structure reachability. The SCAN algorithm finds all clusters w.r.t <italic>ε </italic>and <italic>μ</italic>, however, there might be some isolated vertices that are not assigned to clusters. If this is the case, we categorize each of those vertices either as hub or outlier. When an isolated vertex <italic>v ∈ V </italic>has neighbors belonging to two or more different clusters, it is labeled as hub vertex. Otherwise, isolated vertex would be an outlier.</p>", "<p>In practice, the definitions of a hub and an outlier are flexible. It may be more useful to regard hubs as a special kind of outlier, since both are isolated vertices. The more clusters in which an outlier has neighbors, the more strongly that vertex acts as a hub between those clusters. This point will be discussed further when we consider actual networks.</p>", "<title>Algorithm scan</title>", "<p>In this section, we describe the algorithm SCAN which implements the search for clusters, hubs and outliers in PPI network. The search begins by first visiting each vertex once to find structure-connected clusters, and then visiting the isolated vertices to identify them as either a hub or an outlier.</p>", "<p>The pseudo code of the algorithm SCAN is presented in Figure ##FIG##6##7## with graphical representation in Figure ##FIG##7##8## and Figure ##FIG##8##9##. SCAN performs one pass of a network and finds all structure-connected clusters for a given parameter setting. At the beginning all vertices are labeled as unclassified. The SCAN algorithm classifies each vertex either a member of a cluster or a non-member. For each vertex that is not yet classified, SCAN checks whether this vertex is a core (STEP 1). If the vertex is a core, a new cluster is expanded from this vertex (STEP 2.1). Otherwise, the vertex is labeled as a non-member (STEP 2.2). To find a new cluster, SCAN starts with an arbitrary core <italic>v </italic>and search for all vertices that are structure-reachable from <italic>v </italic>in STEP 2.1. This is sufficient to find the complete cluster containing vertex <italic>v</italic>, due to given definitions. In STEP 2.1, a new cluster ID is generated which will be assigned to all vertices found in STEP 2.1. SCAN begins by inserting all vertices in <italic>ε</italic>-neighborhood of vertex <italic>v </italic>into a queue. For each vertex in the queue, it computes all directly structure-reachable vertices and inserts those vertices into the queue which are not yet classified. This is repeated until the queue is empty.</p>", "<p>The non-member vertices can be further classified as hubs or outliers in STEP 3. If an isolated vertex has edges to two or more clusters, it is classified as a hub. Otherwise, it is an outlier. This final classification is done according to what is appropriate for the network. As mentioned earlier, the more clusters in which an outlier has neighbors, the more strongly that vertex acts as a hub between those clusters. Likewise, a vertex might bridge only two clusters, but how strongly it is viewed as a hub may depend on how aggressively it bridges them.</p>", "<p>As discussed before, the results of SCAN do not depend on the order of processed vertices, i.e. the obtained clustering of network (number of clusters and association of cores to clusters) is determinate.</p>", "<title>Complexity analysis</title>", "<p>In this section, we present an analysis of the computation complexity of the algorithm SCAN. Given a graph with <italic>m </italic>edges and <italic>n </italic>vertices, SCAN first finds all structure-connected clusters w.r.t. a given parameter setting by checking each vertex of the graph (STEP 1). This entails retrieval of all the vertex's neighbors. Using an adjacency list, a data structure where each vertex has a list of which vertices it is adjacent to, the cost of a neighborhood query is proportional to the number of neighbors, that is, the degree of the query vertex. Therefore, the total cost is <italic>O(deg(v</italic>₁<italic>)+deg(v</italic>₂<italic>)+...deg(v</italic>_{<italic>n</italic>}<italic>))</italic>, where <italic>deg(v</italic>_{<italic>i</italic>}<italic>), i = 1,2,..., n </italic>is the degree of vertex <italic>v</italic>_{<italic>i</italic>}. If we sum all the vertex degrees in <italic>G</italic>, we count each edge exactly twice: once from each end. Thus the running time is <italic>O(m)</italic>.</p>", "<p>We also derive the running time in terms of the number of vertices, should the number of edges be unknown. In the worst case, each vertex connects to all the other vertices for a complete graph. The worst case total cost, in terms of the number of vertices, is <italic>O(n(n-1))</italic>, or <italic>O(n2)</italic>. However, real networks generally have sparser degree distributions. In the following we derive the complexity for an average case, for which we know the probability distribution of the degrees. One type of network is the random graph, studied by Erdös and Rényi [##UREF##9##25##]. Random graphs are generated by placing edges randomly between vertices. Random graphs have been employed extensively as models of real world networks of various types, particularly in epidemiology. The degree of a random graph has a poisson distribution:</p>", "<p></p>", "<p>which indicates that most nodes have approximately the same number of links (close to the average degree <italic>E(k) = z</italic>). In the case of random graphs the complexity of SCAN is <italic>O(n)</italic>.</p>", "<p>Many real networks, such as social networks, biological networks and the WWW follow a power-law degree distribution. The probability that a node has <italic>k </italic>edges, <italic>P(k)</italic>, is on the order k-<italic>α</italic>, where <italic>α </italic>is the degree exponent. A value between 2 and 3 was observed for the degree exponent for most biological and non-biological networks studied by the Faloutsos brothers [##UREF##10##26##] and Barabási and Oltvai [##REF##11258382##5##]. The expected value of degree is <italic>E(k) = α/(α-1)</italic>. In this case the average cost of SCAN is again <italic>O(n)</italic>.</p>", "<p>Therefore, the complexity in terms of the number of edges in the graph for SCAN algorithm is in general linear. The complexity in terms of the number of vertices is quadratic in the worst case of a complete graph. For real networks like biological networks, social networks, and computer networks, SCAN expects linear complexity with respect to the number of vertices as well.</p>" ]

[ "<title>Results and discussion</title>", "<title>Protein-Protein Interaction Network</title>", "<p>Hand-curated databases of PPI in <italic>Saccharomyces cerevisiae </italic>have been studied earlier in the literature [##REF##15048975##16##, ####REF##15374873##17##, ##REF##12089522##18####12089522##18##] and are proven to be invaluable resources for bioinformatics research. For this study, PPI network is downloaded from the <italic>Saccharomyces </italic>Genome Database (SGD) [##UREF##6##19##] on January 21, 2008. After cleaning unrelated interaction, we chose only Affinity Capture-MS and Affinity Capture-Western proteins, which account for 26751 interactions between 4030 proteins.</p>", "<title>Validation metric based on Gene Ontology</title>", "<p>The Gene Ontology (GO) database provides controlled vocabularies for the description of the 1) molecular function, 2) biological process, and 3) cellular component of gene products. The ontologies are continuously updated by GO Consortium, and new versions are made available on monthly basis. Of three ontologies, molecular function describes the tasks performed by individual gene products, such as enzyme activator activity and RNA binding; biological process refers broad biological goals, such as chromatin remodeling or mRNA capping; and cellular component covers subcellular structures, locations, and macromolecular complexes, such as intracellular or cytoplasm.</p>", "<p>The ontologies of GO database are manually created by many scientists. GO database is accepted as ground-truth and used for comparison and validation purposes. Thus, in our analysis we used GO ontologies to test if the resulting clusters are related to any known functional modules. Simply relying on number of proteins that have same annotation will be misleading since distributions of genes among various GO categories are not uniform.</p>", "<p>P-value is the probability that a given set of proteins is enriched by a given functional group by random chance. It is usually used as a criteria to assign each cluster to a known function [##REF##15180928##20##,##REF##14960460##21##]. The smaller the p-value, the more evidence the clustering is not random. In terms of GO annotations, a group of genes with smaller p-value is more significant than the one with a higher p-value.</p>", "<p>Consider a cluster with size n, m proteins sharing a particular annotation <italic>A</italic>. Also assume that there are <italic>N </italic>proteins in the PPI database, and <italic>M </italic>of them are known to have annotation <italic>A</italic>. Given that, the probability of observing m or more proteins that are annotated with <italic>A </italic>out of <italic>n </italic>protein is:</p>", "<p></p>", "<p>Based on above formulation, p-value is calculated for each of three ontologies. However we cannot find always three p-values for a cluster since it is not guaranteed that each cluster has at least one member associated with each of ontology. For instance, protein trm2 does not have any association for cellular component, whereas protein mms1 has two entries which are both from biological process ontology. Assuming a cluster has only two members, trm2 and mms1, we cannot calculate a p-value of the cluster for cellular component ontology. Therefore, it would be correct to claim that we calculate at least one p-value for each cluster. In the case of multiple annotations from same ontology, the one with the smaller p-value is assigned to the cluster as functional annotation. That being said, the p-value without any restriction is not enough to label clusters as significant. Hence we use the recommended cutoff value of <italic>0.05 </italic>in order to select significant clusters within each ontology.</p>", "<p>Let <italic>C </italic>be a cluster including k annotations, and <italic>A</italic>^{<italic>C </italic>}denote proteins having annotation <italic>A </italic>in cluster <italic>C</italic>. The cluster is labeled with a functional annotation , 1 ≤ <italic>t </italic>≤ <italic>k</italic>, iff p-value of is the smallest one among others in cluster <italic>C </italic>and less than cutoff value. After all, we call a cluster insignificant if it has no functional annotation.</p>", "<p>While functional annotations, backed by the statistical evidence, are good interpretation for a single cluster, they do not have much impact to quantify the overall clustering accuracy. Therefore we employ a measure called clustering score [##UREF##7##22##] to compare two clustering layouts.</p>", "<p></p>", "<p>where <italic>n</italic>_{<italic>s </italic>}and <italic>n</italic>_{<italic>i </italic>}denotes the number of significant and insignificant clusters, respectively and min(p_{<italic>i</italic>}) represents the smallest p-value of a significant cluster. Note that <italic>min(p</italic>_{<italic>i</italic>}<italic>) </italic>equals to the p-value of functional annotation at the same time. Clustering score is calculated for three different categories of the GO Ontology, molecular function, biological process, and cell component. In Figure ##FIG##0##1## clustering scores are shown for SCAN and CNM. Please refer Additional file ##SUPPL##0##1## for annotation of each cluster.</p>", "<p>Furthermore, to show how SCAN clearly outperforms CNM, we listed top-10 clusters having the smallest p-values with corresponding GO categories in Table ##TAB##0##1## and Table ##TAB##1##2##. For the category of biological process, SCAN finds clusters with smaller p-values. Also note that p-values of clusters in SCAN are increasing gradually from first to tenth cluster (4.45E-98 to 9.29E-28). In contrast, CNM results start with greater p-value (2.10E-61) and spot clusters with larger size. Recall that the smaller p-value is the better to annotate a cluster with certain function. However, some clusters of CNM with smaller p-values are still hard-to-interpret because of their enormous size, such as cluster 14 having 220 proteins, cluster 12 with 919 proteins, and cluster 48 with 549 proteins.</p>", "<p>For molecular function, similar to biological process, there is a significant difference between two algorithms in terms of both p-values and size of clusters. While SCAN clusters have p-values between 5.64E-71 and 1.37E-17 and average cluster-size of 44, CNM yields clusters with p-values ranging between 1.71E-26 and 2.52E-09 and average size of 329 (10 to 1408). Size problem for molecular function seems even worse than biological process.</p>", "<p>In the category of cellular component, group of top-10 clusters starts with cluster 10, p-value 3.67E-66, size of 107. It is good start against SCAN, however, p-values of CNM do not show regular increase as seen in SCAN clusters. Additionally, regarding CNM results, fluctuation in clustering size arises once again and makes the evaluation intricate. Thus, we randomly picked a few clusters and analyzed the accuracy manually.</p>", "<title>Validation based on manual comparisons</title>", "<p>To judge the significance of a cluster, we manually analyzed whether the function of each member corresponds to cluster's assigned function from three different GO Ontologies, biological processes, cellular components, and molecular functions. We chose cluster sizes ranging from 10 to 30 members since most functional complexes contain the comparable numbers of protein components.</p>", "<p>For biological process, SCAN assigned all the anaphase promoting complex proteins (apc1; apc11; apc2; apc4; apc5; apc9; cdc16; cdc23; cdc26; cdc27; doc1; mnd2; swm1) into one cluster. The clustering result is shown in Figure ##FIG##1##2##. In this graph, interactions only for cluster members are shown to keep the network visually readable. Proteins that are members of found cluster are denoted in different color. Although SCAN identified most subunits of the APC/C complex, it missed a few key components such as cdc20 and cdh1. This is due to the multi-valent interactions of these proteins. SCAN assigned these two proteins in the hub category. Unlike SCAN, CNM merged these proteins into a big cluster with 75 members, which also includes proteins involved in translation initiation, ergosterol synthesis, and other cellular processes. SCAN also identified the vacuolar H+- ATPase complex (rav1, rav2, vma10, vph1, etc) and grouped them into a cluster with 16 members. In this cluster, there are three proteins that have not been reported to directly function in the assembly of the vacuolar H+-ATPase. Xdj1p is a chaperone-like protein. Yig1p is involved in anaerobic glycerol production, which may indirectly affect the cytosolic proton homeostasis. Ymr027wp is a protein with unknown function so far. From our clustering results, we predict that xdj1p and ymr027wp are also involved in the assembly of the vacuolar H+-ATPase. Contrary to the SCAN results, CNM grouped these vacuolar H+-ATPase proteins into a huge cluster with 1416 members, which is not insightful for any predictions.</p>", "<p>For cellular components, SCAN accurately identified the exocyst complex (exo70, exo84, sec3, sec15, etc, please refer Figure ##FIG##2##3##), and the DNA replication preinitiation complex (cdc45, dpb11, mcm10 etc). On the contrary, CNM put the exocyst complex into a group with 551 members and the DNA replication preinitiation complex into a group with 919 members. Although SCAN missed protein Sld2p in the DNA replication preinitiation complex and included two extra proteins (srp101p, srp102p), SCAN predicted a new function for erv2p. Erv2p is an ER lumen protein required for the formation of disulfide bonds. Our clustering result suggests that erv2p helps the proper folding of the proteins required for DNA replication. Both SCAN and CNM detected the peroxisome membrane protein complex (SCAN cluster-176 with 9 members; CNM cluster-19 with 11 members). Please refer Additional file ##SUPPL##1##2## and ##SUPPL##2##3## for detailed clustering results. The two extra proteins (pex3, pex19) in CNM cluster 19 are also members of the peroxisomal membrane complex.</p>", "<p>For molecular functions, SCAN found translation initiation complex (cluster-105 with 10 members; gcd1, gcd11, ist1, mrf1, etc). The found clustering result is depicted in Figure ##FIG##3##4##. CNM again assigned these proteins into a huge cluster with 1416 members. The function of protein glycosylation is achieved by many proteins in multiple subcellular locations including the lumen of ER and golgi. After glycosylation, the proteins will be transported to their right locations by secretion or other sub-cellular transport systems. SCAN assigned 35 proteins to one cluster. This cluster contains the oligosaccharyltransferase complex proteins (ost1, ost2, ost3, etc) of the ER lumen, the golgi mannosyltransferase complex proteins (mnn9, mnn10, mnn11), and secretion proteins (sec61, sec62, sec63, etc). Grouping these proteins into one cluster emphasizes the potential of applying SCAN in analysis of dynamic biological networks.</p>", "<p>With regard to assigning proteins with similar functions to a cluster, SCAN out-weighted CNM in 6 out of 7 clusters we analyzed manually. Moreover, smaller size clusters found by SCAN enable us to predict the function of each cluster more accurately.</p>" ]

[ "<title>Results and discussion</title>", "<title>Protein-Protein Interaction Network</title>", "<p>Hand-curated databases of PPI in <italic>Saccharomyces cerevisiae </italic>have been studied earlier in the literature [##REF##15048975##16##, ####REF##15374873##17##, ##REF##12089522##18####12089522##18##] and are proven to be invaluable resources for bioinformatics research. For this study, PPI network is downloaded from the <italic>Saccharomyces </italic>Genome Database (SGD) [##UREF##6##19##] on January 21, 2008. After cleaning unrelated interaction, we chose only Affinity Capture-MS and Affinity Capture-Western proteins, which account for 26751 interactions between 4030 proteins.</p>", "<title>Validation metric based on Gene Ontology</title>", "<p>The Gene Ontology (GO) database provides controlled vocabularies for the description of the 1) molecular function, 2) biological process, and 3) cellular component of gene products. The ontologies are continuously updated by GO Consortium, and new versions are made available on monthly basis. Of three ontologies, molecular function describes the tasks performed by individual gene products, such as enzyme activator activity and RNA binding; biological process refers broad biological goals, such as chromatin remodeling or mRNA capping; and cellular component covers subcellular structures, locations, and macromolecular complexes, such as intracellular or cytoplasm.</p>", "<p>The ontologies of GO database are manually created by many scientists. GO database is accepted as ground-truth and used for comparison and validation purposes. Thus, in our analysis we used GO ontologies to test if the resulting clusters are related to any known functional modules. Simply relying on number of proteins that have same annotation will be misleading since distributions of genes among various GO categories are not uniform.</p>", "<p>P-value is the probability that a given set of proteins is enriched by a given functional group by random chance. It is usually used as a criteria to assign each cluster to a known function [##REF##15180928##20##,##REF##14960460##21##]. The smaller the p-value, the more evidence the clustering is not random. In terms of GO annotations, a group of genes with smaller p-value is more significant than the one with a higher p-value.</p>", "<p>Consider a cluster with size n, m proteins sharing a particular annotation <italic>A</italic>. Also assume that there are <italic>N </italic>proteins in the PPI database, and <italic>M </italic>of them are known to have annotation <italic>A</italic>. Given that, the probability of observing m or more proteins that are annotated with <italic>A </italic>out of <italic>n </italic>protein is:</p>", "<p></p>", "<p>Based on above formulation, p-value is calculated for each of three ontologies. However we cannot find always three p-values for a cluster since it is not guaranteed that each cluster has at least one member associated with each of ontology. For instance, protein trm2 does not have any association for cellular component, whereas protein mms1 has two entries which are both from biological process ontology. Assuming a cluster has only two members, trm2 and mms1, we cannot calculate a p-value of the cluster for cellular component ontology. Therefore, it would be correct to claim that we calculate at least one p-value for each cluster. In the case of multiple annotations from same ontology, the one with the smaller p-value is assigned to the cluster as functional annotation. That being said, the p-value without any restriction is not enough to label clusters as significant. Hence we use the recommended cutoff value of <italic>0.05 </italic>in order to select significant clusters within each ontology.</p>", "<p>Let <italic>C </italic>be a cluster including k annotations, and <italic>A</italic>^{<italic>C </italic>}denote proteins having annotation <italic>A </italic>in cluster <italic>C</italic>. The cluster is labeled with a functional annotation , 1 ≤ <italic>t </italic>≤ <italic>k</italic>, iff p-value of is the smallest one among others in cluster <italic>C </italic>and less than cutoff value. After all, we call a cluster insignificant if it has no functional annotation.</p>", "<p>While functional annotations, backed by the statistical evidence, are good interpretation for a single cluster, they do not have much impact to quantify the overall clustering accuracy. Therefore we employ a measure called clustering score [##UREF##7##22##] to compare two clustering layouts.</p>", "<p></p>", "<p>where <italic>n</italic>_{<italic>s </italic>}and <italic>n</italic>_{<italic>i </italic>}denotes the number of significant and insignificant clusters, respectively and min(p_{<italic>i</italic>}) represents the smallest p-value of a significant cluster. Note that <italic>min(p</italic>_{<italic>i</italic>}<italic>) </italic>equals to the p-value of functional annotation at the same time. Clustering score is calculated for three different categories of the GO Ontology, molecular function, biological process, and cell component. In Figure ##FIG##0##1## clustering scores are shown for SCAN and CNM. Please refer Additional file ##SUPPL##0##1## for annotation of each cluster.</p>", "<p>Furthermore, to show how SCAN clearly outperforms CNM, we listed top-10 clusters having the smallest p-values with corresponding GO categories in Table ##TAB##0##1## and Table ##TAB##1##2##. For the category of biological process, SCAN finds clusters with smaller p-values. Also note that p-values of clusters in SCAN are increasing gradually from first to tenth cluster (4.45E-98 to 9.29E-28). In contrast, CNM results start with greater p-value (2.10E-61) and spot clusters with larger size. Recall that the smaller p-value is the better to annotate a cluster with certain function. However, some clusters of CNM with smaller p-values are still hard-to-interpret because of their enormous size, such as cluster 14 having 220 proteins, cluster 12 with 919 proteins, and cluster 48 with 549 proteins.</p>", "<p>For molecular function, similar to biological process, there is a significant difference between two algorithms in terms of both p-values and size of clusters. While SCAN clusters have p-values between 5.64E-71 and 1.37E-17 and average cluster-size of 44, CNM yields clusters with p-values ranging between 1.71E-26 and 2.52E-09 and average size of 329 (10 to 1408). Size problem for molecular function seems even worse than biological process.</p>", "<p>In the category of cellular component, group of top-10 clusters starts with cluster 10, p-value 3.67E-66, size of 107. It is good start against SCAN, however, p-values of CNM do not show regular increase as seen in SCAN clusters. Additionally, regarding CNM results, fluctuation in clustering size arises once again and makes the evaluation intricate. Thus, we randomly picked a few clusters and analyzed the accuracy manually.</p>", "<title>Validation based on manual comparisons</title>", "<p>To judge the significance of a cluster, we manually analyzed whether the function of each member corresponds to cluster's assigned function from three different GO Ontologies, biological processes, cellular components, and molecular functions. We chose cluster sizes ranging from 10 to 30 members since most functional complexes contain the comparable numbers of protein components.</p>", "<p>For biological process, SCAN assigned all the anaphase promoting complex proteins (apc1; apc11; apc2; apc4; apc5; apc9; cdc16; cdc23; cdc26; cdc27; doc1; mnd2; swm1) into one cluster. The clustering result is shown in Figure ##FIG##1##2##. In this graph, interactions only for cluster members are shown to keep the network visually readable. Proteins that are members of found cluster are denoted in different color. Although SCAN identified most subunits of the APC/C complex, it missed a few key components such as cdc20 and cdh1. This is due to the multi-valent interactions of these proteins. SCAN assigned these two proteins in the hub category. Unlike SCAN, CNM merged these proteins into a big cluster with 75 members, which also includes proteins involved in translation initiation, ergosterol synthesis, and other cellular processes. SCAN also identified the vacuolar H+- ATPase complex (rav1, rav2, vma10, vph1, etc) and grouped them into a cluster with 16 members. In this cluster, there are three proteins that have not been reported to directly function in the assembly of the vacuolar H+-ATPase. Xdj1p is a chaperone-like protein. Yig1p is involved in anaerobic glycerol production, which may indirectly affect the cytosolic proton homeostasis. Ymr027wp is a protein with unknown function so far. From our clustering results, we predict that xdj1p and ymr027wp are also involved in the assembly of the vacuolar H+-ATPase. Contrary to the SCAN results, CNM grouped these vacuolar H+-ATPase proteins into a huge cluster with 1416 members, which is not insightful for any predictions.</p>", "<p>For cellular components, SCAN accurately identified the exocyst complex (exo70, exo84, sec3, sec15, etc, please refer Figure ##FIG##2##3##), and the DNA replication preinitiation complex (cdc45, dpb11, mcm10 etc). On the contrary, CNM put the exocyst complex into a group with 551 members and the DNA replication preinitiation complex into a group with 919 members. Although SCAN missed protein Sld2p in the DNA replication preinitiation complex and included two extra proteins (srp101p, srp102p), SCAN predicted a new function for erv2p. Erv2p is an ER lumen protein required for the formation of disulfide bonds. Our clustering result suggests that erv2p helps the proper folding of the proteins required for DNA replication. Both SCAN and CNM detected the peroxisome membrane protein complex (SCAN cluster-176 with 9 members; CNM cluster-19 with 11 members). Please refer Additional file ##SUPPL##1##2## and ##SUPPL##2##3## for detailed clustering results. The two extra proteins (pex3, pex19) in CNM cluster 19 are also members of the peroxisomal membrane complex.</p>", "<p>For molecular functions, SCAN found translation initiation complex (cluster-105 with 10 members; gcd1, gcd11, ist1, mrf1, etc). The found clustering result is depicted in Figure ##FIG##3##4##. CNM again assigned these proteins into a huge cluster with 1416 members. The function of protein glycosylation is achieved by many proteins in multiple subcellular locations including the lumen of ER and golgi. After glycosylation, the proteins will be transported to their right locations by secretion or other sub-cellular transport systems. SCAN assigned 35 proteins to one cluster. This cluster contains the oligosaccharyltransferase complex proteins (ost1, ost2, ost3, etc) of the ER lumen, the golgi mannosyltransferase complex proteins (mnn9, mnn10, mnn11), and secretion proteins (sec61, sec62, sec63, etc). Grouping these proteins into one cluster emphasizes the potential of applying SCAN in analysis of dynamic biological networks.</p>", "<p>With regard to assigning proteins with similar functions to a cluster, SCAN out-weighted CNM in 6 out of 7 clusters we analyzed manually. Moreover, smaller size clusters found by SCAN enable us to predict the function of each cluster more accurately.</p>" ]

[ "<title>Conclusion and research directions</title>", "<p>We devised a new methodology called SCAN (Structural Clustering Algorithm for Networks) that can efficiently find clusters or functional modules in complex biological networks as well as hubs and outliers [##UREF##5##15##]. We showed the effectiveness of our methodology using the budding yeast (Saccharomyces cerevisiae) protein-protein interaction network.</p>", "<p>To validate our clustering results, we compared our clusters with the known functions of each protein. Additionally, we compared our algorithm with well-known modularity based clustering algorithm, CNM [##UREF##2##12##]. We successfully showed that SCAN can detect functional groups that are annotated with GO terms. Top-10 clusters with minimum p-values demonstrated that clusters of the newly proposed algorithm are more accurate than those of CNM. Manual interpretations of functional groups found by the new algorithm also showed superior performance over CNM.</p>", "<p>Furthermore, a computational complexity analysis demonstrated a linear running-time of the algorithm, which makes it, to our knowledge, the fastest approach for finding clusters in networks.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Biological systems can be modeled as complex network systems with many interactions between the components. These interactions give rise to the function and behavior of that system. For example, the protein-protein interaction network is the physical basis of multiple cellular functions. One goal of emerging systems biology is to analyze very large complex biological networks such as protein-protein interaction networks, metabolic networks, and regulatory networks to identify functional modules and assign functions to certain components of the system. Network modules do not occur by chance, so identification of modules is likely to capture the biologically meaningful interactions in large-scale PPI data. Unfortunately, existing computer-based clustering methods developed to find those modules are either not so accurate or too slow.</p>", "<title>Results</title>", "<p>We devised a new methodology called SCAN (Structural Clustering Algorithm for Networks) that can efficiently find clusters or functional modules in complex biological networks as well as hubs and outliers. More specifically, we demonstrated that we can find functional modules in complex networks and classify nodes into various roles based on their structures. In this study, we showed the effectiveness of our methodology using the budding yeast (Saccharomyces cerevisiae) protein-protein interaction network. To validate our clustering results, we compared our clusters with the known functions of each protein. Our predicted functional modules achieved very high purity comparing with state-of-the-art approaches. Additionally the theoretical and empirical analysis demonstrated a linear running-time of the algorithm, which is the fastest approach for networks.</p>", "<title>Conclusion</title>", "<p>We compare our algorithm with well-known modularity based clustering algorithm CNM. We successfully detect functional groups that are annotated with putative GO terms. Top-10 clusters with minimum p-value theoretically prove that newly proposed algorithm partitions network more accurately then CNM. Furthermore, manual interpretations of functional groups found by SCAN show superior performance over CNM.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>XX and FT have conceived the study. XX is the inventor of the algorithm SCAN used for this study. NY and MM developed the software and performed data analysis, algorithm testing, and benchmarking. FT interpreted clustering results found by the above mentioned algorithms. MM, FT, XX and NY wrote the manuscript.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>This publication was made possible by NIH Grant # P20 RR-16460 from the IDeA Networks of Biomedical Research Excellence (INBRE) Program of the National Center for Research Resources.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Comparison of clustering scores for three GO categories.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Cluster of anaphase promoting complex proteins.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Cluster of exocyst complex.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>Cluster of translation initiation complex.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>A small network demonstrating two clusters, a hub (vertex 6), and an outlier (vertex 13).</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>A toy network demonstrating structural reachability. Similarities between vertices are given.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>Pseudocode of SCAN Algorithm.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>Graphic diagram of the main body of algorithm SCAN.</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p>Graphic diagram of Find Hubs and Outliers, a subprocedure of algorithm SCAN.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Top-10 SCAN clusters with highest p-values</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Cluster ID</bold></td><td align=\"center\"><bold>P -value</bold></td><td align=\"center\"><bold>GO Term</bold></td><td align=\"center\"><bold>Term Freq. in Network</bold></td><td align=\"center\"><bold>Term Freq. in Cluster</bold></td><td align=\"center\"><bold>Cluster Size</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>Biological Process</bold></td><td align=\"center\">1</td><td align=\"center\">4.45E-98</td><td align=\"left\">nuclear mrna splicing, via spliceosome</td><td align=\"center\">66</td><td align=\"center\">58</td><td align=\"center\">88</td></tr><tr><td/><td align=\"center\">89</td><td align=\"center\">1.01E-65</td><td align=\"left\">translation</td><td align=\"center\">252</td><td align=\"center\">58</td><td align=\"center\">64</td></tr><tr><td/><td align=\"center\">5</td><td align=\"center\">1.16E-52</td><td align=\"left\">ubiquitin-dependent protein catabolic process</td><td align=\"center\">60</td><td align=\"center\">34</td><td align=\"center\">56</td></tr><tr><td/><td align=\"center\">2</td><td align=\"center\">9.04E-40</td><td align=\"left\">transcription from rna polymerase ii promoter</td><td align=\"center\">50</td><td align=\"center\">41</td><td align=\"center\">288</td></tr><tr><td/><td align=\"center\">15</td><td align=\"center\">8.58E-38</td><td align=\"left\">anaphase-promoting complex-dependent proteasomal ubiquitin-dependent protein catabolic process</td><td align=\"center\">13</td><td align=\"center\">13</td><td align=\"center\">13</td></tr><tr><td/><td align=\"center\">22</td><td align=\"center\">1.36E-30</td><td align=\"left\">chromatin remodeling</td><td align=\"center\">46</td><td align=\"center\">20</td><td align=\"center\">40</td></tr><tr><td/><td align=\"center\">192</td><td align=\"center\">5.46E-29</td><td align=\"left\">vacuolar acidification</td><td align=\"center\">23</td><td align=\"center\">13</td><td align=\"center\">16</td></tr><tr><td/><td align=\"center\">13</td><td align=\"center\">6.36E-29</td><td align=\"left\">chromosome segregation</td><td align=\"center\">36</td><td align=\"center\">16</td><td align=\"center\">25</td></tr><tr><td/><td align=\"center\">24</td><td align=\"center\">2.14E-28</td><td align=\"left\">regulation of microtubule polymerization or depolymerization</td><td align=\"center\">10</td><td align=\"center\">10</td><td align=\"center\">12</td></tr><tr><td/><td align=\"center\">30</td><td align=\"center\">9.29E-28</td><td align=\"left\">regulation of cell growth</td><td align=\"center\">10</td><td align=\"center\">10</td><td align=\"center\">13</td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"center\"><bold>Cellular Components</bold></td><td align=\"center\">7</td><td align=\"center\">6.81E-53</td><td align=\"left\">cytosolic large ribosomal subunit</td><td align=\"center\">80</td><td align=\"center\">55</td><td align=\"center\">222</td></tr><tr><td/><td align=\"center\">89</td><td align=\"center\">1.50E-51</td><td align=\"left\">mitochondrial small ribosomal subunit</td><td align=\"center\">33</td><td align=\"center\">29</td><td align=\"center\">64</td></tr><tr><td/><td align=\"center\">1</td><td align=\"center\">2.53E-41</td><td align=\"left\">u4/u6 × u5 tri-snrnp complex</td><td align=\"center\">27</td><td align=\"center\">25</td><td align=\"center\">88</td></tr><tr><td/><td align=\"center\">15</td><td align=\"center\">9.01E-36</td><td align=\"left\">anaphase-promoting complex</td><td align=\"center\">15</td><td align=\"center\">13</td><td align=\"center\">13</td></tr><tr><td/><td align=\"center\">22</td><td align=\"center\">7.15E-31</td><td align=\"left\">rsc complex</td><td align=\"center\">16</td><td align=\"center\">15</td><td align=\"center\">40</td></tr><tr><td/><td align=\"center\">24</td><td align=\"center\">2.14E-28</td><td align=\"left\">dash complex</td><td align=\"center\">10</td><td align=\"center\">10</td><td align=\"center\">12</td></tr><tr><td/><td align=\"center\">38</td><td align=\"center\">2.14E-28</td><td align=\"left\">trapp complex</td><td align=\"center\">10</td><td align=\"center\">10</td><td align=\"center\">12</td></tr><tr><td/><td align=\"center\">185</td><td align=\"center\">7.18E-26</td><td align=\"left\">ribonuclease mrp complex</td><td align=\"center\">9</td><td align=\"center\">9</td><td align=\"center\">11</td></tr><tr><td/><td align=\"center\">155</td><td align=\"center\">5.84E-25</td><td align=\"left\">smc5-smc6 complex</td><td align=\"center\">8</td><td align=\"center\">8</td><td align=\"center\">8</td></tr><tr><td/><td align=\"center\">53</td><td align=\"center\">4.93E-24</td><td align=\"left\">dna replication preinitiation complex</td><td align=\"center\">21</td><td align=\"center\">12</td><td align=\"center\">22</td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"center\"><bold>Molecular Functions</bold></td><td align=\"center\">89</td><td align=\"center\">5.64E-71</td><td align=\"left\">structural constituent of ribosome</td><td align=\"center\">210</td><td align=\"center\">58</td><td align=\"center\">64</td></tr><tr><td/><td align=\"center\">5</td><td align=\"center\">2.75E-45</td><td align=\"left\">endopeptidase activity</td><td align=\"center\">26</td><td align=\"center\">24</td><td align=\"center\">56</td></tr><tr><td/><td align=\"center\">1</td><td align=\"center\">7.12E-45</td><td align=\"left\">contributes_to rna splicing factor activity, transesterification mechanism</td><td align=\"center\">29</td><td align=\"center\">27</td><td align=\"center\">88</td></tr><tr><td/><td align=\"center\">37</td><td align=\"center\">6.44E-41</td><td align=\"left\">snap receptor activity</td><td align=\"center\">24</td><td align=\"center\">23</td><td align=\"center\">74</td></tr><tr><td/><td align=\"center\">192</td><td align=\"center\">4.65E-28</td><td align=\"left\">hydrogen ion transporting atpase activity, rotational mechanism</td><td align=\"center\">12</td><td align=\"center\">11</td><td align=\"center\">16</td></tr><tr><td/><td align=\"center\">185</td><td align=\"center\">7.18E-26</td><td align=\"left\">contributes_to ribonuclease mrp activity</td><td align=\"center\">9</td><td align=\"center\">9</td><td align=\"center\">11</td></tr><tr><td/><td align=\"center\">22</td><td align=\"center\">6.60E-23</td><td align=\"left\">contributes_to dna-dependent atpase activity</td><td align=\"center\">15</td><td align=\"center\">12</td><td align=\"center\">40</td></tr><tr><td/><td align=\"center\">34</td><td align=\"center\">1.30E-22</td><td align=\"left\">protein transporter activity</td><td align=\"center\">24</td><td align=\"center\">14</td><td align=\"center\">46</td></tr><tr><td/><td align=\"center\">15</td><td align=\"center\">2.25E-18</td><td align=\"left\">ubiquitin-protein ligase activity</td><td align=\"center\">44</td><td align=\"center\">10</td><td align=\"center\">13</td></tr><tr><td/><td align=\"center\">8</td><td align=\"center\">1.37E-17</td><td align=\"left\">dolichyl-diphosphooligosaccharide-protein glycotransferase activity</td><td align=\"center\">8</td><td align=\"center\">8</td><td align=\"center\">35</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Top-10 CNM clusters with highest p-values</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Cluster ID</bold></td><td align=\"center\"><bold>P -value</bold></td><td align=\"center\"><bold>GO Term</bold></td><td align=\"center\"><bold>Term Freq. in Network</bold></td><td align=\"center\"><bold>Term Freq. in Cluster</bold></td><td align=\"center\"><bold>Cluster Size</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold>Biological Process</bold></td><td align=\"center\">15</td><td align=\"center\">2.10E-61</td><td align=\"left\">nuclear mrna splicing, via spliceosome</td><td align=\"center\">66</td><td align=\"center\">55</td><td align=\"center\">220</td></tr><tr><td/><td align=\"center\">17</td><td align=\"center\">5.00E-40</td><td align=\"left\">transposition, rna-mediated</td><td align=\"center\">33</td><td align=\"center\">19</td><td align=\"center\">22</td></tr><tr><td/><td align=\"center\">13</td><td align=\"center\">5.11E-31</td><td align=\"left\">rna elongation from rna polymerase ii promoter</td><td align=\"center\">53</td><td align=\"center\">51</td><td align=\"center\">919</td></tr><tr><td/><td align=\"center\">49</td><td align=\"center\">4.30E-25</td><td align=\"left\">ribosomal large subunit assembly and maintenance</td><td align=\"center\">39</td><td align=\"center\">34</td><td align=\"center\">549</td></tr><tr><td/><td align=\"center\">16</td><td align=\"center\">1.09E-23</td><td align=\"left\">anaphase-promoting complex-dependent proteasomal ubiquitin-dependent protein catabolic process</td><td align=\"center\">13</td><td align=\"center\">13</td><td align=\"center\">75</td></tr><tr><td/><td align=\"center\">58</td><td align=\"center\">2.78E-22</td><td align=\"left\">microtubule nucleation</td><td align=\"center\">22</td><td align=\"center\">16</td><td align=\"center\">98</td></tr><tr><td/><td align=\"center\">53</td><td align=\"center\">7.86E-20</td><td align=\"left\">trna processing</td><td align=\"center\">14</td><td align=\"center\">8</td><td align=\"center\">10</td></tr><tr><td/><td align=\"center\">45</td><td align=\"center\">8.35E-18</td><td align=\"left\">negative regulation of gluconeogenesis</td><td align=\"center\">9</td><td align=\"center\">7</td><td align=\"center\">12</td></tr><tr><td/><td align=\"center\">56</td><td align=\"center\">3.15E-17</td><td align=\"left\">ubiquitin-dependent protein catabolic process via the multivesicular body pathway</td><td align=\"center\">13</td><td align=\"center\">11</td><td align=\"center\">91</td></tr><tr><td/><td align=\"center\">63</td><td align=\"center\">1.73E-13</td><td align=\"left\">mrna polyadenylation</td><td align=\"center\">17</td><td align=\"center\">13</td><td align=\"center\">241</td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"center\"><bold>Cellular Components</bold></td><td align=\"center\">11</td><td align=\"center\">3.67E-66</td><td align=\"left\">mitochondrial large ribosomal subunit</td><td align=\"center\">43</td><td align=\"center\">41</td><td align=\"center\">107</td></tr><tr><td/><td align=\"center\">49</td><td align=\"center\">4.72E-42</td><td align=\"left\">cytosolic large ribosomal subunit</td><td align=\"center\">80</td><td align=\"center\">64</td><td align=\"center\">549</td></tr><tr><td/><td align=\"center\">17</td><td align=\"center\">5.00E-40</td><td align=\"left\">retrotransposon nucleocapsid</td><td align=\"center\">33</td><td align=\"center\">19</td><td align=\"center\">22</td></tr><tr><td/><td align=\"center\">58</td><td align=\"center\">1.28E-30</td><td align=\"left\">condensed nuclear chromosome kinetochore</td><td align=\"center\">30</td><td align=\"center\">22</td><td align=\"center\">98</td></tr><tr><td/><td align=\"center\">16</td><td align=\"center\">2.48E-24</td><td align=\"left\">anaphase-promoting complex</td><td align=\"center\">15</td><td align=\"center\">14</td><td align=\"center\">75</td></tr><tr><td/><td align=\"center\">20</td><td align=\"center\">1.58E-23</td><td align=\"left\">peroxisomal membrane</td><td align=\"center\">12</td><td align=\"center\">9</td><td align=\"center\">11</td></tr><tr><td/><td align=\"center\">37</td><td align=\"center\">9.63E-23</td><td align=\"left\">smc5-smc6 complex</td><td align=\"center\">8</td><td align=\"center\">8</td><td align=\"center\">11</td></tr><tr><td/><td align=\"center\">53</td><td align=\"center\">2.36E-22</td><td align=\"left\">ribonuclease mrp complex</td><td align=\"center\">9</td><td align=\"center\">8</td><td align=\"center\">10</td></tr><tr><td/><td align=\"center\">63</td><td align=\"center\">5.21E-18</td><td align=\"left\">mrna cleavage and polyadenylation specificity factor complex</td><td align=\"center\">14</td><td align=\"center\">14</td><td align=\"center\">241</td></tr><tr><td/><td align=\"center\">65</td><td align=\"center\">3.13E-15</td><td align=\"left\">alpha-1,6-mannosyltransferase complex</td><td align=\"center\">6</td><td align=\"center\">6</td><td align=\"center\">18</td></tr><tr><td colspan=\"7\"><hr/></td></tr><tr><td align=\"center\"><bold>Molecular Functions</bold></td><td align=\"center\">17</td><td align=\"center\">1.71E-26</td><td align=\"left\">rna binding</td><td align=\"center\">130</td><td align=\"center\">19</td><td align=\"center\">22</td></tr><tr><td/><td align=\"center\">58</td><td align=\"center\">2.34E-24</td><td align=\"left\">structural constituent of cytoskeleton</td><td align=\"center\">47</td><td align=\"center\">22</td><td align=\"center\">98</td></tr><tr><td/><td align=\"center\">53</td><td align=\"center\">2.36E-22</td><td align=\"left\">contributes_to ribonuclease mrp activity</td><td align=\"center\">9</td><td align=\"center\">8</td><td align=\"center\">10</td></tr><tr><td/><td align=\"center\">13</td><td align=\"center\">9.16E-19</td><td align=\"left\">dna-directed rna polymerase activity</td><td align=\"center\">31</td><td align=\"center\">30</td><td align=\"center\">919</td></tr><tr><td/><td align=\"center\">49</td><td align=\"center\">6.52E-17</td><td align=\"left\">snorna binding</td><td align=\"center\">21</td><td align=\"center\">20</td><td align=\"center\">549</td></tr><tr><td/><td align=\"center\">48</td><td align=\"center\">5.97E-13</td><td align=\"left\">contributes_to protein transporter activity</td><td align=\"center\">7</td><td align=\"center\">5</td><td align=\"center\">10</td></tr><tr><td/><td align=\"center\">30</td><td align=\"center\">1.24E-10</td><td align=\"left\">nad-independent histone deacetylase activity</td><td align=\"center\">4</td><td align=\"center\">4</td><td align=\"center\">15</td></tr><tr><td/><td align=\"center\">65</td><td align=\"center\">2.79E-10</td><td align=\"left\">contributes_to alpha-1,6-mannosyltransferase activity</td><td align=\"center\">4</td><td align=\"center\">4</td><td align=\"center\">18</td></tr><tr><td/><td align=\"center\">33</td><td align=\"center\">1.34E-09</td><td align=\"left\">endopeptidase activity</td><td align=\"center\">26</td><td align=\"center\">24</td><td align=\"center\">1408</td></tr><tr><td/><td align=\"center\">63</td><td align=\"center\">2.52E-09</td><td align=\"left\">contributes_to histone lysine n-methyltransferase activity (h3-k4 specific)</td><td align=\"center\">7</td><td align=\"center\">7</td><td align=\"center\">241</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S19-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>v</mml:mi>\n <mml:mi>a</mml:mi>\n <mml:mi>l</mml:mi>\n <mml:mi>u</mml:mi>\n <mml:mi>e</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mi>M</mml:mi>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mi>i</mml:mi>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>N</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>M</mml:mi>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>i</mml:mi>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n <mml:mrow>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mi>N</mml:mi>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mi>n</mml:mi>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S19-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>C</mml:mi><mml:mo>=</mml:mo><mml:mo>{</mml:mo><mml:msubsup><mml:mi>A</mml:mi><mml:mn>1</mml:mn><mml:mi>C</mml:mi></mml:msubsup><mml:mo>∩</mml:mo><mml:msubsup><mml:mi>A</mml:mi><mml:mn>2</mml:mn><mml:mi>C</mml:mi></mml:msubsup><mml:mo>∩</mml:mo><mml:mn>...</mml:mn><mml:mo>∩</mml:mo><mml:msubsup><mml:mi>A</mml:mi><mml:mi>k</mml:mi><mml:mi>C</mml:mi></mml:msubsup><mml:mo>}</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S19-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>A</mml:mi><mml:mi>t</mml:mi><mml:mi>C</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S19-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>A</mml:mi><mml:mi>t</mml:mi><mml:mi>C</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S19-i4\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>C</mml:mi>\n <mml:mi>l</mml:mi>\n <mml:mi>u</mml:mi>\n <mml:mi>s</mml:mi>\n <mml:mi>t</mml:mi>\n <mml:mi>e</mml:mi>\n <mml:mi>r</mml:mi>\n <mml:mi>i</mml:mi>\n <mml:mi>n</mml:mi>\n <mml:mi>g</mml:mi>\n <mml:mtext> </mml:mtext>\n <mml:mi>S</mml:mi>\n <mml:mi>c</mml:mi>\n <mml:mi>o</mml:mi>\n <mml:mi>r</mml:mi>\n <mml:mi>e</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo>−</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle displaystyle=\"true\">\n <mml:munderover>\n <mml:mo>∑</mml:mo>\n <mml:mrow>\n <mml:mi>i</mml:mi>\n <mml:mo>=</mml:mo>\n <mml:mn>1</mml:mn>\n </mml:mrow>\n <mml:mrow>\n <mml:msub>\n <mml:mi>n</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:munderover>\n <mml:mrow>\n <mml:mi>min</mml:mi>\n <mml:mo>⁡</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>p</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mstyle>\n <mml:mo>+</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>n</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo>*</mml:mo>\n <mml:mi>c</mml:mi>\n <mml:mi>u</mml:mi>\n <mml:mi>t</mml:mi>\n <mml:mi>o</mml:mi>\n <mml:mi>f</mml:mi>\n <mml:mi>f</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>n</mml:mi>\n <mml:mi>i</mml:mi>\n </mml:msub>\n <mml:mo>+</mml:mo>\n <mml:msub>\n <mml:mi>n</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>*</mml:mo>\n <mml:mi>c</mml:mi>\n <mml:mi>u</mml:mi>\n <mml:mi>t</mml:mi>\n <mml:mi>o</mml:mi>\n <mml:mi>f</mml:mi>\n <mml:mi>f</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><italic>Γ </italic><italic>(v) = {w ∈ V | (v, w) ∈ E} ∪ {v}</italic></disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1471-2105-9-S9-S19-i5\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>σ</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>v</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:mi>Γ</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>v</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>∩</mml:mo>\n <mml:mi>Γ</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:msqrt>\n <mml:mrow>\n <mml:mo>|</mml:mo>\n <mml:mi>Γ</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>v</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>|</mml:mo>\n <mml:mo>|</mml:mo>\n <mml:mi>Γ</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>w</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>|</mml:mo>\n </mml:mrow>\n </mml:msqrt>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<disp-formula><italic>N</italic>_{<italic>ε </italic>}(<italic>v</italic>) = {<italic>w </italic>∈ <italic>Γ </italic>(<italic>v</italic>)|<italic>σ </italic>(<italic>v</italic>, <italic>w</italic>) ≥ <italic>ε</italic>}</disp-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1471-2105-9-S9-S19-i6\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>k</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mrow>\n <mml:mo>(</mml:mo>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mi>n</mml:mi>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mi>k</mml:mi>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n <mml:mo>)</mml:mo>\n </mml:mrow>\n <mml:msup>\n <mml:mi>p</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msup>\n <mml:msup>\n <mml:mrow>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mn>1</mml:mn>\n <mml:mo>−</mml:mo>\n <mml:mi>p</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>k</mml:mi>\n </mml:mrow>\n </mml:msup>\n <mml:mo>≈</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msup>\n <mml:mi>z</mml:mi>\n <mml:mi>k</mml:mi>\n </mml:msup>\n <mml:msup>\n <mml:mi>e</mml:mi>\n <mml:mi>z</mml:mi>\n </mml:msup>\n </mml:mrow>\n <mml:mrow>\n <mml:mi>k</mml:mi>\n <mml:mo>!</mml:mo>\n </mml:mrow>\n </mml:mfrac>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional File 1</title><p>Cluster annotations for SCAN and CNM.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S2\"><caption><title>Additional File 2</title><p>Clustering result of CNM for PPI network.</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"S3\"><caption><title>Additional File 3</title><p>Clustering result of SCAN for PPI network.</p></caption></supplementary-material>" ]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S19-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-8\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S19-9\"/>" ]

[ "<media xlink:href=\"1471-2105-9-S9-S19-S1.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-S9-S19-S2.pdf\" mimetype=\"application\" mime-subtype=\"pdf\"><caption><p>Click here for file</p></caption></media>", "<media xlink:href=\"1471-2105-9-S9-S19-S3.xls\" mimetype=\"application\" mime-subtype=\"vnd.ms-excel\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>The advent of DNA microarray technology has revolutionized genomic research and medicine because of its ability to simultaneously determine the expression levels of thousands of genes. However, the interpretation of large amounts of microarray gene expression data, and the ability to derive biologically meaningful conclusions from such data, have always been daunting tasks for statisticians. Because of the high volume and complex characteristics of microarray data, much of the initial work on their analysis has focused on development of data mining or data reduction methods to identify differentially expressed genes. Typically, the <italic>p</italic>-value of a test statistic is calculated for each gene, the genes are ranked according to these <italic>p</italic>-values, and a pre-specified significance criterion, such as the false discovery rate, is used to determine a cut-off which creates a category of differentially expressed genes [##UREF##0##1##, ####UREF##1##2##, ##UREF##2##3####2##3##].</p>", "<p>Attempts to interpret individual genes in a list of significant genes are demanding and laborious. Therefore, recent efforts have focused on discovery of biological pathways rather than on individual gene function. Gene ontology terms (GO terms, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.geneontology.org\"/>) reflect gene groupings based on molecular function, biological process, or cellular structure/organelle. The interpretation of differentially expressed GO groups is generally simpler than the presentation of a list of statistically significant genes, and more resistant to erroneous conclusions that can arise from microarray artefacts.</p>", "<p>Several statistical methods that combine the analysis of differential gene expression with biological databases have been proposed and implemented in computer packages for a more rapid interpretation of genome-wide expression data [##REF##15994189##4##]. However, most such methods are based on a series of univariate statistical tests and do not properly account for the complex structure of gene interactions. The statistical significance of a GO group is commonly assessed by comparing the number of statistically significant genes in the group to the number expected by chance using Fisher's exact test, which is based on the hypergeometric distribution [##REF##12620386##5##]. Fisher's exact test is used to compare these proportions to assess overrepresentation of significant genes in functional categories. This approach is not amenable to correction for correlations among <italic>p</italic>-values, since the test inherently assumes exchangeability among genes, an assumption which is not valid under arbitrary or actual correlation structures [##REF##16174746##6##,##REF##15176478##7##].</p>", "<p>Hotelling's <italic>T</italic>²Statistic and permutation methods address the correlation structure among genes. Hotelling's <italic>T</italic>²statistic is not applicable when the sample size is smaller than the number of genes in a GO term [##REF##16877751##8##]. Permutation methods, although quite valuable under appropriate conditions, are severely compromised by limited numbers of permutable sample pairs. In many cases, the design of microarray studies has a rather complicated structure intended to manage technical variation associated with differences among probes, dyes, and reagent batches by creating treatment blocks within these sources of variation [##UREF##3##9##]. Such cases are not suited to permutation methods.</p>", "<p>A modified meta-analysis method was developed by Delongchamp <italic>et al</italic>. [##REF##17118132##10##] to combine <italic>p</italic>-values, and thus to measure the significance of an overall treatment effect on a group of genes, while taking into account the inter-genic correlation structure. The method is based on the fact that <italic>p</italic>-values follow a uniform distribution under the null hypothesis. Inverse-normal transformed <italic>p</italic>-values have a normal distribution and their sum over a set of genes also would follow a normal distribution, provided that the component <italic>p</italic>-values are independent. The test we have developed to measure the significance of overall treatment effect on genes within a GO category is based on a modification of this statistic, to reflect the actual correlation structure among genes sharing a GO term.</p>", "<p>In this paper, we extend the method from a simple one-class <italic>t</italic>-test with the null hypothesis <italic>H</italic>₀: <italic>μ </italic>= 0 to a test for pair-wise contrast in a fixed-effects linear model. In the following sections, we describe in detail the extension of the methodology, with validation through computer simulations and application to two toxicogenomics studies designed to evaluate treatment effects on the levels of mRNA transcripts involved in mitochondrial function. We thus demonstrate that this methodology provides a practical approach to testing the significance of the treatment effects on gene classes defined by GO terms, and by extension on any other prior categorization of genes into functional subsets. Because many microarray experiments measure treatment effects under complicated design structures and with small sample sizes [##UREF##3##9##], we used a simulation study to determine whether the method gives reasonable results under these conditions.</p>", "<p>Specific applications to toxicogenomics studies showed that the methodology has improved specificity in choosing significantly altered pathways or functional categories, and may thus assist in the understanding of molecular mechanisms of mitochondrial toxicity in the liver induced by anti-HIV drugs [##REF##17526437##11##,##REF##18313992##12##] and in assessing effects on mitochondrial function of weight-reducing dietary supplements, such as usnic acid [##UREF##4##13##].</p>" ]

[ "<title>Methods</title>", "<title>Measurement of a treatment effect for each gene</title>", "<p>Under a fixed-effects linear model, gene expression data for an arbitrary gene can be written as <bold>y </bold>= <bold>Xβ </bold>+ <bold>ε</bold>, where <bold>y </bold>and <bold>ε </bold>are <italic>n </italic>× 1 random vectors, X is a known <italic>n </italic>× <italic>p </italic>design matrix of rank <italic>r</italic>, and <bold>β </bold>is a <italic>p </italic>× 1 vector representing unknown parameters. The vector <bold>y </bold>denotes an observed measurement of expression, suitably transformed, for <italic>n </italic>biological samples, and <bold>ε </bold>is an error vector, distributed as <italic>N</italic>_{<italic>n </italic>}(0, <italic>σ</italic>²<bold>I</bold>_{<italic>n</italic>}), where <italic>σ</italic>²denotes the unknown within-treatment variance among samples. The parameters <bold>β </bold>and <italic>σ</italic>²are assumed to be gene-specific. Statistical analyses are applied to one gene at a time, with a common design matrix, <bold>X</bold>. The unbiased estimators of <bold>β </bold>and <italic>σ</italic>²are</p>", "<p></p>", "<p>In many toxicogenomic studies, the significance of a treatment effect is tested under the null hypothesis <italic>H</italic>₀: <bold>cβ </bold>= 0, where <bold>cβ </bold>is a pair-wise contrast among treatments. Under the null hypothesis, has a <italic>t</italic>-distribution with <italic>n </italic>- <italic>r </italic>degrees of freedom, and the <italic>p</italic>-value to assess the significance of a treatment is calculated from this statistic.</p>", "<title>Test for a gene group</title>", "<p>A modified meta-analysis method of combining <italic>p</italic>-values was developed to measure the significance of an overall treatment effect on any group of genes by a one-class <italic>t</italic>-test [##REF##17118132##10##]. The <italic>p</italic>-value calculated from the null hypothesis is a random variable with uniform distribution, which can be transformed to a suitable probability distribution. Inverse-normal transformed <italic>p</italic>-values, <italic>z</italic>_{<italic>k </italic>}= Φ^-1(1 - <italic>p</italic>_{<italic>k</italic>}) ~<italic>N</italic>(0, 1), <italic>k </italic>= 1, ⋯, <italic>m </italic>have a normal distribution and their sum, , is also normally distributed when <italic>p</italic>-values are independent. Here, <italic>p</italic>_{<italic>k </italic>}represents a <italic>p</italic>-value for a gene in a GO group comprising <italic>m </italic>genes. The <italic>p</italic>-value for the sum of inverse-transformed <italic>p</italic>-values, , gives the overall significance of the treatment effect on the GO group. We refer to this as the naïve estimate because it assumes independence among <italic>p</italic>-values.</p>", "<p>In reality, genes in a GO group are likely to be functionally related and thus not independent. When the correlation structure among genes is known, we can make a simple adjustment of the naïve estimate. In this case, the test statistics <italic>T </italic>still has a standard normal distribution and we denote it as , for the <italic>k</italic>-th gene in a GO group. A common contrast vector, c, is used through all genes since we are measuring same contrast for each gene. The summary statistic for a GO group, is also normally distributed and its variance is , where <bold>1 </bold>is m vector of 1s and <bold>R </bold>is the correlation matrix of (<bold>y</bold>_{<bold>1</bold>}, <bold>y</bold>_{<bold>2</bold>}, ⋯, <bold>y</bold>_{<bold>m</bold>}). Note that</p>", "<p></p>", "<p>where <italic>r</italic>_{<italic>s</italic>,<italic>t </italic>}is the <italic>s</italic>-th row and <italic>t</italic>-th column element of <bold>R</bold>. Therefore, is the appropriate <italic>p</italic>-value which corrects the naïve <italic>p</italic>-value, .</p>", "<p>It follows that , where is the average of off-diagonal elements of <bold>R</bold>. The correction depends only on the average correlation, , and the correction tends to give a reduced significance when &gt; 0. When <bold>R </bold>is unknown, we estimated the covariance, to provide an average correlation coefficient for the correction.</p>", "<p>The correlation structure of <italic>p</italic>-values is different for a two-sided <italic>t</italic>-test, which allows gene expression changes in both directions, than for a one-sided situation. A two-tailed test, in which <italic>p </italic>= 2(1 - Φ(|<italic>z</italic>|)), requires a different correction method, since the correlation among |<italic>z</italic>_{<italic>k</italic>}|, <italic>k </italic>= 1, ⋯, <italic>m </italic>differs from a one-sided test in which <italic>p </italic>= 1 - Φ(<italic>z</italic>). The null distribution of the summary statistics <bold>1'</bold>|<bold>z</bold>| can be generated through Monte Carlo sampling from the null distribution of <bold>z</bold>, MVN(0, cov(<bold>z</bold>)). When <bold>z</bold>₁, ⋯, <bold>z</bold>_{<italic>n </italic>}are random samples from MVN(0, cov(<bold>z</bold>)), the <italic>p</italic>-value for the observed value, Ψ = <bold>1'</bold>|<bold>z</bold>|, is computed as , where I(<italic>A</italic>) is an indicator function which gives 1 if <italic>A </italic>is true, or 0 otherwise. Here, cov(<bold>z</bold>) has to be estimated from the data. The estimated correlation, and its variation, , which has for off-diagonal elements, are used to estimate cov(<bold>z</bold>).</p>" ]

[ "<title>Results</title>", "<title>Simulation</title>", "<p>The derivation of the method presented in the previous section is based on the known correlation matrix of the vector of dependent variable Y. When the correlation matrix is not known, we use an estimate of the correlation matrix. In reality, the correlation matrix is always unknown. The proposed method produces an approximately correct p-value for a group of genes. To demonstrate that the method gives not perfect but acceptably correct p-values, we present simulation results in this section. The validation is done by checking the cumulative distribution of p-values from the proposed methods under the null distribution. The <italic>p</italic>-values must have a uniform distribution, which should form a diagonal line in the following figures if <italic>p</italic>-values are correctly calculated.</p>", "<p>The simulation is conducted under a fairly common set of conditions for microarray studies, comprising three treatments with three samples (arrays) per treatment. Samples are generated from <italic>N</italic>(<italic>μ</italic>_{<italic>i</italic>}, <bold>Σ</bold>) for each treatment, <italic>i </italic>= 1, 2 3,. In this simulation, samples for two treatment have the same average values, <italic>μ</italic>₁= <italic>μ</italic>₂= 1, and samples for the other treatment have twice that average value <italic>μ</italic>₃= 2. <italic>P</italic>-values for the pair-wise contrast are calculated under the null hypothesis, H₀: <italic>μ</italic>₁= <italic>μ</italic>₂. A GO term is composed of <italic>m </italic>= 20 genes which have correlation structure generated randomly between 0.36 and 0.55. The standard deviations, <italic>σ</italic>_{<italic>i</italic>}, <italic>i </italic>= 1, ⋯,<italic>m </italic>for the genes, which are the diagonal elements of <bold>Σ</bold>, are generated randomly between 0.01 and 0.25. We iterated this procedure at least 10,000 times to observe the distribution of calculated <italic>p</italic>-values.</p>", "<p>Figure ##FIG##0##1## plots the cumulative distribution of <italic>p</italic>-values from a one-sided test when the number of samples is <italic>n </italic>= 9, i.e., 3 groups with 3 samples for each group. The naïve <italic>p</italic>-values, shown by the red line, clearly deviate from the diagonal line. Almost 30% of <italic>p</italic>-values are estimated to be less than 0.05, indicating that the naïve <italic>p</italic>-values lead to a very high false-discovery rate. The corrected <italic>p</italic>-values (dashed blue line) fall very near the diagonal line. The corrected <italic>p</italic>-values thus have more specificity in choosing altered functional gene groups than the naïve <italic>p</italic>-values.</p>", "<p>Figure ##FIG##1##2## shows the simulation result for a two-sided case. <italic>P</italic>-values are calculated from the null distribution generated from Monte Carlo samples from MVN(0, cov(<bold>z</bold>)). Two estimates of cov(<bold>z</bold>) are used for the sampling. When is used, the empirical distribution of <italic>p</italic>-values is closer to a uniform distribution than when is used. The estimate of the average correlation, , is more robust than that of each element of when sample size is small (n = 3 for each group).</p>", "<p>Figure ##FIG##2##3## shows the distribution of <italic>p</italic>-values for a case with larger samples. The simulation for one-class <italic>t</italic>-test with sample size n = 25 was conducted as above, to compare several methods for two sided tests. The empirical distributions of <italic>p</italic>-values were generated from 500,000 iterations. We looked at small <italic>p</italic>-values between 0.1 and 0.001 on a log scale. Figure ##FIG##2##3## shows that the Hotelling <italic>T</italic>²test gives the smallest difference from the uniform distribution. The Hotelling <italic>T</italic>²test is applicable when the number of sample is larger than the number of genes in a group. When we have a reasonably correct estimate of <bold>R</bold>, the method is a little better than the method which uses an approximation of <bold>R</bold>. Both the method and the method give quite accurate p-values with reference to the p-values from the true correlation matrix, <bold>R</bold>.</p>", "<title>Examples</title>", "<p>We present two real-world examples based on data from Mitochip, a mitochondria-specific mouse oligonucleotide microarray which was developed by Dr. Varsha Desai at the National Center for Toxicological Research [##REF##17526437##11##]. Mitochip measures the levels of mRNA for 542 mitochondrial and nuclear genes associated with mitochondrial structure and function. Each Mitochip includes 9 housekeeping genes and 9 <italic>Arabidopsis </italic>plant genes to serve as positive and negative control genes, respectively. We considered 317 relevant GO groups related to mitochondrial functions, based on a database from Mouse Genome Informatics (MGI, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.informatics.jax.org\"/>).</p>", "<p>Table ##TAB##0##1## shows the design of an experiment to test the effects of zidovudine (AZT) and lamivudine (3TC) on mouse-liver gene expression. AZT is an anti-HIV drug used to reduce mother-to-child transmission of the virus. AZT is reported to produce severe adverse effects, and shows more toxicity when AZT is applied in combination with 3TC. Adverse effects are believed to be due to drug-induced mitochondrial disfunction [##REF##12668570##14##].</p>", "<p>Oxidative phosphorylation is a key mitochondrial function that requires the electron transport assembly of four protein complexes (I, II, III, IV) to catalyze sequential oxidation/reduction reactions, and complex V to generate ATP. Several clinical and animal studies have investigated the effect of nucleoside reverse transcriptase inhibitors (NRTIs), analogs such as AZT, on mitochondrial respiratory chain complexes. These studies suggest that there is a deficit in one of the components of complexes I and IV in skeletal muscle of children perinatally exposed to antiretroviral nucleoside analogues [##REF##12891063##15##].</p>", "<p>Table ##TAB##1##2## shows <italic>p</italic>-values indicating the significance of treatment effects on the GO groups related to oxidative phosphorylation and apoptosis. The two-sided correction method detects significant effects on genes encoding components of complexes III and IV, whereas the naïve method finds that genes in all 5 complexes are significantly affected. This demonstrates that the two-sided correction method is more specific in finding significantly affected gene groups, although of course the \"true\" answer is not known <italic>a priori</italic>. Although Fisher's exact test also detects significant alteration in genes of complex IV, this test appears to be detecting a gene group that is different from the other groups, rather than registering treatment effects directly. The one-sided test is not applicable since it seems that gene expression changes in both directions after AZT and 3TC treatment.</p>", "<p>Usnic acid is a lichen metabolite used as a weight-loss dietary supplement due to its uncoupling action on mitochondria. However, its use has been associated with severe liver disorders in many individuals. Animal studies conducted thus far have evaluated effect of usnic acid on mitochondria, primarily by measuring the rate of oxygen consumption and/or ATP generation. Generation of ATP requires tight coupling of electron transport with oxidative phosphorylation, maintained through a proton gradient across the inner mitochondrial membrane. An important finding of the study is a lack of usnic acid effect on complex V, despite a significant up-regulation of all four complexes of the electron transport chain. Usnic acid is a known uncoupler that is highly lipophilic in both neutral and anionic forms due to its numerous carbonyl groups that absorb the negative charge of the anion by resonance stabilization. This lipophilicity of usnic acid and the usniate anion allows easy passage of both entities through the mitochondrial membranes by passive diffusion into the matrix where it is ionized, releasing a proton into the matrix. The resulting usniate anion can then diffuse back into the inter-membrane space where it binds to the proton on the acidic side of the inner membrane to re-form usnic acid which can then diffuse back into the matrix. The resulting cycle causes proton leakage that eventually can dissipate the proton gradient across the inner membrane, disrupting the tight coupling between electron transport and ATP synthesis. This model would explain the absence of gene-expression changes associated with complex V in usnic acid-treated mice, despite the increased electron transport by complexes I – IV. It may also explain the decline in ATP level in spite of increased oxygen consumption.</p>", "<p>In Table ##TAB##2##3##, only the two-sided correction method enables us to explain the function of usnic acid as described above. The one-sided correction method gives <italic>p</italic>-values similar to those in the two-sided correction method, but this is likely to be due to most of the gene expression changes entailing up-regulation. When the direction of gene expression change due to a treatment is known, then the one-sided correction method is the appropriate choice; it also needs less computation time than the two-sided correction method which employs Monte Carlo sample generation.</p>" ]

[ "<title>Discussion</title>", "<p>In many studies that use microarray data, the number of samples is small as in the first example shown above. While the number of samples in the simulations is only 3 for each group, the distribution of corrected <italic>p</italic>-values approximates a uniform distribution. The estimation for the correlation, , might not be very close to the true correlation, <bold>R</bold>. However, the correction methods that depend only on the average correlation, , are more robust because the estimation of is more robust.</p>", "<p>For the one-sided test, the correction for the correlation depends only on , the average of off-diagonal elements of the correlation matrix. The corrections using or are equivalent for the one-sided test method. Important points in choosing a correlation estimate for the two-sided test are the following; 1) The correlation estimate should be robust for small sample sizes, and 2) The correlation estimate should preserve . satisfies these two conditions.</p>", "<p>When the direction of gene expression change is pre-specified, the one-sided test is a good choice since it is easy and fast to calculate p-values. However, the two-sided test is the one we have to use in most cases, because it is usually not possible to pre-specify how individual genes will respond to treatment in the exploratory context. When we have a small number of samples to estimate the correlation, the method gives a robust result. Since misrepresents the true correlation, and gives biased p-values, the method works better for larger sample sizes. This is seen in Figure ##FIG##2##3##, where the method is better than the method. We hesitate to present a specific threshold sample size as sufficient for a converged correlation estimate, since it varies with respect to several conditions, such as the number of genes in a group, the variation of the elements of the correlation matrix, etc. The simulation result in Figure ##FIG##2##3## shows that both methods give quite accurate p-values compared to the p-values from the true correlation matrix, <bold>R</bold>. The Hotelling <italic>T</italic>²test is the best choice whenever it is applicable.</p>", "<p>In Table ##TAB##0##1##, the distribution of animals from different treatment groups (A-E) in three batches (1–3) gives no permutable pairs. In this case, randomization methods are not applicable. Even though randomization methods inherently take into account the correlation structure among genes, they may not be practical when the design of the experiment is complicated and the number of samples per group is small, reducing the numbers of permutable pairs.</p>", "<p>Randomization methods that permute class labels can adjust p-values for the correlation structure among genes. Randomization methods choose a summary statistic (e.g. enrichment score (ES) in [##REF##16199517##16##], average z-score in [##UREF##5##17##]), which reflects the degree to which a set of genes is enriched. When the significance of the summary statistic is measured by permuting class labels, the method preserves gene-gene correlations and when applicable, would give similar result to the presented method. Randomization methods that permute gene labels, such as Fisher's exact test, do not preserve the correlation structure and misrepresent the group's significance.</p>" ]

[ "<title>Conclusion</title>", "<p>We have presented a method to test the significance of expression changes within a group of genes, while considering the correlation structure among genes in each group. This method will enable the rapid detection of microarray evidence indicating altered cell functions or pathways, and will facilitate the interpretation of microarray outcomes. Application of the method to real data shows that it is an improved, practical method to evaluate the effects of treatments on functional classes of genes such as those based on Gene Ontology descriptors.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<p>In studies that use DNA arrays to assess changes in gene expression, it is preferable to measure the significance of treatment effects on a group of genes from a pathway or functional category such as gene ontology terms (GO terms, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.geneontology.org\"/>) because this facilitates the interpretation of effects and may markedly increase significance. A modified meta-analysis method to combine <italic>p</italic>-values was developed to measure the significance of an overall treatment effect on such functionally-defined groups of genes, taking into account the correlation structure among genes. For hypothesis testing that allows gene expression to change in both directions, <italic>p</italic>-values are calculated under the null distribution generated by a Monte Carlo method.</p>", "<p>As a test of this procedure, we attempted to distinguish altered pathways in microarray studies performed with Mitochips, oligonucleotide microarrays specific to mitochondrial DNA-encoded transcripts. We found that our analytic method improves the specificity of selection for altered pathways, due to incorporation of the inter-gene correlation structure in each pathway. It is thus a practical method to measure treatment effects on GO groups. In many actual applications, microarray experiments measure treatment effects under complicated design structures and with small sample sizes. For such applications to real data of limited statistical power, and also in computer simulations, we demonstrate that our method gives reasonable test results.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>TL conducted the simulations, analyzed the data and wrote the manuscript. VGD conducted the microarray experiments using Mitochip and made the biological interpretations. CV reviewed the literature. RJSR gave valuable suggestions on the preparation of the manuscript. RRD directed the methodology development, data analysis, and manuscript preparation. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>TL was supported by an Oak Ridge Institute of Science and Education (ORISE) fellowship at NCTR.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Cumulative distribution of p-values for one-sided test case with sample size n = 9</bold>. The naïve <italic>p</italic>-values (dashed red line) deviate from the diagonal line. Almost 30% of <italic>p</italic>-values are estimated to be less than 0.05. The corrected <italic>p</italic>-values (dashed blue line) fall very near the diagonal line.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Cumulative distribution of p-values for two-sided test case with sample size n = 9</bold>. <italic>P</italic>-values calculated from random samples based on (dashed blue line) and (dashed green line) give reliable corrections, while the naïve <italic>p</italic>-value (dashed red line) overstates the significance of the test.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Comparison of two-sided tests with sample size n = 25</bold>. Hotelling <italic>T</italic>²test (dashed blue line) gives the smallest difference from the uniform distribution. When we have enough number of samples to have a reasonably correct estimate of <bold>R</bold>, the method (dashed red line) is a little better than the method (dashed green line). Both the method and the method give quite accurate p-values compared to the p-values from the true correlation matrix, <bold>R </bold>(dashed cyan line).</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Experimental design for the AZT and 3TC effects on mouse-liver gene expression.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Genotype</bold></td><td align=\"center\" colspan=\"3\"><bold>(+/-)</bold></td><td align=\"center\" colspan=\"2\"><bold>(+/+)</bold></td></tr></thead><tbody><tr><td align=\"center\">Treatment</td><td align=\"center\">Vehicle</td><td align=\"center\">AZT 240 mg/kg/d</td><td align=\"center\">AZT+3TC 160+100 mg/kg/d</td><td align=\"center\">Vehicle</td><td align=\"center\">AZT+3TC 160+100 mg/kg/d</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\">Batch 1</td><td align=\"center\">A1</td><td align=\"center\">B1</td><td align=\"center\">C1</td><td align=\"center\">D1</td><td align=\"center\">E1</td></tr><tr><td align=\"center\">Batch 2</td><td align=\"center\">A2</td><td align=\"center\">B2</td><td align=\"center\">C2</td><td align=\"center\">D2</td><td align=\"center\">E2</td></tr><tr><td align=\"center\">Batch 3</td><td align=\"center\">A3</td><td align=\"center\">B3</td><td align=\"center\">C3</td><td align=\"center\">D3</td><td align=\"center\">E3</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Effects of AZT and 3TC on oxidative phosphorylation and apoptosis.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>gene group</bold></td><td align=\"left\"><bold># genes</bold></td><td align=\"left\"><bold>up</bold></td><td align=\"left\"><bold># P&lt;0.05</bold></td><td align=\"left\"><bold>Fisher's exact</bold></td><td align=\"left\"><bold>Not corrected</bold></td><td align=\"left\"><bold>One-sided correction</bold></td><td align=\"left\"><bold>Two-sided correction</bold></td></tr></thead><tbody><tr><td align=\"left\">Complex1</td><td align=\"left\">29</td><td align=\"left\">18</td><td align=\"left\">10</td><td align=\"left\">0.460</td><td align=\"left\">7.41E-6</td><td align=\"left\">0.600</td><td align=\"left\">0.049</td></tr><tr><td align=\"left\">Complex2</td><td align=\"left\">3</td><td align=\"left\">2</td><td align=\"left\">1</td><td align=\"left\">0.688</td><td align=\"left\">0.017</td><td align=\"left\">0.828</td><td align=\"left\">0.043</td></tr><tr><td align=\"left\">Complex3</td><td align=\"left\">7</td><td align=\"left\">5</td><td align=\"left\">3</td><td align=\"left\">0.401</td><td align=\"left\">5.3E-5</td><td align=\"left\">0.559</td><td align=\"left\">0.002</td></tr><tr><td align=\"left\">Complex4</td><td align=\"left\">13</td><td align=\"left\">8</td><td align=\"left\">8</td><td align=\"left\">0.026</td><td align=\"left\">1.25E-07</td><td align=\"left\">0.641</td><td align=\"left\">0.001</td></tr><tr><td align=\"left\">Complex5</td><td align=\"left\">14</td><td align=\"left\">6</td><td align=\"left\">6</td><td align=\"left\">0.273</td><td align=\"left\">0.0003</td><td align=\"left\">0.982</td><td align=\"left\">0.022</td></tr><tr><td align=\"left\">apoptosis</td><td align=\"left\">18</td><td align=\"left\">10</td><td align=\"left\">7</td><td align=\"left\">0.347</td><td align=\"left\">2.92E-5</td><td align=\"left\">0.592</td><td align=\"left\">0.005</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Effects of usnic acid on phosphorylation and apoptosis.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">gene group</td><td align=\"left\"># genes</td><td align=\"left\">up</td><td align=\"left\"># P&lt;0.05</td><td align=\"left\">Fisher's exact</td><td align=\"left\">Not corrected</td><td align=\"left\">One-sided correction</td><td align=\"left\">Two-sided correction</td></tr></thead><tbody><tr><td align=\"left\">Complex1</td><td align=\"left\">37</td><td align=\"left\">31</td><td align=\"left\">12</td><td align=\"left\">0.517</td><td align=\"left\">1.32E-12</td><td align=\"left\">0.027</td><td align=\"left\">0.022</td></tr><tr><td align=\"left\">Complex2</td><td align=\"left\">3</td><td align=\"left\">2</td><td align=\"left\">1</td><td align=\"left\">0.680</td><td align=\"left\">0.035</td><td align=\"left\">0.029</td><td align=\"left\">0.014</td></tr><tr><td align=\"left\">Complex3</td><td align=\"left\">7</td><td align=\"left\">5</td><td align=\"left\">2</td><td align=\"left\">0.704</td><td align=\"left\">0.019</td><td align=\"left\">0.055</td><td align=\"left\">0.042</td></tr><tr><td align=\"left\">Complex4</td><td align=\"left\">18</td><td align=\"left\">17</td><td align=\"left\">11</td><td align=\"left\">0.008</td><td align=\"left\">7.08E-10</td><td align=\"left\">0.006</td><td align=\"left\">0.003</td></tr><tr><td align=\"left\">Complex5</td><td align=\"left\">17</td><td align=\"left\">13</td><td align=\"left\">4</td><td align=\"left\">0.838</td><td align=\"left\">0.001</td><td align=\"left\">0.044</td><td align=\"left\">0.051</td></tr><tr><td align=\"left\">apoptosis</td><td align=\"left\">19</td><td align=\"left\">14</td><td align=\"left\">11</td><td align=\"left\">0.014</td><td align=\"left\">8.05E-10</td><td align=\"left\">0.008</td><td align=\"left\">0.0004</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S20-i1\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mtable columnalign=\"left\">\n <mml:mtr>\n <mml:mtd>\n <mml:mover accent=\"true\">\n <mml:mi>β</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mo>=</mml:mo>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>y</mml:mi>\n </mml:mstyle>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:msup>\n <mml:mover accent=\"true\">\n <mml:mi>σ</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mn>2</mml:mn>\n </mml:msup>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mrow>\n <mml:mi>n</mml:mi>\n <mml:mo>−</mml:mo>\n <mml:mi>r</mml:mi>\n </mml:mrow>\n </mml:mfrac>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>y</mml:mi>\n </mml:mstyle>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:mover accent=\"true\">\n <mml:mi>β</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mi>y</mml:mi>\n </mml:mstyle>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:mover accent=\"true\">\n <mml:mi>β</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>.</mml:mo>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S20-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>T</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>c</mml:mi></mml:mstyle><mml:mover accent=\"true\"><mml:mi>β</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mrow><mml:mrow><mml:mover accent=\"true\"><mml:mi>σ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:msqrt><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>c</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msup><mml:mi>X</mml:mi><mml:mo>′</mml:mo></mml:msup><mml:mi>X</mml:mi></mml:mstyle><mml:msup><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mo stretchy=\"false\">)</mml:mo></mml:mstyle><mml:mrow><mml:mo>−</mml:mo><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mn>1</mml:mn></mml:mstyle></mml:mrow></mml:msup><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mi>c</mml:mi><mml:mo>′</mml:mo></mml:msup></mml:mstyle></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S20-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup><mml:mrow><mml:mrow><mml:mrow><mml:msub><mml:mi>z</mml:mi><mml:mi>k</mml:mi></mml:msub></mml:mrow><mml:mo>/</mml:mo><mml:mrow><mml:msqrt><mml:mtext>m</mml:mtext></mml:msqrt></mml:mrow></mml:mrow></mml:mrow></mml:mstyle><mml:mo>~</mml:mo><mml:mi>N</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mn>0</mml:mn><mml:mo>,</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1471-2105-9-S9-S20-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>Φ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup><mml:mrow><mml:mrow><mml:mrow><mml:msub><mml:mi>z</mml:mi><mml:mi>k</mml:mi></mml:msub></mml:mrow><mml:mo>/</mml:mo><mml:mrow><mml:msqrt><mml:mtext>m</mml:mtext></mml:msqrt></mml:mrow></mml:mrow></mml:mrow></mml:mstyle></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1471-2105-9-S9-S20-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>T</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>c</mml:mi></mml:mstyle><mml:msub><mml:mover accent=\"true\"><mml:mi>β</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>k</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mi>σ</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:msqrt><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>c</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msup><mml:mi>X</mml:mi><mml:mo>′</mml:mo></mml:msup><mml:mi>X</mml:mi></mml:mstyle><mml:msup><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mo stretchy=\"false\">)</mml:mo></mml:mstyle><mml:mrow><mml:mo>−</mml:mo><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mn>1</mml:mn></mml:mstyle></mml:mrow></mml:msup><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mi>c</mml:mi><mml:mo>′</mml:mo></mml:msup></mml:mstyle></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1471-2105-9-S9-S20-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mn>1</mml:mn><mml:mo>′</mml:mo></mml:msup><mml:mi>z</mml:mi></mml:mstyle><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>z</mml:mi><mml:mi>k</mml:mi></mml:msub></mml:mrow></mml:mstyle><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup><mml:mrow><mml:msup><mml:mi>Φ</mml:mi><mml:mrow><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:msup><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:msub><mml:mi>p</mml:mi><mml:mi>k</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:msubsup><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:msubsup><mml:mrow><mml:msub><mml:mi>T</mml:mi><mml:mi>k</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1471-2105-9-S9-S20-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>var</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mn>1</mml:mn><mml:mo>′</mml:mo></mml:msup><mml:mi>z</mml:mi></mml:mstyle><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>k</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>m</mml:mi></mml:munderover><mml:mrow><mml:mi>var</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>T</mml:mi><mml:mi>k</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>+</mml:mo><mml:mn>2</mml:mn></mml:mrow></mml:mstyle><mml:mstyle displaystyle=\"true\"><mml:munder><mml:mo>∑</mml:mo><mml:mrow><mml:mi>s</mml:mi><mml:mo>&gt;</mml:mo><mml:mi>t</mml:mi></mml:mrow></mml:munder><mml:mrow><mml:mi>cov</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>T</mml:mi><mml:mi>s</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>T</mml:mi><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo></mml:mrow></mml:mstyle><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mn>1</mml:mn><mml:mo>'</mml:mo><mml:mi>R</mml:mi><mml:mn>1</mml:mn></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1471-2105-9-S9-S20-i8\" overflow=\"scroll\">\n <mml:semantics>\n <mml:mrow>\n <mml:mtable>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mi>cov</mml:mi>\n <mml:mo>⁡</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mi>T</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:msub>\n <mml:mi>T</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mo>=</mml:mo>\n <mml:mi>cov</mml:mi>\n <mml:mo>⁡</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n </mml:mstyle>\n <mml:msub>\n <mml:mover accent=\"true\">\n <mml:mi>β</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>s</mml:mi>\n </mml:msub>\n </mml:mrow>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:msqrt>\n <mml:mrow>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>c</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n </mml:mrow>\n </mml:msqrt>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>,</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n </mml:mstyle>\n <mml:msub>\n <mml:mover accent=\"true\">\n <mml:mi>β</mml:mi>\n <mml:mo>^</mml:mo>\n </mml:mover>\n <mml:mi>t</mml:mi>\n </mml:msub>\n </mml:mrow>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:msqrt>\n <mml:mrow>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>c</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n </mml:mrow>\n </mml:msqrt>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>c</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mi>cov</mml:mi>\n <mml:mo>⁡</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n <mml:msub>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>y</mml:mi>\n </mml:mstyle>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msub>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>y</mml:mi>\n </mml:mstyle>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mn>1</mml:mn>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>c</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n <mml:mi>cov</mml:mi>\n <mml:mo>⁡</mml:mo>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msub>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>y</mml:mi>\n </mml:mstyle>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:mo>,</mml:mo>\n <mml:msub>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>y</mml:mi>\n </mml:mstyle>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:mo stretchy=\"false\">)</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>c</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mrow>\n <mml:mi>s</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>t</mml:mi>\n </mml:mrow>\n </mml:msub>\n </mml:mrow>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>c</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n </mml:mrow>\n </mml:mfrac>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mi>c</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo stretchy=\"false\">(</mml:mo>\n <mml:msup>\n <mml:mi>X</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n <mml:mi>X</mml:mi>\n </mml:mstyle>\n <mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mo stretchy=\"false\">)</mml:mo>\n </mml:mstyle>\n <mml:mrow>\n <mml:mo>−</mml:mo>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:mn>1</mml:mn>\n </mml:mstyle>\n </mml:mrow>\n </mml:msup>\n <mml:mstyle mathsize=\"normal\" mathvariant=\"bold\">\n <mml:msup>\n <mml:mi>c</mml:mi>\n <mml:mo>′</mml:mo>\n </mml:msup>\n </mml:mstyle>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n <mml:mtr>\n <mml:mtd>\n <mml:mrow>\n <mml:mo>=</mml:mo>\n <mml:mfrac>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mrow>\n <mml:mi>s</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>t</mml:mi>\n </mml:mrow>\n </mml:msub>\n </mml:mrow>\n <mml:mrow>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>s</mml:mi>\n </mml:msub>\n <mml:msub>\n <mml:mi>σ</mml:mi>\n <mml:mi>t</mml:mi>\n </mml:msub>\n </mml:mrow>\n </mml:mfrac>\n <mml:mo>=</mml:mo>\n <mml:msub>\n <mml:mi>r</mml:mi>\n <mml:mrow>\n <mml:mi>s</mml:mi>\n <mml:mo>,</mml:mo>\n <mml:mi>t</mml:mi>\n </mml:mrow>\n </mml:msub>\n <mml:mo>,</mml:mo>\n </mml:mrow>\n </mml:mtd>\n </mml:mtr>\n </mml:mtable>\n </mml:mrow>\n \n </mml:semantics>\n </mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1471-2105-9-S9-S20-i9\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>Φ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mn>1</mml:mn><mml:mo>′</mml:mo></mml:msup><mml:mi>z</mml:mi></mml:mstyle></mml:mrow><mml:mrow><mml:msqrt><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mn>1</mml:mn><mml:mo>'</mml:mo><mml:mi>R</mml:mi><mml:mn>1</mml:mn></mml:mstyle></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M10\" name=\"1471-2105-9-S9-S20-i10\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>Φ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mn>1</mml:mn><mml:mo>′</mml:mo></mml:msup><mml:mi>z</mml:mi></mml:mstyle></mml:mrow><mml:mrow><mml:msqrt><mml:mtext>m</mml:mtext></mml:msqrt></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M11\" name=\"1471-2105-9-S9-S20-i11\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>Φ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mn>1</mml:mn><mml:mo>′</mml:mo></mml:msup><mml:mi>z</mml:mi></mml:mstyle></mml:mrow><mml:mrow><mml:msqrt><mml:mi>m</mml:mi></mml:msqrt></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:msqrt><mml:mrow><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:mfrac></mml:mrow></mml:msqrt></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M12\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M13\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M14\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M15\" name=\"1471-2105-9-S9-S20-i13\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>σ</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mrow><mml:mi>s</mml:mi><mml:mo>,</mml:mo><mml:mi>t</mml:mi></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mi>n</mml:mi><mml:mo>−</mml:mo><mml:mi>r</mml:mi></mml:mrow></mml:mfrac><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mo stretchy=\"false\">(</mml:mo></mml:mstyle><mml:msub><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>y</mml:mi></mml:mstyle><mml:mi>s</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>X</mml:mi></mml:mstyle><mml:msub><mml:mover accent=\"true\"><mml:mi>β</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>s</mml:mi></mml:msub><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>′</mml:mo></mml:msup><mml:mo stretchy=\"false\">(</mml:mo></mml:mstyle><mml:msub><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>y</mml:mi></mml:mstyle><mml:mi>t</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>X</mml:mi></mml:mstyle><mml:msub><mml:mover accent=\"true\"><mml:mi>β</mml:mi><mml:mo>^</mml:mo></mml:mover><mml:mi>t</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M16\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M17\" name=\"1471-2105-9-S9-S20-i14\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mi>r</mml:mi></mml:mfrac><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>l</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>r</mml:mi></mml:munderover><mml:mrow><mml:mtext>I</mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>Ψ</mml:mi><mml:mo>&gt;</mml:mo><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:msup><mml:mn>1</mml:mn><mml:mo>′</mml:mo></mml:msup></mml:mstyle><mml:mo>|</mml:mo><mml:msub><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mi>z</mml:mi></mml:mstyle><mml:mi>l</mml:mi></mml:msub><mml:mo>|</mml:mo></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M18\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M19\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M20\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M21\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M22\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M23\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M24\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M25\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M26\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M27\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M28\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M29\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M30\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M31\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M32\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M33\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M34\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M35\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M36\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M37\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M38\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M39\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M40\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M41\" name=\"1471-2105-9-S9-S20-i12\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mover accent=\"true\"><mml:mi>r</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M42\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M43\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M44\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M45\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M46\" name=\"1471-2105-9-S9-S20-i15\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>^</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M47\" name=\"1471-2105-9-S9-S20-i16\" overflow=\"scroll\"><mml:semantics><mml:mstyle mathsize=\"normal\" mathvariant=\"bold\"><mml:mover accent=\"true\"><mml:mi>R</mml:mi><mml:mo>¯</mml:mo></mml:mover></mml:mstyle></mml:semantics></mml:math></inline-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Fifteen samples are collected and assayed for gene expression using the experimental design</p></table-wrap-foot>", "<table-wrap-foot><p><italic>P</italic>-values calculated from four methods are presented for a comparison. The number of genes, up-regulated genes, and significantly expressed genes in each gene group are presented</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S20-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S20-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S20-3\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>MCBIOS 2008 was held February 23–24, 2008 in Oklahoma City, Oklahoma at the Cox Convention Center in Bricktown. It was the best attended in the series of MCBIOS conferences (140 registrants) with the most participation (68 posters submitted). Informative and engaging keynote talks were delivered by Dr. Bruce Roe and Dr. Edward Dougherty. The full agenda is online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.okbios.org\"/>.</p>", "<p>Student poster award winners were: Vinay Ravindrakumar of University of Arkansas for Medical Sciences (1st place), Quan Shi of Little Rock Central High School (2nd) and Brian Roux of the University of New Orleans (UNO) (3rd), with honorary mentions going to Murat Eren of UNO and Prashanti Manda of Mississippi State University (MSU). Student talk winners were: Daniel Quest of the University of Nebraska Medical Center (1st place), Nan Wang of MSU (2nd), and William Sanders of MSU (3rd).</p>" ]

[]

[]

[]

[]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>" ]

[ "<title>Proceedings summary</title>", "<p>This year, 19 out of 27 submitted papers were accepted for inclusion in the official conference proceedings (70%), similar to the number published from MCBIOS 2007 [##UREF##0##1##, ####REF##18047732##2##, ##REF##18047731##3##, ##REF##18047730##4##, ##REF##18047729##5##, ##REF##18047728##6##, ##REF##18047727##7##, ##REF##18047726##8##, ##REF##18047725##9##, ##REF##18047724##10##, ##REF##18047723##11##, ##REF##18047722##12##, ##REF##18047721##13##, ##REF##18047720##14##, ##REF##18047719##15##, ##REF##18047718##16##, ##REF##18047717##17##, ##REF##18047716##18##, ##REF##18047715##19##, ##REF##18047714##20##, ##REF##18047713##21##, ##REF##18047712##22##, ##REF##18047711##23##, ##REF##18047710##24##, ##REF##18047709##25##, ##REF##18047708##26####18047708##26##]. Each paper was peer-reviewed by at least two reviewers. Our goal in peer-review for the Proceedings is to be inclusive enough to accurately reflect the scope of scientific work presented at the conference yet rigorous enough such that only the highest quality work presented is selected for inclusion in the official proceedings. The general themes of this year's proceedings papers fall into five categories, discussed below.</p>", "<title>Systems biology</title>", "<p>Biological systems can be modeled as complex systems with many interactions between the components. One goal of emerging systems biology is to analyze very large complex biological networks such as protein-protein interactions, metabolism, and regulation to identify functional modules and to assign the functions to certain components of the system. Mutlu Mete <italic>et al</italic>. [##REF##18793464##27##] devised a new methodology called SCAN (Structural Clustering Algorithm for Networks) that can efficiently find clusters or functional modules in complex biological networks, as well as hubs and outliers. In addition, nodes can be classified into various roles based on their structures. Interpretations of functional groups found by SCAN showed superior performance over CNM, a well-known modularity-based clustering algorithm.</p>", "<p>Analysis of microarray gene expression data is challenging and may lead to biased or incomplete biological interpretations. To gain a more holistic (i.e., systemic) picture, it is essential to integrate a careful statistical approach with biological knowledge from various sources into the analysis. Mikhail Dozmorov <italic>et al</italic>. [##REF##18793468##28##] present an integrative approach to microarray analysis and demonstrate how the various steps in their process support each other and refine the current model of cell-matrix interaction. With their method, they were able to identify inflammation and G-protein signaling as processes affected by the extracellular matrix.</p>", "<p>Metastases are responsible for the majority of cancer fatalities. The molecular mechanisms governing metastasis are poorly understood, hindering early diagnosis and treatment. Unlike most previous studies, a study by Andrey Ptitsyn <italic>et al</italic>. [##REF##18793472##29##] proposes an approach that puts into focus gene interaction networks and molecular pathways rather than separate marker genes. This study indicates that regardless of the tissue of origin, all metastatic tumors share a number of common features related to changes in basic energy metabolism, cell adhesion/cytoskeleton remodeling, antigen presentation and cell cycle regulation.</p>", "<p>Circadian rhythm is a crucial factor in orchestration of plant physiology, keeping it in synchrony with the daylight cycle. Previous studies reported approximately 16% of plant genes behaved in a circadian fashion, while studies in mammals suggested circadian baseline oscillation in nearly 100% of genes. Andrey Ptitsyn [##REF##18793463##30##] presents a comprehensive analysis of periodicity in two independent <italic>Arabidopsis thaliana </italic>data sets. This study indicates a more pervasive role of gene expression oscillation in the molecular physiology of plants than previously believed. Application of advanced algorithms identified circadian baseline oscillation in almost all plant genes as well as a complex orchestration of gene expression timing in important biological pathways.</p>", "<title>OMICS</title>", "<p>Chromatography coupled to mass spectrometry is a powerful way to resolve and compare the relative abundance of chemical compounds within heterogeneous biological samples. However the resulting data sets are 2 or 3-dimensional, presenting formidable obstacles to peak alignment – a process required to ensure sample comparison is conducted appropriately. The first dimension of separation is chromatographic elution time, which varies from run to run for each molecular species. To solve this problem, Minho Chae <italic>et al</italic>. [##REF##18793460##31##] developed an iterative block-shifting approach that adjusts for variation in retention time without distorting peak area. They first matched chemically identical peaks based on both retention-time and mass-spectral information. Non-peak regions of each chromatogram were stretched or compressed to align peaks with a reference chromatogram, thus preserving the shapes of matched peaks. Their approach compared favorably to other approaches, and was superior in preservation of peak area.</p>", "<p>Also, in the proceedings, Tianxiao Huan <italic>et al</italic>. describe Proteolens, a new tool to navigate and visualize biological networks [##REF##18793469##32##].</p>", "<title>Microarray studies</title>", "<p>Microarrays are a powerful technology and an area of active research interest in bioinformatics, with a focus on the development of novel methods for analysis and interpretation of experiments [##REF##17553857##33##, ####REF##17537751##34##, ##REF##17646348##35##, ##REF##17646292##36##, ##REF##17553854##37##, ##REF##17463023##38##, ##REF##17463021##39##, ##REF##17890270##40##, ##REF##17537754##41##, ##REF##17724061##42##, ##REF##17646307##43##, ##REF##17267426##44##, ##REF##17138587##45##, ##REF##17158516##46##, ##REF##17118957##47##, ##REF##17496320##48##, ##REF##17872912##49####17872912##49##]. This year's proceedings reflect this area of active research interest with several reports that focus on the development of methods and analysis of microarray data.</p>", "<p>Microarray-based molecular signatures have played an increasing role in diagnosis, prognosis and risk/safety assessments, the first step of which is to identify a set of informative genes. Zhenqiang Su <italic>et al</italic>. [##REF##18793473##50##] investigate a new gene selection approach to identify informative genes. The rationale of the approach is that informative genes should consistently be significantly differentially expressed for different variations of sample size. Genes exhibiting significance throughout the iterations are considered a Very Important Pool (VIP) of genes. It was found that the genes identified by the VIP method, but not by the p-value ranking approach, are also related to the disease investigated, and these genes are part of the pathways derived from the common genes shared by both the VIP and p-ranking methods. Moreover, the binary classifiers built from these genes are statistically equivalent to those built from the top 50 p-value ranked genes in distinguishing different types of samples. Therefore, the VIP gene selection approach could identify additional subsets of informative genes that would not always be selected by the p-value ranking method.</p>", "<p>The paper by Taewon Lee <italic>et al</italic>. [##REF##18315852##51##] presents a method to test the significance of expression changes within a group of genes, while considering the correlation structure among genes in each group. This method enables the rapid detection of gene expression changes, indicating altered cell functions or pathways, and facilitates the interpretation of the data. Application of the method to real data shows that it is an improved, practical method to evaluate the effects of treatments on functional classes of genes, such as those based on Gene Ontology descriptors.</p>", "<p>Also in the proceedings, Arun Rawat <italic>et al</italic>. report on a method of microarray graph mining to derive co-expressed genes [##REF##18793471##52##], and Leming Shi <italic>et al</italic>. report on an impressively large study of the reproducibility of gene lists for microarray experiments, and conclude with recommendations for detecting significant differential expression [##REF##18793455##53##].</p>", "<title>Genomic analysis</title>", "<p>As more and more genomes become fully sequenced in the coming years, gene identification is still a limiting factor to scientific discovery. Since a significant proportion of genes exist as members of families of genes with related functions, Ronald Frank <italic>et al</italic>. [##REF##18793465##54##] have employed a strategy to identify these gene family members using patterns indicating negative selection pressure on the coding region. The authors tested the strategy on several well-characterized gene families from Arabidopsis thaliana and report their success in correctly identifying several members of each gene family starting with one known member and using only EST data.</p>", "<p>Highly accurate and reproducible genotype calling are paramount for genome-wide association studies (GWAS), since errors introduced by calling algorithms can lead to inflation of false associations between genotype and phenotype. Most genotype calling algorithms currently used for GWAS are based on multiple arrays, consisting of many samples. Huixiao Hong <italic>et al</italic>. [##REF##18793462##55##] observed that batch size and composition affect the genotype calling results in GWAS using the algorithm BRLMM. The larger the differences in batch sizes, the larger the effect. The more homogenous the samples in the batches, the more consistent the genotype calls. The inconsistency propagates to the lists of significantly associated single nucleotide polymorphisms identified in downstream association analysis. Thus, uniform and large batch sizes should be used to make genotype calls for GWAS. In addition, samples of high homogeneity should be placed into the same batch.</p>", "<p>The cellular machinery by which genes are expressed is both complex and an active area of recent bioinformatics research [##REF##17893090##56##, ####REF##17488754##57##, ##REF##17893089##58##, ##REF##17660203##59##, ##REF##17890736##60##, ##REF##17846037##61##, ##REF##17392330##62##, ##REF##16397004##63##, ##REF##16873497##64##, ##REF##17021159##65##, ##REF##16873509##66####16873509##66##]. A first step in understanding this process is to locate the binding positions of transcription factors over the chromosome. Since the search space is large, advanced computational tools play a central role in solving this problem. Despite the development of nearly two hundred tools to elucidate transcription factor binding sites, much controversy still remains on how to build methods with high sensitivity and specificity. Central in this debate is determining the factors that will improve the quality of computational predictions. The paper by Daniel Quest <italic>et al</italic>. [##REF##18793470##67##], presents a novel benchmarking strategy to automate and evaluate methods designed to detect transcription factor binding sites. The strategy allows researchers, for the first time, to evaluate transcription factor detection methods on the genome scale. In particular, researchers can vary the data, algorithms, parameters and transcription factor binding site representations to determine the method best suited to their problem of interest. The proposed platform allows for rapid evaluation of deficits in current models and paves the way to develop new tools to overcome these problems.</p>", "<p>Also, the Garner Lab extends their work on predicting the impact of single nucleotide polymorphisms (SNPs) in a paper by Vinayak Kulkarni <italic>et al</italic>. [##REF##18793467##68##], and Jerzy Zielinski <italic>et al</italic>. report on a method of analyzing genomic sequences by a time-dependent autoregressive moving average [##REF##18793459##69##].</p>", "<title>Miscellaneous</title>", "<p>Text-mining is an area of bioinformatics whereby identification and analysis of trends in text is done computationally [##REF##17855415##70##, ####REF##17463015##71##, ##REF##17314123##72##, ##REF##17947625##73##, ##REF##17600104##74##, ##REF##16799122##75##, ##REF##17460124##76##, ##REF##17329723##77##, ##REF##17600094##78####17600094##78##]. To this end, Cory Giles and Jonathan Wren developed a method of identifying directional relationships within text (e.g., chemical X increases heart rate, or gene Y elevates inflammation) using natural language processing (NLP) [##REF##18793456##79##]. Their goals were also to make their system scalable to large bodies of text (e.g. MEDLINE has 18 million records and counting), as well as understanding how much apparent contradiction takes place when attempting to extract isolated facts from within a greater context from these huge bodies of text.</p>", "<p>Christopher Bottoms and Dong Xu study atom-naming conventions in the Protein Data Bank and find that some names are assigned ad hoc, resulting in duplicate names and creating problems for standardization and data-mining [##REF##18793461##80##].</p>", "<p>In [##REF##18793457##81##], Roux and Winters-Hilt describe Hybrid SVM/HMM structural sensors for use in analysis of stochastic sequential data. They begin with a novel approach to classification using Support Vector Machines and Markov Models with application to detecting Intron-Exon and Exon-Intron (5' and 3') splice sites. The approach also includes the application of Shannon Entropy based analysis of the stochastic datasets to detect minimal data components for feature extraction. Results are presented for a variety of eukaryotic species.</p>", "<p>In the Winters-Hilt group, work continues on developing nanopore detector signal analysis via machine learning methods for classification and knowledge discovery. In [##REF##18793458##82##], Churbanov and Winters-Hilt describe the application of a distributed Mixture of Hidden Markov Models (MHMMs) to the problem of channel current blockade clustering and associated analyte classification. The distributed MHMM provides a feature extraction that is equivalent to that of the sequential HMM with a speedup factor approximately equal to the number of independent CPUs operating on the data.</p>", "<title>Future meetings</title>", "<p>The Sixth annual MCBIOS Conference will be held in Starkville, Mississippi in early spring, 2009. See <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.MCBIOS.org\"/> for further information on MCBIOS and future meetings. MCBIOS and OKBIOS are both regional affiliates of the International Society for Computational Biology <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ISCB.org\"/>.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>All authors served as co-editors for these proceedings, with JDW serving as Senior Editor. All authors helped write this editorial. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the Food and Drug Administration.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We thank the Conference Committee, Program Committee, student volunteers and sponsors for their help in organizing MCBIOS 2008. We would like to especially thank Jim Mason and Frank Waxman for their sponsorship of MCBIOS 2008 as well as OKBIOS 2004 and OKBIOS 2005. We also thank our peer-reviewers for their quality efforts to review submitted manuscripts.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>A significant proportion of genes that make up a genome are part of larger families of related genes resulting from duplications of individual genes [##REF##15568988##1##], genomic segments [##UREF##0##2##], or even whole genomes ([##UREF##1##3##,##UREF##2##4##]. The accumulation of mutations in duplicates (paralogs) leads to either loss of function for one (death), altered function (subfunctionalization), or a new function (neofunctionalization). The study of the molecular processes by which functional innovation occurs interests not only evolutionary biologists, but protein engineers and medical and agricultural biologists. A clearer understanding of the extent to which gene families contribute to the selected traits in our most important crop species will help guide decisions regarding future improvements. Many studies are aimed at the diversity of function, expression, and regulation among gene family members in many species (reviewed in [##UREF##3##5##]). Others have spawned computational methods to analyze and predict the evolution of gene families in a phylogenetic context [##UREF##4##6##] or determine clinically relevant sites in a protein sequence where amino acid replacements are likely to have a significant effect on phenotype, including those that may cause genetic diseases [##REF##16522197##7##].</p>", "<p>Therefore, it is not surprising that research aimed at the identification of specific gene families and their constituent members has proliferated in the last few decades. Although experimental approaches using degenerate primers for PCR and oligofingerprinting [##REF##12213199##8##] and cDNA library screening [##REF##17216438##9##] generally produce the most reliable results, they can be time consuming and labor-intensive. Many strategies of gene family identification are computational approaches that take advantage of database mining and analysis tools to increase the capability and improve the efficiency of dealing with large amounts of sequenced data. Naturally, if a significant amount of a genome is sequenced computational methods can be somewhat more exhaustive in their search and identification [##REF##15799777##10##, ####REF##12952558##11##, ##REF##16455358##12##, ##REF##16407444##13##, ##REF##15316287##14##, ##REF##17407609##15####17407609##15##]. However, complete genomic data is available for only a limited number of species. Expressed sequence tags (ESTs) on the other hand, are short, unedited, randomly selected single-pass sequences. They can be easily and inexpensively obtained directly from cDNA libraries. Although they were initially used for human gene discovery [##REF##2047873##16##,##REF##1538749##17##], exponential growth in the generation and accumulation of EST data for many diverse organisms has occurred in the last decade. The National Center for Biotechnology Information (NCBI) has a database for ESTs from over 1300 species totaling more than 48 million ESTs (as of 14 December 2007). Sixty-three species have more than 100,000 ESTs in the database making computational analyses more fruitful but complex. Because the number of ESTs in databases is increasing, computational techniques, including BLAST and its variants for comparative analysis and CAP3 [##REF##10508846##18##] for sequence assembly, can be used to speed up gene or gene family identification processes and improve the feasibility of extracting meaningful information from a large and redundant database [##REF##16772268##19##] when parameters are properly selected. These EST-based gene family identification strategies are valuable in species without fully sequenced genomes [##REF##12713273##20##,##REF##10207159##21##]. Caution must be exercised when assembling contigs from EST sequences because contigs not representative of real genes can result from chimera formation during cDNA cloning, errors in single-pass high-throughput sequencing of ESTs, or similarity between protein domains of unrelated sequences. Our group has developed a simple but novel method using evidence of negative selection pressure during divergence of the coding sequences to filter artifactual contigs from those potentially representing actual gene family members. Molecular evolution researchers studying divergence between well-characterized orthologs or paralogs often employ an estimation of the number of synonymous base substitutions per synonymous site versus the number of nonsynonymous base substitutions per nonsynonymous site [##REF##3444411##22##,##REF##8078400##23##]. A dS/dN ratio &gt; 1 indicates purifying or negative selection (lower fitness) that tends to keep amino acid sequences the same if changes were deleterious. A ratio equal to 1 indicates changes that were neutral to fitness, while a dS/dN ratio &lt; 1 would indicate adaptive or positive selection presumably because natural selection favored the amino acid changes. Differences between contigs that are artifactual should be proportionally distributed among synonymous and nonsynonymous sites, whereas differences between contigs that represent paralogs will often exhibit negative selection, dS/dN &gt; 1.</p>", "<p>We understand that negative selection may not be uniform over entire coding regions even assuming that purifying selection was at work in a given gene family. And not all gene families will exhibit negative selection between members. However, we believe that the number of gene families that can be detected by this approach is significant. Evidence has been found for a model whereby complementary deleterious mutations in regulatory elements between duplicate genes partitions the original function resulting in sub-functions [##REF##10101175##24##]. It has also been discovered that the number of shared regulatory elements between duplicated genes in yeast decreases with evolutionary time [##REF##12902158##25##]. The age of the duplicates was estimated by the accumulation of synonymous substitutions in the coding regions. Clearly, some forms of subfunctionalization can occur by changes in regulatory elements whereby some degree of negative selection has maintained protein function. Coding regions of paralogs that have subfunctionalized via changes to regulatory elements should exhibit a bias toward synonymous substitutions. In plants, a significantly greater proportion of genes belong to gene families than in animals or other major taxa [##REF##15680516##26##]. Either gene duplication events have been more common in plants, or more duplicates have been retained during the evolutionary history of plants [##REF##16166257##27##]. If this is the case, there should exist a significant number of gene families that can be identified by a bias toward synonymous substitutions when contigs are assembled from a significantly large database of ESTs. We have demonstrated previously that a simple strategy to detect negative selection patterns (NSP) among assembled ESTs provides a good screen for real versus artifactual contigs [##REF##17118140##28##]. We have modified the filtering criterion to an empirically determined dS/dN threshold and decided to test the negative selection pattern (NSP) strategy on an EST database for which a large percentage of the ESTs have already been mapped to a fully-sequenced genome, <italic>Arabidopsis thaliana</italic>.</p>", "<p>In this article we demonstrate the NSP strategy and report how well it was able to identify ESTs representing distinct family members in a genome where it is testable.</p>" ]

[ "<title>Methods</title>", "<title>Gene family identification by NSP method</title>", "<p>The five gene families chosen to validate the NSP strategy were, eukaryotic release factor 1 (<italic>RF1</italic>), ribosomal protein L6 (<italic>L6</italic>), cinnamyl alcohol dehydrogenase (<italic>CAD</italic>), phenylalanine ammonia-lyase (<italic>PAL</italic>), and an FtsH protease (<italic>FtsH</italic>). One member of the selected gene family was chosen as query for a tblastn search of the <italic>Arabidopsis thaliana </italic>dbEST. All hits with an E value &lt; 1 × 10^-10(maximum of 150 sequences) were selected and the resulting EST sequences were assembled using a contig assembly program (AssemblyLIGN, Oxford Molecular) with 100% match over a minimum 100 nucleotide overlap. The largest open reading frame greater than 100 codons was identified in each resulting non-singleton contig (MacVector, Accelrys). Open reading frames were translated and the resulting polypeptides aligned using ClustalX. The PAL2NAL program [##REF##16845082##29##] produced a codon alignment of all contig open reading frames, and the SNAP program [##UREF##5##30##] at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.hiv.lanl.gov\"/> was used to calculate dS/dN for all pairwise comparisons of contig open reading frames.</p>", "<p>The empirically determined threshold for dS/dN was set to 2.00 and all pairs of contigs with a dS/dN ratio greater than this were classified as putative paralogs. A graph was constructed using vertices to represent contigs, and edges to represent whether pairs of contigs are putative paralogs. In such a graph, the largest fully connected sub-graph (the maximum clique) will be made up of vertices that represent markers (contigs) to the members of the same gene family as the query protein. This sub-graph was determined using a brute-force algorithm. A brute-force algorithm works by checking every possible sub-graph for connectedness. This operation is computationally expensive, and its time complexity increases exponentially, as the factorial of the number of vertices. Fortunately, the contigs that these vertices represent are usually quite few in number. Some contigs can also be excluded from the graph since they do not pass the dS/dN threshold to pair with any other contig. This can be observed in Figure ##FIG##0##1## where only 5 pairwise comparisons of contigs obtained a dS/dN of more than 2.00.</p>", "<p>Figure ##FIG##0##1## shows the dS/dN ratios between contigs generated using the PAL1 gene as the protein query. Note that there are two maximum cliques in this graph. When there are more than one maximum cliques, we arbitrarily choose one of these cliques. The contigs represented by the vertices belonging to this clique are then identified as members of the same gene family. Any vertices that are not part of this clique are classified as either a possible marker to a distinct gene, or as a duplicate marker to an identified gene family member (in the maximum clique) which was different enough to be assembled into a different contig. In the case of Contig3 and Contig9 from Figure ##FIG##0##1##, it was found that these contigs were extremely similar to each other. They were later found to be duplicate markers to the PAL4 gene.</p>", "<p>The representative contig for each putative gene family member identified was then compared to each of the actual gene family member sequences (NCBI) using bl2seq [##REF##10339815##31##] to determine how closely contigs filtered through NSP represented the gene family. Either all or a subset of ESTs from each NSP-identified contig were checked on MapViewer (NCBI) to determine if ESTs from the same contig mapped to different gene family members or if ESTs from different contigs mapped to the same gene family member.</p>" ]

[ "<title>Results</title>", "<title>Phenylalanine ammonia-lyase gene family</title>", "<p>The tblastn search of using <italic>AtPAL1 </italic>protein as query resulted in ESTs and contigs reported previously [##REF##17118140##28##]. Here we report the refinement of using dS/dN ratio rather than a tally of 1^st, 2^nd, and 3^rdposition differences as well as the MapViewer results that validate the accuracy of gene family member identification. The dS/dN data for the assembled contigs are shown in Table ##TAB##0##1## and the resulting maximum clique graph indicating putative paralog relationships is shown in Figure ##FIG##0##1##. The 2.0 dS/dN threshold was established empirically by dS/dN measurements among actual members of several Arabidopsis gene families. Pairwise comparison of contigs 1, 3, and 4 with the actual Arabidopsis gene sequences, reported previously [##REF##17118140##28##] indicate that these three contigs represent AtPAL1, AtPAL4, and AtPAL2, respectively with greater than 96% similarity. The contigs selected by NSP as representative of real gene family members were further validated by checking to see if each EST comprising a single contig is assigned to a single gene family member on the Arabidopsis genome by NCBI MapViewer. Table ##TAB##1##2## shows that all ESTs that comprise a single contig map to the same gene locus and confirms that contigs 1, 3, and 4 represent the PAL1, PAL4, and PAL2 genes of Arabidopsis, respectively.</p>", "<p>For the following four additional NSP-identified gene families only the validating data is shown, not the dS/dN data or maximum clique graphs.</p>", "<title>Ribosomal protein L6 gene family</title>", "<p><italic>AtRPL6A </italic>protein was used as query for the tblastn search of <italic>A. thaliana </italic>dbEST yielding 150 EST sequences that assembled into eight contigs ranging from 449 to 953 bases and 2 to 36 ESTs each. Following ORF identification the 28 pairwise codon alignments and subsequent dS/dN values were analyzed to sort contigs into putative gene family members. From that analysis contig1, contig3 and contig8 were assigned to putative geneA, contig2, contig4, and contig5 to geneB, and contig6 to geneC. Table ##TAB##2##3## shows that each of these contig groups identified, by greater than 98% similarity, a different member of the Arabidopsis ribosomal protein <italic>L6 </italic>gene family when aligned to the actual gene sequences.</p>", "<p>In Table ##TAB##1##2## for ribosomal protein L6 it can be seen that all ESTs from the same contig as well as all ESTs from the same gene grouping are assigned to the same gene locus. Also, in no instances did ESTs belonging to different gene groupings by NSP ever map to the same gene locus.</p>", "<title>Cinnamyl alcohol dehydrogenase gene family</title>", "<p>The tblastn search of using <italic>AtCAD5 </italic>protein as query resulted in 150 EST sequences. The ESTs assembled into eight contigs ranging from 592 to 1248 bases and 2 to 21 ESTs each. Following ORF identification the 28 pairwise codon alignments and subsequent dS/dN values were analyzed to sort contigs into putative gene family members as described above (data not shown). The eight contigs assorted into four groups based on their negative selection pattern with each other contig. These four groups were arbitrarily designated GeneA represented by contig4, contig6, and contig8, GeneB, represented by contig3, contig5, and possibly contig7, GeneC represented by contig1, and GeneD represented by contig2.</p>", "<p>The results of the comparison of representative contigs to the actual gene sequences for the CAD gene family of Arabidopsis are shown in Table ##TAB##3##4##. Each contig group identified, by greater than 99% similarity, a different member of the CAD gene family. MapViewer analysis for the CAD gene family (Table ##TAB##1##2##) shows that all ESTs from the same contig are assigned to the same gene locus, and no ESTs belonging to different contigs map to the same gene locus. Contigs validated by alignment to actual genes but not shown in Table ##TAB##1##2## are comprised of ESTs that have not yet been mapped to specific loci by MapViewer.</p>", "<title>Release Factor 1 gene family</title>", "<p><italic>AtRF1-3 </italic>protein was used as query for the tblastn search of <italic>A. thaliana </italic>dbEST yielding 109 EST sequences that assembled into six contigs ranging from 591 to 930 bases and three to 19 ESTs each. Following ORF identification the 15 pairwise codon alignments by the NSP program resulted in three contigs exhibiting NSP. These were arbitrarily assigned as contig1 representing geneA, contig3 representing geneB, and contig6 representing geneC. Each of these contigs identified, by greater than 97% similarity, a different member of the Arabidopsis RF1 gene family when aligned to the actual gene sequences (Table ##TAB##4##5##). MapViewer results again show that ESTs comprising NSP-selected contigs are unambiguous in the gene locus to which they have been assigned (Table ##TAB##1##2##).</p>", "<title>FtsH protease gene family</title>", "<p>The TBLASTN search of using <italic>AtFtsH8 </italic>protein as query resulted in 150 EST sequences. The ESTs assembled into six contigs ranging from 526 to 1217 bases and 2 to 33 ESTs each. Following ORF identification the 15 pairwise alignments by the NSP program resulted in two contig groups exhibiting NSP. Contig1 and contig5 represent geneA, and contig3 and contig6 represent geneB. Each of these contig groups identified, by greater than 97% similarity, a different member of the Arabidopsis <italic>FtsH </italic>gene family when aligned to the actual gene sequences, as shown in Table ##TAB##5##6##. MapViewer results again show that ESTs comprising NSP-selected contigs are unambiguous in the gene locus to which they have been assigned (Table ##TAB##1##2##).</p>" ]

[ "<title>Discussion</title>", "<p>It has been observed for some time that contig assembly from EST sequences can produce artifactual sequences resulting from relatively high error in EST sequences, chimeras generated in cDNA cloning, and regions of highly conserved domains interspersed in related genes. Therefore, it is necessary that strategies involving the generation of contigs from ESTs employ some criterion for either eliminating unauthentic coding regions or selecting for authentic ones. We have found that contigs representing gene families where the paralogous coding regions have been constrained by negative (purifying) selection pressure can be identified by screening for amino acid substitution patterns indicative of such (NSP, Negative Selection Patterns). However, if differences between contigs are artifacts no pattern among codon positions should be exhibited. If no negative selection pattern is detected we do not conclude that the contigs necessarily represent the same gene. Our goal is only to identify contigs that represent different genes of the same family. We do not expect that all members of a particular family will be detectable by this method. Other members may be identified with iterative searches using previously identified contigs.</p>", "<p>To illustrate that this method can identify members of a gene family with some accuracy using only EST data we tested it on five well-characterized gene families in Arabidopsis. Each case resulted in successful identification of one to three additional gene family members distinct from the member used as initial query. Of the eight initial contigs generated from EST hits when <italic>AtCAD5 </italic>was used as query the NSP strategy identified those representing <italic>AtCAD1</italic>, <italic>AtCAD2</italic>, and <italic>AtCAD4</italic>, in addition to one representing <italic>AtCAD5 </italic>(Table ##TAB##3##4##). Moreover, each of these contigs exhibited less than 87% similarity to other actual members of the gene family. No contigs generated at the parameters specified in the assembly program represented <italic>AtCAD3</italic>, 6, 7, 8 or 9. This could be the result of relative expression levels of those genes, limits on the necessary similarity between gene family members, or limitations on the method which are discussed elsewhere [##REF##17118140##28##]. Similarly, when ribosomal protein <italic>L6A </italic>was used as query the NSP strategy identified contigs accurately representing all three genes of the family, <italic>L6A</italic>, <italic>L6B</italic>, and <italic>L6C </italic>(Table ##TAB##2##3##). Furthermore, all three members of the <italic>RF1 </italic>gene family were accurately represented by NSP-screened contigs (Table ##TAB##4##5##), as were <italic>AtFtsH2 </italic>and <italic>AtFtsH8 </italic>of that 12-member gene family (Table ##TAB##5##6##). We previously reported the accurate identification of <italic>AtPAL1</italic>, 2, and 4 of phenylalanine ammonia-lyase gene family and show here further validation that the contigs identified the appropriate gene family members.</p>", "<p>In addition, we were able to show that all the ESTs of a single contig defined the same actual gene family member according to MapViewer (Table ##TAB##1##2##), i.e., all ESTs of a single contig mapped to the same locus, and perhaps more importantly, no ESTs from different contigs of the same gene family ever mapped to the same locus. This would suggest that although the initial assembly of related ESTs may indeed generate non-valid contigs, screening by NSP allows one to determine which contigs represent real gene loci.</p>", "<p>A limitation to the NSP strategy is the fact that only paralogs that exhibit purifying selection can be identified and that selection pattern must be evident in the portion of the coding region reconstructed by contig assembly, roughly the 3' two-thirds of the protein by our experience. For this reason the NSP strategy in it current phase will only identify a subset of gene families. However, when we consider that estimates of the number of gene families in a plant species may be 10–12,000 [##REF##16899135##32##], that subset may comprise a significant portion in which NSP can detect two to three additional family members. Our goal is to broaden the NSP approach to identify as many gene families as possible without sacrificing the accuracy reported here. We have already automated the four basic steps, 1) BLAST collection of related ESTs, 2) contig assembly, 3) ORF identification, and 4) NSP screening of contigs, such that the input is a query protein of a potential gene family member and the output is contigs representing at least two gene family members. Since the query can be an orthologous sequence, we are currently working on identifying, in <italic>Glycine max</italic>, every gene family for which at least one member has been identified in another plant species. The specific objectives for accomplishing this are to:</p>", "<p>1) use all known <italic>Glycine max </italic>mRNAs as queries to identify other family members, if any.</p>", "<p>2) use mRNAs from related species as queries to identify gene families not identified above.</p>", "<p>3) use Arabidopsis gene families as queries (currently about 1000 gene families in TAIR).</p>", "<p>4) use other Arabidopsis genes, not currently associated with a family as queries to identify potential genes existing as a family in soybean but not so in Arabidopsis.</p>", "<p>5) use all <italic>Glycine max </italic>ESTs not included in contigs from above searches in clustering experiments to potentially identify novel gene families.</p>", "<p>Objectives 1–4 above are identical in protocol. They differ only in the species of origin for the protein query. There are currently about 1350 known <italic>Glycine max </italic>gene sequences in the NCBI database, mostly mRNA sequences but some genomic. Some of these already represent multiple members of the same gene family (e.g. glycinin and conglycinin seed storage proteins, uricase, ascorbate peroxidase, lipoxygenase, rubicase small subunit, phosphoenolpyruvate carboxylase, etc) [##REF##16026604##33##]. Objective 1 will use all known genes of soybean as queries to identify other members of the gene family. Objective 2 involves genes from species more closely related to soybean than Arabidopsis. These include other eurosids I and particularly other legumes that have significant sequence data available such as <italic>Pisum sativum, Phaseolus vulgaris</italic>, and <italic>Medicago truncatula</italic>. Objective 3 will involve queries chosen from Arabidopsis genes that are known to exist as part of a gene family. Currently, The Arabidopsis Information Resource (TAIR) has genomic, coding region, and amino acid sequence data for 996 gene families comprised of 8,331 genes. Objective 4 will use as initial queries all remaining Arabidopsis genes not already identified in soybean and not associated with a gene family in TAIR. It is possible that of the remaining 16,000 genes of Arabidopsis there could be some that are associated with a family in <italic>Glycine max</italic>. Objective 5 does not start with a query sequence but rather a set of ESTs clustered by similarity to each other. Several clustering algorithms could be used for this, UniGene (at NCBI), PACE [##REF##12771222##34##], or one developed in our laboratory several years ago. The majority of UniGene clusters are annotated with \"strongly similar to,\" moderately similar to,\" or \"weakly similar to\" gene or protein functions of other organisms. Others are labeled simply as \"Transcribed locus\" to indicate that they represent RNA sequences that do not show similarity to any currently known gene or protein (Build #31 has 6812 such clusters). We have run a few of these clusters through the NSP strategy and found that some will generate contigs that indicate the cluster may represent ESTs from distinct members of a gene family. More work in this direction will allow us to expand the strategy to include identification of yet undiscovered gene families.</p>" ]

[ "<title>Conclusion</title>", "<p>Although the NSP strategy is not a global gene family identification protocol, our tests on the Arabidopsis EST dataset indicate that it performs well in distinguishing contigs that represent real genes from contigs that are artifacts. Every EST tested, from contigs that NSP predicted to be distinct gene family members, mapped to the appropriate gene in Arabidopsis. Further expansion of the strategy to clustered ESTs eliminating the need for individual query sequences and further automation of the steps will allow the identification of a significant proportion of gene families with reliable accuracy.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Gene family identification from ESTs can be a valuable resource for analysis of genome evolution but presents unique challenges in organisms for which the entire genome is not yet sequenced. We have developed a novel gene family identification method based on negative selection patterns (NSP) between family members to screen EST-generated contigs. This strategy was tested on five known gene families in Arabidopsis to see if individual paralogs could be identified with accuracy from EST data alone when compared to the actual gene sequences in this fully sequenced genome.</p>", "<title>Results</title>", "<p>The NSP method uniquely identified family members in all the gene families tested. Two members of the FtsH gene family, three members each of the PAL, RF1, and ribosomal L6 gene families, and four members of the CAD gene family were correctly identified. Additionally all ESTs from the representative contigs when checked against MapViewer data successfully identify the gene locus predicted.</p>", "<title>Conclusion</title>", "<p>We demonstrate the effectiveness of the NSP strategy in identifying specific gene family members in Arabidopsis using only EST data and we describe how this strategy can be used to identify many gene families in agronomically important crop species where they are as yet undiscovered.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>RLF participated in the conception and design of the study, carried out the gene family identification via NSP including BLAST, contig assembly, ORF identification, alignment and dS/dN analysis, and drafted the manuscript. CK developed scripts to construct graphical output of dS/dN results and performed all genome locus identification studies using MapViewer. FE participated in the conception, design, and development of the computational aspects of data generation. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Graph representing potential paralogs with dS/dN &gt;= 2</bold>. Edges are labeled with the dS/dN ratios, followed by the number of substitutions (Sd+Sn) seen An edge indicates dS/dN&gt;=2.00; No edge indicates dS/dN&lt;2.00 OR dS/dN=NA.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>dS/dN calculations for phenylalanine ammonia-lyase (PAL) contigs</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\" colspan=\"2\"><bold>Comparison</bold></td><td align=\"right\"><bold>Sd^a</bold></td><td align=\"right\"><bold>Sn</bold></td><td align=\"right\"><bold>S</bold></td><td align=\"right\"><bold>N</bold></td><td align=\"right\"><bold>ps</bold></td><td align=\"right\"><bold>pn</bold></td><td align=\"right\"><bold>ds</bold></td><td align=\"right\"><bold>dn</bold></td><td align=\"right\"><bold>ds/dn</bold></td><td align=\"right\"><bold>ps/pn</bold></td></tr></thead><tbody><tr><td align=\"center\">Contig1</td><td align=\"center\">Contig4</td><td align=\"right\">38.50</td><td align=\"right\">18.50</td><td align=\"right\">62.83</td><td align=\"right\">219.17</td><td align=\"right\">0.61</td><td align=\"right\">0.08</td><td align=\"right\">1.27</td><td align=\"right\">0.09</td><td align=\"right\">14.22</td><td align=\"right\">7.26</td></tr><tr><td align=\"center\">Contig1</td><td align=\"center\">Contig3</td><td align=\"right\">42.17</td><td align=\"right\">39.83</td><td align=\"right\">65.17</td><td align=\"right\">216.83</td><td align=\"right\">0.65</td><td align=\"right\">0.18</td><td align=\"right\">1.49</td><td align=\"right\">0.21</td><td align=\"right\">7.07</td><td align=\"right\">3.52</td></tr><tr><td align=\"center\">Contig1</td><td align=\"center\">Contig9</td><td align=\"right\">42.33</td><td align=\"right\">36.67</td><td align=\"right\">65.50</td><td align=\"right\">216.50</td><td align=\"right\">0.65</td><td align=\"right\">0.17</td><td align=\"right\">1.48</td><td align=\"right\">0.19</td><td align=\"right\">7.73</td><td align=\"right\">3.82</td></tr><tr><td align=\"center\">Contig1</td><td align=\"center\">Contig6</td><td align=\"right\">52.50</td><td align=\"right\">138.50</td><td align=\"right\">63.33</td><td align=\"right\">197.67</td><td align=\"right\">0.83</td><td align=\"right\">0.70</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.18</td></tr><tr><td align=\"center\">Contig1</td><td align=\"center\">Contig8</td><td align=\"right\">53.50</td><td align=\"right\">138.50</td><td align=\"right\">63.33</td><td align=\"right\">197.67</td><td align=\"right\">0.84</td><td align=\"right\">0.70</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.21</td></tr><tr><td align=\"center\">Contig1</td><td align=\"center\">Contig7</td><td align=\"right\">45.17</td><td align=\"right\">153.83</td><td align=\"right\">62.83</td><td align=\"right\">210.17</td><td align=\"right\">0.72</td><td align=\"right\">0.73</td><td align=\"right\">2.39</td><td align=\"right\">2.80</td><td align=\"right\">0.85</td><td align=\"right\">0.98</td></tr><tr><td align=\"center\">Contig4</td><td align=\"center\">Contig3</td><td align=\"right\">211.83</td><td align=\"right\">133.17</td><td align=\"right\">286.33</td><td align=\"right\">985.67</td><td align=\"right\">0.74</td><td align=\"right\">0.14</td><td align=\"right\">3.22</td><td align=\"right\">0.15</td><td align=\"right\">21.64</td><td align=\"right\">5.48</td></tr><tr><td align=\"center\">Contig4</td><td align=\"center\">Contig9</td><td align=\"right\">143.00</td><td align=\"right\">105.00</td><td align=\"right\">191.67</td><td align=\"right\">639.33</td><td align=\"right\">0.75</td><td align=\"right\">0.16</td><td align=\"right\">3.94</td><td align=\"right\">0.19</td><td align=\"right\">21.27</td><td align=\"right\">4.54</td></tr><tr><td align=\"center\">Contig4</td><td align=\"center\">Contig6</td><td align=\"right\">152.50</td><td align=\"right\">490.50</td><td align=\"right\">201.50</td><td align=\"right\">665.50</td><td align=\"right\">0.76</td><td align=\"right\">0.74</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.03</td></tr><tr><td align=\"center\">Contig4</td><td align=\"center\">Contig8</td><td align=\"right\">80.33</td><td align=\"right\">225.67</td><td align=\"right\">99.83</td><td align=\"right\">314.17</td><td align=\"right\">0.80</td><td align=\"right\">0.72</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.12</td></tr><tr><td align=\"center\">Contig4</td><td align=\"center\">Contig7</td><td align=\"right\">65.00</td><td align=\"right\">233.00</td><td align=\"right\">93.17</td><td align=\"right\">326.83</td><td align=\"right\">0.70</td><td align=\"right\">0.71</td><td align=\"right\">2.00</td><td align=\"right\">2.25</td><td align=\"right\">0.89</td><td align=\"right\">0.98</td></tr><tr><td align=\"center\">Contig3</td><td align=\"center\">Contig9</td><td align=\"right\">1.50</td><td align=\"right\">14.50</td><td align=\"right\">197.17</td><td align=\"right\">633.83</td><td align=\"right\">0.01</td><td align=\"right\">0.02</td><td align=\"right\">0.01</td><td align=\"right\">0.02</td><td align=\"right\">0.33</td><td align=\"right\">0.33</td></tr><tr><td align=\"center\">Contig3</td><td align=\"center\">Contig6</td><td align=\"right\">160.50</td><td align=\"right\">486.50</td><td align=\"right\">206.83</td><td align=\"right\">660.17</td><td align=\"right\">0.78</td><td align=\"right\">0.74</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.05</td></tr><tr><td align=\"center\">Contig3</td><td align=\"center\">Contig8</td><td align=\"right\">81.83</td><td align=\"right\">222.17</td><td align=\"right\">102.83</td><td align=\"right\">311.17</td><td align=\"right\">0.80</td><td align=\"right\">0.71</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.11</td></tr><tr><td align=\"center\">Contig3</td><td align=\"center\">Contig7</td><td align=\"right\">70.50</td><td align=\"right\">228.50</td><td align=\"right\">96.00</td><td align=\"right\">324.00</td><td align=\"right\">0.73</td><td align=\"right\">0.71</td><td align=\"right\">2.90</td><td align=\"right\">2.11</td><td align=\"right\">1.37</td><td align=\"right\">1.04</td></tr><tr><td align=\"center\">Contig9</td><td align=\"center\">Contig6</td><td align=\"right\">150.00</td><td align=\"right\">454.00</td><td align=\"right\">196.17</td><td align=\"right\">613.83</td><td align=\"right\">0.76</td><td align=\"right\">0.74</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.03</td></tr><tr><td align=\"center\">Contig9</td><td align=\"center\">Contig8</td><td align=\"right\">81.33</td><td align=\"right\">220.67</td><td align=\"right\">103.17</td><td align=\"right\">310.83</td><td align=\"right\">0.79</td><td align=\"right\">0.71</td><td align=\"right\">NA</td><td align=\"right\">0.00</td><td align=\"right\">NA</td><td align=\"right\">1.11</td></tr><tr><td align=\"center\">Contig9</td><td align=\"center\">Contig7</td><td align=\"right\">71.67</td><td align=\"right\">229.33</td><td align=\"right\">96.33</td><td align=\"right\">323.67</td><td align=\"right\">0.74</td><td align=\"right\">0.71</td><td align=\"right\">3.61</td><td align=\"right\">2.17</td><td align=\"right\">1.66</td><td align=\"right\">1.05</td></tr><tr><td align=\"center\">Contig6</td><td align=\"center\">Contig8</td><td align=\"right\">2.00</td><td align=\"right\">4.00</td><td align=\"right\">108.33</td><td align=\"right\">305.67</td><td align=\"right\">0.02</td><td align=\"right\">0.01</td><td align=\"right\">0.02</td><td align=\"right\">0.01</td><td align=\"right\">1.42</td><td align=\"right\">1.41</td></tr><tr><td align=\"center\">Contig6</td><td align=\"center\">Contig7</td><td align=\"right\">68.33</td><td align=\"right\">200.67</td><td align=\"right\">97.50</td><td align=\"right\">301.50</td><td align=\"right\">0.70</td><td align=\"right\">0.67</td><td align=\"right\">2.04</td><td align=\"right\">1.64</td><td align=\"right\">1.25</td><td align=\"right\">1.05</td></tr><tr><td align=\"center\">Contig8</td><td align=\"center\">Contig7</td><td align=\"right\">68.33</td><td align=\"right\">199.67</td><td align=\"right\">97.50</td><td align=\"right\">301.50</td><td align=\"right\">0.70</td><td align=\"right\">0.66</td><td align=\"right\">2.04</td><td align=\"right\">1.61</td><td align=\"right\">1.27</td><td align=\"right\">1.06</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>MapViewer locus for ESTs of NSP generated contigs</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Putative gene family</bold></td><td align=\"center\"><bold>Gene group by NSP</bold></td><td align=\"center\"><bold>Contig</bold></td><td align=\"center\"><bold>EST accession</bold></td><td align=\"center\"><bold>MapViewer locus</bold></td><td align=\"center\"><bold>MapViewer Gene Name</bold></td></tr></thead><tbody><tr><td align=\"center\"><italic>CAD</italic></td><td align=\"center\">GeneB</td><td align=\"center\">contig3</td><td align=\"right\">CK121258</td><td align=\"right\">AT4G39330</td><td align=\"right\">AtCAD1</td></tr><tr><td/><td/><td/><td align=\"right\">CB074210</td><td align=\"right\">AT4G39330</td><td align=\"right\">AtCAD1</td></tr><tr><td/><td align=\"center\">GeneC</td><td align=\"center\">contig1</td><td align=\"right\">BP561562</td><td align=\"right\">ELI3-1</td><td align=\"right\">AtCAD4</td></tr><tr><td/><td/><td/><td align=\"right\">BP796450</td><td align=\"right\">ELI3-1</td><td align=\"right\">AtCAD4</td></tr><tr><td/><td/><td/><td align=\"right\">CD530744</td><td align=\"right\">ELI3-1</td><td align=\"right\">AtCAD4</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\"><italic>RF1</italic></td><td align=\"center\">GeneA</td><td align=\"center\">contig1</td><td align=\"right\">AV823314</td><td align=\"right\">ERF1-3</td><td align=\"right\">AteRF1-3</td></tr><tr><td/><td align=\"center\">GeneB</td><td align=\"center\">contig3</td><td align=\"right\">AV822373</td><td align=\"right\">ERF1-2</td><td align=\"right\">AteRF1-2</td></tr><tr><td/><td/><td/><td align=\"right\">BP803175</td><td align=\"right\">ERF1-2</td><td align=\"right\">AteRF1-2</td></tr><tr><td/><td/><td/><td align=\"right\">Z18188</td><td align=\"right\">ERF1-2</td><td align=\"right\">AteRF1-2</td></tr><tr><td/><td align=\"center\">GeneC</td><td align=\"center\">contig6</td><td align=\"right\">AV825957</td><td align=\"right\">ERF1-1</td><td align=\"right\">AteRF1-1</td></tr><tr><td/><td/><td/><td align=\"right\">BE845168</td><td align=\"right\">ERF1-1</td><td align=\"right\">AteRF1-1</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\"><italic>PAL</italic></td><td align=\"center\">GeneA</td><td align=\"center\">contig1</td><td align=\"right\">8720101</td><td align=\"right\">PAL1</td><td align=\"right\">AtPAL1</td></tr><tr><td/><td/><td/><td align=\"right\">8736225</td><td align=\"right\">PAL1</td><td align=\"right\">AtPAL1</td></tr><tr><td/><td align=\"center\">GeneB</td><td align=\"center\">contig3</td><td align=\"right\">8722848</td><td align=\"right\">AT3G10340</td><td align=\"right\">AtPAL4</td></tr><tr><td/><td/><td/><td align=\"right\">8723431</td><td align=\"right\">AT3G10340</td><td align=\"right\">AtPAL4</td></tr><tr><td/><td/><td/><td align=\"right\">8728745</td><td align=\"right\">AT3G10340</td><td align=\"right\">AtPAL4</td></tr><tr><td/><td/><td/><td align=\"right\">8730514</td><td align=\"right\">AT3G10340</td><td align=\"right\">AtPAL4</td></tr><tr><td/><td/><td/><td align=\"right\">9780248</td><td align=\"right\">AT3G10340</td><td align=\"right\">AtPAL4</td></tr><tr><td/><td/><td/><td align=\"right\">9788228</td><td align=\"right\">AT3G10340</td><td align=\"right\">AtPAL4</td></tr><tr><td/><td align=\"center\">GeneC</td><td align=\"center\">contig6</td><td align=\"right\">8690351</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">8724245</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">8725529</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">19869024</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">19869200</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">37426635</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td align=\"center\">GeneC</td><td align=\"center\">contig8</td><td align=\"right\">9786707</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">37426640</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td align=\"center\">GeneC</td><td align=\"center\">contig4</td><td align=\"right\">8719100</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">14580232</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">19855615</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">49165014</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td/><td/><td/><td align=\"right\">59667557</td><td align=\"right\">PAL2</td><td align=\"right\">AtPAL2</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\"><italic>L6</italic></td><td align=\"center\">GeneA</td><td align=\"center\">contig1</td><td align=\"right\">5761694</td><td align=\"right\">AT1G18540</td><td align=\"right\">AtL6A</td></tr><tr><td/><td/><td/><td align=\"right\">8724065</td><td align=\"right\">AT1G18540</td><td align=\"right\">AtL6A</td></tr><tr><td/><td/><td/><td align=\"right\">19802678</td><td align=\"right\">AT1G18540</td><td align=\"right\">AtL6A</td></tr><tr><td/><td align=\"center\">GeneB</td><td align=\"center\">contig4</td><td align=\"right\">19868834</td><td align=\"right\">AT1G74060</td><td align=\"right\">AtL6B</td></tr><tr><td/><td/><td/><td align=\"right\">23303389</td><td align=\"right\">AT1G74060</td><td align=\"right\">AtL6B</td></tr><tr><td/><td align=\"center\">GeneC</td><td align=\"center\">Contig6</td><td align=\"right\">8714872</td><td align=\"right\">AT1G74050</td><td align=\"right\">AtL6C</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\"><italic>FtsH</italic></td><td align=\"center\">GeneA</td><td align=\"center\">contig1</td><td align=\"right\">AV518555</td><td align=\"right\">VAR2</td><td align=\"right\">AtFtsH2</td></tr><tr><td/><td/><td/><td align=\"right\">AV558102</td><td align=\"right\">VAR2</td><td align=\"right\">AtFtsH2</td></tr><tr><td/><td/><td/><td align=\"right\">AV800962</td><td align=\"right\">VAR2</td><td align=\"right\">AtFtsH2</td></tr><tr><td/><td/><td/><td align=\"right\">BP785237</td><td align=\"right\">VAR2</td><td align=\"right\">AtFtsH2</td></tr><tr><td/><td align=\"center\">GeneB</td><td align=\"center\">contig6</td><td align=\"right\">BP626558</td><td align=\"right\">FTSH8</td><td align=\"right\">AtFtsH8</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Percent similarity for NSP generated contigs aligned with actual ribosomal protein L6 genes</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"3\"><bold>GeneA</bold></td><td align=\"center\" colspan=\"3\"><bold>GeneB</bold></td><td align=\"center\"><bold>GeneC</bold></td></tr></thead><tbody><tr><td/><td align=\"center\"><bold>contig1</bold></td><td align=\"center\"><bold>contig3</bold></td><td align=\"center\"><bold>contig8</bold></td><td align=\"center\"><bold>contig2</bold></td><td align=\"center\"><bold>contig4</bold></td><td align=\"center\"><bold>contig5</bold></td><td align=\"center\"><bold>contig6</bold></td></tr><tr><td colspan=\"8\"><hr/></td></tr><tr><td align=\"center\"><italic>AtL6A</italic></td><td align=\"center\">100</td><td align=\"center\">98</td><td align=\"center\">98</td><td align=\"center\">82</td><td align=\"center\">83</td><td align=\"center\">82</td><td align=\"center\">84</td></tr><tr><td align=\"center\"><italic>AtL6B</italic></td><td align=\"center\">83</td><td align=\"center\">83</td><td align=\"center\">82</td><td align=\"center\">99</td><td align=\"center\">99</td><td align=\"center\">99</td><td align=\"center\">93</td></tr><tr><td align=\"center\"><italic>AtL6C</italic></td><td align=\"center\">84</td><td align=\"center\">84</td><td align=\"center\">83</td><td align=\"center\">93</td><td align=\"center\">93</td><td align=\"center\">93</td><td align=\"center\">99</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Percent similarity for NSP generated contigs aligned with actual CAD genes</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>GeneA</bold></td><td align=\"center\"><bold>GeneB</bold></td><td align=\"center\"><bold>GeneC</bold></td><td align=\"center\"><bold>GeneD</bold></td></tr></thead><tbody><tr><td/><td align=\"center\"><bold>contig8</bold></td><td align=\"center\"><bold>contig3</bold></td><td align=\"center\"><bold>contig1</bold></td><td align=\"center\"><bold>contig2</bold></td></tr><tr><td colspan=\"5\"><hr/></td></tr><tr><td align=\"left\"><italic>AtCAD-1</italic></td><td align=\"center\">NSS^a</td><td align=\"center\">99</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td></tr><tr><td align=\"left\"><italic>AtCAD-2</italic></td><td align=\"center\">99</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td></tr><tr><td align=\"left\"><italic>AtCAD-3</italic></td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">78</td><td align=\"center\">82</td></tr><tr><td align=\"left\"><italic>AtCAD-4</italic></td><td align=\"center\">NSS</td><td align=\"center\">76</td><td align=\"center\">100</td><td align=\"center\">87</td></tr><tr><td align=\"left\"><italic>AtCAD-5</italic></td><td align=\"center\">NSS</td><td align=\"center\">72</td><td align=\"center\">84</td><td align=\"center\">100</td></tr><tr><td align=\"left\"><italic>AtCAD-6</italic></td><td align=\"center\">79</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td></tr><tr><td align=\"left\"><italic>AtCAD-7</italic></td><td align=\"center\">NSS</td><td align=\"center\">78</td><td align=\"center\">72</td><td align=\"center\">NSS</td></tr><tr><td align=\"left\"><italic>AtCAD-8</italic></td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td></tr><tr><td align=\"left\"><italic>AtCAD-9</italic></td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Percent similarity for NSP generated contigs aligned with actual release factor genes</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>GeneA</bold></td><td align=\"center\"><bold>GeneB</bold></td><td align=\"center\"><bold>GeneC</bold></td></tr></thead><tbody><tr><td/><td align=\"center\"><bold>contig1</bold></td><td align=\"center\"><bold>contig3</bold></td><td align=\"center\"><bold>contig6</bold></td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\"><italic>AtRF1-1</italic></td><td align=\"center\">82</td><td align=\"center\">83</td><td align=\"center\">99</td></tr><tr><td align=\"left\"><italic>AtRF1-2</italic></td><td align=\"center\">88</td><td align=\"center\">97</td><td align=\"center\">83</td></tr><tr><td align=\"left\"><italic>AtRF1-3</italic></td><td align=\"center\">99</td><td align=\"center\">85</td><td align=\"center\">82</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T6\"><label>Table 6</label><caption><p>Percent similarity for NSP generated contigs aligned with actual FtsH genes</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"2\"><bold>GeneA</bold></td><td align=\"center\" colspan=\"2\"><bold>GeneB</bold></td></tr></thead><tbody><tr><td/><td align=\"center\"><bold>contig1</bold></td><td align=\"center\"><bold>contig5</bold></td><td align=\"center\"><bold>contig3</bold></td><td align=\"center\"><bold>contig6</bold></td></tr><tr><td colspan=\"5\"><hr/></td></tr><tr><td align=\"right\"><italic>AtFtsH1</italic></td><td align=\"center\">NSS</td><td align=\"center\">71</td><td align=\"center\">73</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH2</italic></td><td align=\"center\">100</td><td align=\"center\">97</td><td align=\"center\">86</td><td align=\"center\">83</td></tr><tr><td align=\"right\"><italic>AtFtsH3</italic></td><td align=\"center\">NSS</td><td align=\"center\">78</td><td align=\"center\">70</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH4</italic></td><td align=\"center\">NSS</td><td align=\"center\">79</td><td align=\"center\">78</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH5</italic></td><td align=\"center\">NSS</td><td align=\"center\">73</td><td align=\"center\">73</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH6</italic></td><td align=\"center\">72</td><td align=\"center\">73</td><td align=\"center\">69</td><td align=\"center\">77</td></tr><tr><td align=\"right\"><italic>AtFtsH7</italic></td><td align=\"center\">NSS</td><td align=\"center\">68</td><td align=\"center\">73</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH8</italic></td><td align=\"center\">88</td><td align=\"center\">85</td><td align=\"center\">99</td><td align=\"center\">100</td></tr><tr><td align=\"right\"><italic>AtFtsH9</italic></td><td align=\"center\">NSS</td><td align=\"center\">68</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH10</italic></td><td align=\"center\">NSS</td><td align=\"center\">75</td><td align=\"center\">74</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH11</italic></td><td align=\"center\">NSS</td><td align=\"center\">77</td><td align=\"center\">77</td><td align=\"center\">NSS</td></tr><tr><td align=\"right\"><italic>AtFtsH12</italic></td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td><td align=\"center\">NSS</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>SNAP output results for all 21 pairwise comparisons of 7 contigs in which an ORF was identified. A ds/dn value greater than 2.00 was chosen as threshold to indicate contigs that potentially represent distinct gene family members.</p><p>a – See <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.hiv.lanl.gov\"/> for explanation of abbreviations and calculations.</p></table-wrap-foot>", "<table-wrap-foot><p>Individual ESTs of representative contigs for putative gene family members of the 5 Arabidopsis families tested were located to a specific locus by NCBI MapViewer.</p></table-wrap-foot>", "<table-wrap-foot><p>Representative contigs for 3 putative gene family members, GeneA, GeneB, and GeneC, identified by the NSP method were aligned with actual Arabidopsis gene family members and percent similarity determined.</p></table-wrap-foot>", "<table-wrap-foot><p>Representative contigs for 4 putative gene family members, GeneA, GeneB, GeneC, and GeneD identified by the NSP method were aligned with actual Arabidopsis gene family members and percent similarity determined.</p><p>a – No significant similarity as returned by bl2seq program.</p></table-wrap-foot>", "<table-wrap-foot><p>Representative contigs for 3 putative gene family members, GeneA, GeneB, and GeneC, identified by the NSP method were aligned with actual Arabidopsis gene family members and percent similarity determined.</p></table-wrap-foot>", "<table-wrap-foot><p>Representative contigs for 2 putative gene family members, GeneA and GeneB, identified by the NSP method were aligned with actual Arabidopsis gene family members and percent similarity determined.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S2-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The Human Genome Project [##REF##11237011##1##] was driven by the hope that characterization of the human genome would elucidate the molecular etiology of human disease. This resulted into an abundant amount of data regarding the genetic variation across the human genome as amended by the HapMap project [##REF##12029063##2##], which catalogs the location and linkage information for many human genetic variants. The most common variations are single nucleotide polymorphisms (SNPs), single base pair positions in the genome at which different sequence alternatives (alleles) exist. They occur approximately once every 1,000 bases unevenly distributed across the human genome, principally in non-coding regions presumably due to higher selection pressure in coding regions [##REF##10395891##3##]. While approximately half of the SNPs in coding regions are silent [##REF##10391209##4##,##REF##10391210##5##], the other half result in missense mutations (change in the encoded protein sequence) that may be neutral or involved in a disease or phenotype.</p>", "<p>SNPs like other genetic variations may be indicators of susceptibility to polygenic diseases [##REF##9872978##6##,##REF##10217129##7##] and could provide a basis for diagnostic and optimal therapeutic choices. A major challenge in realizing these expectations is to identify variants likely to be disease related. While the characterization of all SNPs through disease association studies is economically and practically unrealistic, computational methods to rank SNPs based on their potential impact would help to select and focus on those base positions predicted to be strongly associated with disease. To this end, a variety of approaches with different philosophies have been proposed. Some are purely based on sequence information including conservation in inter-species homologous proteins (orthologs), natural selective pressure at the residue level and the nature of the residue change [##REF##11689479##8##, ####REF##11337480##9##, ##REF##12824425##10##, ##REF##16584746##11####16584746##11##]. Other methods combine protein sequence with other physico-chemical properties including protein thermodynamic stability and structure, adding more functionally-relevant knowledge but restricting the applicability of those algorithms to particular cases (known proteins, known structures...) [##REF##11337480##9##,##REF##11230178##12##, ####REF##12079393##13##, ##REF##12270722##14##, ##REF##11812146##15##, ##REF##11254390##16####11254390##16##]. An interesting example has been proposed by Fleming <italic>et al</italic>. [##REF##12531920##17##] where the authors categorize missense mutations in the exon 11 of the BRAC1 gene by assigning a probabilistic score representing their likelihood to disrupt the gene function, potentially resulting in breast cancer. The authors used the SIFT (Separating Intolerant from Tolerant) method, considered a gold standard for the prediction of functional effects of mutations [##REF##11337480##9##]. However, this effort has limitations. For instance, SIFT is based on amino acid substitutions and cannot be directly applied to sequences that are transcribed but not translated also called non-coding RNA (ncRNA). These ncRNAs, have been shown to play diverse functions (mRNA splicing, RNA modification, translational regulation etc...) and represent the majority of the transcriptional output in higher eukaryotes [##REF##11713189##18##]. Futhermore, most methods including SIFT, predict effects of non-synonymous substitutions while recent studies have shown that synonymous substitutions, although not resulting in a change in the encoded transcripts may none-the-less cause a measurable phenotypic change and sometimes disease [##REF##17185560##19##].</p>", "<p>In this study we propose a predictive method primarily based on sequence and phylogenic information with the aim to apply it to the entire human genome. This method like others hinges on the assertion that evolutionarily conserved nucleotide bases are important for gene function and that single base mutations at these conserved positions are likely to represent disease alleles [##REF##12682377##20##]. Our goal is to predict the impact of a mutation appearing in any gene and at any position, including transcribed but not translated genes, and provide the disease likelihood associated with such a change. Along with the level of interspecies conservation and residue change, we also investigated several factors known or suspected to be disease markers in other published studies such as the location in the protein and nucleotide position in the triplet type of substitution. We have incorporated all these factors into the design of a Support Vector Machine (SVM) classifier that was trained on disease related mutations as found in Human Gene Mutation Database (HGMD) [##REF##8882888##21##] and mutations found in dbSNP [##REF##11125122##22##] for the vast majority of which no disease linkage has been established and are likely to be neutral or of minor phenotypic impact. The SVM classifier appropriately scales these factors relative to their importance and quantifies probable disease-causing substitutions within the human genome regardless of whether they have been previously observed and annotated as such.</p>" ]

[ "<title>Methods</title>", "<title>Overview and basic data sets</title>", "<p>Our predictive method is based on a Support Vector Machine (SVM) algorithm that one can easily implement using the popular 'R' statistical software tool <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.R-project.org\"/> and more importantly that satisfactorily addresses the non parametric nature of the data distribution for a number of individual input factors [##UREF##0##25##]. Those factors, including different inter-species conservation parameters, were chosen for their ability to identify positions in the genome where mutations may have a significant phenotypic impact (i.e. disease). Figure ##FIG##0##1## represents the process established to calculate the conservation score for every nucleotide in the human genome.</p>", "<p>Genes were obtained from the Reference Sequence (RefSeq) Database (Release 21) [##REF##10592200##26##]. Single base mutations were obtained from Human Gene Mutation Database (HGMD)[##REF##12754702##27##], forming a set of disease causing mutations (last available public version, June 2005, 21,964 mutations) and from dbSNP [##REF##11125122##22##] (build 126, 97,102 mutations in the coding regions) that is assumed to contain for the vast majority neutral mutations. The mutations were extracted from ENTREZ SNPs page <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/sites/entrez\"/> setting the limits parameters to {Organism : Homo Sapiens, Functional class : coding non synonymous, coding synonymous, mRNA UTR and only reference SNPs, SNP class : SNP}. A subset of 998 redundant mutations between both was separated into the 'Common SNPs' set and later used as one of the tests of the SVM classifier. Mutations in dbSNP were partitioned into synonymous (33,871 mutations) and non-synonymous (42,961 mutations).</p>", "<title>Inter-species alignment and conservation score</title>", "<title>Finding putative human gene orthologs in other species for optimal alignment</title>", "<p>For every human RefSeq entry, its corresponding ortholog was identified using megaBLAST [##REF##10802712##28##] in a reciprocal manner <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/Homology/\"/>. Briefly, each human RefSeq was BLASTed against eight comparable vertebrate genomes Table ##TAB##1##2##, and the highest scoring similar sequence for each genome was reverse-BLASTed against the human RefSeq database. Only those non-human sequences for which the original human sequence was the highest scoring match were retained as putative orthologs (Table ##TAB##1##2##). Each human RefSeq was subsequently aligned with its putative orthologs using CLUSTALW [##REF##12824352##29##].</p>", "<title>Evaluating nucleotide conservation</title>", "<p>Ideally, a conservation scoring method would result in significantly higher scores at 'sensitive' positions where disease causing mutations occur comparatively to random positions. For each position in each gene, a score was calculated as the weighted average of the conservation obtained from the alignment of this gene and its orthologs. The weighting parameters were the UCSC browser phylogenetic distances [##REF##14656959##30##] for the genomes where the given nucleotide is conserved and were kindly provided by the UCSC genome browser group (Table ##TAB##1##2##). In this way, a nucleotide conserved in a distant genome weighs more than the same nucleotide conserved only in closely related species. This score has also been normalized to allow comparisons between genes (for each gene, the number of orthologs will be different) (Figure ##FIG##5##6##).</p>", "<p>To ensure that conservation scores significantly correlate with disease alleles, normalized conservation scores were also computed for randomly selected base positions in genes containing mutations listed in HGMD or dbSNP databases (1,260 and 14,449 genes, respectively, Table ##TAB##2##3##). As shown in Figure ##FIG##1##2##, disease-causing mutations occur at highly conserved positions and usually are the most conserved positions within the gene. Conversely, and as anticipated, dbSNPs mutations tend to occur in the least conserved positions.</p>", "<title>Mapping SNPs from HGMD and dbSNP to their gene location</title>", "<p>Each and every mutation in the two datasets HGMD and dbSNP, were mapped to their gene location in three steps. 1) For each mutation, 50 base flanking regions upstream and downstream were BLASTed against the human genome. Only hits with 100% identity and no gaps were kept. 2) The BLAST coordinates were used to map the exact location of the given mutation nucleotide within the gene. 3) In order to ensure accurate mapping of mutation positions to the genes, mutations obtained from dbSNP were cross-validated using the Genbank XML flat file and mutations obtained from HGMD were cross-validated using upstream and downstream flanking regions. Using this procedure, a total of 20,094 and 76,832 unique mutations from HGMD and dbSNP, respectively, were accurately mapped, as shown in Table ##TAB##2##3##.</p>", "<title>SVM design</title>", "<p>A number of factors that we hypothesized may quantify the effect of a mutation, including conservation, location of the mutation within the gene, substitution type (synonymous or non-synonymous), type of residue change in terms of charge, mass, volume, hydrophobicity, and favorability of a substitution as represented in the BLOSUM and PAM matrices were used in the SVM. For these SVMs, the radial kernel function was used to design the classifier due to the non-linear nature of the parameters with cost and gamma parameters set as 1 and 0.125 respectively to enhance the classifier performance.</p>", "<title>Conservation</title>", "<p>The conservation information used in the SVM is based on seven different metrics: (1) nucleotide raw conservation score, (2) normalized conservation score to the maximum possible score per position, (3) maximum conservation score per gene, (4) average conservation score per gene (5) percentage of bases with conservation scores below or (6) equal to the nucleotide position of interest, and (7) comparison of nucleotide conservation score with the average conservation score within the gene. Figure ##FIG##6##7## shows that in contrast to dbSNP mutations or random positions, HGMD mutations tend to occur at positions that are significantly more conserved than the gene itself (as measured by the conservation averaged over all its positions). Although these different conservation metrics have redundancy, analysis using the 'leave one out' technique will identify which of these metrics provide the best ability to separate disease causing base variants (from HGMD) from those base positions unlikely to be associated with disease (from dbSNP).</p>", "<title>Parameters relying on availability of translation information</title>", "<p>Although the method we propose is not dependent upon translation information, various parameters including amino-acid substitution matrixes were considered in order to assess their relative importance in predicting mutations impact at various positions. Furthermore, the use of translation information (when available) facilitates the comparison of at least a portion our method with SIFT.</p>", "<p>Two sets of widely used matrixes, BLOSUM [##REF##1438297##31##] and PAM [##UREF##1##32##] matrices (BLOSUM30, BLOSUM60, BLOSUM62, BLOSUM80, BLOSUM90, BLOSUM100 PAM10, PAM50, PAM100, PAM250, and PAM500) were tested for every mutation in both dbSNP and HGMD datasets to determine whether distant or close relationship-based matrixes are more appropriate for identification of base positions likely to cause a disease if mutated. Again, it is reasonable to expect that disease-causing mutations found in HGMD would more often result in unfavorable substitutions than mutations in dbSNP.</p>", "<p>A number of physico-chemical matrixes have been used to take into account possible aberrant changes in the volume, mass or charge of the original amino acid that could potentially disrupt the protein structure. Any extreme change in any of the above factors could have a disproportionately high impact and therefore be overrepresented in disease causing alleles (i.e., mutations in HGMD relative to mutations in dbSNP).</p>", "<p>Each individual mutation was also annotated for its location within the codon. Mutations at the first position are more likely to modify the protein sequence than mutations occurring at the second and third positions. Consequently mutations at the first codon position were expected to occur more often in HGMD than in dbSNP.</p>", "<title>SVM training and testing</title>", "<p>All the above-mentioned factors were individually evaluated for their potential to identify diseased alleles and statistically segregate HGMD from dbSNP mutations. The set of 19,096 HGMD mutations (20,094 minus the 998 redundant) was randomly split in two subsets for training (80% or 15,277 mutations) and testing (20% or 3,819 mutations). An equal number of mutations from dbSNP were also randomly chosen and mixed with the corresponding HGMD subset. This 80–20 proportion has been chosen as a good comprise between the size of the testing and training sets, and to allow us to obtain a testing set big enough for comparison with SIFT. The obtained accuracy with these proportions is 84.3% (Figure ##FIG##7##8##).</p>", "<p>The importance of each factor was weighed individually against the final prediction outcome by employing the 'leave one factor out' method (Figure ##FIG##2##3##). Only factors that significantly contributed to the prediction outcome were kept to establish a final SVM classifier.</p>", "<title>SVM validation</title>", "<p>The final classifier was applied to all coding bases within the human genome to identify high impact positions, where mutations might have a higher probability for disease-causation. Literature, ontologies and pathway relationships for genes with highest and lowest base-averaged SVM scores were further inspected. GeneSifter (VizX Labs, Seattle, WA), a software program typically applied to the analysis and interpretation of gene expression data, was used to generate GO Ontology information. Pathways analysis reports were obtained from the Kyoto Encyclopedia of Genes and Genomes (KEGG) [##REF##11752249##33##]. Disease associations were also based on annotations provided by the NCBI, the Stanford SOURCE database [##REF##12519986##34##], the Comparative Toxicogenomic database [##REF##16675512##35##] and published literature. Similar analyses were performed with genes randomly selected as a control, thus allowing the calculation of a z-score.</p>" ]

[ "<title>Results and discussion</title>", "<title>Inter-species conservation as a measure of disease susceptibility</title>", "<p>Using the process detailed in Figure ##FIG##0##1##, we have exhaustively confirmed that known disease mutations in HGMD are disproportionately distributed in the most conserved nucleotide positions within genes (Figure ##FIG##1##2##). The average normalized conservation score calculated for HGMD mutations is 0.88 ± 0.24. Importantly, 87% of HGMD mutations score higher than the gene (average of all positions) in which they occur. Conversely, and as anticipated, dbSNPs mutations tend to occur in the least conserved positions (0.57 ± 0.40, and 0.63 ± 0.40 if synonymous mutations are excluded). Only half of the mutations in dbSNP score higher than the gene in which they occur (57% if synonymous mutations are not considered). The conservation score distribution for randomly selected positions is similar to dbSNPs mutations: the average score is 0.61 ± 0.38 and the fraction of positions scoring higher than the gene is 56%. Since dbSNP and HGMD databases are not cross referenced, it is unknown if the mutations in dbSNP are entirely unique or were co-deposited in both databases. While it seems counter-intuitive that random position have a slightly higher score than dbSNP mutations, the difference is minor compared to what is seen with HGMD mutations or the common set of mutations. The average conservation score calculated for the 998 mutations that were found in both sets (0.80 ± 0.32) confirms the relationship between conservation and sensitivity of a position.</p>", "<p>As an interesting aside, we discovered by inspection that the conservation score decreased in the base positions on either side of the nucleotides that were annotated as disease-causing in HMGD. This trend was observed for 83% of HMGD SNPs for up to 5 bases (the maximum number inspected) on either side of the disease-causing base. This suggests that disease-causing alleles have stronger selection pressure relative to the surrounding sequence space.</p>", "<p>Finally and as a limitation, under the scoring scheme used, we do not take into account positions in distant species that appear to be conserved, but were in fact mutated multiple times to ultimately revert back to the same nucleotide base. In this particular case, our scoring method doesn't factor in such 'back and forth' mutations.</p>", "<title>Translation information as a measure of disease susceptibility</title>", "<p>This study was predominantly aimed at identifying genome-wide factors that could be used to predict disease-like mutations across all genes. Factors associated with mutation position in the codon, residue-change effects, such as alteration of the shape, size or charge of the residue were found to be very weak predictors. However these factors may be of interest in certain subsets of genes. For instance, hydrophobic to hydrophilic substitutions captured in the PHAT matrix [##REF##11108698##23##] constructed specifically to study membrane proteins would presumably be more appropriate to study genes coding for membrane proteins [##REF##15784611##24##].</p>", "<p>The BLOSUM and PAM matrices computed from genome-wide alignments of different families of proteins were also found to be useful to segregate sensitive from insensitive base positions. The BLOSUM80 matrix was found to be the most efficient at identifying disease-causing mutations. 94% of HGMD mutations had a non-favorable substitution versus only 45% for dbSNP. This is consistent with the assumption that non-favorable changes may significantly impact the protein function.</p>", "<title>Important predictors and SVM performance</title>", "<p>Inter-species conservation information was found to be the strongest predictor among the ones that were selected to identify disease allele. This information was provided to the SVM via 7 different metrics (see Methods). We evaluated the relative importance of these factors using the \"leave one factor out technique\" (Figure ##FIG##2##3##). Removal of these factors individually did not significantly affect the classifier performance due to redundant information in the factors. However, removing all of them resulted in a significant decrease in performance. Despite their redundancy, the use of these factors in concert produced the strongest SVM classifier, confirming that inter-species conservation is a crucial parameter for identifying single base mutations that correlate with phenotype alteration.</p>", "<p>The other factors found to be good predictors are the BLOSUM 80 substitution matrix, the number of orthologs found for a gene, and the position of the base within the codon (Figure ##FIG##2##3##). Several random factors were incorporated into the design of the SVM as control parameters (length of the gene, location within the gene). The insignificance of these random factors was confirmed when their removal from the SVM model did not impact its performance.</p>", "<p>The final classifier yielded 83% sensitivity and 84% specificity in segregating the HGMD-like mutations from the dbSNP-like mutations. If parameters not related to translation information are removed (codon position, substitution type, substitution matrix) the classifier still achieves 64% sensitivity and 72% specificity. For the majority of misclassified HGMD mutations, we observed that although they had high normalized conservation scores, they were located in genes with few orthologs (less than 3). The classifier is likely to achieve better performance as the number of sequenced genomes and identified orthologs increase. Another set of misclassified HGMD mutations was the 49 synonymous mutations, presumably due to the low number of such mutations in the HGMD training sets. Classification of synonymous mutations will likely improve as more become available in the HGMD.</p>", "<p>Unfortunately and although HGMD has now doubled to over 40,782 mutations, it is no longer openly available <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.hgmd.cf.ac.uk/ac/index.php\"/> limiting us to the last edition publicly available (2005). Finally, when applied to dbSNP, 16% of the SNPs (12,393) were found to have HGMD-like signatures. While these are technically misclassified, they nonetheless remain potential candidates for further investigation. Due to the absence of disease linkage information in dbSNP, we have assumed that most mutations in dbSNPs may not have a significant phenotypic impact. However absence of evidence is not evidence of absence and dbSNP very likely contains a number of disease causing mutations that have as yet to be identified. Identifying them and refining the training sets could also significantly improve the algorithm effectiveness.</p>", "<title>Comparison with SIFT</title>", "<p>To assess the performance of the SVM classifier, we compared it with SIFT, a popular method to predict mutations impacts. The last public version of SIFT predictions on dbSNP mutations was downloaded and predictions compared with ours. Out of the 27,279 mutations present in SIFT, only 19,276 were common between the two versions of the database. We also compared the predictions of SIFT for the subset of SNPs present in HGMD, to evaluate the sensitivity and specificity of both methods.</p>", "<p>Our SVM, complete and with conservation only were compared to SIFT. Mutations used for testing were different from those used to train our SVMs (Table ##TAB##0##1##). On these sets, the complete SVM achieves the highest Sensitivity (71% versus 39% for SIFT), at the expense of a lower specificity (62% versus 73%). Overall the best balance of specificity and specificity as measured with the F-measure appears to be achieved when using our SVM, although under these experimental conditions, SIFT achieves the best accuracy (71.9% versus 62.7%). A combination of the SVM-complete and the SIFT method may represent the best alternative. The SVM with conservation only performs at the lowest level.</p>", "<p>Reasons for the diminished performance for our SVM when applied to the specific subset of mutations present in both our data set and that used in the SIFT study is unclear. One contributing factor is the very small overlap in the data for the disease causing mutations, i.e. those found in HGMD, common to both SIFT and our SVM. Since SIFT is computed only on dbSNP mutations, only that small subset of mutations (998 out of 21,946) that appear in both databases can be used for testing and comparison. We specifically excluded these common mutations while training our SVM so that we could best differentiate between and classify mutations that are disease causing (HGMD) and those presumed to be non-disease causing (dbSNP). When compared to the results obtained during the testing of our SVM, the SVM complete achieves an accuracy of 83.5%, with a testing set that contained many more mutations and presumably more illustrative of the actual performance of our SVM (Figure ##FIG##3##4##).</p>", "<title>Genome wide application and validation</title>", "<p>In its present form, the SVM, through the use parameters like BLOSUM80 substitution scores or the base position in the codon, partially depends on translation information limiting its application to genes with well defined expressed sequences (introns/exons, coding frames etc...). Furthermore, the use of BLOSUM80 requires prior knowledge of mutation transitions from their wild type state to their mutated states. Finally, for the 5,073 out of the 39,218 known RefSeq genes that do not encode proteins, it would be inconsistent and unreliable to use codon transitions for predictions.</p>", "<p>Parameters depending on translation information have been removed from a second version of our SVM that has been applied to all the coding regions within the human genome. Each base position has been scored in accordance to the disease likelihood associated with a mutation occurring at this particular position. The calculated score provides a quantitative measure scaled from 0 to 100 instead of a qualitative prediction (disease/non-disease). Figure ##FIG##4##5## shows the cumulative frequency distribution for all HGMD and dbSNP mutations as a function of HGMD like score (with 100% being the more HGMD like). The distribution for all coding bases in the human genome is also shown for comparison. The similarity between the dbSNP accumulated distribution and all coding bases, suggest that most bases, if altered, will not result in disease causation (i.e. are not HGMD-like). The HGMD distribution curve was used to identify a threshold corresponding to the inflection point with maximum slope (maxima of the derivative). This threshold was found to be 94%. This 6% range of conservation (from 94% to 100%) is enriched with over half of the mutations in HGMD, while it includes only 13% of nucleotide bases (or 12,038,565 bases) in the entire coding genome that could potentially play a causative role in disease/phenotype alteration. Mutations and SNPs occurring within this range of conservation score, and for which no disease linkage has been established, have the highest potential value for researchers designing disease association studies.</p>", "<p>Using our methodology 26,874 gene transcripts in the human genome found at least one ortholog in comparable genomes. These genes were then ranked according to their likelihood of being associated with disease if mutated, based on our disease probability score obtained with our SVM based on conservation information only. Out of the 30 top scoring genes, which are very sensitive to variation, 21 are known to be directly associated with a disease from published literature (Additional file ##SUPPL##0##1##). Conversely, only one gene out of the 30 with the lowest scores were found to be associated with disease in the published literature. Many of those genes with lower scores are hypothetical, likely due to the low similarity with known genes (consistently with low conservation and low scores). The results strongly suggest that the highest scoring genes are indeed enriched with those that might contribute to disease if mutated.</p>" ]

[ "<title>Results and discussion</title>", "<title>Inter-species conservation as a measure of disease susceptibility</title>", "<p>Using the process detailed in Figure ##FIG##0##1##, we have exhaustively confirmed that known disease mutations in HGMD are disproportionately distributed in the most conserved nucleotide positions within genes (Figure ##FIG##1##2##). The average normalized conservation score calculated for HGMD mutations is 0.88 ± 0.24. Importantly, 87% of HGMD mutations score higher than the gene (average of all positions) in which they occur. Conversely, and as anticipated, dbSNPs mutations tend to occur in the least conserved positions (0.57 ± 0.40, and 0.63 ± 0.40 if synonymous mutations are excluded). Only half of the mutations in dbSNP score higher than the gene in which they occur (57% if synonymous mutations are not considered). The conservation score distribution for randomly selected positions is similar to dbSNPs mutations: the average score is 0.61 ± 0.38 and the fraction of positions scoring higher than the gene is 56%. Since dbSNP and HGMD databases are not cross referenced, it is unknown if the mutations in dbSNP are entirely unique or were co-deposited in both databases. While it seems counter-intuitive that random position have a slightly higher score than dbSNP mutations, the difference is minor compared to what is seen with HGMD mutations or the common set of mutations. The average conservation score calculated for the 998 mutations that were found in both sets (0.80 ± 0.32) confirms the relationship between conservation and sensitivity of a position.</p>", "<p>As an interesting aside, we discovered by inspection that the conservation score decreased in the base positions on either side of the nucleotides that were annotated as disease-causing in HMGD. This trend was observed for 83% of HMGD SNPs for up to 5 bases (the maximum number inspected) on either side of the disease-causing base. This suggests that disease-causing alleles have stronger selection pressure relative to the surrounding sequence space.</p>", "<p>Finally and as a limitation, under the scoring scheme used, we do not take into account positions in distant species that appear to be conserved, but were in fact mutated multiple times to ultimately revert back to the same nucleotide base. In this particular case, our scoring method doesn't factor in such 'back and forth' mutations.</p>", "<title>Translation information as a measure of disease susceptibility</title>", "<p>This study was predominantly aimed at identifying genome-wide factors that could be used to predict disease-like mutations across all genes. Factors associated with mutation position in the codon, residue-change effects, such as alteration of the shape, size or charge of the residue were found to be very weak predictors. However these factors may be of interest in certain subsets of genes. For instance, hydrophobic to hydrophilic substitutions captured in the PHAT matrix [##REF##11108698##23##] constructed specifically to study membrane proteins would presumably be more appropriate to study genes coding for membrane proteins [##REF##15784611##24##].</p>", "<p>The BLOSUM and PAM matrices computed from genome-wide alignments of different families of proteins were also found to be useful to segregate sensitive from insensitive base positions. The BLOSUM80 matrix was found to be the most efficient at identifying disease-causing mutations. 94% of HGMD mutations had a non-favorable substitution versus only 45% for dbSNP. This is consistent with the assumption that non-favorable changes may significantly impact the protein function.</p>", "<title>Important predictors and SVM performance</title>", "<p>Inter-species conservation information was found to be the strongest predictor among the ones that were selected to identify disease allele. This information was provided to the SVM via 7 different metrics (see Methods). We evaluated the relative importance of these factors using the \"leave one factor out technique\" (Figure ##FIG##2##3##). Removal of these factors individually did not significantly affect the classifier performance due to redundant information in the factors. However, removing all of them resulted in a significant decrease in performance. Despite their redundancy, the use of these factors in concert produced the strongest SVM classifier, confirming that inter-species conservation is a crucial parameter for identifying single base mutations that correlate with phenotype alteration.</p>", "<p>The other factors found to be good predictors are the BLOSUM 80 substitution matrix, the number of orthologs found for a gene, and the position of the base within the codon (Figure ##FIG##2##3##). Several random factors were incorporated into the design of the SVM as control parameters (length of the gene, location within the gene). The insignificance of these random factors was confirmed when their removal from the SVM model did not impact its performance.</p>", "<p>The final classifier yielded 83% sensitivity and 84% specificity in segregating the HGMD-like mutations from the dbSNP-like mutations. If parameters not related to translation information are removed (codon position, substitution type, substitution matrix) the classifier still achieves 64% sensitivity and 72% specificity. For the majority of misclassified HGMD mutations, we observed that although they had high normalized conservation scores, they were located in genes with few orthologs (less than 3). The classifier is likely to achieve better performance as the number of sequenced genomes and identified orthologs increase. Another set of misclassified HGMD mutations was the 49 synonymous mutations, presumably due to the low number of such mutations in the HGMD training sets. Classification of synonymous mutations will likely improve as more become available in the HGMD.</p>", "<p>Unfortunately and although HGMD has now doubled to over 40,782 mutations, it is no longer openly available <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.hgmd.cf.ac.uk/ac/index.php\"/> limiting us to the last edition publicly available (2005). Finally, when applied to dbSNP, 16% of the SNPs (12,393) were found to have HGMD-like signatures. While these are technically misclassified, they nonetheless remain potential candidates for further investigation. Due to the absence of disease linkage information in dbSNP, we have assumed that most mutations in dbSNPs may not have a significant phenotypic impact. However absence of evidence is not evidence of absence and dbSNP very likely contains a number of disease causing mutations that have as yet to be identified. Identifying them and refining the training sets could also significantly improve the algorithm effectiveness.</p>", "<title>Comparison with SIFT</title>", "<p>To assess the performance of the SVM classifier, we compared it with SIFT, a popular method to predict mutations impacts. The last public version of SIFT predictions on dbSNP mutations was downloaded and predictions compared with ours. Out of the 27,279 mutations present in SIFT, only 19,276 were common between the two versions of the database. We also compared the predictions of SIFT for the subset of SNPs present in HGMD, to evaluate the sensitivity and specificity of both methods.</p>", "<p>Our SVM, complete and with conservation only were compared to SIFT. Mutations used for testing were different from those used to train our SVMs (Table ##TAB##0##1##). On these sets, the complete SVM achieves the highest Sensitivity (71% versus 39% for SIFT), at the expense of a lower specificity (62% versus 73%). Overall the best balance of specificity and specificity as measured with the F-measure appears to be achieved when using our SVM, although under these experimental conditions, SIFT achieves the best accuracy (71.9% versus 62.7%). A combination of the SVM-complete and the SIFT method may represent the best alternative. The SVM with conservation only performs at the lowest level.</p>", "<p>Reasons for the diminished performance for our SVM when applied to the specific subset of mutations present in both our data set and that used in the SIFT study is unclear. One contributing factor is the very small overlap in the data for the disease causing mutations, i.e. those found in HGMD, common to both SIFT and our SVM. Since SIFT is computed only on dbSNP mutations, only that small subset of mutations (998 out of 21,946) that appear in both databases can be used for testing and comparison. We specifically excluded these common mutations while training our SVM so that we could best differentiate between and classify mutations that are disease causing (HGMD) and those presumed to be non-disease causing (dbSNP). When compared to the results obtained during the testing of our SVM, the SVM complete achieves an accuracy of 83.5%, with a testing set that contained many more mutations and presumably more illustrative of the actual performance of our SVM (Figure ##FIG##3##4##).</p>", "<title>Genome wide application and validation</title>", "<p>In its present form, the SVM, through the use parameters like BLOSUM80 substitution scores or the base position in the codon, partially depends on translation information limiting its application to genes with well defined expressed sequences (introns/exons, coding frames etc...). Furthermore, the use of BLOSUM80 requires prior knowledge of mutation transitions from their wild type state to their mutated states. Finally, for the 5,073 out of the 39,218 known RefSeq genes that do not encode proteins, it would be inconsistent and unreliable to use codon transitions for predictions.</p>", "<p>Parameters depending on translation information have been removed from a second version of our SVM that has been applied to all the coding regions within the human genome. Each base position has been scored in accordance to the disease likelihood associated with a mutation occurring at this particular position. The calculated score provides a quantitative measure scaled from 0 to 100 instead of a qualitative prediction (disease/non-disease). Figure ##FIG##4##5## shows the cumulative frequency distribution for all HGMD and dbSNP mutations as a function of HGMD like score (with 100% being the more HGMD like). The distribution for all coding bases in the human genome is also shown for comparison. The similarity between the dbSNP accumulated distribution and all coding bases, suggest that most bases, if altered, will not result in disease causation (i.e. are not HGMD-like). The HGMD distribution curve was used to identify a threshold corresponding to the inflection point with maximum slope (maxima of the derivative). This threshold was found to be 94%. This 6% range of conservation (from 94% to 100%) is enriched with over half of the mutations in HGMD, while it includes only 13% of nucleotide bases (or 12,038,565 bases) in the entire coding genome that could potentially play a causative role in disease/phenotype alteration. Mutations and SNPs occurring within this range of conservation score, and for which no disease linkage has been established, have the highest potential value for researchers designing disease association studies.</p>", "<p>Using our methodology 26,874 gene transcripts in the human genome found at least one ortholog in comparable genomes. These genes were then ranked according to their likelihood of being associated with disease if mutated, based on our disease probability score obtained with our SVM based on conservation information only. Out of the 30 top scoring genes, which are very sensitive to variation, 21 are known to be directly associated with a disease from published literature (Additional file ##SUPPL##0##1##). Conversely, only one gene out of the 30 with the lowest scores were found to be associated with disease in the published literature. Many of those genes with lower scores are hypothetical, likely due to the low similarity with known genes (consistently with low conservation and low scores). The results strongly suggest that the highest scoring genes are indeed enriched with those that might contribute to disease if mutated.</p>" ]

[ "<title>Conclusion</title>", "<p>We have developed a comparative genomic analysis method for genome-wide identification of genome positions with a greater likelihood of being important to gene function. Mutations occurring at these sites have a higher probability of representing disease alleles. New single base mutations or SNPs can then be scored for their potential to cause a disease, helping direct SNP discovery efforts.</p>", "<p>There is an underlying signature for disease mutations, as evidenced by their occurrence in the most conserved nucleotide positions, the favorability of residue substitution observed from BLOSUM scores, codon bias and other factors. This signature was exploited to identify and flag putative disease-causing mutations in all coding human bases, including some in dbSNP, irrespective of the existence of annotation as causing a disease or clinical phenotype. Additional genetic (or other) factors, if discovered in the future to be causative of disease, can be easily incorporated in the prediction algorithm, i.e. the SVM classifier. This makes this tool and approach versatile, enabling one to quantitatively test the strength of new factors or metrics for their potential for disease causation.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Single Nucleotide Polymorphisms (SNPs) are the most abundant form of genomic variation and can cause phenotypic differences between individuals, including diseases. Bases are subject to various levels of selection pressure, reflected in their inter-species conservation.</p>", "<title>Results</title>", "<p>We propose a method that is not dependant on transcription information to score each coding base in the human genome reflecting the disease probability associated with its mutation. Twelve factors likely to be associated with disease alleles were chosen as the input for a support vector machine prediction algorithm. The analysis yielded 83% sensitivity and 84% specificity in segregating disease like alleles as found in the Human Gene Mutation Database from non-disease like alleles as found in the Database of Single Nucleotide Polymorphisms. This algorithm was subsequently applied to each base within all known human genes, exhaustively confirming that interspecies conservation is the strongest factor for disease association. For each gene, the length normalized average disease potential score was calculated. Out of the 30 genes with the highest scores, 21 are directly associated with a disease. In contrast, out of the 30 genes with the lowest scores, only one is associated with a disease as found in published literature. The results strongly suggest that the highest scoring genes are enriched for those that might contribute to disease, if mutated.</p>", "<title>Conclusion</title>", "<p>This method provides valuable information to researchers to identify sensitive positions in genes that have a high disease probability, enabling them to optimize experimental designs and interpret data emerging from genetic and epidemiological studies.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>Vinayak Kulkarni developed the pipeline for SNP analysis, including prepping all the data from different datasets, annotating the SNPs with additional information from various sources and designing and fine-tuning the classifier for optimal prediction. Mounir Errami analyzed the data, lead the project and the preparation of this article. Robert Barber provided with valuable inputs at various time points of the project which helped in validating the hypothesis and build a robust method. Harold Garner provided guidance and organizational support for the successful completion of the project.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>We would like to thank Dr. John W Fondon III and Dr Guanghua Xiao, Dr Cristi Galindo, M. Mark Burkart and Dr Wayne Fisher, for discussion about this manuscript and Linda Gunn for administrative assistance. This work was supported by the P.O'B. Montgomery Distinguished Chair, the Hudson Foundation and the National Institute of Health/National Cancer Institute SPORE grant (50CA70907).</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Ortholog identification with reciprocal MegaBLAST, and alignment calculation.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Distribution of mutations by dataset and conservation score. The bins represent the normalized conservation score levels</bold>. The bins were set in to order to best simplify the figure. The random pools were position selected from that alignments obtained with each pool. For instance for random HGMD: for each HGMD mutation, a position was randomly selected within the same gene (to avoid an alignment bias) leading to a same number of random positions (20094). The common group represents mutations that were found in both datasets dbSNP and HGMD.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Leave one factor technique to identify the important factors in the SVM classifier</bold>. Abbreviations include Nt = nucleotide and cons = conservation. Bottom left square: Receiving operating characteristic (ROC) graph. The number next to each point represents the parameter left out from the SVM.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>ROC graph for the comparison of our SVM performance during testing (empty shapes) and during comparison with SIFT (full shapes)</bold>. The points labeled 'Max' represent the performances obtained during testing of our SVM on larger samples of mutations that were not used during training.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>Cumulative frequency distribution as a function of the HGMD-like probability score, with 100% being the most HGMD-like.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>Conservation score calculated as the weighted average of phylogenic distances in conserved genomes.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>Proportion of mutations with a conservation score higher than the average score of the gene in which they occur.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>SVM accuracy as a function of the training set size.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>SVM performance during testing (labeled with the keyword 'Max') and comparison with SIFT predictions.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"2\">HGMD</td><td align=\"center\" colspan=\"2\">dbSNP</td><td/><td/><td/><td/></tr><tr><td/><td colspan=\"4\"><hr/></td><td/><td/><td/><td/></tr><tr><td/><td align=\"center\"><bold>TP</bold></td><td align=\"center\"><bold>FN</bold></td><td align=\"center\"><bold>TN</bold></td><td align=\"center\"><bold>FP</bold></td><td align=\"center\"><bold>Sensitivity</bold></td><td align=\"center\"><bold>Specificity</bold></td><td align=\"center\">Accuracy</td><td align=\"center\">F-measure</td></tr></thead><tbody><tr><td/><td align=\"center\" colspan=\"8\"><bold>\n <italic>Testing set created using random mutations not used for training</italic>\n </bold></td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td align=\"center\"><bold>SVM Complete ; Max</bold></td><td align=\"center\">16527</td><td align=\"center\">3567</td><td align=\"center\">64378</td><td align=\"center\">12454</td><td align=\"center\">82.2</td><td align=\"center\">83.8</td><td align=\"center\">83.5</td><td align=\"center\">67.4</td></tr><tr><td align=\"center\"><bold>SVM Conservation Only ; Max</bold></td><td align=\"center\">14467</td><td align=\"center\">5627</td><td align=\"center\">49178</td><td align=\"center\">27659</td><td align=\"center\">72.0</td><td align=\"center\">64.0</td><td align=\"center\">65.7</td><td align=\"center\">46.5</td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\" colspan=\"8\"><bold>\n <italic>Testing set created to allow comparison with SIFT</italic>\n </bold></td></tr><tr><td colspan=\"9\"><hr/></td></tr><tr><td align=\"center\"><bold>SIFT</bold></td><td align=\"center\">159</td><td align=\"center\">245</td><td align=\"center\">13995</td><td align=\"center\">5281</td><td align=\"center\">39.4</td><td align=\"center\">72.6</td><td align=\"center\">71.9</td><td align=\"center\">5.4</td></tr><tr><td align=\"center\"><bold>SVM Complete</bold></td><td align=\"center\">287</td><td align=\"center\">117</td><td align=\"center\">12049</td><td align=\"center\">7227</td><td align=\"center\">71.0</td><td align=\"center\">62.5</td><td align=\"center\">62.7</td><td align=\"center\">7.2</td></tr><tr><td align=\"center\"><bold>SVM Conservation Only</bold></td><td align=\"center\">230</td><td align=\"center\">174</td><td align=\"center\">10216</td><td align=\"center\">9060</td><td align=\"center\">56.9</td><td align=\"center\">53.0</td><td align=\"center\">53.1</td><td align=\"center\">4.7</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Human RefSeq orthologs found in other species using reciprocal MegaBLAST. The number of orthologs found are roughly inversely proportional to the phylogenetic distances (from Human), imported from the UCSC browser.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Organism</bold></td><td align=\"left\"><bold>Total number of Reference Sequences</bold></td><td align=\"left\"><bold>Putative Orthologs found in Human</bold></td><td align=\"left\"><bold>Phylogenetic distance from Humans</bold></td></tr></thead><tbody><tr><td align=\"left\"><italic>Homo sapiens</italic></td><td align=\"left\">39,218</td><td align=\"left\">-</td><td align=\"left\">0</td></tr><tr><td align=\"left\"><italic>Pan troglodytes</italic></td><td align=\"left\">57,924</td><td align=\"left\">22,743</td><td align=\"left\">1.4</td></tr><tr><td align=\"left\"><italic>Macaca mulatta</italic></td><td align=\"left\">43,198</td><td align=\"left\">19,907</td><td align=\"left\">6.4</td></tr><tr><td align=\"left\"><italic>Canis familaris</italic></td><td align=\"left\">33,644</td><td align=\"left\">15,783</td><td align=\"left\">33.5</td></tr><tr><td align=\"left\"><italic>Bos taurus</italic></td><td align=\"left\">26,501</td><td align=\"left\">15,519</td><td align=\"left\">34.2</td></tr><tr><td align=\"left\"><italic>Mus musculus</italic></td><td align=\"left\">50,569</td><td align=\"left\">14,665</td><td align=\"left\">45.3</td></tr><tr><td align=\"left\"><italic>Rattus norvegicus</italic></td><td align=\"left\">40,672</td><td align=\"left\">14,168</td><td align=\"left\">46.1</td></tr><tr><td align=\"left\"><italic>Gallus gallus</italic></td><td align=\"left\">19,131</td><td align=\"left\">5,191</td><td align=\"left\">108.7</td></tr><tr><td align=\"left\"><italic>Danio rerio</italic></td><td align=\"left\">35,695</td><td align=\"left\">1,789</td><td align=\"left\">182.9</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Total mutations mapped and genes represented in both datasets.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"left\"><bold>HGMD</bold></td><td align=\"left\"><bold>dbSNP</bold></td></tr></thead><tbody><tr><td align=\"left\">Total coding mutations in the database</td><td align=\"left\">21,964</td><td align=\"left\">97,102</td></tr><tr><td align=\"left\">Total mutations correctly mapped</td><td align=\"left\">20,094</td><td align=\"left\">76,832</td></tr><tr><td align=\"left\">Genes represented</td><td align=\"left\">1,260</td><td align=\"left\">14,449</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p>Supplementary Table: Genes with the highest and lowest HGMD like scores involved in a disease, as per published literature. The score is the disease association probability (the maximum being 100).</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><p>Abbreviations: TP = True Positives; FP False Positives; TN = True Negatives; FN = False Negatives.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S3-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S3-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S3-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S3-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S3-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S3-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S3-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S3-8\"/>" ]

[ "<media xlink:href=\"1471-2105-9-S9-S3-S1.doc\" mimetype=\"application\" mime-subtype=\"msword\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Microarrays are widely used to overview gene expression landscapes under different experimental conditions. Since their initial appearance microarrays developed into very dependable tools with good inter- and intra-platform reproducibility [##REF##17428341##1##,##REF##16964229##2##]. Although numerous attempts to unify microarray analysis workflow were made, each manufacturer has its own methods for processing large quantities of data, and there is no general consensus as to the best means to analyze microarray data, and probably never will be. Each experimental situation is different, and different designs may be necessary for hypothesis testing as compared to hypothesis generation. With the former, one biological system is compared with another, and the significance of differences is statistically tested using a t-test, generally with the assumption that each biological sample is homogeneous. In such experiments statistical power becomes the driving consideration. In hypotheses generating experiments, a number of biological situations are compared, for example a series of different cell lines, a time course study or a dose-response study. The biological samples may not be homogeneous. Cost becomes a major consideration because the number of replicates needed to test hypotheses may make experiments prohibitive. Thus, there is a need for analytical approaches to use under hypothesis-generating conditions that are based on sound statistical principles but which nonetheless reduce the number of replicates needed to assemble at least a preliminary global picture of the effect of a particular biological situation on gene expression [##REF##10475062##3##].</p>", "<p>We present here a statistically robust approach for analyzing the changes in the transcriptome that is driven by the underlying biology. Previous work by I. Dozmorov showed that approaches based on separating variability in expression of genes into biological and technical sources provide an alternative means of identifying \"genes of interest\" for further analysis [##REF##12538240##4##, ####REF##15514108##5##, ##REF##15128432##6##, ##REF##15271778##7####15271778##7##]. Under the assumption that in any experiment most genes do not change expression, the F-test is used to identify genes that express more variability than the overall technical variability of the system. This set of genes is referred to as \"hypervariable genes,\" and has been assumed to reflect the relevant biological variables in the system. In this communication we have added a number of in silico tests based upon properties of these genes that do not depend upon expression. These additional analyses confirmed that at the level of transcriptional regulatory networks this approach does identify important genes that can then be assembled into networks of functions, transcriptional regulation and with previous knowledge. This represents a further extension of work published in our laboratory that included only <italic>in silico </italic>analyses [##REF##16412233##8##,##REF##17291344##9##].</p>", "<p>We applied this method to examining the effect of cancer remodeled extracellular matrix (crECM) on bladder papilloma-derived cell line (RT4) as they grow over the course of several days on a crECM after having been transferred from culture on plastic [##REF##12926044##10##]. Papillomas represent a very early premalignant change, and determining how a crECM can drive them toward malignancy could identify novel targets for therapy. Genes exhibiting major changes in expression introduced by crECM were selected and their functions examined. Two overlapping canonical pathways were identified as the main targets. Finally, transcription factors regulating the genes of interest were found and their validity proven by additional experimental method. Such integrative approach may reveal new roles of unknown genes [##REF##11245483##11##], new drug targets [##REF##10475062##3##], and lead to clinical tests [##REF##16849555##12##].</p>" ]

[ "<title>Methods</title>", "<p>The RT4 bladder transitional cell papilloma cell line (American Type Culture Collection, Manassas Virginia) was cultured on plastic and on cancer-remodeled ECM, Matrigel™ (crECM), and RNA was isolated as described previously using the RNAeasy kit (QIAGEN Inc., Valencia, CA) [##REF##16412233##8##,##REF##12954494##19##].</p>", "<p>Microarray data were obtained using a spotted array from cells cultured on plastic (two arrays) and across 9 days time course of growth on crECM, as previously described [##REF##17291344##9##]. Cy3 labeled cDNA was synthesized and hybridized onto glass arrays spotted with 22,464 long oligos (~70 mers) from the UniGene database of functionally known genes.</p>", "<p>Expression data were normalized to the variability around the zero point, as described previously [##REF##15514108##5##,##REF##15271778##7##]. Genes were considered to be expressed if their expression normalized to the S.D. of zero point exceeded 3.0 (p &lt; 0.001). The arrays were then globally adjusted to each other by robust linear regression. Genes expressing higher variability than the technical variability of the system (hypervariable, or HV genes) were identified as described previously [##REF##12538240##4##] and as shown in Figure ##FIG##2##3B##. Two thresholds were used. HV genes showed relative standard deviations &gt; 3.8 SD above that of the population mode (p &lt; 6 .7 × 10^-5). This value is 1/N, where N is the number of expressed genes (~15,000). A more stringent criterion of &gt; 5 SD (p &lt; 2.87 × 10^-7) was also used to identify a subset of very HV, or \"beacon\" genes. Their expression profiles were clustered by the Cluster 3.0 program <ext-link ext-link-type=\"uri\" xlink:href=\"http://bonsai.ims.u-tokyo.ac.jp/~mdehoon/software/cluster/software.htm\"/>[##REF##9843981##20##]. Hierarchical clustering identified major patterns of gene expression changes; further clustering of selected \"beacon\" genes was done using K-means clustering, k = 3, which was selected as described below. Significantly over-represented gene ontologies were identified using the Database for Annotation, Visualization and Integrated Discovery (DAVID, <ext-link ext-link-type=\"uri\" xlink:href=\"http://david.abcc.ncifcrf.gov/\"/>) [##REF##12734009##16##]. Biologically relevant networks were assembled from identified clusters and groups of common genes by using Ingenuity Pathways Analysis (IPA, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ingenuity.com/\"/>). Each gene identifier was mapped to its corresponding gene object in the Ingenuity Pathways Knowledge Base. Genes were not weighted by expression levels, and biological networks were built on this assumption. Analysis of common TREs shared by genes in each ontological cluster was performed by using the web-based program PAINT <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.dbi.tju.edu/dbi/tools/paint/\"/>[##REF##14583114##17##] against whole list of genes in microarray.</p>", "<p>Additional assessment of the DNA binding activity of transcription factors using TranSignal Protein/DNA Combo Arrays (spin column version, # MA1215, Panomics, Redwood City, CA) was conducted according to the manufacturer's protocol. Briefly, cell nuclear extracts were incubated with biotin-labeled oligonucleotides that possess consensus DNA binding sites for 345 transcription factors. The protein bound probes were isolated by using a spin column and then hybridized to the DNA/protein array. After the DNA/protein array was washed, the array was incubated with detection solution and images of the chemiluminescent signal were captured using an Alpha Imager (Alpha Innotech, San Leandro, CA) and quantitated by using AlphaEase software and standardized against biotinylated DNA spots on the membrane. The results are linear between 0 and 100 units. In order to detect low expression transcription factor activity, two exposure times were used, 8 and 25 min. Values were adjusted for exposure so that all values were measured within the linear range of the assay. A histogram of expression data was plotted and was found to be bimodal, with one mode centered about zero. Background subtraction was performed by calculating the standard deviation of this distribution and subtracting 3 standard deviations above the mode from all expression values, approximately 6 units.</p>" ]

[ "<title>Results</title>", "<p>The workflow chart of our approach is shown in Figure ##FIG##0##1##. The normalization process is presented because it is different from most approaches using normalization to the mean, median, or housekeeping genes. The frequency histogram of the un-normalized expression values yielded a bimodal, right-skewed curve as shown on Figure ##FIG##1##2A##[##REF##17291344##9##]. The distribution around the peak near zero was fitted to a Gaussian curve, providing a measure of the variability around zero that can be used to identify genes expressed significantly above zero. The zero point itself is slightly above zero because of non-specific binding. Interestingly, the array contained completely blank spots, which show up as a sharp peak exactly around zero, as well as a large number of spots of the solvent (3× SSC) the long oligo clones are contained in. The average of those points corresponds almost exactly to the zero point established from the entire array by the method above. The standard deviation of this peak is used to normalize all the expression data. Normalizing expression to the uncertainty in zero allows for a ready determination for the threshold of non-zero expression. Thus, expression value of \"3\" corresponds to 3 standard deviations (SD) above the zero point and corresponds to the valley in the total distribution. With this value, the p-value of a false positive assignment of expression vs. non-expression is &lt; 0.001. With a threshold of 5 SD, the p-value for a false positive is p &lt; 2.87 × 10^-7. The arrays were then Log10 transformed and globally adjusted to each other by robust linear regression, which assumes that the expression of most genes is not altered in the experiment and down-weights the effect on global expression of those that do change. Box plots (Figure ##FIG##1##2B##) graphically show this adjustment. Full sets of raw and transformed data are available on Gene Expression Omnibus (GEO, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/geo/\"/>), accession number GSE9291.</p>", "<title>Gene expression dynamics</title>", "<p>Genes expressed 5 SD above background were considered highly significantly expressed. While this choice is arbitrary, 5 SD above background was selected to focus on highly expressed genes and to minimize false positives by eliminating the noisy low-expressed genes. Of the 21,308 unique probes represented on the array, a total of 15,287 were expressed at 5 SD above background on the two arrays from cells grown on plastic and on the nine arrays from cells grown on crECM. The next step is to identify genes whose expression level responds to crECM. The total variance in expression of any gene is the sum of the technical variance (variance due to the measurement itself), Vt, the relevant biology, Vr, and the irrelevant biology Vi. Vi may be attributed to deviations in preparation of biological samples or other biological factors not related to the biology of interest. Genes that do not respond to the biological variables will express only technical variability and can effectively serve as \"housekeeping genes\" or a reference set against which genes that do respond can be identified with an F-test. Genes responsive to the biology generally show high variability that likely is systematic across the experiment. The overall mean technical variability was determined from the mode of the frequency distribution histogram of standard deviation of the data set (Figure ##FIG##2##3A##) and was 8.41% with a standard deviation of 6.75%. Thus, genes with relative standard deviations exceeding 3 SD above the mean (28.56%) were defined as hypervariable. A total of 3462 hypervariable (HV) genes were identified. However, the lack of replicates leads to an excessive sensitivity to single outliers. A second screening by a leave-one-out method [##REF##16954463##13##] minimized false positives due to a single data point. A total of 2743 HV genes were judged to be hypervariable (p &lt; 6.7 × 10^-5) and, presumably, represent all the genes that respond to the presence of a crECM. This approach has proven reliable in identifying a set of genes of interest with minimal need for replicates as compared to methods based on t-tests [##REF##12538240##4##,##REF##17291344##9##].</p>", "<title>Visualization of gene expression changes</title>", "<p>For easy visualization of up- and downregulated genes their expression values were antilog-transformed and normalized around zero with standard deviation equal to 1. They were organized into 3 clusters by K-means clustering with 1000 runs and a similarity metric as correlation (uncentered). Increasing number of clusters yielded similar clusters differing only in amplitude, while using 2 clusters failed to distinguish the obvious dynamics. The results were visualized with Java TreeView program <ext-link ext-link-type=\"uri\" xlink:href=\"http://jtreeview.sourceforge.net/\"/>[##REF##15180930##14##,##REF##18220245##15##] (data not shown). The clustering of hypervariable genes demonstrated that the vast majority of the biological variability was between cells grown on plastic and on the crECM, and with the main pattern being a change in state of expression, i.e. from off to on or vice versa. After the first 12 hours on the crECM, little variability in gene expression was observed. This indicates the changes of gene expression level occurred within the first 12 hours in cells grown on crECM, which in turn drives phenotypic changes (Figure ##FIG##3##4##). That later time points are not different from the earlier ones indicates no new biological processes such as cell death were introduced over the time course of the experiment.</p>", "<p>This represents a very large set of genes of interest, and other than determining ontologies, such a large number is inconvenient to interpret. We therefore focused on the genes showing the largest changes, considering they could serve as \"beacons\" to draw our attention to the changes in underlying processes. Because the dynamics of gene expression exhibit mainly a change in state between cells growing on plastic vs. crECM, the set of hypervariable genes was filtered to identify the genes that are expressed below noise level on plastic and highly expressed on crECM (\"off-on\" genes) and vice versa (\"on-off\" genes) (Figure ##FIG##3##4A##). The two arrays of gene expression on plastic and nine arrays of timecourse on crECM were each averaged and genes with average expression level on plastic &lt; 0 and on crECM &gt; 1 (Log10 scale, average identifies geometric mean) were selected as \"off-on\" genes. The opposite criteria were applied to identify \"on-off\" genes. Using these stringent criteria, a total of 877 unique \"off-on\" genes were turned on by the crECM, whereas a total of 74 unique \"on-off\" genes were shut down by the crECM. The validity of this approach was tested in the next section.</p>", "<p>Beside the main dynamic of state change genes a smaller number of genes showed change in level. Figure ##FIG##2##3B## shows the distribution of the ratio of expression of genes from cells grown on plastic to those grown on crECM. The standard deviation of this ratio is 0.2, and 3 standard deviations (0.6) corresponds to a 3-fold difference. A total of 241 genes were identified that were expressed at least 5 SD above background in cells grown on plastic and showed at least a 3-fold increase in expression. Only 67 genes showed the opposite pattern being highly expressed on plastic and decreasing 3-fold but still expressed above noise.</p>", "<title>Gene ontology analysis and visualization</title>", "<p>The ontologies of the genes of interest were examined using the Database for Annotation, Visualization and Integrated Discovery (DAVID, <ext-link ext-link-type=\"uri\" xlink:href=\"http://david.abcc.ncifcrf.gov/\"/>) [##REF##12734009##16##] tool, which examines all the functions represented by each gene in a gene list and identifies groups that share ontologies. The over-represented ontologies form the basis for identifying functional processes represented in the change of state induced by a crECM. Several parameters can be adjusted to achieve a reasonable and comprehensive set of ontologies and associated genes. For the 877 \"off-on\" genes the following parameters were set: Similarity term overlap: 5; Similarity threshold: 0.5; Initial group membership: 5; Final group membership: 5; Multiple linkage threshold: 0.5, which is equal to the \"Highest\" stringency setting in DAVID. After examining the results provided by different stringencies, the above set was selected because the picture presented overall affinities without too many groups but provided sufficient detail to build a conceptual model of the effect of crECM on progressing urothelial cells.</p>", "<p>Out of the 877 \"off-on\" genes 190 clustered into 12 clusters of ontologies at highest stringency and 86 did not have recognized ontologies. These 86 unannotated genes likely represent either novel processes not currently identified or genes whose participation in known processes has not yet been discovered [##REF##11245483##11##]. The remaining 601 were predominantly distributed among \"related genes\" that shared some ontological features with one or more of the 12 clusters but did not rise above the threshold of significance. Some were entirely irrelevant and showed no similarity to any of the clusters. This step is illustrated in Table ##TAB##0##1## along with significantly over-represented TREs shared by all members of each cluster.</p>", "<p>Examination of 241 level-change genes increased on crECM by medium stringency ontological analysis of these genes found five clusters of functions, three of which were similar to those for state change.</p>", "<p>The 74 genes that were shut off and the 64 genes that were 3-fold down-regulated on crECM were less informative than were those that were turned on or up-regulated. At same stringency as was used for \"off-on\" genes, one over-represented cluster was identified in the genes that were shut off and consisted of 6 transcription factors sharing homeobox ontology. Decreasing the stringency to \"medium\" (the default for DAVID) increased the number of genes in the cluster to 10 but did not add clusters. Genes that were down-regulated at least 3-fold yielded two clusters under medium stringency.</p>", "<p>The validity of the selection of \"beacon genes\" was tested by comparing the ontological clusters observed with the entire set of 2743 HV genes. A total of 17 clusters was seen, all of which were identified using the \"beacon\" genes. This demonstrates that the smaller data set of state- and level-change genes will identify all the processes seen in the larger set of HV genes.</p>", "<title>Pathway analysis</title>", "<p>Having preliminary understanding of functions represented by state change genes, they were probed for membership in canonical pathways by Ingenuity^©Pathway Analysis (IPA, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ingenuity.com/\"/>). IPA maps each gene identifier to its corresponding gene object in the Ingenuity^©Pathways Knowledge Base, and generates multiple biological networks with associated ontologies from a list of focus genes, as well as general gene ontologies overrepresented. IPA canonical pathways analysis identified the most significant known biological pathways for a given set of genes. For identification of a significant canonical pathway or pre-defined network of genes it is only necessary to identify a single member as significant, hence the term \"beacon\" genes. The participation of other members of the network is checked manually against the entire data set to ensure they are expressed, or show a smaller change than the \"beacon\" genes threshold [##REF##17291344##9##]. Pathways or connections involving genes that are not expressed are deleted. This process is particularly helpful in cases where a large number of genes of interest has been discovered. The smaller, more tractable set of \"beacon\" genes are used to draw attention to processes, and all the details are filled in with the entire data set as is shown below.</p>", "<p>Of the 877 \"off-on\" genes 165 failed to map and represent unannotated genes about which little or nothing is known. The difference in annotation with the DAVID is due to IPA being curated. Of the 86 not annotated by DAVID 74 also were not annotated by IPA. Of the 714 mapped genes, 151 fit into various cell signaling processes and 133 were involved with cellular growth and proliferation. More informative were interconnected canonical pathways, many of which overlap. Any one gene may exhibit multiple functions and participate in multiple pathways. The most significant canonical pathways identified were the interconnected G-protein and NF-κB signaling networks (22 \"beacon\" genes combined). An NF-κB network is shown in Figure ##FIG##4##5## with the gene expression dynamics indicated with a color code. When compared back against the set of HV genes and expressed genes, every member of the network was expressed, and many were found in the set of HV genes, which meant they showed smaller changes than the \"beacons.\"</p>", "<title>Identification of potential transcriptional networks</title>", "<p>Genes sharing similar ontologies may be regulated by one or more common transcription factors. \"Off-on\" genes clustered by DAVID were tested for the presence of common transcription regulatory elements (TREs) upstream of the transcribed genes in each ontological cluster by the web-based program Promoter Analysis and Interaction Network Toolset, v.3.5 (PAINT, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.dbi.tju.edu/dbi/tools/paint/\"/>) [##REF##14583114##17##]. PAINT queries the Transfac™ database and calculates the probability that the TREs identified in a given list of genes differ from TREs in a random sample of genes. In this case, the basis of partitioning into ontological clusters being driven by particular transcription factors was tested by comparing the TREs found in each cluster against the entire list of 21 K genes present in the microarray. This provided a map of TREs significantly overrepresented in a given cluster against a significance threshold of p &lt; 0.05. For more reliable results filtering with a false discovery rate (FDR) &lt; 0.3 criterion was used, when specified. An example identifying the significantly overrepresented TREs in cluster 1 is shown in Figure ##FIG##5##6A##. The over-represented TREs are also summarized in Table ##TAB##0##1## by cluster. The probability of a random collection of genes sharing a common TRE is less than 0.05. Thus the finding that a set of genes contains common TREs or fit into known networks supports that they are neither randomly selected by chance, nor the product of technical error to within the limits of statistical testing.</p>", "<p>The predictions of PAINT were tested using an independent experimental assay that measured the DNA binding activity of transcription factors using TranSignal Protein/DNA Combo Arrays. This array allows estimation of binding activity of 345 TFs. We found a number of up-regulated transcription factors on crECM. The bar diagram in Figure ##FIG##5##6B## compares the activity of selected transcription factors in cells grown on plastic vs. crECM. The majority of transcription factors that showed large changes in activity were also identified by PAINT as driving up- or down-regulated clusters of genes, the results are shown in Table ##TAB##0##1## with TREs confirmed by Panomics array highlighted in bold. Depending on stringency, between 5 and 13 transcription factors were shut off and between 25 and 40 were activated by the crECM. Supplemental table S1 in Additional file ##SUPPL##0##1## shows DNA binding activity of all transcription factors in cells grown on plastic or crECM. Most of the changes were \"off-on,\" as was observed for the mRNAs of the downstream genes. Some transcription factors were active in cells growing under both conditions.</p>" ]

[]

[ "<title>Conclusion</title>", "<p>In this study we present a self-guided approach for analyzing a complex biological change by microarrays and illustrate its use to describe the complex change in gene expression that occur when papilloma cells are placed on a crECM. The flowchart of each step of this approach is shown in Figure ##FIG##0##1##. We also confirm the validity of the integrated approach by independent verification of the predictions of transcriptional regulatory networks. With a very complex biological system mobility more than 2000 genes identified as significantly varying, we show that the essential elements of the change in the large scale picture of the biology can be captured in a smaller subset of \"beacon\" genes. Analysis of this more concise set of \"beacons\" facilitates mapping the gene expression dynamics onto known processes [##REF##16420732##18##]. We wish to emphasize that the resulting biological picture does not derive solely from indentifying only a few key genes. This approach also requires that all members of a pathway be expressed, which is determined by comparing putative networks or canonical pathways against the entire dataset of expressed genes. Genes showing smaller changes than shown by the \"beacons\" are identified against the set of HV genes.</p>", "<p>The approach also is statistically robust. Expression is judged against the uncertainty of the zero point, and the threshold can be selected either to minimize false negatives or false positives. The need for replicates, and therefore the cost of experiments, is minimized using a global F-test against the variance of the system as a whole with a p-value standard of 1/N that minimizes false positive identification of significant genes. The HV gene approach is best suited to providing an overall description for hypothesis generation with multiple biological variables as opposed to hypothesis-testing in a two-state system (e.g. treated and control).</p>", "<p>In summary, this article demonstrates an approach to microarray analysis that organizes the findings into a biologically based model that should in turn, facilitate generation and testing of hypotheses because the analysis itself is structured around the properties of biological system. In this case, the findings suggest that G-protein signaling plays a major role in the modulation of phenotype by crECM, that the cells are differentiated and acquire specialized functions (e.g. immune function and transmembrane proteins) and that several transcription factors regulate the process.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<p>A statistically robust and biologically-based approach for analysis of microarray data is described that integrates independent biological knowledge and data with a global F-test for finding genes of interest that minimizes the need for replicates when used for hypothesis generation. First, each microarray is normalized to its noise level around zero. The microarray dataset is then globally adjusted by robust linear regression. Second, genes of interest that capture significant responses to experimental conditions are selected by finding those that express significantly higher variance than those expressing only technical variability. Clustering expression data and identifying expression-independent properties of genes of interest including upstream transcriptional regulatory elements (TREs), ontologies and networks or pathways organizes the data into a biologically meaningful system. We demonstrate that when the number of genes of interest is inconveniently large, identifying a subset of \"beacon genes\" representing the largest changes will identify pathways or networks altered by biological manipulation. The entire dataset is then used to complete the picture outlined by the \"beacon genes.\" This allow construction of a structured model of a system that can generate biologically testable hypotheses. We illustrate this approach by comparing cells cultured on plastic or an extracellular matrix which organizes a dataset of over 2,000 genes of interest from a genome wide scan of transcription. The resulting model was confirmed by comparing the predicted pattern of TREs with experimental determination of active transcription factors.</p>" ]

[ "<title>List of abbreviations used</title>", "<p>cDNA – complementary deoxyribonucleic acid; crECM – cancer-remodeled extracellular matrix; DAVID – Database for Annotation, Visualization and Integrated Discovery; DNA – deoxyribonucleic acid; ECM – extracellular matrix; FDR – false discovery rate; HV – hypervariable genes; IPA – Ingenuity^©Pathway Analysis; mRNA -messenger ribonucleic acid; PAINT – Promoter Analysis and Interaction Network Toolset; SD – standard deviation; TF – transcription factor; TRE – transcription regulatory element; SSC – Saline-Sodium Citrate</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Supplementary Material</title>" ]

[ "<title>Acknowledgements</title>", "<p>The authors gratefully acknowledge the technical assistance of Jean Coffman. This work was supported in part by a grant from the NIH to REH, R01 DK069808</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Schematic diagram of steps in microarray analysis.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Identification of noise level and microarray adjustment</bold>. A) Frequency histogram of gene expression level from one microarray dataset. The first peak in the bimodal distribution represents the normal distribution of system noise centered around zero. Genes expressed 3 SD above noise level are defined as expressed genes. B) Box plots of microarray datasets before and after linear regression, values are log10 transformed.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Identification of hypervariable and differentially expressed genes</bold>. A) Frequency histogram of variances of genes across timecourse. The normal distribution of low-variable genes identified from the left part of the histogram; the white vertical line marks the threshold for hypervariable genes expressed 3.8 SD above distribution of constant genes expressing only technical variability. B) Log10 ratio of average gene expression of cells grown on plastic and crECM is presented as a frequency histogram. Ratio values &gt; 2 or &lt; -2 were truncated and set to 2 and -2, respectively.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Visualization of changes in gene expression level identified by microarray analysis</bold>. A) Schematic graphs of main changes observed in the system from clustering of hypervariable genes. B) Part of clustering heatmap of \"beacon\" state change genes: P 1, P 2 – duplicate gene expression profiles of RT4 cells grown on plastic; M 0.5, M 1, M 2 etc. – cells grown on crECM for the indicated number of days. Red/green intensity indicates level of gene expression, up-/downregulated, respectively.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>The NFκB canonical signaling pathway from IPA</bold>. Dark red &gt; 3-fold increase in gene expression; light red &lt; 3-fold increase in gene expression; dark green – &gt; 3-fold decrease in gene expression; light green – &lt; 3-fold decrease in gene expression; gray – unchanged gene expression; no color – gene not in array. Gene symbols with a single border represent single genes. Double border represent a complex of genes or the possibility that alternative genes might act in the pathway.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Transcription factor activity identified <italic>in silico </italic>and <italic>in vivo</italic></bold>. A) Example of TREs overrepresented in first ontological cluster. Several genes (vertical) share common TREs (horizontal), highlighted by red. Results were filtered to show only TREs overrepresented at p &lt; 0.05 and FDR &lt; 0.3. TREs in bold show a significant increase in expression on crECM compared to plastic confirmed by transcription factor array experiment. B) Example of changes in binding activity of a few TFs on plastic and crECM. Gray/black bars show binding activity on plastic/crECM, respectively.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Major functional groups overrepresented among state-change genes and corresponding overrepresented TREs. The groups are presented according to the order of significance identified by DAVID. Overrepresented TREs marked in bold are either \"off-on\" TFs or increased their level; regular – not present in Panomics set; <italic>italics </italic>– not present under either condition.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Functional group</bold></td><td align=\"center\"><bold>Number of genes</bold></td><td align=\"center\"><bold>Major Gene Ontologies</bold></td><td align=\"center\"><bold>Overrepresented TREs</bold></td></tr></thead><tbody><tr><td align=\"left\" colspan=\"4\"><italic>Gene expression, RNA processing and protein synthesis</italic></td></tr><tr><td align=\"center\">12</td><td align=\"center\">73</td><td align=\"left\">Transcription factors</td><td/></tr><tr><td align=\"center\">7</td><td align=\"center\">5</td><td align=\"left\">Translation initiation factors</td><td align=\"left\"><bold>v-Maf</bold>, SOX-9, FOXJ2, CP2, <bold>HFH-3</bold>, <bold>Elk-1</bold>, NRF-2, <italic>AREB6</italic></td></tr><tr><td align=\"center\">9</td><td align=\"center\">6</td><td align=\"left\">RNA processing – ribosome biogenesis</td><td align=\"left\">NGFI-C, GR, <italic>HNF-4</italic>, <bold>YY1</bold>, <bold>Elk-1</bold>, NRF-2, <bold>v-Myb</bold>, <bold>NF-κB</bold>, TATA, <italic>c-Myc</italic></td></tr><tr><td align=\"center\">11</td><td align=\"center\">8</td><td align=\"left\">Zinc binding</td><td align=\"left\">MIF-1, <bold>Tax/CREB</bold></td></tr><tr><td align=\"left\" colspan=\"4\"><italic>Cell signaling proteins</italic></td></tr><tr><td align=\"center\">3</td><td align=\"center\">16</td><td align=\"left\">G-protein receptor</td><td/></tr><tr><td align=\"center\">2</td><td align=\"center\">7</td><td align=\"left\">GABA receptor/ion channels</td><td align=\"left\">N-Myc</td></tr><tr><td align=\"center\">6</td><td align=\"center\">9</td><td align=\"left\">Ion channels, K</td><td/></tr><tr><td align=\"left\" colspan=\"4\"><italic>Post-translational modification and regulatory control</italic></td></tr><tr><td align=\"center\">8</td><td align=\"center\">5</td><td align=\"left\">Glycosyltransferases</td><td/></tr><tr><td align=\"center\">10</td><td align=\"center\">28</td><td align=\"left\">Kinases</td><td align=\"left\"><bold>CDP</bold>, CR3+HD, <italic>CRE-BP1</italic>, <bold>CCAAT</bold></td></tr><tr><td align=\"left\" colspan=\"4\"><italic>Cell-ECM adhesion</italic></td></tr><tr><td align=\"center\">4</td><td align=\"center\">4</td><td align=\"left\">Cadherins</td><td align=\"left\"><bold>HNF-3β</bold>, <bold>CDP </bold>CR3+HD, <bold>E2</bold>, <bold>NF-κB</bold>, <bold>USF</bold></td></tr><tr><td align=\"left\" colspan=\"4\"><italic>Immune function associated with suppression of effector T-cells</italic></td></tr><tr><td align=\"center\">1</td><td align=\"center\">15</td><td align=\"left\">Transmembrane immunoglobulin-like proteins</td><td align=\"left\"><bold>NF-κB</bold>, <bold>v-Maf</bold>, <bold>RSRFC4</bold>, FOXJ2, <bold>AP-1</bold>, <bold>Pax-4</bold>, <bold>USF</bold>, <bold>CDP</bold>, Brn-2</td></tr><tr><td align=\"left\" colspan=\"4\"><italic>Transmembrane proteins of unknown significance</italic></td></tr><tr><td align=\"center\">5</td><td align=\"center\">14</td><td align=\"left\">Transmembrane proteins</td><td align=\"left\"><bold>AP-1</bold></td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"S1\"><caption><title>Additional file 1</title><p>Supplemental Table S1. Activity of 345 transcription factors in cells grown on Matrigel and on Plastic. Values were obtained as described in Methods and are arbitrary units, but are normalized to protein content and exposure time against an assay standard of biotinylated DNA.</p></caption></supplementary-material>" ]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S4-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S4-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S4-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S4-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S4-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S4-6\"/>" ]

[ "<media xlink:href=\"1471-2105-9-S9-S4-S1.doc\" mimetype=\"application\" mime-subtype=\"msword\"><caption><p>Click here for file</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The concept of networks is ubiquitous in systems biology. In the past decade, high-throughput experimental techniques such as yeast 2-hybrid systems and mass spectrometry-based proteomics led to an influx of biomolecular interaction data in curated databases such as HPRD [##REF##14525934##1##], DIP [##REF##11752321##2##], and BIND [##REF##12519993##3##]. Computational methods to predict protein interactions with domain interaction profiles [##REF##12368246##4##], co-expression patterns [##REF##15797912##5##], and term co-occurrences based on text mining [##REF##11301305##6##] have also led to the development of databases such as OPHID [##REF##15657099##7##], InterNetDB [##REF##17112386##8##], UniHi [##REF##17158159##9##], HAPPI [##UREF##0##10##], and STRING [##REF##15608232##11##]. These databases support the transformation of biological network studies into essential biological data analysis tasks that include inferring global protein functions [##REF##12740586##12##], assembling protein modules [##REF##12461819##13##], integrating different Omics data sets [##UREF##1##14##], reconstructing biological pathways [##REF##17854489##15##], predicting disease-relevant genes/proteins [##UREF##2##16##] and developing panel biomarkers [##REF##17922014##17##].</p>", "<p>Many network visualization software tools have been developed recently to help biological researchers visually query, annotate and analyze biomolecular network data. For example, Cytoscape [##REF##14597658##18##] is one of the most commonly used software platforms that contains all basic functions for visualizing and annotating a network graph derived from protein-protein or protein-DNA interaction data. It has a robust graph layout engine that allows for automatic layout and manual control of network graph nodes and edges attributes corresponding to user annotation data. Cytoscape adopts an open and flexible software architecture that supports software plug-ins, which extends the core functionality of Cytoscape through third-party software extensions. VisANT [##REF##15980487##19##] competes with Cytoscape by offering several built-in statistical functions to help users calculate several key network topological parameters and perform global real-time network analysis. WebInterViewer [##REF##15215357##20##] uses a ultra-fast graph-layout algorithm that can scale up for manipulating the layout of a biomolecular interaction network up to tens of thousands of nodes on a desktop computer, while providing several network abstraction and comparison operators. The most recent feature-rich network data analysis software tool, Biological Networks [##REF##16845051##21##], enables advanced bioinformatics users to integrate microarray data analysis with biomolecular interaction network analysis over a diverse set of database choices through powerful template-based query interfaces. Pathway Studio [##REF##14594725##22##], which is available commercially, also uses powerful visualization engine and query interfaces, and allows its users to manage and access data stored in relational databases and to integrate biomolecular interaction data from its PubMed literature mining engine with other sources. In summary, current development trend is to equip users with extended ability to query and interpret existing experimental data, particularly those from \"Omics\" platforms, in the emerging context of biomolecular interaction networks.</p>", "<p>Recent research in network biology has expanded beyond the study of protein-protein interactions or protein-DNA interactions, therefore presenting new challenges and opportunities for biological network visualization and analysis software. These networks are more complex, with heterogeneous types of biological entities spanning broad range of scales from molecular (i.e., DNA, proteins, metabolites), to super-molecular (i.e., gene ontology categories, protein complexes, pathways), to intercellular (i.e., signaling between different cell types), to tissue and physiological (i.e., individual disorder types) levels. For example, Goh <italic>et al</italic>. explored all known associations of disease phenotypes by representing disease phenotypes instead of molecular entities as nodes in a network graph [##REF##17502601##23##]. They described two new types of biological networks, \"disease interaction network\" and \"disease-gene network\", in which the former represented disease names as nodes and disease associations at the molecular level (sharing &gt; 1 disease genes between associated diseases) as edges, while the latter represented genes as nodes and gene associations shared in a common disease (shared &gt; 1 diseases between associated genes) as edges. To characterize the global relationships between protein targets and all chemical drug compounds available today, Yildirim <italic>et al</italic>. [##REF##17921997##24##] built a drug-target association network representing all known drugs and their targets recorded in the DrugBank database [##REF##18048412##25##]. The network offered an intriguing view with \"hot\" drug intervention points (popular drug targets) and multi-targeted drugs clearly displayed. Analyzing the data in multi-scale biological networks is inherently more challenging than that of biomolecular interaction networks, primarily because the heterogeneous interacting biological entities may differ significantly in size, quality, complexity and annotation requirements, making it combinatorial more difficult to develop user interfaces that preserve usability and robustness at the same time. Few existing tools today can empower users to perform \"visual analytics\"–to discover novel information through visualization–for multi-scale biological networks.</p>", "<p>To support multi-scale biological network visual analytics studies, new software tools must meet three basic requirements. First, the bulk of data should be managed by robust backend engines that support rich schemas such as relational database management systems (e.g., PostgreSQL, Oracle) or XML/RDF data stores (e. g, Jena, Piazza). Flat files quickly become unsustainable beyond one or two spreadsheets of custom user input data, due to lack of a standard schema and difficulty in combining information from separate spreadsheets. Second, iterative, exploratory and bi-directional data analysis capabilities to save temporary results and build visualization sessions on top of one another should be a pre-requisite. Many current software tools support only one-way information flow from data sheets to visualization, and therefore should be referred to as \"visual annotation\" or \"visual display\" tools instead of \"visual analytic\" tools. Third, visual querying languages, even if borrowed directly from SQL in relational database querying or SPARQL in semantic web based data querying, will become quite beneficial to advanced users, who have to filter different facets of biological networks and manipulate complex network analysis tasks, by automating tasks that are \"menu-driven\" or \"mouse-click intensive\". As Suderman <italic>et al </italic>recently surveyed, none of the 35 commonly used biological network visualization tools supported such query languages embedded directly [##REF##17720984##26##].</p>", "<p>We developed ProteoLens as a new visual analytic software platform for creating, annotating and analyzing multi-scale biological networks. When compared with existing biological network visualization tools, ProteoLens introduced a new set of design choices, which made it easy for bioinformatics expert data analysts work on large sets of biological networks and Omics data. There are three primary characteristics that distinguish it from existing network visualization tools. First, it supports direct database connectivity to Oracle and PostgreSQL database and SQL statements including both Data Definition Languages (DDL) and Data Manipulation Languages (DML). Users of ProteoLens can use the tool to iteratively prepare data stored in relational databases without leaving the visual analytic environment. Data from different tables in a complex relational database schema can also be queried on the fly to create networks at the appropriate level for exploration. Second, ProteoLens supports graph/network represented data expressed in standard Graph Modeling Language (GML) formats. Therefore, visual layouts performed in comparable software tools can interoperate with ProteoLens as long as they also support GML standards. This allows users to perform visual network analysis for data from heterogeneous sources that are syntactically represented in non-relational format. Third, it supports the decoupling of complex user interfaces for network visualization into two separate functional layers: data annotation and data visualization. The concepts of \"node association rules\" and \"edge association rules\" provide users with significant flexibility in choosing what data attributes (e.g., score, rank, description) to map to nodes or edges, and association visualization display options allow to select visual effects to represent values of these attributes later.</p>", "<p>In the next several sections, we first describe ProteoLens implementation and then demonstrate how it can be used to enable multi-scale biological network-based research through three case studies.</p>" ]

[]

[ "<title>Results and discussion</title>", "<title>New features</title>", "<p>Figure ##FIG##1##2## illustrates the core functionalities of ProteoLens. We compared ProteoLens with several existing popular visualization software tools, including Cytoscape, VisANT and BiologicalNetworks, and summarized the comparison in Table ##TAB##0##1##. We describe the features of ProteoLens in detail in this section.</p>", "<title>a. Relational and XML data support</title>", "<p>Currently, ProteoLens supports two types of physical data sources: tab-delimited text files on the local file system and tables/views in relational tables managed by Oracle 10 g or PostgreSQL 8.x database management systems. A user can manage and query the data stored in the relational database, create network data association rules from the view, and immediately make the rule available for visual annotation. Since only meta-data are stored when a data association is created, the query execution can be performed in real time of visualization against the underlying complex data structure in the database. This design allows for an efficient data retrieval and analysis, and saving of the data file and workspace resources. Almost infinite configurations of data views can be created from multiple underlying data sources, and be used for building complex integrated visualizations. ProteoLens also supports semi-structured data format in Graph Modeling Language (GML) – the standard file format in the Graphlet graph editor system, for non-relational graphs. Network visualization is created in a view can be saved in a GML file, thus allowing for reopening and further editing in a new session, or data exchange without relational databases. The network view can be exported as a JPEG or PNG file. The user can import and manipulate any network data using standard GML file formats in addition to structured data stored in the relational databases. ProteoLens stores every data association in a session configuration XML file. Users can save the session and recommence their analysis at any time.</p>", "<title>b. SQL-based visual data analysis</title>", "<p>ProteoLens supports direct database connectivity through Java Database Connectivity (JDBC) to Oracle and PostgreSQL database tables and views, and the entire set of database operations can be specified using full SQL statements including Data Definition Languages (DDL) and Data Manipulation languages (DML). This extends the range of data that expert users may bring into later network visualizations for annotation and visual exploration tasks. Users of ProteoLens can iteratively prepare data stored in relational databases without leaving the visual analytic environment. Data from different tables in a complex relational database schema can also be queried on the fly to create networks at the appropriate level for exploration. SQL queries are also used to present \"views\" of different underlying database tables for network data associations, therefore making it possible for users to perform all visualization pre-analysis without leaving the ProteoLens platform.</p>", "<title>c. Flexible network data visual annotation</title>", "<p>In ProteoLens, users can explicitly define network data association rules. As described earlier, the IDs of the nodes in the network are used for attaching multiple attributes in the rule for subsequent visual annotations. The annotation of an edge is based on the mapping of attributes in the rule identified by two interacting node IDs for each edge. Graphical attributes currently available for automatic node annotations are: label, size, shape, and fill color. The latter annotation allows mapping multiple or multiple-valued properties (multiple colors per node object), in which case pie-chart style filling will be drawn. Graphical attributes currently available for automatic edge annotation are: line style, width, color, and text label. There are two types of mappings between attributes and visual properties: 1) the <italic>categorical mapping </italic>type that allows displaying attributes with discrete set of specific property values (for instance, using protein ID → GO molecular function association rule, and requesting all nodes annotated as a \"kinase\" or a \"phosphatase\" to be colored in red); and 2) the <italic>continuous-range mapping </italic>type that allows displaying properties with continuous numerical values (<italic>e.g</italic>. expression levels, or interaction confidence scores), using color gradients, shape sizes or line widths.</p>", "<p>The use of declarative SQL to specify how data should be managed, pre-processed, associated, and then subsequently mapped to visual properties is characteristic of ProteoLens. A user can use SQL queries to specify and store \"associations\" between nodes/edges and other attributes. These associations can be used to visually annotate large displayed networks using node/edge shape, size, weight, color and text. This gives users more choices and flexibility than any custom-built annotation user interface for complex visual network analysis.</p>", "<title>d. Sub-network manipulations</title>", "<p>Users of ProteoLens can conveniently specify sub-networks based on existing networks to conduct studies in a specific biological context. Users could specify what types of nodes or edges to include in the sub-network according to set of qualifying conditions. For example, a user may retrieve all the interactions where at least one of the partners is annotated as \"cell-cycle related\" proteins. This does not require bringing into the view a huge biomolecular network in its entirety and then filtering/zooming in onto its part. Neither this specification requires users to prepare the data outside the ProteoLens visualization software platform; instead, users could write a SQL statement to first create a relational view between protein nodes and gene ontology annotations and then to create a new network data association of \"nodes\", with which a new sub-network could be retrieved from the original network and annotated. This approach can be imagined as a huge underlying biomolecular network stored in the backend databases and/or files and integrated (logically) by the virtue of data associations. Only the relevant part of the network is physically pulled into the visualization layer for detailed examination. Such visualized sub-networks are not \"final\" either but could be gradually updated by iteratively using additional queries to bring more data from different sources into the view, until the final complex network is built. In comparison, Cytoscape does not support a network filtering feature and requires the entire network for visual data analysis to be pre-formatted properly and imported from the input data files. A Cytoscape software plug-in, Bubble Router, has became available recently; however, it allows only creating a sub-network with one-pass filters, which cannot be iteratively extended to fit the exploratory nature of visual analytic operations.</p>", "<title>Case studies</title>", "<p>To demonstrate the new functionalities of ProteoLens, we show several case studies that demonstrate how the software is used to solve real-world biological problems.</p>", "<title>Case study 1: human cancer association network</title>", "<p>Decade-long study of disease-causing genes has generated a comprehensive set of \"disease disorders – genes\" relationship pairs (also referred to as the \"diseasome\"), which are represented in the OMIM morbidity map [##REF##15608251##27##]. Goh <italic>et al</italic>. recently showed a global view of the \"human disease network\" (HDN), which included 22 disease disorder classes, 1284 disease disorders, and 1777 disease genes [##REF##17502601##23##]. In the HDN, nodes represented disease disorders, while edges represented the presence of at least one common gene between two connected disease disorders. The study of HDN showed how diseases related to one another and formed major disease clusters connected by underlying shared molecular entities. The initial construction of the network for HDN, however, was labor-intensive, since preparation of the disease-gene association file and other gene/disease annotation files needed to be processed separately with different software tools before visualizations were to be performed.</p>", "<p>In this case study, we show how to reproduce a similar human disease network with ProteoLens, using data for 13 common types of cancers derived from the OMIM database, papers by Goh <italic>et al</italic>, and public biological databases [##REF##17502601##23##,##REF##15608251##27##]. The visual analytic process can be divided into four steps, network data pre-processing, network data association rule specifications, initial network visualization, and iterative visual data analysis, all without leaving the software platform:</p>", "<title>Network data pre-processing</title>", "<p>At the beginning of the analysis, only one database table–the GENE_DISEASE_MAP table–is available. The table contains the pairing relationships between disease disorders and genes. From this table, we could define the relationship of two individual diseases as \"associated\" if and only they shared at least one common gene implicated in both diseases. The SQL to create such a specification is quite straightforward: (Figure ##FIG##2##3##)</p>", "<p>The constructed view put together three attributes. The first and second attributes represent paired cancer diseases that share at least one common disorder gene. The third attribute represent the total count of shared genes between the shared diseases. To create an annotation for cancers and total count of genes implicated in each cancer, we can write the following SQL statements inside ProteoLens: (Figure ##FIG##3##4##)</p>", "<title>Network data association rule specifications</title>", "<p>Here, we assign visual attributes of interest to either network nodes or edges, using network data association rules. In this case study, we can create the following sample association rules using ProteoLens:</p>", "<p>• <italic>Disease_gene_implicated</italic>: {Disease_Name} → {Cnt_Gene}</p>", "<p>• <italic>D2D_gene_shared</italic>: {Disease_A, Disease_B} → {Cnt_Gene_Shared}</p>", "<p>Note that network data association rules must involve mapping from either the node (identified by Disease_Name) or edge (identified by Disease_A, Disease_B) to an annotation attribute such as \"Cnt_Gene\" or \"Cnt_Gene_Shared\". The former data association rule is \"node-styled\" and the latter data association rule is \"edge-styled\", since they provide annotations (attributes) for nodes and edges, respectively.</p>", "<title>Initial network visualization</title>", "<p>The final construction of the human cancer association network in ProteoLens is now simplified. First, we lay out the basic network disease pairing information from \"D2D_INTERACTION\"; then, we apply all the node-style network data association rules to the annotation of \"nodes\" and all the edge-style network data association rules to the annotation of \"edges\". In this case study, we chose to represent the total numbers of genes implicated in a given disease as sizes of the disease nodes, and the total numbers of genes shared between two diseases as the edge widths.</p>", "<title>Iterative visual data analysis</title>", "<p>ProteoLens supports iterative visual analysis by allowing additional visual information to be captured as network nodes/edges annotations later. In this case study, after examining that breast cancer is well studied (with many mapped genes) and lung cancer is not, we decided to further incorporate first-time incident frequency during 2007 in the U.S., by retrieving relevant statistics from the American Cancer Society [##UREF##3##28##] and annotating disease names represented as network nodes with color gradients from white to red. The final view of the annotated network is shown in Figure ##FIG##4##5##. The figure reveals intriguing insights into the relations of the human cancers to each other. Interestingly, lung cancer studies are found to be under-represented for all cancers, given its large number of new incident rates; the discovery opportunities for lung cancer seem huge.</p>", "<title>Case study 2: compound-target interaction network</title>", "<p>Different from conventional network biology studies or case study #1, this case study is concerned with two different types of biological entities–chemical compounds and drug target proteins. We attempt to understand the specificity of drug compound actions and to visualize potential drug targets for major diseases.</p>", "<p>Using ProteoLens, we created drug-target network visualization, using hierarchical layout and two different node shapes to represent drugs and protein targets separately. In Figure ##FIG##3##4A##, we show a snapshot of the network, which contains all the drug compounds developed for ACM2_HUMAN and its direct interacting protein partners. ACM2_HUMAN is an acetylcholine binding receptor and a member of the G-protein-coupled receptors (GPCR) protein family–a major class of current drug targets that accounts for more than 50% of known contemporary drug compounds [##REF##12362358##29##].</p>", "<p>The entire protein-drug interaction data was downloaded from DrugBank [##REF##18048412##25##] and stored in Oracle 10 g database tables. It is interesting to note that target proteins could be visually clustered, and the clustering relationships correlate well with the evolutionary relationship defined by separate alignments of protein sequences (see Figure ##FIG##3##4B##). While a few ACM2_HUMAN interacting proteins such as SC6A1_HUMAN, SC6A2_HUMAN and ACHA2_HUMAN are not members of GPCRs, all other proteins belonged to GPCRs. ACM1_HUMAN to ACM5_HUMAN are acetylcholine receptors. As shown in Figure ##FIG##5##6##, the results suggest that proteins with similar phylogenetic profiles tend to share similar core set of drug compounds, perhaps due to similarities of underlying protein structures in the proximity of the functional site. The visualization of such drug-target network opens up new \"network pharmacology\" study opportunities, in which a drug may be evaluated for its ability to find multiple \"targets\" related to a specific biological sub-network [##REF##17921993##30##]; while the effects of drug compounds may also be evaluated in the context of common structures of all interacting target proteins. This type of visual network studies could help users develop novel perspectives for drug designs and/or protein target validations.</p>", "<title>Case study 3: peptide-protein mapping networks</title>", "<p>In this case study, we apply ProteoLens to the study of mass spectrometry (MS) based proteomics. In each tandem mass-spectrometry experiment, many partially Trypsin-digested peptides can be detected by the MS/MS spectrometers and identified by MS analysis software. The software normally aims to identify all the peptides from MS spectra and to map them unambiguously to proteins in the sample. Traditional MS analysis software identifies proteins from peptides by directly mapping them to the most common protein isoforms found in the pre-computed MS search database; therefore, incomplete results may arise, especially in cases where common peptides may be shared by two or more protein isoforms.</p>", "<p>In Figure ##FIG##6##7##, we show how ProteoLens could be used to help establish all the relationships between found peptides and possible protein isoforms that they may link to, in the HIP2 database–an online database that collects all experimentally identified proteins and peptide-mapping evidence in normal human plasma[##REF##18439290##31##]. Two proteins, A1AG1_HUMAN and A1AG2_HUMAN, are shown in the visualization network. By writing SQL inside ProteoLens, we identify all the potential protein-peptide relationships: (Figure ##FIG##7##8##).</p>", "<p>In the peptide-protein mapping network shown, the common peptides identified and mapped to either protein in the original experiments are colored green, whereas newly mapped peptides are colored yellow. We also used different node shapes to annotate peptides identified from different MS instrument types, e.g., IMS-MS instruments, LC-MS/MS instruments or MALDI-MS instruments. By visualizing the raw data in the protein-peptide network, we can see that both protein isoforms are found in human plasma, since both common peptides and protein-specific peptides are found and mapped. Interestingly, IMS-MS instruments are seen to resolve only one protein isoform \"A1AG1_HUMAN\", suggesting that either the platform could be biased towards identifying certain types of peptides or the search database used for this experiment might have not contained A1AG2_HUMAN. ProteoLens makes it easy for users to explore different hypothesis and continue scientific explorations through iterative network visual data analysis.</p>" ]

[ "<title>Results and discussion</title>", "<title>New features</title>", "<p>Figure ##FIG##1##2## illustrates the core functionalities of ProteoLens. We compared ProteoLens with several existing popular visualization software tools, including Cytoscape, VisANT and BiologicalNetworks, and summarized the comparison in Table ##TAB##0##1##. We describe the features of ProteoLens in detail in this section.</p>", "<title>a. Relational and XML data support</title>", "<p>Currently, ProteoLens supports two types of physical data sources: tab-delimited text files on the local file system and tables/views in relational tables managed by Oracle 10 g or PostgreSQL 8.x database management systems. A user can manage and query the data stored in the relational database, create network data association rules from the view, and immediately make the rule available for visual annotation. Since only meta-data are stored when a data association is created, the query execution can be performed in real time of visualization against the underlying complex data structure in the database. This design allows for an efficient data retrieval and analysis, and saving of the data file and workspace resources. Almost infinite configurations of data views can be created from multiple underlying data sources, and be used for building complex integrated visualizations. ProteoLens also supports semi-structured data format in Graph Modeling Language (GML) – the standard file format in the Graphlet graph editor system, for non-relational graphs. Network visualization is created in a view can be saved in a GML file, thus allowing for reopening and further editing in a new session, or data exchange without relational databases. The network view can be exported as a JPEG or PNG file. The user can import and manipulate any network data using standard GML file formats in addition to structured data stored in the relational databases. ProteoLens stores every data association in a session configuration XML file. Users can save the session and recommence their analysis at any time.</p>", "<title>b. SQL-based visual data analysis</title>", "<p>ProteoLens supports direct database connectivity through Java Database Connectivity (JDBC) to Oracle and PostgreSQL database tables and views, and the entire set of database operations can be specified using full SQL statements including Data Definition Languages (DDL) and Data Manipulation languages (DML). This extends the range of data that expert users may bring into later network visualizations for annotation and visual exploration tasks. Users of ProteoLens can iteratively prepare data stored in relational databases without leaving the visual analytic environment. Data from different tables in a complex relational database schema can also be queried on the fly to create networks at the appropriate level for exploration. SQL queries are also used to present \"views\" of different underlying database tables for network data associations, therefore making it possible for users to perform all visualization pre-analysis without leaving the ProteoLens platform.</p>", "<title>c. Flexible network data visual annotation</title>", "<p>In ProteoLens, users can explicitly define network data association rules. As described earlier, the IDs of the nodes in the network are used for attaching multiple attributes in the rule for subsequent visual annotations. The annotation of an edge is based on the mapping of attributes in the rule identified by two interacting node IDs for each edge. Graphical attributes currently available for automatic node annotations are: label, size, shape, and fill color. The latter annotation allows mapping multiple or multiple-valued properties (multiple colors per node object), in which case pie-chart style filling will be drawn. Graphical attributes currently available for automatic edge annotation are: line style, width, color, and text label. There are two types of mappings between attributes and visual properties: 1) the <italic>categorical mapping </italic>type that allows displaying attributes with discrete set of specific property values (for instance, using protein ID → GO molecular function association rule, and requesting all nodes annotated as a \"kinase\" or a \"phosphatase\" to be colored in red); and 2) the <italic>continuous-range mapping </italic>type that allows displaying properties with continuous numerical values (<italic>e.g</italic>. expression levels, or interaction confidence scores), using color gradients, shape sizes or line widths.</p>", "<p>The use of declarative SQL to specify how data should be managed, pre-processed, associated, and then subsequently mapped to visual properties is characteristic of ProteoLens. A user can use SQL queries to specify and store \"associations\" between nodes/edges and other attributes. These associations can be used to visually annotate large displayed networks using node/edge shape, size, weight, color and text. This gives users more choices and flexibility than any custom-built annotation user interface for complex visual network analysis.</p>", "<title>d. Sub-network manipulations</title>", "<p>Users of ProteoLens can conveniently specify sub-networks based on existing networks to conduct studies in a specific biological context. Users could specify what types of nodes or edges to include in the sub-network according to set of qualifying conditions. For example, a user may retrieve all the interactions where at least one of the partners is annotated as \"cell-cycle related\" proteins. This does not require bringing into the view a huge biomolecular network in its entirety and then filtering/zooming in onto its part. Neither this specification requires users to prepare the data outside the ProteoLens visualization software platform; instead, users could write a SQL statement to first create a relational view between protein nodes and gene ontology annotations and then to create a new network data association of \"nodes\", with which a new sub-network could be retrieved from the original network and annotated. This approach can be imagined as a huge underlying biomolecular network stored in the backend databases and/or files and integrated (logically) by the virtue of data associations. Only the relevant part of the network is physically pulled into the visualization layer for detailed examination. Such visualized sub-networks are not \"final\" either but could be gradually updated by iteratively using additional queries to bring more data from different sources into the view, until the final complex network is built. In comparison, Cytoscape does not support a network filtering feature and requires the entire network for visual data analysis to be pre-formatted properly and imported from the input data files. A Cytoscape software plug-in, Bubble Router, has became available recently; however, it allows only creating a sub-network with one-pass filters, which cannot be iteratively extended to fit the exploratory nature of visual analytic operations.</p>", "<title>Case studies</title>", "<p>To demonstrate the new functionalities of ProteoLens, we show several case studies that demonstrate how the software is used to solve real-world biological problems.</p>", "<title>Case study 1: human cancer association network</title>", "<p>Decade-long study of disease-causing genes has generated a comprehensive set of \"disease disorders – genes\" relationship pairs (also referred to as the \"diseasome\"), which are represented in the OMIM morbidity map [##REF##15608251##27##]. Goh <italic>et al</italic>. recently showed a global view of the \"human disease network\" (HDN), which included 22 disease disorder classes, 1284 disease disorders, and 1777 disease genes [##REF##17502601##23##]. In the HDN, nodes represented disease disorders, while edges represented the presence of at least one common gene between two connected disease disorders. The study of HDN showed how diseases related to one another and formed major disease clusters connected by underlying shared molecular entities. The initial construction of the network for HDN, however, was labor-intensive, since preparation of the disease-gene association file and other gene/disease annotation files needed to be processed separately with different software tools before visualizations were to be performed.</p>", "<p>In this case study, we show how to reproduce a similar human disease network with ProteoLens, using data for 13 common types of cancers derived from the OMIM database, papers by Goh <italic>et al</italic>, and public biological databases [##REF##17502601##23##,##REF##15608251##27##]. The visual analytic process can be divided into four steps, network data pre-processing, network data association rule specifications, initial network visualization, and iterative visual data analysis, all without leaving the software platform:</p>", "<title>Network data pre-processing</title>", "<p>At the beginning of the analysis, only one database table–the GENE_DISEASE_MAP table–is available. The table contains the pairing relationships between disease disorders and genes. From this table, we could define the relationship of two individual diseases as \"associated\" if and only they shared at least one common gene implicated in both diseases. The SQL to create such a specification is quite straightforward: (Figure ##FIG##2##3##)</p>", "<p>The constructed view put together three attributes. The first and second attributes represent paired cancer diseases that share at least one common disorder gene. The third attribute represent the total count of shared genes between the shared diseases. To create an annotation for cancers and total count of genes implicated in each cancer, we can write the following SQL statements inside ProteoLens: (Figure ##FIG##3##4##)</p>", "<title>Network data association rule specifications</title>", "<p>Here, we assign visual attributes of interest to either network nodes or edges, using network data association rules. In this case study, we can create the following sample association rules using ProteoLens:</p>", "<p>• <italic>Disease_gene_implicated</italic>: {Disease_Name} → {Cnt_Gene}</p>", "<p>• <italic>D2D_gene_shared</italic>: {Disease_A, Disease_B} → {Cnt_Gene_Shared}</p>", "<p>Note that network data association rules must involve mapping from either the node (identified by Disease_Name) or edge (identified by Disease_A, Disease_B) to an annotation attribute such as \"Cnt_Gene\" or \"Cnt_Gene_Shared\". The former data association rule is \"node-styled\" and the latter data association rule is \"edge-styled\", since they provide annotations (attributes) for nodes and edges, respectively.</p>", "<title>Initial network visualization</title>", "<p>The final construction of the human cancer association network in ProteoLens is now simplified. First, we lay out the basic network disease pairing information from \"D2D_INTERACTION\"; then, we apply all the node-style network data association rules to the annotation of \"nodes\" and all the edge-style network data association rules to the annotation of \"edges\". In this case study, we chose to represent the total numbers of genes implicated in a given disease as sizes of the disease nodes, and the total numbers of genes shared between two diseases as the edge widths.</p>", "<title>Iterative visual data analysis</title>", "<p>ProteoLens supports iterative visual analysis by allowing additional visual information to be captured as network nodes/edges annotations later. In this case study, after examining that breast cancer is well studied (with many mapped genes) and lung cancer is not, we decided to further incorporate first-time incident frequency during 2007 in the U.S., by retrieving relevant statistics from the American Cancer Society [##UREF##3##28##] and annotating disease names represented as network nodes with color gradients from white to red. The final view of the annotated network is shown in Figure ##FIG##4##5##. The figure reveals intriguing insights into the relations of the human cancers to each other. Interestingly, lung cancer studies are found to be under-represented for all cancers, given its large number of new incident rates; the discovery opportunities for lung cancer seem huge.</p>", "<title>Case study 2: compound-target interaction network</title>", "<p>Different from conventional network biology studies or case study #1, this case study is concerned with two different types of biological entities–chemical compounds and drug target proteins. We attempt to understand the specificity of drug compound actions and to visualize potential drug targets for major diseases.</p>", "<p>Using ProteoLens, we created drug-target network visualization, using hierarchical layout and two different node shapes to represent drugs and protein targets separately. In Figure ##FIG##3##4A##, we show a snapshot of the network, which contains all the drug compounds developed for ACM2_HUMAN and its direct interacting protein partners. ACM2_HUMAN is an acetylcholine binding receptor and a member of the G-protein-coupled receptors (GPCR) protein family–a major class of current drug targets that accounts for more than 50% of known contemporary drug compounds [##REF##12362358##29##].</p>", "<p>The entire protein-drug interaction data was downloaded from DrugBank [##REF##18048412##25##] and stored in Oracle 10 g database tables. It is interesting to note that target proteins could be visually clustered, and the clustering relationships correlate well with the evolutionary relationship defined by separate alignments of protein sequences (see Figure ##FIG##3##4B##). While a few ACM2_HUMAN interacting proteins such as SC6A1_HUMAN, SC6A2_HUMAN and ACHA2_HUMAN are not members of GPCRs, all other proteins belonged to GPCRs. ACM1_HUMAN to ACM5_HUMAN are acetylcholine receptors. As shown in Figure ##FIG##5##6##, the results suggest that proteins with similar phylogenetic profiles tend to share similar core set of drug compounds, perhaps due to similarities of underlying protein structures in the proximity of the functional site. The visualization of such drug-target network opens up new \"network pharmacology\" study opportunities, in which a drug may be evaluated for its ability to find multiple \"targets\" related to a specific biological sub-network [##REF##17921993##30##]; while the effects of drug compounds may also be evaluated in the context of common structures of all interacting target proteins. This type of visual network studies could help users develop novel perspectives for drug designs and/or protein target validations.</p>", "<title>Case study 3: peptide-protein mapping networks</title>", "<p>In this case study, we apply ProteoLens to the study of mass spectrometry (MS) based proteomics. In each tandem mass-spectrometry experiment, many partially Trypsin-digested peptides can be detected by the MS/MS spectrometers and identified by MS analysis software. The software normally aims to identify all the peptides from MS spectra and to map them unambiguously to proteins in the sample. Traditional MS analysis software identifies proteins from peptides by directly mapping them to the most common protein isoforms found in the pre-computed MS search database; therefore, incomplete results may arise, especially in cases where common peptides may be shared by two or more protein isoforms.</p>", "<p>In Figure ##FIG##6##7##, we show how ProteoLens could be used to help establish all the relationships between found peptides and possible protein isoforms that they may link to, in the HIP2 database–an online database that collects all experimentally identified proteins and peptide-mapping evidence in normal human plasma[##REF##18439290##31##]. Two proteins, A1AG1_HUMAN and A1AG2_HUMAN, are shown in the visualization network. By writing SQL inside ProteoLens, we identify all the potential protein-peptide relationships: (Figure ##FIG##7##8##).</p>", "<p>In the peptide-protein mapping network shown, the common peptides identified and mapped to either protein in the original experiments are colored green, whereas newly mapped peptides are colored yellow. We also used different node shapes to annotate peptides identified from different MS instrument types, e.g., IMS-MS instruments, LC-MS/MS instruments or MALDI-MS instruments. By visualizing the raw data in the protein-peptide network, we can see that both protein isoforms are found in human plasma, since both common peptides and protein-specific peptides are found and mapped. Interestingly, IMS-MS instruments are seen to resolve only one protein isoform \"A1AG1_HUMAN\", suggesting that either the platform could be biased towards identifying certain types of peptides or the search database used for this experiment might have not contained A1AG2_HUMAN. ProteoLens makes it easy for users to explore different hypothesis and continue scientific explorations through iterative network visual data analysis.</p>" ]

[ "<title>Conclusion</title>", "<p>We developed ProteoLens as a multi-scale network visual analytic software tool for advanced network biology studies. It is built on robust software architecture that supports flexible network data association specification using rules, integrates data processing through relational databases and GML data files, and scalable data visualization through layered annotations. It is intended for advanced bioinformatics users who manage large existing sets of biological data in the Oracle or PostgreSQL databases, and who are skilled in SQL programming. ProteoLens is by far the first bio-molecular network visualization software with full SQL support. ProteoLens enables iterative visual layout, annotation and exploration of bio-molecular networks. It effectively liberates advanced data analysts from the burden of data preparation and processing prior to generating visualization, and thus helps to better concentrate on the scientific visualization itself. The support for both \"network browsing\" and \"network querying\" operations makes ProteoLens a promising visual data analytic and visual data mining tool for hypothesis-driven network biology studies. With future releases of ProteoLens, we plan to add open Application Program Interfaces (APIs) so that 1) Proteolens can interoperate with other software tools in bioinformatics, and 2) third-party plug-ins could be developed to accommodate expanding user community needs.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>New systems biology studies require researchers to understand how interplay among myriads of biomolecular entities is orchestrated in order to achieve high-level cellular and physiological functions. Many software tools have been developed in the past decade to help researchers visually navigate large networks of biomolecular interactions with built-in template-based query capabilities. To further advance researchers' ability to interrogate global physiological states of cells through multi-scale visual network explorations, new visualization software tools still need to be developed to empower the analysis. A robust visual data analysis platform driven by database management systems to perform bi-directional data processing-to-visualizations with declarative querying capabilities is needed.</p>", "<title>Results</title>", "<p>We developed ProteoLens as a JAVA-based visual analytic software tool for creating, annotating and exploring multi-scale biological networks. It supports direct database connectivity to either Oracle or PostgreSQL database tables/views, on which SQL statements using both Data Definition Languages (DDL) and Data Manipulation languages (DML) may be specified. The robust query languages embedded directly within the visualization software help users to bring their network data into a visualization context for annotation and exploration. ProteoLens supports graph/network represented data in standard Graph Modeling Language (GML) formats, and this enables interoperation with a wide range of other visual layout tools. The architectural design of ProteoLens enables the de-coupling of complex network data visualization tasks into two distinct phases: 1) creating network data association rules, which are mapping rules between network node IDs or edge IDs and data attributes such as functional annotations, expression levels, scores, synonyms, descriptions etc; 2) applying network data association rules to build the network and perform the visual annotation of graph nodes and edges according to associated data values. We demonstrated the advantages of these new capabilities through three biological network visualization case studies: human disease association network, drug-target interaction network and protein-peptide mapping network.</p>", "<title>Conclusion</title>", "<p>The architectural design of ProteoLens makes it suitable for bioinformatics expert data analysts who are experienced with relational database management to perform large-scale integrated network visual explorations. ProteoLens is a promising visual analytic platform that will facilitate knowledge discoveries in future network and systems biology studies.</p>" ]

[ "<title>Implementation</title>", "<p>ProteoLens is a standalone software tool written in Java programming language. Its software architecture consists of two separate functional layers – a data processing layer at the backend and a data visualization layer at the frontend–connected by a network data association engine (Figure ##FIG##0##1##). Different from conventional network visualization software, where data preparation prior to visualizations is usually not supported by the software itself, the ProteoLens data processing layer allows network data to be pre-processed and integrated using built-in database management utilities. The data visualization layer at the frontend enables users to iteratively build and layout query-based (sub)networks and overlay them with visually displayed annotations as additional data sets are brought in. The network data association engine bridges the two functional layers by creating network data association rules (to be described next) between pre-processed data elements and identifiers of network nodes or edges. This design enables users to navigate between data management and data visualization iteratively until useful insights from the proper visualization are established.</p>", "<title>Network data association rules: the concept</title>", "<p>Network data association rules represent a basic concept in ProteoLens design. A network data association rule defines a relationship between a network data attribute such as an edge or a node and other non-network data attributes such as a computed score or an expression value. Such an association rule establishes the mapping between data in the data processing layer and data in the data visualization layer. There are two types of association rules:</p>", "<p>1) <italic>Graph Node Association Rules</italic>. For example, Rule <italic>X</italic>: {<underline>Protein ID</underline>} → {Protein Name}defines the network data attribute \"Protein ID\" as an identifying attribute for a network node further annotated with a node attribute \"Protein Name\". Note that the network attribute \"Protein ID\" and associated non-network attribute \"Protein Name\" may not necessarily be stored in the same database table and may be defined using a complex SQL query for visual data mapping purposes only.</p>", "<p>2) <italic>Graph Edge Association Rules</italic>. For example, Rule <italic>X</italic>: {<underline>Protein ID A, Protein ID B</underline>} → {} defines the combination of two network data attributes, \"Protein ID A\" and \"Protein ID B\", as identifying attributes for a network edge without further annotations; whereas Rule <italic>Y</italic>: {<underline>Protein ID A, Protein ID B</underline>} → {Interaction Score} defines the combination of two network data attributes, \"Protein ID A\" and \"Protein ID B\", as identifying attributes for a network edge further annotated with an edge attribute \"Interaction Score\". Similar to graph node association rules, the network attributes and non-network attributes may come from different physical data structures.</p>", "<title>Data processing layer</title>", "<p>The data processing layer is the place where biological data from different sources, including flat files, XML data and tabular data in relational databases, can be managed and converted from one format into another for subsequent analysis. In ProteoLens, users could specify the sources of data, pre-process data and make certain subsets of data available to the subsequent data analysis. Unlike conventional visualization software tools, ProteoLens supports full Structured Query Language (SQL)–including both Data Definition Languages (DDL) and Data Manipulation languages (DML)–for these tasks. The combination of DDL and DML is particularly powerful for network biological studies, since many network data association rules may require selected data sets (via DML) and nested definition of complex data structures (via DDL) by pulling data from many physical table locations. In ProteoLens, the data processing layer is implemented with the combination of GML data handler and Oracle 11 g/PostgreSQL relational database engines.</p>", "<title>Data visualization layer</title>", "<p>The data visualization layer is the place where specified network data attributes and data association rules are converted to network layouts and network visual properties. The data visualization layer accepts network data association rules, lays out the drawing of networks as graphs, and visualizes network nodes and edges using graphical attributes defined in the network data association rules. In ProteoLens, the data visualization layer is implemented with a fully functional graph editor, which supports laying out the nodes and edges in the network and editing their graphical attributes such as colors and shapes with rules defined in network data association rules. ProteoLens supports multiple independent network views. In a network view, each associated attribute specified by the association rule can be added either as a node attribute or an edge attribute, depending on the association rule type. Any numbers of associations can be selected as annotation sources to modify the appearance of network nodes and edges. In ProteoLens, the graph layout is extended from yWorks Java package 3.0, a commercially available graph layout library.</p>", "<title>Availability and requirements</title>", "<p><bold>Project name: </bold>ProteoLens</p>", "<p>• <bold>Project home page: </bold><ext-link ext-link-type=\"uri\" xlink:href=\"http://bio.informatics.iupui.edu/proteolens/\"/></p>", "<p>• <bold>Operating system(s): </bold>The software is platform independent and can run anywhere Java Virtual Machine runtime is available. An installer is provided for Windows NT/XP/2003/Vista users.</p>", "<p>• <bold>Programming language: </bold>Java</p>", "<p>• <bold>Other requirements: </bold>Java Runtime Environment (JRE) version 1.5 or above is required.</p>", "<p>• <bold>License: </bold>free software license to all users.</p>", "<p>• <bold>Any restrictions to use by non-academics: </bold>Non-academic users can freely use the software for research purposes. Non-academic users cannot redistribute, modify, reverse-engineer, or resell the software for commercial purposes.</p>", "<title>List of abbreviations used</title>", "<p><bold>GINY</bold>- Graph Interface library; <bold>SQL </bold>-Structural query language; <bold>PSI-MI </bold>– Proteomics Standards Initiative – Molecular Interactions.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>JYC is the principal architect and custodian of the software. JYC and AS conceptualized the multi-scale biological network visualization ideas together, designed computing architecture, and implemented the software in 2003–2006. TH and SHH took over the software improvement and maintenance tasks since then by rewriting significant portions of the software, developing new documentations, testing it on different platforms, developing a set of case studies. TH outlined the paper and all authors have read and approved the final version of the manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We thank Indiana University – Purdue University Indianapolis (IUPUI) RSFG fund, IUPUI Signature Center Fund, and China scholarship council for providing partial funding support for this work. We thank Josh Hayen, Yan Zhong, Dr. Mark Goebl, and Dr. Zengliang Bai for providing invaluable user feedbacks during extensive testing and development of the software since 2004. We thank members of the discovery informatics and computing laboratory and Indiana Center for Systems Biology and Personalized Medicine for providing computational support and timely discussions that led to the completion of the project.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>An overview of the ProteoLens core architecture</bold>. The design of ProteoLens decoupled the data processing and visualization presenting in two layers and the two layers communicated by the abstract Data Associations. The major components of ProteoLens are SQL data retrieving engine, network layout engine and graph attributes editing engine.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>The overview core functionalities of ProteoLens</bold>. Some of the core functionalities of ProteoLens labelled in this figure: a) ProteoLens can access both the relational database and local file system, b) the SQL statement can be edited and run in the software environment for data association building, c) SQL-like for building the sub network by retrieving particular characters of nodes/edges, d) convenient quick query of the nodes in the network view and sub-network retrieving, and e) flexible and comprehensive annotation adding.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>SQL statement for identifying disease-disease association</bold>. The SQL statement created a view recording the relationship of two individual diseases as \"associated\" if they shared at least one common disorder gene.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>SQL statement for counting the genes involving in every disease.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>Disease-disease association network</bold>. This is a sub network of the cancer disease association network, built by retrieving 13 kinds of popular cancer. In this representation, the node is a kind of cancer, and if two kinds of cancer have common genetic disorder genes, there is an edge connecting them. The size of nodes indicates the number of cancerogenic disorder genes and the color of nodes indicates the number of cases in 2007 in the U.S; dark red indicates more cases, light red indicates less number, and white indicates less statistic data. The width of edge indicates the number of common genetic disorder genes of two kinds of cancer disease.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p><bold>Compound- protein target interaction network</bold>. a) The compound-protein target interaction network is drawn by the ProteoLens hierarchical layout. b) The evolutionary tree of all the target proteins is drawn by ClustalW2 <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ebi.ac.uk/Tools/clustalw2/index.html\"/>. For every protein sub family marked with a color, the node color in A corresponds to the color bar in B.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p><bold>MS proteomic peptide-protein mapping network</bold>. The blue color marking nodes are the original common peptides of the two proteins, and the yellow ones are newly discovered common peptides. The peptides' nodes are marked in three kinds of shape indicating different MS experimental platforms.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>SQL statement for identifying the protein-peptide relationship.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Compare ProteoLens against Cytoscape, VisANT and BiologicalNetworks.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"left\"><bold>ProteoLens</bold></td><td align=\"left\"><bold>Cytoscape</bold></td><td align=\"left\"><bold>VisANT</bold></td><td align=\"left\"><bold>BiologicalNetworks</bold></td></tr></thead><tbody><tr><td align=\"left\"><bold>Graph manipulation</bold></td><td align=\"left\">yFiles</td><td align=\"left\">yFiles and GINY</td><td align=\"left\">In house</td><td align=\"left\">Cytoscape</td></tr><tr><td align=\"left\"><bold>Laying out Network Algorithm</bold></td><td align=\"left\">force-directed, Radial Layout hierarchical, circular, orthogonal</td><td align=\"left\">More than 13 kinds of layout styles.</td><td align=\"left\">force-directed</td><td align=\"left\">grid, circular, force-directed</td></tr><tr><td align=\"left\"><bold>Drawing appearance</bold></td><td align=\"left\">node shape/colour/border/label, edge colour/style/direction/label</td><td align=\"left\">node shape/colour/border/label, edge colour/style/direction/label</td><td align=\"left\">node shape/colour/size</td><td align=\"left\">node shape/colour/border/label, edge colour/style/direction/label</td></tr><tr><td align=\"left\"><bold>Filters</bold></td><td align=\"left\">select nodes/links according to properties or using SQL statement for table attributes selecting directly</td><td align=\"left\">select nodes/links according to properties (SQL-like)</td><td align=\"left\">Several 'select' filters available</td><td align=\"left\">select nodes/links according to properties (SQL-like)</td></tr><tr><td align=\"left\"><bold>Expand/Collapse nodes</bold></td><td align=\"left\">expand node neighbours</td><td align=\"left\">Plug-in</td><td align=\"left\">expand node neighbours</td><td/></tr><tr><td align=\"left\"><bold>Database Incline</bold></td><td align=\"left\">Common Relational database</td><td align=\"left\">Plug-in</td><td align=\"left\">Predictome</td><td align=\"left\">PathSys</td></tr><tr><td align=\"left\"><bold>System Requirements</bold></td><td align=\"left\">Java stand-alone</td><td align=\"left\">Java applet or stand-alone</td><td align=\"left\">Java applet</td><td align=\"left\">JSP (Java Server Pages)</td></tr><tr><td align=\"left\"><bold>Save</bold></td><td align=\"left\">GML,XML session</td><td align=\"left\">GML,SIF</td><td align=\"left\">network with layout</td><td align=\"left\">save all work as projects</td></tr><tr><td align=\"left\"><bold>Imports</bold></td><td align=\"left\">Text, GML, XML, Oracle or PostgreSQL</td><td align=\"left\">text, GML, expression matrix, OBO</td><td align=\"left\">PSI-MI, BioPAX, KGML, network relations (text)</td><td align=\"left\">microarray data (Stanford,Affymetrix,TIGR,GenePix), SBML, SIF, PSI-MI, BioPAX</td></tr><tr><td align=\"left\"><bold>Exports</bold></td><td align=\"left\">JPEG,BMP, GML network relations, Node lists, selections node lists (text)</td><td align=\"left\">graphical file, SVG, GML, network relations (text)</td><td align=\"left\">PSI-MI, BioPAX, SVG, JPEG, network relations (text)</td><td align=\"left\">GIF, JPEG, SWF, PDF, PNG, PostScript, RAW, SVG, BMP</td></tr><tr><td align=\"left\"><bold>Comments &amp; other features</bold></td><td align=\"left\">Embedding the SQL query make its software more flexible to suit powerful bioinformatics experts usage</td><td align=\"left\">The importance of Cytoscape is its solid support for plug-in, growing number of which is available.</td><td align=\"left\">Statistics ability for topological characteristic analysis and integrating several biological database</td><td align=\"left\">Integrated visualization and analysis of expression data.</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>*A summary of attributes of Cytoscape, VisANT and BiologicalNetworks as presented in detail by Matthew Suderman <italic>et al </italic>in review. [##REF##17720984##26##]</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1471-2105-9-S9-S5-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S5-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S5-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S5-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S5-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S5-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S5-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S5-8\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The regulation of gene expression in the cell is controlled by regulatory proteins and functional RNAs that interact specifically with binding locations on the DNA. In the past century, molecular biologists have developed many innovative techniques to identify sites on the DNA that are bound and regulated by functional RNA and proteins. As the number of sequenced organisms increases, traditional techniques such as gel shift assays in combination with DNAse foot-printing assays can not keep pace with the explosion of available sequences. While accurate, traditional methods require a great deal of time and expertise. Traditional methods also require that the binding energy of the protein or RNA is great enough to maintain contact though the course of the assay. Reductionist approaches can be problematic when faced with non-specific binding sites or with sites that require the formation of a complex with adjacent units in order to function. It is for these reasons that complementary approaches on the computer have been developed in recent years to increase success in discovering and annotating cis-regulatory modules.</p>", "<p>Computational discovery of functional subsequences has been around for some time. In 1984, Galas, Eggert, and Waterman proposed a method for finding and characterizing binding positions for the σ 70 protein in <italic>E. coli </italic>promoter sequences [##REF##3908689##1##]. In its most basic form, the problem is identifying sequence signatures, or motifs, that exist in a set of sequences that share the same property (e.g. promoters bound by protein <italic>A</italic>), but does not exist in a set of similar sequences that does not have the same property (e.g. promoters not bound by protein <italic>A</italic>). Computational approaches have the advantage that they can reduce the guess work and cost associated with pure biochemical approaches. Consequently, many computational methods such as Gibbs [##REF##8211139##2##,##REF##12824370##3##], MEME [##REF##7584402##4##], Consensus [##REF##10487864##5##,##REF##2193692##6##], BioProspector [##REF##11262934##7##], AlignAce [##REF##10698627##8##], Ann-Spec [##REF##10902194##9##], MTAP: The Motif Tool Assessment Platform.</p>", "<p>Glam [##REF##14704356##10##], and Weeder [##REF##11473011##11##,##UREF##0##12##] have been developed in recent years. Each of these methods employs different algorithmic insights and scoring functions. The scoring function and algorithm parameters impact the representation and rank of the motifs discovered by the algorithm. In recent years, it has become clear that each of these computational methods has distinct advantages and that the best performance is achieved by trying multiple methods over the same data set. The bioinformatics community is currently seeking to simplify this situation by providing approaches that integrate multiple methods into one tool such as BEST [##REF##15814553##13##] and EMD [##REF##16839417##14##]. Approaches that integrate additional sources of information have also emerged in recent years. For example, approaches such as PhyME [##UREF##1##15##], PhyloGibbs [##REF##16477324##16##], and WeederH [##REF##17286865##17##] integrate sequences from regulatory regions of related organisms. Other approaches, such as REDUCE [##REF##12824350##18##], integrate expression values from gene expression arrays. In addition, the advent of high density arrays and ChIP-chip technology has necessitated methods that integrate genome wide binding data [##REF##15759653##19##].</p>", "<p>While there are many advantages in integrating additional sources of information, the steps required also add additional complexity and cost. Each step along the integration pipeline opens a new question: what is the best way to represent and integrate this new data? One has to wonder if additional data is always better in practice. It could be that for some regulatory modules, the additional data also includes additional noise making it more difficult to recover the binding sites. To make matters more confounding, there currently exist more than 150 methods with thousands of possible pre- and post-processing steps and alternative runtime procedures (we have compiled a list of many popular methods here: <ext-link ext-link-type=\"uri\" xlink:href=\"http://biobase.ist.unomaha.edu\"/>). It is clear that benchmarking technology is needed to map cis-regulatory motif discovery methods to data where the method has the best performance. In other words, given a method <italic>M </italic>and a set of co-regulated genes with regulatory sequences <italic>T </italic>= {<italic>t</italic>₁, <italic>t</italic>₂,..., <italic>t</italic>_{<italic>n</italic>}}, find a mapping <italic>M </italic>→ <italic>T </italic>with expected performance over some threshold. A <italic>T </italic>that has no method over the threshold is particularly interesting in that such data can tell us more about the current limitations of regulatory motif discovery programs so that we can propose improvements.</p>", "<p>In recent years, researchers have begun to solve this problem by creating benchmarking datasets. In 2004, Tompa <italic>et al. </italic>published the first assessment of 13 regulatory motif discovery algorithms over Fly, Human, Mouse and Yeast [##REF##15637633##20##]. This work was seminal in that it provided methods for comparing regulatory motif detection software and that it benchmarked a large number of popular tools [##REF##15814553##13##]. In 2007, Sandve <italic>et al. </italic>improved benchmarking technology over the Tompa dataset using a machine learning approach [##REF##17559676##21##]. While these advances opened the important debate on how to benchmark algorithm performance, it remains to be seen if a selection of so few regulatory binding sites is large enough to form a representative set (The assessment included 8 sites from <italic>Saccharomyces cerevisiae</italic>, 6 sites from <italic>Drosophila melanogaster</italic>, 12 sites from <italic>Mus musculus</italic>, and 26 sites from <italic>Homo sapiens</italic>). In response, Klepper <italic>et al. </italic>proposed a larger test set and used it to evaluate several composite motif discovery tools [##REF##18302777##22##]. This is an important first step in making benchmarking datasets more comprehensive. To date, the most comprehensive test set was over the entire <italic>E. coli </italic>genome [##REF##16284194##23##] using known transcription factors found in RegulonDB [##REF##14681419##24##]. While bacterial transcription regulation is very different than regulation in eukaryotes, the density of the RegulonDB annotation is greater than for any other organism and this coverage provides us with unique benchmarking opportunities. Unfortunately, this assessment only considered 5 tools and neglected to include the tool most successful in the Tompa Benchmark (i.e. Weeder). To date, there are no tools that can assess algorithms that incorporate additional sources of data such as sequences from phylogentically related species.</p>", "<p>Currently, cis-regulatory motif tools are assessed using the statistics and methods from gene prediction. However, <italic>in silico </italic>gene annotation differs from cis-regulatory module annotation in that in gene prediction there exists an mRNA library for reference. Unlike regulatory binding assays, mRNA library sequencing is high throughput and thus provides a large and diverse benchmarking dataset for gene prediction tools. Consequently, gene prediction and cis-regulatory module prediction are very different problems. We currently have very few completely annotated regulatory modules and therefore motif prediction tools have very few cases to train. In addition, regulatory modules are thought to evolve at a much faster rate than the genes they regulate [##UREF##2##25##]. If this theory is true, the diversity of regulatory regions is expected to be far greater than the coding regions they regulate. This means that the parameters, data, transcription module representation and algorithm are all important factors in evaluating motif detection tools.</p>", "<p>In this work, we expand upon our parallel architecture for regulatory motif prediction [##UREF##3##26##] and propose a method for evaluation and discovery of data algorithm mappings. MTAP is a platform that allows one to vary the way scientists process data and algorithm parameters to fine tune the representation of a regulatory module in method <italic>M</italic>. The goal is to discover the best possible mapping of <italic>M </italic>→ <italic>D</italic>. Our platform also integrates phylogentically related regulatory sequences via computing downstream orthologs from other species. It is clear that phylogentic footprinting has great potential, however assessing methods that incorporate sequence from closely related species is not practical or accurate without automation (mainly because of the rapid availability of new organisms: as new organism sequences become available, benchmarks need to change to take into account the new information). For these reasons, there is a place for adaptive and automatic cis-regulatory motif prediction and benchmarking.</p>" ]

[]

[ "<title>Results</title>", "<p>In this section we will provide illustrative examples of how our benchmarking technology can be used to evaluate several important parameters in cis-regulatory motif discovery. To construct a series of tests for evaluation, we extracted 2247 known transcription factor binding positions from RegulonDB [##REF##14681419##24##] corresponding to known positions from <italic>Escherichia coli K12</italic>, and 680 known motifs from DBTBS [##REF##11125112##27##] corresponding to positions from <italic>Bacillus subtilis</italic>. Then, we extracted 522 positions from Eukaryote dataset developed by Tompa <italic>et al.</italic>[##REF##15637633##20##]. Because RegulonDB has the most comprehensive coverage, we used RegulonDB to assess the impact of including two additional genomes (<italic>E. coli </italic>strain W3110, and <italic>E. coli strain </italic>UTI89 in addition to <italic>E. coli </italic>K12). We used our approach to include regulatory regions from these strains in evaluating PhyME and PhyloGibbs. Our primary goal is to show how MTAP can be used to evaluate each of the central questions in cis-regulatory motif discovery. To illustrate the capabilities of MTAP we will use it to evaluate the impact of four of the key factors for regulatory motif discovery introduced earlier in this paper, mainly: (1) the length of the sequences in the positive and negative sets, (2) the number of sequences in the positive and negative sets, (3) the distribution of transcription factor binding sites in the positive set, (4) the relative entropy of the transcription factor binding motif. Through these illustrative examples we intend to show the exploratory power of MTAP towards discovering <italic>M </italic>→ <italic>T </italic>mappings.</p>", "<title>Benchmark automation</title>", "<p>Our first goal was to illustrate that our system of automation can provide similar results to manual runs. To do this, we downloaded the benchmark results from the Tompa assessment and compared these results to results obtained from our pipelines for AlignACE, Ann-Spec, Glam, MEME, and Weeder. Results for nSn, nSp, and sSn for our platform versus the Tompa benchmarks are shown in Figure ##FIG##5##6##. Overall, sensitivity over our dataset is higher, and specificity suffers slightly. There are a few reasons for this. First, occasionally experts in the Tompa assessment pick a TFBS that is not the highest scoring motif. We think they do this because of their experience with known TFBS in Transfac. Also, MTAP allows the top <italic>c </italic>(three in this case) predictions to be scored as suggested as an improvement by Tompa to increase sensitivity. The original assessment only allowed the top prediction to be scored. In many cases, high specificity is obtained by the tool not making a prediction.</p>", "<p>Consequently, we feel pipelines that increase sensitivity at a low cost to specificity provide a good trade-off. Overall performance is similar over this dataset. This provides evidence that our automation pipelines work well relative to manual runs by experts. However, there are still many things better understood by the experts and we continue to refine our pipelines as more information becomes available.</p>", "<title>Automated assessment</title>", "<p>We next used MTAP to produce a benchmark over the sites annotated in RegulonDB. We chose to run MTAP in 'cr' mode over upstream sequences of 400 bp (shown in Figure ##FIG##6##7##). Overall specificity over this dataset is quite high. This data indicates that tools such as MEME and Weeder achieve higher sensitivity (at both the site level and nucleotide level) without substantial losses to specificity on <italic>E. coli </italic>TFBS. Summed over all TFBS, nCC ranged from -0.03 for PhyloGibbs to 0.06 for MEME. ELPH and Ann-Spec showed the least correlation in this test with a nCC value of 0.01 each. Overall correlation is extremely weak for any tool in the test.</p>", "<p>Positive predictive value ranged from 0.13 (PhyME) to 0.28 (Glam) at the nucleotide level and 0.1 (PhyME) to 0.35 (Glam) at the site level. The low values for nPPV and sPPV for PhyME can be attributed to the large number of sites that PhyME did not predict any binding positions. This behaviour is most likely explained by the large amount of sequence conservation found between the upstream regions in these different strains of <italic>E. coli</italic>. It is likely that the multiple sequence alignment step employed by PhyME did not encounter enough sequence divergence in this test set to distinguish between regulatory binding positions and background sequence conservation. The PhyloGibbs algorithm did not appear to encounter the same difficulties. However, PhyloGibbs did not appear to gain a substantial performance gain over Weeder even though it had regulatory sequences from related strains and Weeder did not. AlignACE and Ann-Spec differ from the other programs in that they sacrifice specificity slightly for increased sensitivity. Over many of the regulatory regions both AlignACE and Ann-Spec provided correct predictions somewhere in the list of predicted sites when other tools did not.</p>", "<p>Figure ##FIG##6##7## provides additional evidence that implementation and user parameters are not as important as the algorithm approach and discrimination function. In this graph, the two programs (ELPH and Gibbs) that use weight matrices for motif discrimination and gibbs sampling as the algorithmic optimization procedure have almost identical performance profiles despite the fact that the parameters provided to each of these tools are quite different in our pipelines.</p>", "<p>In the original Tompa Assessment, Weeder had more discrimination power than other approaches. While still quite good, Weeder does not appear to have the same advantages in this test. We feel that this gives further evidence that the organism and type of regulatory mechanism greatly impact the expected performance of a tool.</p>", "<title>Number of upstream sequences</title>", "<p>As the upstream size increases relative to the size of the transcription factor binding sites, the background signals found in the dataset also increase. This makes discrimination of true transcription factor binding positions more default as the number of regulatory regions and length of each region increases. We wanted to explore the relationship over known transcription factor binding sites. To discover the relationship between |<italic>T</italic>_{<italic>l</italic>, <italic>i</italic>}| and |Σ_{<italic>i </italic>}<italic>T</italic>_{<italic>l</italic>, <italic>i</italic>}| we set |<italic>T</italic>_{<italic>l</italic>, <italic>i</italic>}| = 400 <italic>bp </italic>and ran pipelines for AlignACE, AnnSpec, Elph, Glam, Gibbs, MEME, PhyME, Phylogibbs, and Weeder (the notation |<italic>x</italic>| means the sequence length of <italic>x</italic>). Figure ##FIG##7##8## shows nCC versus |Σ_{<italic>i </italic>}<italic>T</italic>_{<italic>l</italic>, <italic>i</italic>}|. As the number of sequences increases, we expect the absolute value of nCC for a tool to increase once the number of co-regulated sequences containing the same signal surpasses some threshold. Once the signal is detected and we continue to increase the number of sequences in the upstream dataset, the absolute value of nCC should decrease as the 'noise' introduced for each added sequence far exceeds the signals. If this is the case for the regulatory regions in <italic>E. coli </italic>K12, then the data indicates that 3 co-regulated sequences provides enough signal to be detected by many of the tools tested in this assessment. While nCC is quite low; the performance over this test set does not indicate that regulatory binding sites are more easily detected if we have more instances of them (as would be suggested by statistical learning theory; e.g. if we have more recorded instances of a phrase uttered by more people an HMM can detect the phrase more easily). It could be that global regulators that have more binding positions over the genome also have more variability in their binding sites. This makes sense if one considers that each instance of a global regulatory binding site must have a different binding energy to control each of the many genes regulated at different rates. The regulatory binding positions with the most occurrences indicate a higher nCC averaged over the motif detection tools. This could be explained by a superior ability of the algorithms to recognize transcription factor binding sites once the number of binding instances is large relative to the length of <italic>G</italic>_{<italic>j</italic>}.</p>", "<p>Figures ##FIG##8##9## and ##FIG##9##10## show the site level sensitivity and nucleotide specificity for AlignACE, AnnSpec, Elph, Glam, Gibbs, MEME, PhyME, Phylogibbs, and Weeder with |<italic>t</italic>_{<italic>k</italic>, <italic>i</italic>}| = 400 <italic>bp</italic>. Overall, specificity of these tools maintains a consistent level or increases as the number of sequences in <italic>T</italic>_{<italic>k </italic>}increase. The inverse is true for sensitivity. As the number of sequences increase, increased instances of regulatory signal does not lead to increased tool sensitivity. This data indicates that as the number of regulatory signals in the foreground increases linearly, the background 'noise' increases quadratically. High specificity is most likely the result of increased reluctance on the part of tools to make predictions as the number of sequences increases. As the number of sequences increases, the number of co-occuring motif instances also increases. This makes it more likely that multiple occurrences of motif for a related transcription factor may occur in the same upstream set. This motif cross-talk may play a significant roll in defeating current detection methods.</p>", "<title>Length of upstream sequences</title>", "<p>To further explore the impact of the size of <italic>T</italic>_{<italic>k</italic>}, we used MTAP to extract 500 bp and 200 bp upstream regions from motifs found in RegulonDB and ran pipelines for each of the tools. Most all of the tools did not show any significant correlation between size of <italic>T</italic>_{<italic>k </italic>}and prediction performance (data not shown). We believe that this indicates a problem with these classification algorithms over the datasets and not to variation in the size of <italic>T</italic>_{<italic>k</italic>}. It could be that these algorithms only classify certain subclasses of regulatory binding profiles accurately. Here we present the results from Weeder to demonstrate the impact of varying length of the upstream file. For Weeder, there is a slight impact on performance if the size of <italic>T</italic>_{<italic>k </italic>}is varied. To demonstrate this point, we calculated sensitivity and specificity over the 10 largest and smallest upstream files at 200 bp and 500 bp, respectively (Tables ##TAB##2##3## and ##TAB##3##4##). This data shows that Weeder nucleotide specificity is not greatly impaired by the size of the dataset in these tests. However, we do see a marked decrease in sensitivity both at the nucleotide and site level given larger datasets. The largest dataset has an average sSp of 0.41 while the smallest dataset has a average sSp 0.58 – a substantial difference. While nSn increases as a trend from smaller to larger tests, the predicted window size is on average much smaller than the motif size resulting in many missed predicted nucleotides. Weeder predicts individual transcription factor binding locations can be detected fairly well up to 33066 bp over this dataset. Increasing the dataset size further precipitates a steady drop in sensitivity until predictions are no longer useful.</p>", "<p>To further understand the impact of upstream length on motif detection performance. For each tool we ran MTAP and generated ROC graphs for lengths 20 bp, 50 bp, 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, and 800 bp upstream of the gene for DBTBS and RegulonDB. To understand the roll of data generation methods, we generated both completely-realistic ('cr') and semi-realistic ('sr') data. Here we provide the results for ANN-Spec in Figure ##FIG##10##11## which is illustrative of these results. The most important characteristic of note is between the performance curves of DBTBS (Figure ##FIG##10##11a## and ##FIG##10##11b##) and RegulonDB (Figure ##FIG##10##11c## and ##FIG##10##11d##). Figure ##FIG##10##11c## and ##FIG##10##11d## are more smooth than Figures ##FIG##10##11a## and ##FIG##10##11b##. This is because the number of sites in the regulonDB dataset is much greater than the number of sites annotated in DBTBS.</p>", "<p>Commonly, researchers would like to know what motif discovery program is best suited to a particular organism. These findings suggest that this question can not be addressed currently because of the different amounts of coverage found in each dataset. If the coverage of TFBS over the genome were greater in DBTBS, the curves in Figure ##FIG##10##11## would show an accurate comparison of the sensitivity-specificity tradeoff in running the Ann-Spec pipeline on each organism. Figure ##FIG##10##11a## and ##FIG##10##11c## refer to semi-realistic data generation (the known binding site is in the middle of the upstream sequence). As the window size increases, there is a precipitous drop in Ann-Spec's ability to correctly recover the site. High nSn and nSp are expected at 20 bp 'sr' as most any prediction will overlap the true TFBS. As the window size increases, we expect the performance to remain the same for tools with high recovery rate, but performance should decrease for tools that have poor accuracy. Figure ##FIG##10##11c## shows a performance drop in nSn and nSp as the length of the upstream sequence increases.</p>", "<p>Ann-Spec does appear to recover many sites regardless of the upstream length as noted by the close cluster of performance graphs on the right hand side of Figure ##FIG##10##11c##. Although similar, 'cr' generated data appears to have higher recovery rates for short windows. We believe that these recovery rates are most likely related to the relationship between the location of the signal for the TFBS and the location of the signal for the <italic>σ</italic>70 binding site – but this requires more exploration.</p>", "<title>Site distribution</title>", "<p>Algorithm practitioners commonly work from the assumption that TFBS have more information than the surrounding sequence. If this is so, the total number of TFBS in <italic>T</italic>_{<italic>l </italic>}(or the site density) should impact the performance of a tool. One would expect that a low density of sites would result in higher recovery rates (and vice versa for a high density of sites). To test this, we computed the total number of sites in <italic>T</italic>_{<italic>l </italic>}∀ <italic>l </italic>as annotated by <italic>RED</italic>. The site density for <italic>T</italic>_{<italic>l </italic>}is the number of sites from <italic>RED </italic>that exist in <italic>T</italic>_{<italic>l </italic>}over the number of sequences in <italic>T</italic>_{<italic>l</italic>}. We graphed sSn, nSp and nCC versus density for each of the tools. For all of the tools, site density did not appear to have any effect on sSn, nSp and nCC over 400 bp upstream sequences from RegulonDB. Figure ##FIG##11##12## shows no apparent decrease in nCC as the site density increases. It could be that there does not exist enough complex regulatory regions in <italic>E. coli </italic>to notice an impact on performance. It is likely that we would see a different result for organisms with more complex regulatory mechanisms, so we can not rule out site density as a factor in accurate regulatory motif prediction. These results do indicate that the 'background' signal is far more complex than originally thought and every program has difficulty distinguishing the foreground TFBS from interfering background signals.</p>", "<title>Site entropy</title>", "<p>If the background signals are simple and the binding sites are complex because they must be conserved by evolution, the relative information content of the TFBS should be greater than the information content found in the background signal. If this is the case, there should be a relationship between the information content of the binding site and prediction accuracy. To test this, we calculated information content for each site in RegulonDB using BioPython and plotted it against nSn, nSp, sSn, (not shown) and nCC (Figure ##FIG##12##13##). Information content of the site alone does not appear to be a determinative factor in how well these programs can recover the site. Perplexed by this result, we plotted nSn, nSp, sSn, and nCC versus information content divided by the number of upstream sequences in <italic>T</italic>_{<italic>k </italic>}(total number of bp in <italic>T</italic>_{<italic>k</italic>}). The result for sSn is shown in Figure ##FIG##13##14##.</p>", "<p>Figure ##FIG##13##14## shows that for some tools, the ratio of information content of the TFBS to the number of sequences in the upstream file can play a roll in sSn. Stronger information content and less background information implies better performance for Weeder, MEME and AlignACE. For the other tools, it is not clear if this relationship is present.</p>" ]

[ "<title>Discussion</title>", "<p>The most practical outcome of this work is an ability to rank motif prediction tools based on a known TFBS dataset. Tools with favourable performance characteristics can then be used to discover additional binding sites in closely related genomes or used in conjunction with experimental validation to improve the quality and comprehensiveness of existing TFBS databases. In RegulonDB, for example, of the methods tested Weeder, MEME and AlignACE present advantages over the other tools. AlignACE presents a more diverse list with more false positives whereas Weeder and MEME present true motif instances more often than the other tools. Motif prediction tools are composed of both a motif scoring function and a discrimination algorithm. The scoring function accepts a motif representation (e.g. a probability weight matrix) and then calculates the motif prediction candidates based on a discrimination function (e.g. maximum likelihood). Discrimination algorithms present a computational strategy to approximate the multiple sequence alignment of predicted binding positions relative to all multiple sequence alignments found in the background signal. Both Weeder and AlignACE have original scoring functions that could explain their utility on RegulonDB. MEME on the other hand uses expectation maximization as its discrimination algorithm. It could be that MEME benefits from this strategy over programs utilizing gibbs sampling. On the other hand, it is also likely that the predictions provided by these programs happen to be better over RegulonDB by random chance.</p>" ]

[ "<title>Conclusion</title>", "<p>In this paper we have presented a general method, MTAP, for evaluating cis-regulatory motif discovery tools. MTAP is novel and completely different from other approaches in that it allows both algorithm practitioners and users the flexibility to dynamically change attributes of data collection, algorithm parameters, and assessment. Our results indicate a clear need toward improvements in each of these areas. In our results, we explored four of the most commonly attributed factors to prediction accuracy: upstream file size, length of upstream sequences, TFBS density, and TFBS information content. The results obtained by MTAP in this assessment do not point toward any of these individual factors as playing a critical roll in finding TFBS. The results do indicate that the ratio of information content over upstream file size may have an influence on performance for some tools.</p>", "<p>The primary innovation in MTAP is not that we produce additional tools or additional benchmarks, but it is that we produce a platform that can be used to improve tools and the benchmarking process. The results presented in this paper indicate that the methods used to prepare upstream data, the algorithm, the parameters, and the method used in evaluation all play important rolls in how we look at the cis-regulatory motif discovery problem.</p>", "<p>In the past, many authors have dismissed bacterial regulatory motif detection as a far simpler problem than eukaryote regulatory motif detection (e.g. Xing <italic>et. al </italic>[##REF##15272436##35##]). While it is true that the annotated bacteria regulatory modules do not have the same level of complexity and combinatorial control, our results indicate that even for this 'simple' problem, regulatory motif detection methods have substantial room for improvement.</p>", "<p>Unlike other approaches, MTAP allows for the integration of regulatory regions from other species through an automated procedure. It remains to be seen which integration procedure and what combination of closely related and distantly related species improves performance for tools that incorporate regulatory regions from phylogentically related species. This is the subject of future work. At the moment, it does not appear that two closely related strains are enough to improve performance over conventional single sequence approaches. It would be interesting to extend our current implementation of MTAP and assess tools that integrate data from expression arrays and ChIP-chip arrays. Such approaches should lead to an increase in performance, but the parameters and procedures for this are not currently clear.</p>", "<p>It is not currently understood what features of the data make the problem of finding TFBS so difficult. The key advantage of MTAP is that it allows us to explore these features and propose new models that are more accurate and robust. It is important to understand the performance characteristics of the models that have been proposed in the past before we integrate additional information. In this way, we can understand more completely if the relationships found by more sophisticated techniques are real or if they could have occurred by random chance. Further exploration into motif representation, motif scoring, and the relationship between binding sites is still necessary if we are to accurately predict regulatory binding sites on the computer.</p>" ]

[ "<p>This is an open access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>In recent years, substantial effort has been applied to de novo regulatory motif discovery. At this time, more than 150 software tools exist to detect regulatory binding sites given a set of genomic sequences. As the number of software packages increases, it becomes more important to identify the tools with the best performance characteristics for specific problem domains. Identifying the correct tool is difficult because of the great variability in motif detection software. Consequently, many labs spend considerable effort testing methods to find one that works well in their problem of interest.</p>", "<title>Results</title>", "<p>In this work, we propose a method (MTAP) that substantially reduces the effort required to assess de novo regulatory motif discovery software. MTAP differs from previous attempts at regulatory motif assessment in that it automates motif discovery tool pipelines (something that traditionally required many manual steps), automatically constructs orthologous upstream sequences, and provides automated benchmarks for many popular tools. As a proof of concept, we have run benchmarks over human, mouse, fly, yeast, <italic>E. coli </italic>and <italic>B. subtilis</italic>.</p>", "<title>Conclusion</title>", "<p>MTAP presents a new approach to the challenging problem of assessing regulatory motif discovery methods. The most current version of MTAP can be downloaded from <ext-link ext-link-type=\"uri\" xlink:href=\"http://biobase.ist.unomaha.edu/\"/></p>" ]

[ "<title>Problem description</title>", "<p>A cis-regulatory motif discovery pipeline, <italic>M</italic>_{<italic>i</italic>}, contains a series of steps to separate transcription factor binding sites from 'background noise'. Motif discovery pipelines require collecting the positive example sequences, collecting negative example sequences, collecting relevant supporting data, running a separation filter to rank relevant sites based on an objective function and finally evaluating or verifying putative sites within the context of each regulatory module. Several factors outside of the pipeline itself contribute to the potential success of the discovery process, mainly: (1) the length of the sequences in the positive and negative sets, (2) the number of sequences in the positive and negative sets, (3) the distribution of transcription factor binding sites in the positive set, (4) the relative entropy of the transcription factor binding motif, and (5) the fraction of null sequences (ones that do not contain a binding site) in the positive set [##REF##15637633##20##]. It is likely there are additional unknown variables that impact the accuracy of an approach. For example, better background sequence models or sequences from closely related species (phylogenetic footprinting) are known to affect performance. Our approach to finding the variables, parameters, and algorithms that are best suited to annotating regulatory regions is an adaptive data collection and analysis platform (Figure ##FIG##0##1##).</p>", "<p>Our hypothesis is that a platform that automates each operation in cis-regulatory motif discovery enables discovery of better methods for cis-regulatory motif prediction. Such a platform enables practitioners and users the necessary flexibility to change characteristics of the data and algorithm parameters, and to isolate and understand known challenge cases. In addition, known methods with many manual steps can be made more explicit. Presently, it is difficult to benchmark existing approaches because they have a great variety of algorithmic parameters that allow the user to interactively optimize the discovery process. While this approach is advantageous because it provides flexibility in the discovery process, it also presents a challenge in algorithm benchmarking: how to formalize data collection, parameter fine tuning, and other intuitive steps taken by the most experienced users of these methods. From a more practical perspective, one has to wonder what evidence experienced practitioners use to determine a set of algorithm refinement steps that will result in better performance over the new dataset (i.e. what qualities in the data indicate a usage procedure). Our hypothesis is that the motif discovery process can be greatly improved if tool parameters are fine tuned based on previous performance benchmarks. We present algorithm benchmarking over a set of known and related transcription factor binding sites (TFBS) as a method to uncover the likely performance on the current dataset. This approach allows the refinement and modification of motif prediction pipelines. In particular, such a benchmarking approach allows for interactive modification of known TFBS in incomplete datasets, modification of the procedures used in collecting positive, negative, and phylogenatically related sequence samples, modification of methods for finding TFBS within the samples, and even modification of the methods used to score motif discovery pipelines.</p>", "<p>Our central hypothesis is that by looking at all components of data and software along the cis-regulatory motif discovery process, we can refine our understanding of regulation and discover pipelines with high accuracy. The way we look at this complex problem is a result of how we collect the data, how we build algorithms for discovering regulatory motifs, the biological and manual processes and data structures used to create comprehensive annotations of cis-regulatory regions, and the methods we choose to use in grading these motif discovery tools. We view each of these components as parts that can and should be modified as the specifics of the biological problem at hand become clear. In this paper, we present a open source prototype for cis-regulatory motif discovery algorithm evaluation.</p>", "<title>Implementation</title>", "<p>MTAP provides an automated method for ranking regulatory motif detection algorithms. The underlying principle is to create a 'test', <italic>T</italic>, and an 'answer key', <italic>K </italic>= {<italic>k</italic>₁, <italic>k</italic>₂,...}. <italic>K </italic>is generated by parsing the raw database management systems (DBMS) of RegulonDB [##REF##14681419##24##], DBTBS [##REF##11125112##27##], PRODORIC [##REF##12519998##28##], RegTransBase [##REF##17142223##29##] and Transfac [##REF##8594589##30##]. The annotated genome, <italic>G</italic>_{<italic>j</italic>}, is then parsed for each regulatory binding site annotated in <italic>K</italic>. <italic>T</italic>_{<italic>k </italic>}is a collection of sequences corresponding to the surrounding regions of known binding sites for transcription factor <italic>k </italic>in <italic>G</italic>_{<italic>j</italic>}. <italic>T</italic>_{<italic>k </italic>}is composed of <italic>l </italic>instances of binding positions <italic>T</italic>₁, <italic>T</italic>₂,..., <italic>T</italic>_{<italic>l </italic>}in <italic>G</italic>_{<italic>j </italic>}as annotated by the database. To score a method, <italic>M</italic>_{<italic>i</italic>}, we construct an automated pipeline via a scripting language (Perl/Python) that runs the method. Each method has a different algorithmic approach and has different requirements of pre- and post-processing. For each method, we construct background probability models, standardize input datasets, install program dependencies, and provide conversion and utility scripts so that each method can be graded fairly. We consider a fair test to be a test where each program has access to the same information and must mark the transcription factor binding positions in a standard way. This standard marking is then assessed by comparing the annotation of regulatory sites provided by the algorithm with the known annotation in the database.</p>", "<p>A schematic overview of our assessment method is presented in Figure ##FIG##1##2##. Because databases contain many types of binding positions (not just for transcription factors but also for sigma factor binding sites, ncRNA interaction sites and other binding proteins) these evaluations are an indication of how well each algorithm can recover the binding positions for each element in the regulatory network and not just transcription factor binding positions.</p>", "<title>Generating upstream sequences</title>", "<p>Varying upstream length allows us to explore the trade off between detecting long range interactions (large <italic>n</italic>) and high prediction accuracy (small <italic>n</italic>). MTAP contains two genome substringing methods for generating upstream sequences as shown in Figure ##FIG##2##3##. 'Completely-realistic' (cr) data generation is best suited to problems that convert gene lists to binding locations (such as a micro-array). 'Semi-realistic' (sr) data generation is best suited to problems that generate a set of sequences (for example when we know the regions of proposed binding sites). In most cases 'sr' constructed data produces a more fair comparison between pattern finding methods, but is not realistic when the tools are applied to a set of co-regulated loci.</p>", "<p>Completely-realistic data generation is most often the natural choice, but is limited because the upstream file may not contain the motif for long range interactions. Also, the 'cr' method may select a different downstream gene as part of the data construction procedure. These two issues are realistic in the cis-regulatory motif discovery process and are representative of current problems in cis-regulatory motif discovery. This method is therefore representative of current methods used in constructing co-regulated upstream sequences. Future work could be done to make automated upstream data generation more sensitive to these types of issues.</p>", "<p>The transformation function <italic>t</italic>(<italic>a</italic>, <italic>n</italic>) applies one of the substring operations (currently 'cr' or 'sr') to the reference genome <italic>G</italic>_{<italic>j </italic>}to produce <italic>T</italic>_{<italic>k</italic>}. <italic>T</italic>_{<italic>k </italic>}is generated by selecting regulator <italic>k </italic>and generating an upstream sequence <italic>u</italic>_{<italic>k</italic>, <italic>n </italic>}for all instances of <italic>k </italic>in <italic>G</italic>_{<italic>j</italic>}. Figure ##FIG##3##4## provides a diagram of several of the stages used to construct <italic>T </italic>= <italic>T</italic>₁, <italic>T</italic>₂,... for all instances of <italic>k </italic>in <italic>G</italic>_{<italic>j</italic>}. Despite the fact that <italic>T</italic>_{<italic>k </italic>}is constructed by considering only TF <italic>k</italic>, all of the other transcription factors in the database are marked and scored if they fall within sequence indices for any sequence in <italic>T</italic>_{<italic>k</italic>}.</p>", "<title>Generating orthologous upstreams</title>", "<p>After the upstream file, <italic>U</italic>_{<italic>k</italic>, <italic>n</italic>}, is constructed, MTAP collects regulatory sequences from closely related genomes <italic>G</italic>_{<italic>j</italic>+1}, <italic>G</italic>_{<italic>j</italic>+2},..., by using downstream orthologs. To do this, MTAP constructs a list of all proteins in each genome <italic>G</italic>_{<italic>j</italic>+1}, <italic>G</italic>_{<italic>j</italic>+2},.... Before the upstream is created, MTAP uses as orthalog detection method to create an orthalog table, <italic>O </italic>(our current ortholog detection methods are best bi-directional blast hits and RSD [##REF##15593400##31##]). <italic>O </italic>contains a list of protein products from <italic>G</italic>_{<italic>j </italic>}and <italic>G</italic>_{<italic>j</italic>+1}and a confidence score <italic>O</italic>_{<italic>c </italic>}corresponding to the confidence in the orthology relationship as determined by the ortholog detection method. For each sequence in <italic>U</italic>_{<italic>k</italic>}, we select the nearest downstream gene, <italic>g</italic>_{<italic>i</italic>, <italic>j </italic>}∈ <italic>G</italic>_{<italic>j</italic>}. MTAP then looks up any entries for <italic>g</italic>_{<italic>i</italic>, <italic>j </italic>}∈ <italic>O</italic>. If <italic>O</italic>_{<italic>c </italic>}&gt; <italic>τ </italic>for some constant <italic>τ</italic>, MTAP appends a region <italic>n </italic>bp upstream of <italic>g</italic>_{<italic>j</italic>+1}. If multiple entries exist for <italic>g</italic>_{<italic>i</italic>, <italic>j</italic>}, MTAP appends the region upstream <italic>g</italic>_{<italic>j</italic>+1, <italic>i </italic>}such that <italic>O</italic>_{<italic>c </italic>}is greatest. MTAP continues this procedure for all genomes <italic>G</italic>_{<italic>j</italic>}, <italic>G</italic>_{<italic>j</italic>+1},.... This procedure is then repeated to construct <italic>T</italic>₁, <italic>T</italic>₂,.... An example result is found in Figure ##FIG##3##4D##.</p>", "<p>Several important points should be made about phylogentic foot-printing. First, the existence of a set of regulatory binding sites exists in <italic>G</italic>_{<italic>j </italic>}does not imply the existence of the same set of regulatory binding positions in <italic>G</italic>_{<italic>j</italic>+1}in the same positions. In our example in Figure ##FIG##3##4##, <italic>T</italic>_{<italic>k </italic>}does not contain a binding position for <italic>K</italic>₂∈ <italic>G</italic>_{<italic>j</italic>+2}. It is possible that our criteria to classify orthologs is too stringent to find an actual ortholog in <italic>G</italic>_{<italic>j</italic>+1}. Notice that in our example <italic>T</italic>_{<italic>l </italic>}contains no upstream corresponding to <italic>g</italic>₁in <italic>G</italic>_{<italic>j</italic>+2}. In our example, this occurred because the homology threshold criteria excluded <italic>g</italic>₂from <italic>G</italic>_{<italic>j</italic>+2}. Again, these problems will exist in any current automated pipeline used for regulatory motif detection. Some of these issues can be resolved within the algorithms themselves: many algorithms incorporate the hypothesis that zero or many instances of the cis-regulatory binding motif exists within the upstream sequence. However, the prospect is very real that we could integrate sequences that are related but do not contain binding sites for transcription factor <italic>k</italic>. Currently, we have very little understanding of the relationship between the evolution of regulatory sequences and coding sequences. It is quite possible that phylogenetically related sequences introduce additional 'noise' with very little signal in regulatory sequences for genes that do not provide critical functions. These complex relationships warrant an in-depth study of regulatory evolution as it relates to coding sequence evolution – which is the subject of future work. For now, we wish to use this approach to benchmark single genome methods as they compare to multiple genome based methods. To our knowledge, MTAP is the first method that allows additional sources of information (sequence conservation) to be automatically integrated as part of the evaluation process.</p>", "<title>Constructing background sequences</title>", "<p>Once <italic>T</italic>_{<italic>k </italic>}has been constructed for all <italic>K</italic>, we present three possible background sequence files to motif discovery pipelines: (1) all upstream sequences of length <italic>n </italic>in <italic>G</italic>_{<italic>j</italic>}, (2) all upstream sequences of length <italic>n </italic>that exist in some <italic>T</italic>_{<italic>k</italic>}, and (3) a fasta formated sequence of <italic>G</italic>_{<italic>j</italic>}. Programs that incorporate phylogeny have different background sequence requirements. Such programs require a background phylogenetic tree constructed from extracting 16S rRNA from each genome in the study. Some programs require pre-processing steps to calculate an HMM or GC content of the test sequences. Other programs require a background probability distribution of all upstream sequences in <italic>G</italic>_{<italic>j</italic>}. We compute each of these requirements for each pipeline in our pre-processing stage and provide the files to each pipeline.</p>", "<title>Compensating for unknown TFBS</title>", "<p>Unknown TFBS that exist in <italic>T</italic>_{<italic>k </italic>}complicate the assessment process. Tools could predict true sites that are currently unknown or un-annotated. To exclude unknown sites from <italic>T</italic>_{<italic>k</italic>}, we construct a Markov chain, <italic>MC</italic>_{<italic>m</italic>}, (of depth <italic>m</italic>). For each sequence, <italic>S</italic>_{<italic>i</italic>}, in <italic>T</italic>_{<italic>k</italic>}, <italic>s</italic>_{<italic>i </italic>}is sampled with <italic>MC</italic>_{<italic>m</italic>}. We then generate an alternative sequence, <italic>S'</italic>_{<italic>i</italic>}, through a random walk through the sates in <italic>MC</italic>_{<italic>m</italic>}. For all TFBS in <italic>K </italic>that overlap <italic>s</italic>_{<italic>i</italic>}, we insert the true TFBS sequence from <italic>s</italic>_{<italic>i </italic>}into <italic>s'</italic>_{<italic>i</italic>}. In this way, we use <italic>MC</italic>_{<italic>m </italic>}to 'scramble' the upstream sequences in the test and then re-insert the known motifs back into the sequences at the same positions found in the source genome. It is also informative to have instances of true negative sequences produced via <italic>MC</italic>_{<italic>m</italic>}, so we produce instances of <italic>T</italic>_{<italic>k </italic>}with no inserted TFBS. Orthologous upstream sequences need not be scrambled if they are not scored. These synthetic sequences serve to make a ' more fair' test in those cases where very few of the known motifs are marked. However, such sequences may not correctly incorporate the biological process that generates true sequences. To accommodate this, we also insert TFBS into a random sequence from the set of all upstream sequences in <italic>G</italic>_{<italic>j</italic>}.</p>", "<title>Constructing motif discovery pipelines</title>", "<p>In constructing motif discovery pipelines, our intention is to include pipelines for as many tools as possible. However, building parsers, installing scripts, and optimizing a pre-processing, post-processing, and runtime pipeline for each of over 150 programs is extremely labor intensive. Our main obstacle is in finding executables that can run on a variety of architectures in a Linux cluster. In many cases, we attempted to contact authors to arrange a port of their tool to a Linux cluster. Many authors are extremely helpful and we would like to thank them for the advice and guidance of how to use and port their tool. However, we realized that even if we have a stand-alone version of a method working on our architecture, MTAP users will still need to install many of the tools directly from the authors (if we do not have the legal authority to distribute the tool). Our strategy for including a tool is as follows:</p>", "<p>• Include tools that are the most popular.</p>", "<p>• Include tools that present different and novel scoring functions for differentiating background sequences from transcription factor binding sites.</p>", "<p>• Include tools that integrate diverse types of information from public sources.</p>", "<p>• Do not include tools that do not have a downloadable executable or can not be compiled locally, as such tools can not easily be run thousands of times for assessment purposes.</p>", "<p>• Do not include tools that do not have support for different architectures and operating systems (tools we can not run on our computers).</p>", "<p>• As we can not redistribute tools with strict licensing agreements or 'abandonware', inclusion of such tools is left to the user community.</p>", "<p>Our platform is provided open-source for practitioners who would like to develop their own pipelines to integrate their tool into MTAP. This approach has an advantage in that the developer of the tool is also the developer of the pipeline for evaluating the tool. This could provide an edge as the tool developer will understand the limitations and usage procedures best. We provide our assessment pipelines within our framework for open access review and improvement.</p>", "<p>In this work, we developed pipelines for AlignACE [##REF##10698627##8##], Ann-Spec [##REF##10902194##9##], ELPH [##UREF##4##32##], Gibbs [##REF##12824370##3##], Glam [##REF##14704356##10##], MEME [##REF##16845028##33##], PhyloGibbs [##REF##16477324##16##], PhyME [##UREF##1##15##], and Weeder [##REF##11473011##11##,##UREF##0##12##]. For each of these tools in the Tompa <italic>et. al </italic>benchmark, we developed an automated system that was as close to the spirit of the procedure used by the algorithm practitioners as practical. For example, our AlignACE pipeline contains a pre-processing script for calculating GC content of the upstream file and a postprocessing script to mask low complexity repeats using RepeatMasker. The pipeline then parses the raw AlignACE output which results in a ranked list of predicted transcription factor binding sites sorted via the AlignACE MAP score. The pipeline accepts the highest <italic>c </italic>scoring motifs by MAP score and then determines confidence by calculating the group specificity score as provided by CompareACE. A high group specificity score and MAP score indicates a high degree of confidence in the prediction provided by the AlignACE pipeline. For those tools that are not in the Tompa assessment, we carefully followed the usage guides and refined our pipelines using MTAP to produce the best results possible.</p>", "<p>For some methods, the code is not available and the method paper does not make it clear if certain data preparation procedures can be accommodated by the method. For example, some methods account for upstream sequences that are reverse complemented while others do not. Some methods account for zero, one, or many motif instances on one strand while others do not. Some methods allow for variable motif widths, while others require an explicit window (thus we need to write a procedure to run the algorithm over all reasonable multiple widths and then rank and combine the results from multiple runs). We assume that motif discovery pipelines are sufficiently robust to account for these technical details and we attempted to make our pipelines robust in this way. However, in constructing these pipelines in this way, we may have overlooked some aspects of the algorithmic approach that would make our pipelines not representative of the original author intent (i.e. our pipeline may not be representative of the best performance possible by expert manual application of the method). We addressed this issue in two ways. First, we built pipelines for multiple implementations of similar approaches (e.g. Gibbs and ELPH). Second, we refined each pipeline based on the objective function found in the literature and the benchmarks obtained via MTAP. To guard against over-fitting of a pipeline to a certain dataset, we did not allow modification of the code or implementation of any procedures not recommended explicitly by the authors. We performed benchmarks over the Tompa assessment datasets, RegulonDB [##REF##14681419##24##], and DBTBS [##REF##11125112##27##] and selected transcription factors at random to validate if the proposed change improved results. As the search space is large, there may exist a set of pre-processing, post-processing, and runtime steps that may improve the performance of our current pipelines. If used in this way, MTAP provides the framework for method developers and method users to formalize and improve motif discovery pipelines.</p>", "<title>Pipeline evaluation</title>", "<p>Three levels of specificity can be considered when evaluating the accuracy of <italic>M</italic>_{<italic>i</italic>}: (1) <italic>M</italic>_{<italic>i </italic>}correctly predicts the region bound by some transcription factor <italic>k</italic>, (2) <italic>M</italic>_{<italic>i </italic>}correctly predicts the binding site of <italic>k</italic>, and (3) <italic>M</italic>_{<italic>i </italic>}correctly predicts the amino acids in <italic>TF </italic>that interact with specific nucleotides within the regulatory region. <italic>M</italic>_{<italic>i </italic>}may also correctly predict the type and strength of the interactions. Predictions of Type 1 are analogous to a gel shift assay in that we can identify a part of the regulatory region bound by a protein. Predictions of Type 2 are analogous to a DNA foot-printing assay. Predictions of Type 2 are more specific in that each region of interaction between <italic>k </italic>and the DNA is identified. For example, if a transcription factor is a dimer, two interaction sites are identified by predictions of Type 2 whereas only one interaction site is identified by a prediction of Type 1 (a site that contains both sites in a Type 2 prediction). The third type of prediction is analogous to determining the crystal structure interaction points of the transcription factor – DNA complex. While the specificity and information provided by the third Type of prediction is far greater than annotations of Type 1 or Type 2, such data is difficult to obtain and few methods make predictions at this level. We therefore generate two annotation files: a 'redfile' and a 'blackfile' corresponding to the site level and region level respectively. To generate an upstream file, we use the blackfile annotation. To assess the performance of an algorithm we evaluate the predictions versus the redfile annotation.</p>", "<p>The redfile annotation <italic>RED </italic>= <italic>I</italic>₁, <italic>I</italic>₂,... contains a set of intervals <italic>I</italic>_{<italic>k </italic>}= (<italic>u</italic>₁, <italic>v</italic>₁), (<italic>u</italic>₂, <italic>v</italic>₂),... that correspond to the start (<italic>u</italic>_{<italic>h</italic>}) and stop (<italic>v</italic>_{<italic>h</italic>}) positions in <italic>G</italic>_{<italic>j </italic>}corresponding to binding locations for transcription factor <italic>k</italic>. To compare the redfile annotation to the motif tool predictions, <italic>U</italic>_{<italic>i </italic>}(upstream sequence <italic>i </italic>in <italic>T</italic>_{<italic>k</italic>}) is used as a scaffold to place annotation elements. To annotate the known sites at the nucleotide level, we mark each base, <italic>j</italic>, in <italic>U</italic>_{<italic>i </italic>}if <italic>u</italic>_{<italic>h </italic>}≤ <italic>j </italic>≤ <italic>v</italic>_{<italic>h </italic>}∀ <italic>K</italic>. (<italic>v</italic>_{<italic>h </italic>}≤ <italic>j </italic>≤ <italic>u</italic>_{<italic>h </italic>}∀ <italic>K </italic>if <italic>U</italic>_{<italic>i </italic>}is on the opposite strand) The predicted binding locations, , predicted by <italic>M</italic>_{<italic>i </italic>}are parsed and translated into a ranked list of predicted binding sites, each of the form . The ranked list contains elements <italic>TFL</italic>₁, <italic>TFL</italic>₂,... sorted according to the confidence that <italic>M</italic>_{<italic>i </italic>}has in the prediction accuracy. MTAP accepts the top <italic>c </italic>elements from the rank list for evaluation and inserts them onto the upstream scaffold. MTAP then marks each position, <italic>j</italic>, as a predicted nucleotide if there exists some predicted binding site, <italic>B</italic>, that overlaps it. At the nucleotide level, we collect the overlap statistics shown in Table ##TAB##0##1##.</p>", "<p>The first four core statistics (nTP-nucleotide true positives, nFN-nucleotide false negatives, nFP-nucleotide false positives, and nTN-nucleotide true negatives) are collected by summing the number of each of the occurrences shown in Table ##TAB##0##1## in <italic>T</italic>_{<italic>k</italic>}. The site level statistics (sTP-site true positives, sFN-site false negatives, and sFP-site false positives) are the final three core statistics provided by MTAP. A site level statistic encompasses the idea that a group of adjacent nucleotides marked as binding positions for transcription factor <italic>k </italic>by <italic>M</italic>_{<italic>i </italic>}is representative of a binding site annotation. A site is annotated as a true positive if overlaps <italic>I</italic>_{<italic>x </italic>}∀ <italic>x </italic>by more than τ percent of <italic>I</italic>_{<italic>x</italic>}. For example, consider two overlapping sites, a known site <italic>I</italic>_{<italic>x </italic>}of 12 consecutive nucleotides and a predicted site <italic>B </italic>of 8 consecutive nucleotides. Given <italic>τ </italic>= .25 If <italic>I</italic>_{<italic>x </italic>}shares 3 nucleotides with <italic>B </italic>it is annotated as a sTP. If <italic>I</italic>_{<italic>x </italic>}shares only 2 nucleotides with <italic>B </italic>it is annotated as a sFP. Once a site, <italic>I</italic>_{<italic>x</italic>}, overlaps a prediction, it can not be annotated as a sFN. All remaining sites, <italic>I</italic>₁, <italic>I</italic>₂,..., <italic>I</italic>_{<italic>x</italic>}, that do not have an overlapping prediction for tool <italic>M</italic>_{<italic>i </italic>}are annotated as sFN. The site level statistics are in Table ##TAB##1##2##.</p>", "<p>Tompa <italic>et. al </italic>set <italic>τ </italic>to 25%. The logic was that such an overlap makes discovery and refinement of the TFBS possible in the lab. In large scale genome annotations, we find such a threshold to be too strict. For example, many TFBS are not annotated specific enough in databases such as RegulonDB. This results in TFBS that exist in <italic>K </italic>that can be large. Such annotations are actually representative of the region of binding of the transcription factor (a blackfile annotation) and not the binding sites (a redfile annotation). Because many motif discovery programs have fixed motif widths (e.g. 8), a threshold of 25\\% would not be sufficient to mark a sTP (e.g. a site of width 60 and a site prediction of length 8). We could choose to rank site level motifs based on a percentage of the prediction width instead of the regulatory motif width, but this would give an unfair advantage to methods that predict larger sites. Our current approach is to set τ equal to the maximum annotated site width in the dataset divided by the minimum expected motif width predicted by our suite of programs (usually 8) times 25%. Our logic is that a degree of overlap indicates that computational and biological refinement of site predictions can still find the site. That said, manual curation of datasets to ensure binding site annotations rather than region annotations is necessary. Standards across regulatory binding site databases to delineate each of the three levels of biological data would greatly increase evaluation accuracy of motif discovery tools. Also, as not all tools provide a mapping for a set of sites to a putative regulator, these statistics are currently not reflective of which regulator is annotated by a site level prediction.</p>", "<p>Following Tompa <italic>et. al </italic>we define the statistics in Figure ##FIG##4##5## to perform the assessment.</p>", "<p>These statistics enable us to determine the quality of algorithm predictions and therefore infer which tools may be best suited to discover unknown motifs under similar situations. MTAP evaluates each of these statistics by comparing the predictions found in the program output with a set of known binding positions of the same type. For each instance found in the known dataset, a motif prediction tool is run and then parsed. The prediction is compared to the known binding site via the seven key statistics in Table ##TAB##0##1## and Table ##TAB##1##2##. These statistics will then be used to assess the overall performance of the algorithm. MTAP produces an output file for each regulatory binding motif in <italic>K</italic>. Users can sum the raw statistics in these files as they see fit. For this paper, we collect the seven raw performance statistics for each motif in the assessment and then sum these values as if the collection of runs was actually one run.</p>", "<p>In some cases, such sums do not graphically represent the contribution of each element in the set to the total performance score. To address this, we also developed a graph that iterates over all runs in the test (<italic>T</italic>₁, <italic>T</italic>₂,..., <italic>T</italic>_{<italic>k</italic>}) The graph produced is a modified receiver operating characteristic (ROC) curve that combines statistics from multiple runs [##UREF##5##34##]. We use the following algorithm to produce our ROC graphs:</p>", "<p>   for each motif in the dataset:</p>", "<p>      input xTP, xFP, nFN, nTN, sTP, sFP, sFN</p>", "<p>      P &lt; -calculate xSP and xSN for this motif</p>", "<p>   for all motifs in the dataset:</p>", "<p>      totalSP = sum(xSP)</p>", "<p>      totalSN = sum(xSN)</p>", "<p>   sort(P.xSN)</p>", "<p>   for all ties in P.xSN; sort(P.xSP)</p>", "<p>   for i in P:</p>", "<p>      plot(xSN/totalSN, xSP/totalSP)</p>", "<p>This produces a curve that travels straight up and then to the right if all motifs in the dataset are predicted correctly. The curve will travel straight to the right and then up if very few of the motifs are predicted correctly. Finally, if the tool predicts sites correctly as often as it predicts sites incorrectly, a line along the diagonal of the ROC graph will be plotted. However, unlike machine learning algorithms where such a graph is often no better than a random classification of sites, using this method there is some value in graphs along the diagonal because we only allow 3 site predictions to be placed on the scaffold.</p>", "<title>Known motif databases</title>", "<p>In MTAP, we have implemented interfaces to each of the following databases: RegulonDB [##REF##14681419##24##], DBTBS [##REF##11125112##27##], PRODORIC [##REF##12519998##28##], RegTransBase [##REF##17142223##29##] and TRANSFAC [##REF##8594589##30##] (via the Tompa Benchmark Set). Each of these databases was constructed with different goals and none were built explicitly to evaluate motif prediction tools. Therefore, some database cleaning is required to make these datasets more appropriate for algorithm assessment. We provide two procedures: (1) We require that the known binding site occur in at least <italic>nl </italic>locations in <italic>G</italic>_{<italic>j </italic>}(we set this to 3 for the results below). MTAP will not create upstream files for TFBS that do not meet this threshold. (2) We provide a script to analyze multiple sequence alignments and information content of a set of known motifs. We do not require that the TFBS have a consensus sequence as annotated by the database – our logic being that users can eliminate sites without a strong consensus from their analysis by using our information content script and keeping only those sites that exceed a set threshold.</p>", "<p>Duplicate instances of the same upstream file are eliminated to prevent bias before they are processed by the programs (however the originals are stored for those users wishing to do further analysis). MTAP does not accept TFBS that do not contain a start and end position in <italic>G</italic>_{<italic>j</italic>}. For TFBS that are inconsistent, users can eliminate the sites and re-score the TFBS. Our primary goal in collecting data is to provide automated methods that can improve as the datasets become more comprehensive. At this time, both PRODORIC [##REF##12519998##28##] and RegTransBase [##REF##17142223##29##] have very few TFBS with enough binding positions in the same genome <italic>G</italic>_{<italic>j </italic>}for us to provide a comprehensive benchmark (the goal of these databases is more focused on tracking conversation of TFBS across species). Though labor intensive, high annotation density in datasets such as these provides the greatest insight into evaluating computational methods that predict transcription factor binding sites.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>DQ, HA, and DB proposed the design of the system. KD implemented the prediction pipelines and parsers for each tool. DQ and KD integrated the databases. DQ and MS implemented the system and migrated the system to a clustered environment. DQ, KD and MS tested the system. DQ conducted the computational experiments and wrote the manuscript. All authors approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We would like to thank each of the authors of motif prediction tools for making their tools available and for helping us in running and automating tool pipelines including their method. We would like to thank Laura A. Quest for help with the manuscript.</p>", "<p>This research project was made possible by the NSF grant number EPS-0091900 and the NIH grant number P20 RR16469 from the INBRE Program of the National Center for Research Resources.</p>", "<p>This article has been published as part of <italic>BMC Bioinformatics </italic>Volume 9 Supplement 9, 2008: Proceedings of the Fifth Annual MCBIOS Conference. Systems Biology: Bridging the Omics. The full contents of the supplement are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1471-2105/9?issue=S9\"/></p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>The Motif Tool Assessment Platform (MTAP) enables a researcher to automate each of the steps in cis-regulatory motif discovery, evaluate tools, and propose changes. MTAP is built as a platform where the data collection methods, motif discovery pipelines, and known binding sites are all modular components that can be edited and substituted to look at different aspects of the complex problem of cis-regulatory region annotation.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>An overview of the MTAP running procedure. A. All known binding positions <italic>G</italic>_{<italic>j </italic>}in collected into upstream regions corresponding to each CDS in <italic>G</italic>_{<italic>j</italic>}. B. A transformation function <italic>t</italic>(<italic>a</italic>, <italic>n</italic>) creates a test for binding protein <italic>an </italic>bases upstream of each CDS (note that the transcription start site is often unknown or not correctly annotated). C. Background probability information for the entire genome is collected by comparing the upstream regions from the entire genome (or ∀ <italic>k</italic>) to the foreground regions selected by <italic>t</italic>(<italic>a</italic>, <italic>n</italic>). D. Pipeline <italic>p </italic>runs each step of the proposed method <italic>M</italic>_{<italic>i</italic>}. E. <italic>M</italic>_{<italic>i </italic>}creates a marking on the sequences in B that is evaluated against all marked transcription factor binding positions in B to score the performance of <italic>M</italic>_{<italic>i </italic>}in recovering binding sites for transcription factor <italic>a</italic>. This is then repeated for transcription factors <italic>b </italic>and <italic>c</italic>.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Demonstration of two different methods for obtaining the sequence <italic>n </italic>bases surrounding the known regulatory binding site. In the figure 'cr' refers to completely realistic generation where we find the closest downstream CDS location in the genome and extract <italic>n </italic>bases upstream from that CDS. 'sr' refers to extracting <italic>n</italic>/2 bases upstream and downstream of the center of the binding site. We assume that programs reverse complement sequences appropriately as needed by the method as part of the discovery procedure but provide upstream sequences on the positive strand relative to the downstream CDS. The known motif 'blackfile' sequence is represented by a black line over the binding site <italic>k </italic>that refers to the region that is bound by the transcription factor. The red regions in the diagram illustrate the actual binding positions known for motif <italic>k </italic>for those nucleotides that interact with the regulatory protein.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>An overview of the pre-processing steps taken in MTAP: A. The source genome <italic>G</italic>_{<italic>j </italic>}with genes <italic>g</italic>₁, <italic>g</italic>₂, <italic>g</italic>₃with known binding sites <italic>k</italic>₁, <italic>k</italic>₂, <italic>k</italic>₃, <italic>k</italic>₄, <italic>k</italic>₅(found clockwise around the genome from the origin); <italic>K</italic>₁and <italic>K</italic>₂are the known regulatory proteins that bind to the transcription factors B. Two additional genomes, <italic>G</italic>_{<italic>j</italic>+1}and <italic>G</italic>_{<italic>j</italic>+2}, that are phylogentically closely related to <italic>G</italic>_{<italic>j</italic>}. <italic>G</italic>_{<italic>j</italic>+1}and <italic>G</italic>_{<italic>j</italic>+2}also have binding sites C. The background phylogenetic tree constructed via extracting 16S rRNA D. Two tests <italic>T</italic>₁and <italic>T</italic>₂corresponding to known positions from <italic>K</italic>₁and <italic>K</italic>₂.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>Statistics for Evaluating Motif Prediction Algorithm Implementations.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>8 performance statistics for 9 motif prediction pipelines generated by MTAP over RegulonDB 400 bp upstream regulatory sequences.</p></caption></fig>", "<fig position=\"float\" id=\"F7\"><label>Figure 7</label><caption><p>8 performance statistics for 9 motif prediction pipelines generated by MTAP over RegulonDB 400 bp upstream regulatory sequences.</p></caption></fig>", "<fig position=\"float\" id=\"F8\"><label>Figure 8</label><caption><p>nCC versus total number of base pairs in <italic>T</italic>_{<italic>k </italic>}over RegulonDB 400 bp sequences.</p></caption></fig>", "<fig position=\"float\" id=\"F9\"><label>Figure 9</label><caption><p>Site sensitivity versus the number of sequences in the upstream file for each <italic>T</italic>_{<italic>k </italic>}in RegulonDB with more than 2 unique binding positions in <italic>G</italic>_{<italic>j</italic>}.</p></caption></fig>", "<fig position=\"float\" id=\"F10\"><label>Figure 10</label><caption><p>Nucleotide specificity versus the number of sequences in the upstream file for each <italic>T</italic>_{<italic>k </italic>}in RegulonDB with more than 2 unique binding positions in <italic>G</italic>_{<italic>j</italic>}.</p></caption></fig>", "<fig position=\"float\" id=\"F11\"><label>Figure 11</label><caption><p>ROC curves for Ann-Spec (20 (blue), 50 (green), 100 (red), 200 (cyan), 300 (magenta), 400 (yellow), 500 (black), and 800 (lower red) basepairs upstream of the CDS.</p></caption></fig>", "<fig position=\"float\" id=\"F12\"><label>Figure 12</label><caption><p>Density of the binding sites in <italic>T</italic>_{<italic>k </italic>}versus nCC.</p></caption></fig>", "<fig position=\"float\" id=\"F13\"><label>Figure 13</label><caption><p>nCC versus Information Content (IC) for <italic>T</italic>_{<italic>l </italic>}400 bp from RegulonDB.</p></caption></fig>", "<fig position=\"float\" id=\"F14\"><label>Figure 14</label><caption><p>sSn versus information content divided by number of upstream sequences.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Nucleotide Level Statistics. <italic>u</italic>_{<italic>i</italic>, <italic>j </italic>}represents the upstream regulatory sequence <italic>j </italic>at position <italic>i</italic>, <italic>RED </italic>is the set of annotated database positions found in the 'redfile', <italic>B </italic>represents a binding site predicted by method <italic>M</italic>.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Statistic</td><td align=\"left\">Definition</td></tr></thead><tbody><tr><td align=\"left\">nTP</td><td align=\"left\">If <italic>u</italic>_{<italic>i</italic>, <italic>j </italic>}exists in both <italic>RED </italic>and <italic>B</italic>.</td></tr><tr><td align=\"left\">nFN</td><td align=\"left\">If <italic>u</italic>_{<italic>i</italic>, <italic>j </italic>}exists in <italic>RED </italic>and not in <italic>B</italic>.</td></tr><tr><td align=\"left\">nFP</td><td align=\"left\">If <italic>u</italic>_{<italic>i</italic>, <italic>j </italic>}exists in <italic>B </italic>and not in <italic>RED</italic>.</td></tr><tr><td align=\"left\">nTN</td><td align=\"left\">If <italic>u</italic>_{<italic>i</italic>, <italic>j </italic>}exists in neither <italic>RED </italic>nor <italic>B</italic>.</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Site Level Statistics.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Statistic</td><td align=\"left\">Definition</td></tr></thead><tbody><tr><td align=\"left\">sTP</td><td align=\"left\">Number of known sites overlapped by predicted sites.</td></tr><tr><td align=\"left\">sFN</td><td align=\"left\">Number of known sites not overlapped by predicted sites.</td></tr><tr><td align=\"left\">sFP</td><td align=\"left\">Number of predicted sites not overlapped by known sites.</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Sensitivity and specificity for the 10 largest and smallest datasets from a Weeder run over all of RegulonDB at 500 bp.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\" colspan=\"4\">10 Smallest Motif Datasets</td></tr><tr><td align=\"left\">Size</td><td align=\"left\">nSn</td><td align=\"left\">nSp</td><td align=\"left\">sSp</td></tr></thead><tbody><tr><td align=\"left\">1503</td><td align=\"left\">0.35</td><td align=\"left\">0.81</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\">1503</td><td align=\"left\">0.12</td><td align=\"left\">0.92</td><td align=\"left\">0.33</td></tr><tr><td align=\"left\">1503</td><td align=\"left\">0.26</td><td align=\"left\">0.82</td><td align=\"left\">0.67</td></tr><tr><td align=\"left\">1503</td><td align=\"left\">0.24</td><td align=\"left\">0.79</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\">1503</td><td align=\"left\">0.34</td><td align=\"left\">0.81</td><td align=\"left\">0.67</td></tr><tr><td align=\"left\">1503</td><td align=\"left\">0.23</td><td align=\"left\">0.61</td><td align=\"left\">0.67</td></tr><tr><td align=\"left\">2004</td><td align=\"left\">0.32</td><td align=\"left\">0.62</td><td align=\"left\">0.73</td></tr><tr><td align=\"left\">2004</td><td align=\"left\">0.14</td><td align=\"left\">0.72</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\">2004</td><td align=\"left\">0.31</td><td align=\"left\">0.78</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\">2004</td><td align=\"left\">0.19</td><td align=\"left\">0.74</td><td align=\"left\">0.5</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\">Avg:</td><td align=\"left\">0.25</td><td align=\"left\">0.76</td><td align=\"left\">0.56</td></tr><tr><td align=\"center\" colspan=\"4\">10 Largest Motif Datasets</td></tr><tr><td align=\"left\">Size</td><td align=\"left\">nSn</td><td align=\"left\">nSp</td><td align=\"left\">sSp</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\">7515</td><td align=\"left\">0.1</td><td align=\"left\">0.79</td><td align=\"left\">0.31</td></tr><tr><td align=\"left\">8016</td><td align=\"left\">0.4</td><td align=\"left\">0.84</td><td align=\"left\">0.75</td></tr><tr><td align=\"left\">10521</td><td align=\"left\">0.34</td><td align=\"left\">0.73</td><td align=\"left\">0.54</td></tr><tr><td align=\"left\">14028</td><td align=\"left\">0.25</td><td align=\"left\">0.91</td><td align=\"left\">0.44</td></tr><tr><td align=\"left\">16533</td><td align=\"left\">0.18</td><td align=\"left\">0.89</td><td align=\"left\">0.34</td></tr><tr><td align=\"left\">17535</td><td align=\"left\">0.15</td><td align=\"left\">0.87</td><td align=\"left\">0.48</td></tr><tr><td align=\"left\">19038</td><td align=\"left\">0.11</td><td align=\"left\">0.92</td><td align=\"left\">0.24</td></tr><tr><td align=\"left\">22044</td><td align=\"left\">0.1</td><td align=\"left\">0.87</td><td align=\"left\">0.29</td></tr><tr><td align=\"left\">33066</td><td align=\"left\">0.13</td><td align=\"left\">0.83</td><td align=\"left\">0.43</td></tr><tr><td align=\"left\">63126</td><td align=\"left\">0.09</td><td align=\"left\">0.93</td><td align=\"left\">0.28</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\">Avg:</td><td align=\"left\">0.18</td><td align=\"left\">0.86</td><td align=\"left\">0.41</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Sensitivity and specificity for the 10 largest and smallest datasets from a Weeder run over all of RegulonDB at 200 bp.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\" colspan=\"4\">10 Smallest Motif Datasets</td></tr><tr><td align=\"left\">Size</td><td align=\"left\">nSn</td><td align=\"left\">nSp</td><td align=\"left\">sSp</td></tr></thead><tbody><tr><td align=\"left\">603</td><td align=\"left\">0.27</td><td align=\"left\">0.65</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\">603</td><td align=\"left\">0.4</td><td align=\"left\">0.76</td><td align=\"left\">1</td></tr><tr><td align=\"left\">603</td><td align=\"left\">0.06</td><td align=\"left\">0.91</td><td align=\"left\">0.33</td></tr><tr><td align=\"left\">603</td><td align=\"left\">0.06</td><td align=\"left\">0.79</td><td align=\"left\">0.14</td></tr><tr><td align=\"left\">603</td><td align=\"left\">0.19</td><td align=\"left\">0.85</td><td align=\"left\">0.5</td></tr><tr><td align=\"left\">603</td><td align=\"left\">0.18</td><td align=\"left\">0.81</td><td align=\"left\">0.33</td></tr><tr><td align=\"left\">804</td><td align=\"left\">0.47</td><td align=\"left\">0.66</td><td align=\"left\">0.88</td></tr><tr><td align=\"left\">804</td><td align=\"left\">0.27</td><td align=\"left\">0.84</td><td align=\"left\">0.57</td></tr><tr><td align=\"left\">804</td><td align=\"left\">0.08</td><td align=\"left\">0.88</td><td align=\"left\">1</td></tr><tr><td align=\"left\">804</td><td align=\"left\">0.25</td><td align=\"left\">0.86</td><td align=\"left\">0.5</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\">Avg:</td><td align=\"left\">0.22</td><td align=\"left\">0.80</td><td align=\"left\">0.58</td></tr><tr><td align=\"center\" colspan=\"4\">10 Largest Motif Datasets</td></tr><tr><td align=\"left\">Size</td><td align=\"left\">nSn</td><td align=\"left\">nSp</td><td align=\"left\">sSp</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\">3015</td><td align=\"left\">0.3</td><td align=\"left\">0.67</td><td align=\"left\">0.65</td></tr><tr><td align=\"left\">3216</td><td align=\"left\">0.47</td><td align=\"left\">0.84</td><td align=\"left\">0.81</td></tr><tr><td align=\"left\">4221</td><td align=\"left\">0.43</td><td align=\"left\">0.87</td><td align=\"left\">0.64</td></tr><tr><td align=\"left\">5628</td><td align=\"left\">0.31</td><td align=\"left\">0.82</td><td align=\"left\">0.61</td></tr><tr><td align=\"left\">6633</td><td align=\"left\">0.24</td><td align=\"left\">0.81</td><td align=\"left\">0.46</td></tr><tr><td align=\"left\">7035</td><td align=\"left\">0.28</td><td align=\"left\">0.78</td><td align=\"left\">0.82</td></tr><tr><td align=\"left\">7638</td><td align=\"left\">0.25</td><td align=\"left\">0.77</td><td align=\"left\">0.54</td></tr><tr><td align=\"left\">8844</td><td align=\"left\">0.42</td><td align=\"left\">0.74</td><td align=\"left\">0.72</td></tr><tr><td align=\"left\">13266</td><td align=\"left\">0.15</td><td align=\"left\">0.89</td><td align=\"left\">0.37</td></tr><tr><td align=\"left\">25326</td><td align=\"left\">0.16</td><td align=\"left\">0.91</td><td align=\"left\">0.42</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"left\">Avg:</td><td align=\"left\">0.30</td><td align=\"left\">0.81</td><td align=\"left\">0.60</td></tr></tbody></table></table-wrap>" ]

[ "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1471-2105-9-S9-S6-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>B</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>,</mml:mo><mml:mi>u</mml:mi><mml:mo>,</mml:mo><mml:mi>v</mml:mi></mml:mrow><mml:mi>u</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1471-2105-9-S9-S6-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>B</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mn>1</mml:mn><mml:mo>,</mml:mo><mml:mi>u</mml:mi><mml:mn>1</mml:mn><mml:mo>,</mml:mo><mml:mi>v</mml:mi><mml:mn>1</mml:mn></mml:mrow><mml:mi>u</mml:mi></mml:msubsup><mml:mo>,</mml:mo><mml:msubsup><mml:mi>B</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mn>1</mml:mn><mml:mo>,</mml:mo><mml:mi>u</mml:mi><mml:mn>2</mml:mn><mml:mo>,</mml:mo><mml:mi>v</mml:mi><mml:mn>2</mml:mn><mml:mo>,</mml:mo><mml:mn>...</mml:mn></mml:mrow><mml:mi>u</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1471-2105-9-S9-S6-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msubsup><mml:mi>B</mml:mi><mml:mrow><mml:mi>k</mml:mi><mml:mo>,</mml:mo><mml:mi>u</mml:mi><mml:mo>,</mml:mo><mml:mi>v</mml:mi></mml:mrow><mml:mrow><mml:msub><mml:mi>u</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:msubsup></mml:mrow></mml:semantics></mml:math></inline-formula>" ]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1471-2105-9-S9-S6-1\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-2\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-3\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-4\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-5\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-6\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-7\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-8\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-9\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-10\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-11\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-12\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-13\"/>", "<graphic xlink:href=\"1471-2105-9-S9-S6-14\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>RNA interference (RNAi) in <italic>Drosophila</italic> cell cultures is a powerful tool for the identification of proteins involved in mitotic cell division. The addition of a double stranded RNA (dsRNA) to the cell medium leads to rapid downregulation of the corresponding mitotic protein, resulting in a specific and penetrant phenotype ##REF##12134082##[1]##–##REF##17412918##[10]##. Identification of mitotic genes/proteins by RNAi has thus far relied on two general approaches. The first involved genome-wide screens to detect gross changes in cell and nuclear morphology ##REF##14527345##[3]##,##REF##14764878##[6]##,##REF##15547975##[8]##, defects in cytokinesis ##REF##15380073##[7]##,##REF##15547975##[8]## or in spindle and centrosome structure ##REF##17412918##[10]##. Most of these screens were performed using automated microscopy ##REF##14527345##[3]##,##REF##14764878##[6]##,##REF##15547975##[8]## or the visual analysis of a very simple phenotype ##REF##15380073##[7]##. In a second approach, RNAi experiments were performed on selected gene groups, such as those encoding kinesins, actin-binding proteins, kinases or phosphatases ##REF##12975346##[2]##,##REF##12975351##[4]##,##REF##15616552##[5]##,##REF##17306545##[9]##. Cells depleted for these proteins were examined by standard fluorescence microscopy that allowed detection of a wide spectrum of mitotic abnormalities.</p>", "<p>Although genome-wide and gene-specific approaches have identified many mitotic functions, the inventory of such proteins is likely to be largely incomplete. For example, RNAi has never been used to detect genes involved in establishing proper mitotic chromosome morphology or required to maintain mitotic chromosome integrity. We present here a novel approach for the identification of mitotic proteins by RNAi. Using a co-expression-based bioinformatic procedure, we generated a list of 1000 <italic>Drosophila</italic> genes highly enriched in mitotic functions. We then performed RNAi experiments for each of these 1000 genes, and examined both mitotic chromosome structure and spindle morphology in the treated cells. This screen has led to the identification of 142 mitotic genes, 70 of which have not been previously implicated in mitosis.</p>" ]

[ "<title>Methods</title>", "<title>Generation of Co-Expression Lists</title>", "<p>Co-expression analysis was perfomed on a previously described gene expression dataset ##REF##12144710##[14]##. This comprised 267 GeneChip <italic>Drosophila</italic> Genome Arrays (Affymetrix, Santa Clara, CA, USA) that covered 89 different embryonic and adult experimental conditions and contained expression data for 13,166 probesets (corresponding to 12,229 genes in the current FlyBase release). Pearson correlation coefficients (PCC) were calculated on log₂ transformed expression values, averaged for each experimental condition ##REF##12144710##[14]##. For each of the six prototype mitotic genes, we obtained a ranked coexpression list by calculating the PCC of the corresponding probeset with all the other probesets of the gene expression matrix. In the case of two or more probesets referring to the same gene, we considered only those showing the highest ranks. To obtain a ranked consensus co-expression list, we first scored every probeset for its presence in the upper 3000 ranks of the single co-expression lists and calculated the average PCC; we then ordered the probesets for decreasing values of these parameters (##SUPPL##16##Table S1##).</p>", "<p>The first 3,000 genes in this consensus co-expression list are reported in ##SUPPL##16##Table S1##. 56 of the putative genes in the first 1000 positions of the list were either too small (less than 300 bp) or otherwise could not be amplified by PCR using at least two different pairs of primers. These putative genes (highlighted in grey in ##SUPPL##16##Table S1##) were not assayed in our RNAi screen. Thus, to total 1000 genes, we performed RNAi for 56 additional genes that ranked from 1001 to 1056 on the consensus coexpression list (##SUPPL##16##Table S1##).</p>", "<title>Validation of the Coxpression Lists</title>", "<p>To ascertain the predictive value of our coexpression lists, we determined the ranks of 164 known mitotic genes in each list. These genes were selected because their ablation or knockdown by either mutation or RNAi has previously been shown to result in a strong mitotic phenotype. These 164 genes represent most of the <italic>Drosophila</italic> mitotic genes so far identified, but they do not include new mitotic genes detected in a recent RNAi screen performed by Goshima et al. ##REF##17412918##[10]## (see ##SUPPL##23##Text S1## and ##SUPPL##22##Table S7## for a comparison with this screen). As shown in ##SUPPL##17##Tables S2## and ##SUPPL##18##S3##, 46% of these 164 genes are included in the first 1000 genes of our consensus coexpression list, which contains in total more than 13,000 genes. This non-random clustering indicates that there is both a strong tendency for mitotic genes to be transcriptionally coexpressed, as well as a strong positive correlation between the rank of a gene in the list and the probability of its involvement in mitosis. Examination of ##SUPPL##17##Tables S2## and ##SUPPL##18##S3## shows that the mitotic genes that are most tightly coexpressed are those required for proper chromosome structure and/or condensation. In contrast, the genes implicated in cytokinesis exhibit the broadest variation in expression patterns. This variation might reflect the complexity of the cytokinetic process, which requires several functions that are not specific for mitosis. For example, functions involved in regulation of the actin cytoskeleton or membrane trafficking are likely to be required in many cellular processes in addition to cell division.</p>", "<title>dsRNAs used for RNAi</title>", "<p>Recent work has raised the issue of possible off-target effects (OTEs) associated with the use of long dsRNAs for RNAi screens. It has been suggested that short homology stretches of 19 base pairs (bp) within a long dsRNA can target a gene other than the intended one ##REF##16964256##[24]##,##REF##16990807##[59]##,##REF##16496002##[60]##. In designing the primers for dsRNA synthesis, we consistently tried to obtain RNAs longer than 600 bp with a minimum content of OT sequences. The average length of the RNAs used in our screen was 750 bp; none of these RNAs carried strings of CAR triplets that are known to result in OTEs ##REF##16964239##[25]##, but 54% of our dsRNAs contained at least one OT sequence (##SUPPL##19##Table S4##; OT sequences were identified using the program developed by Flockhart et al. ##REF##16381918##[61]##: <ext-link ext-link-type=\"uri\" xlink:href=\"http://flyrnai.org/RNAi_find_frag_free.html\">http://flyrnai.org/RNAi_find_frag_free.html</ext-link>). However, we have several reasons to believe that very few, if any, of these OT sequences contributed to the production of the observed phenotypes. First, except for the “no dividing cells” (NDC) phenotype, we looked at very specific, strong and reproducible mitotic phenotypes, which are unlikely to be elicited by an OT sequence highly diluted by the excess of sequences homologous to the target gene. Second, we never found homology between any OT sequences present in the genes of the same phenocluster. Third, we found that where the data are available, the mitotic phenotypes observed in our screen match those seen in fly mutants and/or in previous RNAi experiments, regardless of whether OT sequences were present in the dsRNAs (31% of the dsRNAs that resulted in expected phenotype contained OT sequences). Finally, examination of 100 randomly chosen RNAs that did not elicit mitotic phenotypes showed that 40% of them contain OT sequences.</p>", "<p>Individual gene sequences were amplified by PCR from a pool of cDNAs obtained from 5 different libraries: 4 embryonic libraries from 0–4, 4–8, 8–12 and 12–24 hr embryos; and an imaginal disc library, all kindly provided by N. Brown ##REF##3199441##[62]##. If this cDNA pool did not provide the desired PCR product, DNA was amplified from genomic DNA. The primers used in the PCR reactions were 35 nt long and all contained a 5′ T7 RNA polymerase binding site (<named-content content-type=\"gene\">5′-TAATACGACTCACTATAGGGAGG-3′</named-content>) joined to a gene-specific sequence. The primers used to amplify the 155 mitotic genes detected in the screen are reported in ##SUPPL##19##Table S4##. dsRNA synthesis and analysis were performed as previously described ##REF##12134082##[1]##.</p>", "<title>Cell cultures and RNAi Treatments</title>", "<p>S2 cells were cultured at 25°C in Shields and Sang M3 medium (Sigma) supplemented with 10% heat-inactivated fetal bovine serum (FBS, Invitrogen). RNAi treatments were carried out according to Somma et al. ##REF##12134082##[1]##. 1×10⁶ cells were plated in 1 ml of serum-free medium in a well of a six-well culture dish (Sarstedt). Each culture was inoculated with 15 µg of dsRNA. After a 1 hr incubation at 25°C, 2 ml of medium supplemented with 15% FBS were added to each culture. Control cultures were prepared in the same way but without addition of dsRNA. Both RNA-treated and control cells were grown for 96 hr at 25°C, and then processed for cytological analyses.</p>", "<title>Cytological Procedures</title>", "<p>RNAi treatments were performed using six-well plates. In a typical experiment, cells in 16 wells from three plates were treated with dsRNAs, while cells in the remaining two wells served as control. After four-day incubation with dsRNA, cells from 3 ml cultures were resuspended and processed in two ways. 2 ml of this suspension were centrifuged at 800 g for 5 min, washed in 10 ml PBS, and fixed for 7 min in 3 ml 3.7% formaldehyde in PBS. Fixed cells were spun down by centrifugation, resuspended in 500 µl PBS, and cytocentrifuged onto a clean slide using a Shandon cytocentrifuge at 900 rpm for 4 min. The slides were immersed in liquid nitrogen, washed in PBS, and incubated in PBT (PBS+0.1% TritonX-100) for 15 min, and then in PBS containing 3% BSA for 20 min. These preparations were immunostained using the following antibodies, all diluted 1∶100 in PBS: anti-α tubulin monoclonal DM1A (Sigma), rabbit anti-ZW10 ##REF##1339459##[20]##, rabbit anti-cyclin B and anti-CENP-C (gifts of Christian Lehner, University of Bayreuth, Germany), and chicken anti-CID ##REF##11483958##[18]##. These primary antibodies were detected by incubation for 1 hr with FITC-conjugated anti-mouse IgG and Cy3-conjugated anti-rabbit IgG (Jackson Laboratories). All slides were mounted in Vectashield with DAPI (Vector) to stain DNA and reduce fluorescence fading.</p>", "<p>To obtain metaphase chromosome preparations, 1 ml of cell suspension was left in its well and treated for 2 hr with colchicine (final concentration 10⁻⁵ M). Colchicine-treated cells were then centrifuged at 800 g for 5 min. Pelleted cells were washed in 10 ml PBS, spun down by centrifugation and resuspended in 5 ml hypotonic solution (0.5 M Na citrate) for 7 min. After further centrifugation, pelleted cells were fixed in 5 ml of methanol: acetic acid (3∶1), spun down again, and resuspended in the small volume of fixative left after the removal of supernatant. 10 µl of this suspension was dropped onto a microscope slide and air-dried. All slides were mounted in Vectashield with DAPI (Vector) to stain DNA.</p>", "<p>All images were captured using a CoolSnap HQ CCD camera (Photometrics; Tucson, AZ) connected to a Zeiss Axioplan fluorescence microscope equipped with an HBO 100W mercury lamp as described previously ##REF##16431372##[63]##. Gray scale digital images were collected separately, converted to Photoshop format, pseudocolored, and merged.</p>", "<title>Phenotypic Analysis</title>", "<p>We considered 18 major phenotypic traits indicated in the headings of ##FIG##1##Figures 2## and ##FIG##2##3##, and ##SUPPL##20##Table S5##. In addition, we observed 13 relatively rare phenotypes that are reported under the OSD (other spindle defects) and OMD (other mitotic defects) headings of ##FIG##1##Figures 2## and ##FIG##2##3##, and ##SUPPL##20##Table S5##. Each individual phenotypic trait was considered as genuine when its frequency in a dsRNA-treated sample was significantly higher than the frequency of that trait in controls with p&lt;0.001, using the χ2 contingency test. The same trait was considered strong when its frequency was at least threefold (in the case of chromosome aberrations) or fivefold (in all the other cases) the control frequency. A phenotypic trait was considered weak when its frequency was significantly different from the control (with p&lt;0.001), but below the above thresholds.</p>", "<p>Cells from control wells without dsRNA were systematically compared with cells treated with dsRNAs that do not elicit mitotic phenotypes. We never observed phenotypic differences between any of these cells, indicating that none of the aberrant traits we detected was due simply to the addition of random dsRNA sequences to the culture.</p>", "<p>Most of the mitotic genes identified in our screen are highly conserved and have putative human orthologs. The known function of these <italic>Drosophila</italic> genes and their human counterparts are reported in ##SUPPL##21##Table S6##, together with the supporting references.</p>" ]

[ "<title>Results</title>", "<title>Construction of a Mitotic-Gene-Enriched Co-Expression List</title>", "<p>Numerous studies indicate that genes involved in the same biological process tend to be transcriptionally co-expressed (see, for example,##REF##9843981##[11]##–##REF##15588312##[13]##. We thus exploited extant microarray data ##REF##12144710##[14]## to rank the complete set of annotated <italic>Drosophila</italic> genes according to their co-expression with six well-characterized genes representative of different aspects of mitosis: <italic>gluon</italic> (<italic>glu</italic>) encodes a condensin ##REF##11267866##[15]##; <italic>ida/APC5</italic>, specifies a subunit of the anaphase promoting complex (APC/C ##REF##11870214##[16]##); <italic>cid/CenpA</italic> is the gene for the centromere-specific histone H3 variant required for kinetochore assembly ##REF##10639145##[17]##,##REF##11483958##[18]##; <italic>Eb1</italic> encodes a microtubule (MT)-associated protein required for spindle assembly ##REF##12213835##[19]##; <italic>zw10</italic> specifies a component of the RZZ complex that helps target cytoplasmic dynein to the kinetochore and that is involved in the spindle checkpoint ##REF##1339459##[20]##; and finally <italic>sti</italic> (<italic>Citron kinase</italic>) encodes a serine/threonin protein kinase required for the completion of cytokinesis ##REF##15371536##[21]##–##REF##15459099##[23]##. Using the Pearson correlation coefficient, the expression of these prototype genes was correlated with the expression levels of most <italic>Drosophila</italic> genes across 89 different microrray experiments ##REF##12144710##[14]##. The 13,166 probesets contained in this dataset were separately ranked for their co-expression with each prototype gene. We then generated a ranked consensus co-expression list by combining the six gene-specific lists (##SUPPL##16##Table S1##).</p>", "<p>To validate our bioinformatic approach, we determined the rank in the consensus co-expression lists of 164 <italic>Drosophila</italic> mitotic genes; these genes represent most of the <italic>Drosophila</italic> mitotic genes so far identified but do not include the genes identified in the recent genome-wide screen performed by Goshima et al. ##REF##17412918##[10]##. As shown in ##FIG##0##Figure 1A## and ##SUPPL##17##Tables S2## and ##SUPPL##18##S3##, the first 1000 genes of our consensus co-expression list include 46% of the 164 known mitotic genes. This implies that the first 1000 genes of the same list should contain roughly half of all mitotic genes, including those that are currently unknown.</p>", "<title>Identification of Genes Involved in Mitotic Division</title>", "<p>To identify new mitotic genes, we synthesized a dsRNA for each of the first 1000 genes in our consensus co-expression list. In designing the primers for such RNAs, we minimized gene overlap to avoid off-target effects of dsRNAs ##REF##16964256##[24]##,##REF##16964239##[25]## (##SUPPL##19##Table S4##). Each dsRNA was then added to S2 cells grown in 3 milliliters (ml) of culture medium. After a 96 h treatment with dsRNA, the cells were split into two aliquots. 2 ml were fixed with formaldehyde and then stained for both tubulin and DNA. The resulting preparations were then blindly scored by at least two independent observers for abnormalities in spindle morphology and chromosome segregation. The remaining 1 ml of cell suspension was incubated with colchicine for two hours, hypotonically swollen, fixed with methanol/acetic acid, and stained with DAPI. The metaphase chromosomes obtained in this way were then blindly examined by at least two observers for abnormalities in chromosome structure and/or the presence of chromosome aberrations. We performed two independent RNAi experiments for each gene. In most of these experiments, we examined 50 colchicine-arrested metaphases and 50 tubulin-stained mitotic figures. If the results of the two experiments were significantly different, we performed additional experiments to define the RNAi phenotype. An RNAi phenotype was considered positive only when the frequency of affected cells was significantly different from controls with p&lt;0.001 (using the χ² contingency test; see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref>).</p>", "<p>Our screen identified 155 genes whose inactivation by RNAi causes a strong mitotic phenotype. Based on phenotypic analysis, these genes can be grouped in seven broad categories that we further subdivided in 18 phenoclusters (PHCs) ##REF##11778048##[26]##: 13 genes required for progression through the cell cycle, identified by dsRNAs that result in a complete (or nearly complete) absence of mitotic figures (PHC: NM); 44 genes required for chromosome integrity, identified by dsRNAs that cause chromosome aberrations (PHC: CA); 11 genes required for proper mitotic chromosome condensation (PHCs: CC1–CC3); 41 genes required for regular chromosome segregation (PHCs: CS1–CS5); 33 genes required for spindle assembly (PHCs: SA1–SA4); 7 genes required for cytokinesis (PHCs: CY1 and CY2); and 6 genes required for multiple mitotic functions (PHCs: SC1 and SC2) (##FIG##0##Figures 1B##, ##FIG##1##2##, ##FIG##2##3##, and ##FIG##3##4C##; ##SUPPL##20##Table S5##; a synopsis on the functions of these genes can be found in ##SUPPL##21##Table S6##). Remarkably, the distribution of these mitotic genes in our co-expression list was clearly nonrandom: their frequency decreased with an increase of their rank, further validating our co-expression approach (##FIG##0##Figure 1C##).</p>", "<title>Genes Required for Progression through the Cell Cycle</title>", "<p>We identified 13 dsRNAs that result in the absence of dividing cells at 96 h after treatment initiation (##SUPPL##20##Table S5##). Six of these genes (<italic>cdc2c</italic>, <italic>cyclinA</italic>, <italic>cyclinE</italic>, <italic>geminin</italic>, <italic>ran</italic> and <italic>string</italic>) are well-known cell-cycle regulators. Two genes, <italic>RpII140 and RpII215</italic>, encode the 140 and 215 kDa subunits of RNA polymerase II, respectively. Three genes are involved in RNA metabolism and encode either canonical splicing factors (Prp8 and SF1) or the small DebB ribonucleoprotein, which is also likely to be involved in RNA splicing. Defects in <italic>PRPF8</italic>, the human homolog of <italic>Prp8</italic>, are one cause of Retinitis pigmentosa. Of the remaining two genes, <italic>CG9273</italic> encodes a protein with similarity to a subunit of DNA replication factor A, and <italic>Bx42</italic> specifies a protein involved in Notch signal transduction (##SUPPL##21##Table S6##).</p>", "<title>Genes Required for Chromosome Integrity</title>", "<p>Although the 1000 genes in our list were selected for their co-expression with mitotic genes, our screen uncovered several functions required for chromosome integrity (##FIG##0##Figures 1## and ##FIG##3##4##; ##SUPPL##20##Tables S5## and ##SUPPL##21##S6##). As shown in ##FIG##3##Figure 4##, we found 44 dsRNAs that significantly increase the frequency of spontaneous chromosome aberrations. Of the 44 genes identified by these dsRNAs, 38 have apparent human orthologs (##FIG##3##Figure 4C##) yet only 4 have previously been implicated in the maintenance of chromosome integrity (##SUPPL##21##Table S6##). These genes can be subdivided in several broad classes, based on their putative functions: (1) genes required for DNA replication, including <italic>Ribonucleotide reductase (RnrS)</italic>, <italic>DNA primase (DNAprim)</italic>, <italic>DNA polymerase alpha (DNApol)</italic>, <italic>Orc5</italic>, <italic>RfC40</italic>, <italic>Rpa70</italic>/<italic>RPA1</italic> and <italic>peterpan (ppan)</italic>; (2) genes involved in both DNA replication and repair, such as <italic>mus209</italic>/<italic>PCNA</italic>, <italic>cul-4</italic>, <italic>thymidilate synthetase (Ts)</italic> and <italic>CG6854/CTP synthase</italic>; (3) genes that mediate different aspects of DNA repair but are not known to participate in DNA replication, such as <italic>DDB1</italic>, <italic>okra/RAD54L</italic>, <italic>CG6197/XAB2</italic> and <italic>CG7003/MSH6</italic>; (4) genes involved in transcription and RNA maturation, including <italic>Dp/TFPD2</italic>, <italic>CG10354/XRN2</italic>, <italic>Taf6</italic>, <italic>l(2)NC136/CNOT3</italic>, <italic>noi/SF3A3</italic>, <italic>CG7757/PRPF3)</italic>, <italic>without children (woc)</italic>, <italic>CG6480/FRG1</italic> and <italic>CG6686/SART1</italic> (##SUPPL##21##Table S6##). Chromosome aberrations were also induced by RNAi against 8 genes whose diverse functions are not easily classified into the four groupings above. These include <italic>BEAF-32</italic>, that encodes a chromatin insulator factor; <italic>dnk</italic>, that specifies a deoxyribonucleotide kinase similar to the human mitochondrial kinase TK2; <italic>Su(var)2-10</italic>, whose product is an E3 SUMO ligase; the <italic>H3.3B</italic> histone variant gene; <italic>SMC1</italic>, that encodes a conserved cohesin involved in the Cornelia de Lange syndrome in humans; <italic>Dcp-1</italic>, that specifies a caspase precursor; <italic>Megator (Mtor)</italic> that encodes a component of the putative spindle matrix; and <italic>CG17446</italic>, whose product is homologous to a subunit of the mammalian Set1 histone methyltransferase complex (##SUPPL##21##Table S6##). Our screen also identified 12 chromosome stability genes without any assigned putative functions, 7 of which are conserved in humans. Together, these results indicate that the maintenance of chromosome stability requires a large number of functions, many of which remain to be identified.</p>", "<p>The analysis of tubulin-stained mitotic figures revealed an interesting phenotype associated with the presence of chromosome aberrations. We observed many metaphase figures with the centric portions of broken chromosomes aligned at the metaphase plate and the acentric fragments near the cell poles (##FIG##3##Figure 4B##). Immunostaining for the kinetochore marker Cenp-C ##REF##16140985##[27]## verified that most chromosome fragments at the poles of these metaphases were indeed devoid of centromere (##FIG##3##Figure 4B##). This phenotype suggests that chromosome fragments severed from their kinetochores are transported to the cell poles. A similar phenomenon has been observed in plants <italic>(Hemanthus)</italic> and in crane fly spermatocytes. In both systems, when a metacentric chromosome is cut with the laser, the resultant acentric fragment moves to the closest cell pole at the same velocity as anaphase chromosomes ##REF##8601587##[28]##,##REF##11739800##[29]##. To explain this phenomenon, it has been suggested that the acentric chromosomes fragments adhere to the lateral surfaces or plus ends of microtubules and are transported poleward by the microtubule flux ##REF##11739800##[29]##. We believe that this mechanism also occurs in <italic>Drosophila</italic> S2 cells. Strong support for this view comes from observations on RPA70-depleted cells, which exhibit extreme chromosome fragmentation but form regular spindles. In these cells, most acentric fragments accumulate at the poles of ana/telophase figures, suggesting that they are driven poleward by microtubule-based forces (##FIG##3##Figure 4B##).</p>", "<title>Genes Required for Accurate Chromosome Segregation</title>", "<p>We identified 41 genes required for regular chromosome segregation. These genes are not required for spindle formation, as cell depleted for their products do not exhibit defects in late prophase/early prometaphase spindles. However, metaphase and ana/telophase spindles are often highly abnormal with respect to spindle morphology and the distribution of chromosomes along the spindle. The genes required for chromosome segregation (CS) can be subdivided into five phenoclusters (CS1, CS2, CS3, CS4 and CS5) based upon differences and similarities in the RNAi phenotypes (##FIG##1##Figure 2##). The CS1 group includes only the <italic>doubleparked</italic> (<italic>dup</italic>) gene. In most <italic>dup</italic> RNAi metaphase-like figures, the chromosomes are not replicated and have the appearance of single chromatid (##FIG##4##Figure 5B##). This is likely to results in a merotelic attachment of the spindle fibers to the kinetochore, leading to an impairment of chromosome movement during anaphase (##FIG##4##Figure 5B##). This phenotype has been previously observed in embryonic cells of <italic>dup</italic> mutants, suggesting that <italic>dup</italic> is required for both DNA replication and the checkpoint that prevents mitosis until completion of S-phase ##REF##10898791##[30]##,##REF##12699620##[31]##. RNAi for the 5 genes in the CS2 group resulted in precocious sister chromatid separation, lack of chromosome congression to the cell equator at metaphase, and unequal or otherwise abnormal sister chromatid segregation (##FIG##4##Figures 5C## and ##SUPPL##0##S1##). Four of the genes included in this CS2 phenocluster (<italic>bub1</italic>, <italic>bub3</italic>, <italic>zw10</italic> and <italic>rod</italic>) are well known components of the spindle checkpoint machinery (##SUPPL##21##Table S6##). The other gene, <italic>dalmatian</italic> (<italic>dmt</italic>) has never been implicated in this checkpoint. However, since studies in <italic>C. elegans</italic> have clearly shown that genes with similar RNAi phenotypes are often required for a common process ##REF##11778048##[26]##, ##REF##12445391##[32]##–##REF##16094371##[34]##, we propose that <italic>dmt</italic> might play a role in the spindle checkpoint.</p>", "<p>Inactivation of the 18 genes in the CS3 phenocluster (##FIG##1##Figure 2##) resulted in a peculiar mitotic phenotype. The chromosomes of metaphase-like figures were not connected to the spindle poles by bundles of kinetochore microtubules (MTs) and thus never congressed to the equator of the spindle. In addition to metaphase-like spindles, the RNAi cells of the CS3 phenocluster also showed many elongated ana/telophase spindles. However, these spindles contained chromosomes with unseparated sisters chromatids; these chromosomes usually appeared to segregate to the poles at random (##FIG##5##Figures 6##, ##FIG##6##7## and S2–S4). Some of these peculiar ana/telophase-like figures displayed both a central spindle and an actin-based contractile ring (##SUPPL##4##Figure S5##). However, most of these structures were morphologically irregular and were thus probably unable to mediate cytokinesis.</p>", "<p>To define their phenotype in more detail, RNAi cells for the CS3 genes were stained for the checkpoint proteins ZW10 and BubR1 and also for the cell cycle marker Cyclin B. In most ana/telophase-like figures, Cyclin B was still high, whereas in control cells it was degraded during anaphase and absent from telophases (##FIG##5##Figure 6B##). In the metaphase-like RNAi figures, ZW10 did not exhibit any streaming towards the cell poles as occurs in normal metaphases (##FIG##6##Figure 7A##), consistent with a defect in microtubule attachments to the kinetochore ##REF##12034769##[35]##. Moreover, the ana/telophase-like figures showed strong ZW10 and BubR1 centromeric signals; these signals were mostly absent from control ana/telophase chromosomes (##FIG##6##Figure 7A## and data not shown). Finally, the chromosomes of the ana/telophase-like cells displayed two centromeric spots after staining for the kinetochore marker Cenp-C (##FIG##6##Figure 7B##). These findings confirm that the chromosomes at the poles of the ana/telophase spindles seen in the CS3 phenocluster are indeed comprised of both sister chromatids.</p>", "<p>The CS3 phenocluster includes the <italic>CG9938/Hec1/Ndc80</italic>, <italic>CG8902/Nuf2</italic> and <italic>CG1558/Nsl1</italic> genes, which encode interacting components of the <italic>Drosophila</italic> kinetochore ##REF##17333235##[36]##,##REF##17534428##[37]##, as well as <italic>cid/Cenp-A</italic>, that encodes the <italic>Drosophila</italic> centromere-specific histone H3 variant ##REF##10639145##[17]##,##REF##11483958##[18]##. Of the remaining 14 genes in the CS3 group, one specifies a conserved product of unknown function <italic>(CG8233)</italic> and 13 encode highly conserved splicing factors (##SUPPL##21##Table S6##). The RNAi phenotypes of the genes in the CS3 phenocluster suggest that their products are required for proper kinetochore-microtubule interactions. We propose that in the absence of these interactions, the spindle checkpoint remains engaged and sister chromatid separation does not occur. The high levels of Cyclin B and the lack of ZW10 streaming in CS3 RNAi cells are both consistent with this hypothesis ##REF##17895365##[38]##. We further posit that the chromosomes are driven to the spindle poles by the same forces that act on the acentric chromosome fragments. As the chromosomes move towards the poles, the spindle elongates so as to resemble an ana/telophase spindle; some of these spindles manage to assemble a defective central spindle and attempt to undergo cytokinesis. Collectively, these results provocatively indicate that in S2 cells typical telophase events, such as central spindle assembly and initiation of cytokinesis, can occur in the absence of sister chromatid separation.</p>", "<p>RNAi for the 9 genes in the CS4 group resulted in a pseudo metaphase-arrest phenotype (##FIG##1##Figures 2##, ##FIG##7##8A##, and ##SUPPL##5##Figure S6##). Most dsRNA-treated cells with spindles of metaphase shape displayed apparently normal kinetochore fibers and normal chromosome congression. However, we also observed many mitotic figures with elongated ana/telophase-like spindles and unsegregated chromosomes at the center of the cell. In these peculiar mitotic figures, the centromeres of most chromosomes had congressed to the middle of the spindle, while the chromosomes arms were parallel to the spindle axis with the telomeres pointing towards the spindle poles. In addition, in many cells with long telophase-like spindles, the chromosomes stuck at the cell equator displayed variable degrees of decondensation, as through they were undergoing the decondensation process that occurs during normal telophase (##FIG##7##Figure 8A## and ##SUPPL##5##Figure S6##). Finally, in most ana/telophase-like figures, Cyclin B remained high, as observed in RNAi cells for the CS3 genes (data not shown).</p>", "<p>The CS4 phenocluster includes the <italic>Separase</italic> and <italic>three rows</italic> (<italic>thr</italic>) genes, which encode interacting proteins required for sister chromatid separation at anaphase (##FIG##1##Figure 2## and ##SUPPL##21##Table S6##). Previous studies have shown that embryonic cells of <italic>thr</italic> mutants display metaphase arrest with congressed chromosomes, followed by an irregular extension of the spindle without chromosome segregation and by chromosome decondensation ##REF##8305737##[39]##,##REF##8270646##[40]##. This phenotype is fully comparable to that we observed in S2 cells after <italic>thr</italic> downregulation by RNAi. The CS4 group also includes the <italic>CyclinB</italic> gene and <italic>Otefin</italic>, a gene encoding a non-conserved protein that may interact with lamin (##SUPPL##21##Table S6##). All the remaining genes in the group are involved in RNA metabolism: one specifies a putative transcription factor while the others encode conserved splicing factors (##SUPPL##21##Table S6##). The genes included in the CS4 phenocluster are likely to be required for sister chromatid separation at the anaphase onset. We propose that upon inactivation of these genes, the opposing forces exerted by the MTs attached to the sister kinetochores keep the centromeres aligned at the metaphase plate. At the same time, however, the same forces that mediate the poleward motion of acentric fragments act on the chromosome arms, orienting them parallel to the spindle axis. Our observations also suggest that the latter forces can occasionally prevail over those exerted by the kinetochore fibers, so that some chromosomes leave the metaphase plate and move towards the poles with unseparated chromatids. The finding that RNAi cells for the CS4 genes undergo spindle elongation and chromosome decondensation while arrested in a metaphase-like state provides further support for the view that in S2 cells telophase events do not require sister chromatid separation.</p>", "<p>RNAi for the 9 genes in the CS5 phenocluster resulted in defective chromosome congression at metaphase and abnormal chromosome segregation at anaphase (##FIG##1##Figure 2##). Knockdowns of the expression of most of these genes caused a partial metaphase arrest characterized by extremely contracted chromosomes. However, even though sister chromatid separation did occur in most of the cases, ana/telophases were severely defective. The segregating chromatids were highly contracted and the two chromatid sets remained close to each other in many cells (##FIG##7##Figure 8B## and ##SUPPL##6##Figure S7##). These unusual ana/telophases resemble very early anaphase figures, which are quite rare in untreated cells. These observations suggest that chromosome movement towards the poles is partially impaired in RNAi cells, resulting in delayed and irregular chromosome segregation.</p>", "<p>The CS5 phenocluster includes <italic>ida/APC5</italic> and <italic>CG11419/APC10</italic>, that encode two subunits of the APC complex; and <italic>fizzy (fzy)/Cdc20</italic>, whose product regulates APC/C activity (##SUPPL##21##Table S6##). This phenocluster also includes <italic>Pros26.4</italic>, that specifies a proteasome subunit; <italic>Klp3A</italic>, that encodes a kinesin-like protein; <italic>CG4266</italic> and <italic>kin17</italic>, whose products are conserved proteins implicated in RNA metabolism and the stress response, respectively; and <italic>CG3221</italic>, that encodes a poorly conserved product of unknown function. The finding that the phenotype elicited by depletion of the APC components is substantially different from that caused by Separase inhibition strongly suggests that the APC/C is required not only for Securin and Cyclin B degradation, but also for the regulation of other aspects of spindle dynamics and spindle-kinetochore interactions.</p>", "<p>Inactivation of the genes in the CS1–CS5 groups often resulted in very elongated ana/telophase spindles (##SUPPL##7##Figure S8##); in some cases, these spindles were twice as long as their counterparts in control cells. Long spindles were often bent or S-shaped, probably due to mechanical constraints imposed by the plasma membrane (##SUPPL##0##Figures S1##, ##SUPPL##1##S2##, ##SUPPL##2##S3##, ##SUPPL##3##S4##, ##SUPPL##4##S5##, ##SUPPL##5##S6##, ##SUPPL##6##S7##, and ##SUPPL##7##S8##). In addition, we observed that the degree of spindle elongation correlates with the presence of scattered chromosomes between the spindle poles. Long spindles have been observed previously in both <italic>Drosophila</italic> and mammalian cells with defective kinetochores ##REF##12975346##[2]##,##REF##17412918##[10]##,##REF##17534428##[37]##,##REF##10398680##[41]##, and have been attributed to a misregulation of tubulin addition at the plus ends of kinetochore MTs ##REF##12975346##[2]##,##REF##10398680##[41]##. We observed long ana/telophase-like spindles in cells containing chromosomes with either functional or nonfunctional kinetochores. Thus, spindle elongation may depend on factors other than kinetochore dysfunction. For example, the chromosomes scattered within the aberrant ana/telophase figures may induce MT growth and/or stabilization ##REF##8684481##[42]##, leading to the formation of particularly long spindles.</p>", "<title>Genes Required to Maintain Proper Chromosome Structure</title>", "<p>We identified 11 dsRNAs that cause defects in chromosome structure without affecting spindle assembly. The phenotypes produced by these RNAs can be grouped into three phenoclusters we call CC1, CC2 and CC3 (##FIG##1##Figure 2##). The CC1 group includes <italic>Minichromosome maintenance 3 (Mcm3)</italic>, <italic>Mcm7</italic> and <italic>cap</italic>. <italic>Mcm3</italic> and <italic>Mcm7</italic> encode the orthologs of two components of the human (MCM)2-7 helicase complex (##SUPPL##21##Table S6##), while <italic>Cap</italic> encodes a protein orthologous to the SMC3 cohesin whose mutant form is responsible for a mild variant of the Cornelia de Lange syndrome ##REF##17273969##[43]##. RNAi for these genes resulted in loss of sister chromatid cohesion in the heterochromatic regions of the chromosomes and defective chromosome congression and segregation (##FIG##8##Figure 9A## and ##SUPPL##8##Figure S9##). A similar phenotype was previously observed in mutants in the <italic>wings-apart like (wapl) Drosophila</italic> gene ##REF##10747063##[44]##; the human ortholog of Wapl interacts with cohesin and regulates its association with chromatin ##REF##17112726##[45]##,##REF##17113138##[46]##.</p>", "<p>The CC2 phenocluster includes 5 genes that encode well-known condensins: SMC2, Gluon/SMC4, CapD2, Cap-G and Barren/CAP-H (##FIG##1##Figure 2## and ##SUPPL##21##Table S6##). RNAi for these genes resulted in very similar phenotypes. In all cases, chromosomes displayed an abnormal mitotic condensation: although their longitudinal axis was shortened normally, their sister chromatids were swollen and fuzzy. In addition, ana/telophase figures displayed frequent lagging chromosomes and chromatin bridges, consistent with a strong defect in sister chromatid resolution during anaphase (##SUPPL##9##Figure S10##).</p>", "<p>In contrast, in RNAi cells for either gene in the CC3 group, <italic>Topoisomerase II (Top2)</italic>, <italic>greatwall (gwl)</italic> (encoding a conserved kinase; see ##SUPPL##21##Table S6##) and <italic>Orc-2</italic>, metaphase chromosomes were abnormally elongated and irregularly condensed, suggesting a defect in chromosome shortening. In RNAi cells for these genes, chromosome congression and segregation were also affected, consistent with previously published results (##FIG##8##Figure 9B##; ##SUPPL##21##Table S6##).</p>", "<title>Genes Required for Spindle Assembly</title>", "<p>RNAi for 33 genes caused defects in spindle structure that were apparent as early as prophase or the beginning of prometaphase. Most of these genes (29/33) can be grouped in three broad phenoclusters (SA1, SA2 and SA3); although the phenotypes associated with the remaining 4 genes do not resemble each other, we assign them to a single miscellaneous group (SA4) for convenience (##FIG##2##Figure 3##). Inactivation of the 18 genes in the SA1 group resulted in the formation of bipolar spindles that were significantly shorter than control spindles. Knockdowns of most of these genes also caused poorly focused spindle poles, monopolar spindles, hypercontracted chromosomes and defects in chromosome congression and segregation (##FIG##2##Figures 3##, ##FIG##9##10##; ##SUPPL##10##Figures S11## and ##SUPPL##11##S12##). The monopolar spindles observed in these RNAi cells might not reflect defective centrosome separation at prophase, but instead be a consequence of the instability of short bipolar spindles. This is suggested by previous observations of Orbit/Mast-depleted S2 cells. Live imaging of these cells has shown that the centrosomes of bipolar minispindles often collapse towards each other during prometaphase to form a monopolar spindle ##REF##16723741##[47]##. The excessive chromosome contraction is the likely outcome of a delayed progression through mitosis and could be responsible for a partial impairment of kinetochore function, resulting in defective chromosome congression and segregation.</p>", "<p>The SA1 phenocluster includes the β-tubulin gene <italic>β-tub56B</italic>, 4 genes that encode MT-interacting proteins [Map60/CP60, Eb1, Minispindles (Msps) and Mars/HURP], the mitotic kinase gene <italic>ik2</italic>, and 3 genes (<italic>CG4865</italic>, <italic>CG14781</italic> and <italic>CG17293</italic>) that encode proteins of unknown function (##SUPPL##21##Table S6##). The remaining 9 genes of the SA1 group are involved in either transcription or translation. <italic>CG8950</italic> encodes a PolII transcription factor; <italic>tho2</italic> specifies a component of the conserved THO complex, which couples splicing and mRNA export (##SUPPL##21##Table S6##). <italic>Trip1/eIF3-S2</italic>, <italic>CG8636</italic>/<italic>eIF3-S4</italic>, <italic>Int6</italic>/<italic>eIF3-S6</italic>, <italic>eIF3-p66</italic> and <italic>eIF3-S10</italic> encode different subunits of the highly conserved eukaryotic translation initiation factor 3, while <italic>Nnp-1</italic> and <italic>CG1234</italic> are involved in ribosome biogenesis or maturation (##SUPPL##21##Table S6##). The <italic>Int6</italic> gene is a frequent integration site of the MMTV virus in mouse mammary tumors and its silencing leads to mitotic defects in human cells (##SUPPL##21##Table S6##). In addition to short spindles, Int6-depleted cells displayed two unusual phenotypic traits: a severe undercondensation of the pericentric regions of the chromosomes and abnormally long astral MTs in telophase figures (##FIG##9##Figure 10A##). Horse tail-like telophase asters were also observed in <italic>msps</italic> and <italic>CG8950</italic> RNAi cells (##FIG##2##Figure 3## and ##SUPPL##10##Figure S11##).</p>", "<p>The SA2 phenocluster includes only 5 genes: <italic>CG11881</italic> and <italic>CG16969</italic>, that encode proteins of unknown function; <italic>Grip75</italic> and <italic>γ tubulin 23C</italic>, that specify components of the gamma tubulin ring complex; and <italic>NippedA</italic>, that encodes a subunit of the conserved TRRAP complex implicated both in transcriptional regulation and DNA repair (##SUPPL##21##Table S6##). Interestingly, loss of the TRRAP complex affects gene expression at mitotic stages ##REF##11544477##[48]##. Downregulation of the genes in the SA2 group results in spindles with a low MT density and poorly focused poles (##FIG##10##Figure 11## and ##SUPPL##12##Figure S13##). Aberrant spindles with low MT density have previously been observed in S2 cells depleted of gamma tubulin ring components ##REF##16476773##[49]##. RNAi cells for the SA2 genes were also defective in chromosome congression and sister chromatid separation, just as those of the CS3 phenocluster (##FIG##10##Figure 11##). This phenotypic profile suggests that the products of the SA2 genes are required for the stability of spindle MTs and for their interaction with the kinetochores.</p>", "<p>RNAi for the 6 the genes in the SA3 phenocluster resulted in anastral and poorly focused spindles with normal MT density (##FIG##11##Figure 12## and ##SUPPL##13##Figure S14##). These genes encode the DSas-4 protein required for centriole duplication; the PCM component Centrosomin (Cnn) required for MT nucleation; the <italic>CG17826</italic> product homologous to the <italic>C. elegans</italic> centrosomal protein Spd2; and Abnormal spindle (Asp), a protein that associates with both the centrosomes and the MT minus ends (##SUPPL##21##Table S6##). The other two genes in the SA3 group encode the NiPp1 inhibitor of protein phosphatase type 1 (PP1); and the <italic>CG6937</italic> product, which is homologous to the human MKI67IP protein that contains an RNA recognition motif and interacts with the Ki-67 mitotic protein (##SUPPL##21##Table S6##). The phenotypic differences between <italic>DSas-4</italic>, <italic>cnn</italic> or <italic>CG17826</italic> RNAi cells (SA3 phenocluster) and those depleted of either Dgrip75 or γ tubulin (SA2) are intriguing; they support the view that the latter proteins are not only involved in MT nucleation from the centrosomes but are also required for either MT stability or chromatin and/or kinetochore-induced MT growth ##REF##16378099##[50]##.</p>", "<p>We have included in the SA4 group 4 genes that are essential for spindle assembly but elicit different phenotypic profiles when inactivated by RNAi. Consistent with previous results, RNAi of <italic>Klp61F</italic> and <italic>ncd</italic> resulted in monopolar spindles and disorganized bipolar or multipolar spindles, respectively, while <italic>Klp67A</italic> downregulation led to abnormally long MTs that are unable to interact properly with the kinetochores (##FIG##2##Figure 3## and ##SUPPL##14##Figure S15##; ##SUPPL##21##Table S6##). The phenotype of cells treated with dsRNA for <italic>cdc2</italic> is reminiscent of that of the CS3 phenocluster, with chromosomes that remain congressed in a metaphase plate even when the spindle assumes an ana/telophase configuration. However, <italic>cdc2</italic> RNAi cells often show an additional phenotype in which the centrosomes/asters are detached from the spindle poles (##FIG##2##Figure 3## and ##SUPPL##14##Figure S15##).</p>", "<title>Genes Required for Chromosome Condensation, Spindle Formation, and/or Cytokinesis</title>", "<p>We identified 6 genes required for both chromosome condensation and spindle formation, which can be subdivided into two phenoclusters (SC1 and SC2). The SC1 phenocluster includes the three components of the chromosome passenger complex (Incenp, Ial/Aurora B and Borealin), as well as chromatin assembly factor 1 (Caf1). Consistent with previous studies, downregulation of these proteins resulted in elongated and poorly condensed chromosomes, disorganized spindles, defective chromosome congression and segregation, and frequent failures in cytokinesis (##FIG##2##Figure 3## and ##SUPPL##15##Figure S16##; ##SUPPL##21##Table S6##).</p>", "<p>The SC2 group includes Polo kinase and the <italic>Drosophila</italic> homolog of the Myb transcriptional activator (##FIG##2##Figure 3## and ##SUPPL##21##Table S6##). In <italic>polo</italic> RNAi cells, the chromosomes were fuzzy and irregularly condensed, while in Myb-depleted cells the chromosomes were overcontracted and swollen with no resolution between sister chromatids. Dowregulation of either of these two proteins disrupts MT-kinetochore interactions, leading to failures of chromosome congression and sister chromatid separation (##FIG##2##Figure 3## and ##SUPPL##15##Figure S16##).</p>", "<title>Genes Required for Cytokinesis</title>", "<p>The RNAi phenotypes of the 7 genes required for cytokinesis (##FIG##2##Figure 3##) have been described previously in greater detail (##SUPPL##21##Table S6##). They can be subdivided into two phenoclusters (CY1 and CY2). Inactivation of the genes in the CY1 group [<italic>fascetto (feo)</italic>, <italic>racGAP50</italic>, <italic>pavarotti (pav)</italic> and <italic>pebble (pbl)</italic>] results in early cytokinetic defects in both the central spindle and the contractile ring. In contrast, ablation of the CY2 genes [<italic>anillin (ani)</italic>, <italic>citron kinase (sti)</italic> and <italic>twinstar (tsr)</italic>] does not affect either central spindle or contractile ring assembly, but it does disrupt the final stages of cytokinesis (##FIG##2##Figure 3## and ##SUPPL##21##Table S6##).</p>" ]

[ "<title>Discussion</title>", "<p>Our RNAi screen for mitotic genes differs from those previously performed in two important ways. First, we used a bioinformatic approach to focus our experiments on a group of genes that was enriched in mitotic functions. Second, we analyzed potential mitotic phenotypes not only by examining cells stained for tubulin and DNA, but also by looking at colchicine-treated chromosome preparations. Since this latter technique allows the analysis of well spread metaphase chromosomes with excellent cytological resolution, we were able to identify 44 genes required to prevent spontaneous chromosome breakage, most of which have not previously been implicated in the maintenance of chromosome integrity. The human orthologs of some of these genes may play roles in carcinogenesis, as shown for many genes required for chromosome stability ##REF##9872311##[51]##,##REF##15126332##[52]##. In addition, examination of colchicine-arrested metaphases led to the detection of phenotypes such as precocious sister chromatid separation (CS2 phenocluster) and a lack of sister chromatid cohesion in the heterochromatic regions (CC1 phenocluster). These phenotypic traits allowed us to distinguish between genes required for proper chromosome segregation, permitting their assignment to different functional groups.</p>", "<p>Although previous RNAi screens were not designed to detect subtle changes in chromosome\nstructure, they identified many genes involved in spindle assembly, chromosome segregation and\ncytokinesis ##REF##12134082##[1]##–##REF##17412918##[10]##. Of particular interest is\na comparison between our screen and a recent genome-wide screen performed by Goshima and coworkers in S2 cells ##REF##17412918##[10]## (see ##SUPPL##23##Text S1## and ##SUPPL##22##Table S7## for details). Goshima <italic>et al.</italic> used automated microscopy to identify 189 genes required for spindle assembly and chromosome alignment at metaphase. Remarkably, 38% of these 189 genes are included in the first 1000 genes of our consensus co-expression list and 50% in the first 2000. We identified 98 genes involved in the same processes, 30 of which were not found in the Goshima et al. screen. However, we failed to detect 17 genes that elicited RNAi phenotypes in their screen. Together, these results further validate our co-expression-based method for the identification of mitotic genes by RNAi. We believe that elaboration of consensus co-expression lists using genes in the same phenocluster will provide many candidate genes for small-scale RNAi screens aimed at completing the inventory of proteins involved in specific mitotic processes.</p>", "<p>One striking and unanticipated finding among our results merits special attention. We identified 17 highly conserved splicing factors that are required for sister chromatid separation at anaphase. RNAi for the genes encoding these factors resulted in two types of aberrant mitotic figures. Downregulation of the genes in the CS3 phenocluster resulted in mitotic cells showing scattered chromosomes without apparent kinetochore-spindle connections. This phenotype was identical to that caused by downregulation of genes encoding well-known kinetochore proteins such as <italic>cid</italic>, <italic>CG9938/Ndc80/Hec1</italic>, <italic>CG8902/Nuf2</italic> or <italic>l(1)G023/Nsl1</italic>. In RNAi cells for genes in the CS4 phenocluster, most chromosomes showed regular connections with the spindle fibers and remained at the center of the cell, a phenotype similar to that produced by downregulation of Separase. However, RNAi for all of the CS4 genes and some of the CS3 genes produced a fraction of cells with an intermediate CS3/CS4 phenotype. These observations raise the possibility that inactivation of the splicing factor genes of both phenoclusters causes the same primary defect in centromere/kinetochore organization. One can envisage that when this defect is strong, both sister chromatid separation and kinetechore MT-interaction are affected; when the defect is weak, sister chromatid separation would be disrupted with little effect on kinetochore function.</p>", "<p>Splicing factors have previously been implicated in mitosis in fission yeast, <italic>Drosophila</italic> and human cells ##REF##17412918##[10]##, ##REF##7865880##[53]##–##REF##15616564##[55]##. However, the precise mitotic function of these splicing factors has never been described. We have clearly shown here that splicing factors are required for sister chromatid separation. However, the mechanisms by which splicing factors regulate centromere/kinetochore function remain unclear. It is possible that these factors mediate the splicing of one or more pre-mRNAs, whose protein products play crucial roles for proper centromere or spindle function. Alternatively, the splicing factors may be involved in the production and/or stabilization of spindle- or centromere-associated structural RNAs. Recent studies have shown that RNA associates with the mitotic spindle and plays a translation-independent role in spindle assembly ##REF##15851029##[56]##. Moreover, there is evidence that maize and human kinetochores are enriched in single-stranded RNAs encoded by centromeric DNA sequences. It has been suggested that these RNAs may facilitate proper assembly of centromere-specific nucleoprotein complexes ##REF##15514020##[57]##,##REF##17623812##[58]##. Deciphering the precise role of splicing factors in centromere/kinetochore assembly and functioning will be a challenging task for future studies.</p>" ]

[]

[ "<p><bold>¤a:</bold> Current address: Institut Gustave Roussy, Villejuif, France</p>", "<p><bold>¤b:</bold> Current address: MCDB Department, University of California Santa Cruz, Santa Cruz, California, United States of America</p>", "<p>Conceived and designed the experiments: MPS MG. Performed the experiments: MPS FC EB VN VDA RP AP LC MGG CP RP GC FV BF. Analyzed the data: MPS FC EB VN VDA MLG FDC MG. Wrote the paper: MPS MG. Performed the co-expression analysis: FDC.</p>", "<p>RNAi screens have, to date, identified many genes required for mitotic divisions of <italic>Drosophila</italic> tissue culture cells. However, the inventory of such genes remains incomplete. We have combined the powers of bioinformatics and RNAi technology to detect novel mitotic genes. We found that <italic>Drosophila</italic> genes involved in mitosis tend to be transcriptionally co-expressed. We thus constructed a co-expression–based list of 1,000 genes that are highly enriched in mitotic functions, and we performed RNAi for each of these genes. By limiting the number of genes to be examined, we were able to perform a very detailed phenotypic analysis of RNAi cells. We examined dsRNA-treated cells for possible abnormalities in both chromosome structure and spindle organization. This analysis allowed the identification of 142 mitotic genes, which were subdivided into 18 phenoclusters. Seventy of these genes have not previously been associated with mitotic defects; 30 of them are required for spindle assembly and/or chromosome segregation, and 40 are required to prevent spontaneous chromosome breakage. We note that the latter type of genes has never been detected in previous RNAi screens in any system. Finally, we found that RNAi against genes encoding kinetochore components or highly conserved splicing factors results in identical defects in chromosome segregation, highlighting an unanticipated role of splicing factors in centromere function. These findings indicate that our co-expression–based method for the detection of mitotic functions works remarkably well. We can foresee that elaboration of co-expression lists using genes in the same phenocluster will provide many candidate genes for small-scale RNAi screens aimed at completing the inventory of mitotic proteins.</p>", "<title>Author Summary</title>", "<p>Mitosis is the evolutionarily conserved process that enables a dividing cell to equally partition its genetic material between the two daughter cells. The fidelity of mitotic division is crucial for normal development of multicellular organisms and to prevent cancer or birth defects. Understanding the molecular mechanisms of mitosis requires the identification of genes involved in this process. Previous studies have shown that such genes can be readily identified by RNA interference (RNAi) in <italic>Drosophila</italic> tissue culture cells. Because the inventory of mitotic genes is still incomplete, we have undertaken an RNAi screen using a novel approach. We used a co-expression–based bioinformatic procedure to select a group of 1,000 genes enriched in mitotic functions from a dataset of 13,166 <italic>Drosophila</italic> genes. This group includes roughly half of the known mitotic genes, implying that it should contain half of all mitotic genes, including those that are currently unknown. We performed RNAi against each of the 1,000 genes in the group. By limiting the number of genes to be examined, we were able to perform a very detailed phenotypic analysis of RNAi cells. This analysis allowed the identification of 70 genes whose mitotic role was previously unknown; 30 are required for proper chromosome segregation and 40 are required to maintain chromosome integrity.</p>" ]

[ "<title>Supporting Information</title>" ]

[ "<p>We thank C. Sunkel, C. Lehner and G. Karpen for kindly providing some of the antibodies used in this study.</p>" ]

[ "<fig id=\"pgen-1000126-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g001</object-id><label>Figure 1</label><caption><title>The functions of 155 <italic>Drosophila</italic> genes detected by an RNAi screen of 1000 genes enriched in mitotic functions by co-expression analysis.</title><p>(A) Distribution of 164 known mitotic genes in co-expression lists with <italic>cid</italic>, <italic>glu</italic>, <italic>eb1</italic>, <italic>zw10</italic>, <italic>ida</italic> and <italic>sti</italic> (<italic>Citron kinase</italic>); ‘consensus’, is a consensus co-expression list constructed by combining the six single gene lists. Numbers in columns are percentages. (B) Quantitative grouping of the 155 genes detected in the screen according to the observed RNAi phenotypes. (C) Distribution of the 155 genes in the consensus co-expression list; note that the frequency of mitotic genes decreases with the increase of the rank in the list.</p></caption></fig>", "<fig id=\"pgen-1000126-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g002</object-id><label>Figure 2</label><caption><title>RNAi phenotypes elicited by the genes detected in the screen (CS and CC phenoclusters).</title><p>Colors refer to the strength of the phenotype: pale blue, weak; blue, strong. The numbers in the CAB column are frequencies of chromosome aberrations per cell. The other numbers refer to relatively rare phenotypic traits. PHC, phenocluster: CS, chromosome segregation; CC, chromosome condensation. Main phenotypic traits: NDC no dividing cells; CAB chromosome aberrations; ACC abnormal chromosome condensation; PCS precocious sister chromatid separation; DCC defective chromosome congression at metaphase; HCC hypercontracted chromosomes; LFA, low relative frequency of anaphases; DCS defective chromosome segregation following sister chromatid separation; NSS no sister chromatid separation with scattered chromosomes; NSC no sister chromatid separation with chromosomes at the center of the cell; CBA chromatin bridges at anaphase; DDA, chromosome decondensation during anatelophase; LTS long anatelophase spindles; MPS monopolar spindles; DSP defective spindle poles (defective asters and/or broad poles); SSP short spindles; LMD, spindles with low MT density; OSD other spindle defects; BIN binucleated cells (cytokinesis failure); OMD, other mitotic defects. Other phenotypic traits: (1) metaphase-like figures contain unreplicated chromosomes; (2) many anaphase-like figures but few telophase-like figures; (3) endoreduplicated metaphases (diplochromosomes); (4) lack of sister chromatid cohesion in the heterochromatic regions of the chromosomes.</p></caption></fig>", "<fig id=\"pgen-1000126-g003\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g003</object-id><label>Figure 3</label><caption><title>RNAi phenotypes elicited by the genes detected in the screen (SA, SC and CY phenoclusters).</title><p>Colors refer to the strength of the phenotype: pale blue, weak; blue, strong. The numbers in the CAB column are frequencies of chromosome aberrations per cell. The other numbers refer to relatively rare phenotypic traits. PHC, phenocluster: SA, spindle assembly; SC, spindle assembly and chromosome condensation; CY, cytokinesis. See legend of ##FIG##1##Figure 2## for main phenotypic traits. Other phenotypic traits: (1) long astral MTs in telophase; (2) drastic undercondensation of the heterochromatic regions of the chromosomes; (3) disorganized spindles; (4): split spindle poles; (5) multipolar spindles; (6) umbrella-like telophase spindles with all chromosomes at the astral pole; (7) centrosomes/asters detached from the spindle poles; (8) early cytokinesis defects: central spindle and contractile ring are both abnormal; (9) late cytokinesis defects: central spindle and contractile ring are normally assembled.</p></caption></fig>", "<fig id=\"pgen-1000126-g004\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g004</object-id><label>Figure 4</label><caption><title>Genes required for the maintenance of chromosome integrity.</title><p>(A) Examples of chromosome aberrations in colchicine/hypotonic-treated cells. Arrows and arrowheads point to chromatid and isochromatid deletions, respectively. Asterisks indicate asymmetric (U-type; one asterisk) and symmetric (X-type; two asterisks) chromatid exchanges. (B) Mitotic cells with acentric chromosome fragments (arrowheads) that have migrated to the cell poles. Staining for Cenp-C shows that the fragment at the pole of the RnrS-depleted metaphase (arrowhead) lacks the kinetochore (see text for further explanation). meta, metaphase; telo, telophase. In the merged figures, DNA is blue, Cenp-C red, and tubulin green. Scale bars, 5 µm. (C) The 44 genes required for chromosome integrity. Colors in the CAB (chromosome aberrations)/cell column denote the strength of the phenotype: light blue, weak; blue, strong. Numbers in this column are the frequencies of CABs per cell (see <xref ref-type=\"sec\" rid=\"s4\">Methods</xref> for details). In control cells, the frequency of spontaneous chromosome aberrations is 0.24±0.014 per cell (n = 941; from 12 independent experiments).</p></caption></fig>", "<fig id=\"pgen-1000126-g005\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g005</object-id><label>Figure 5</label><caption><title>Mitotic phenotypes observed in the CS1 and CS2 phenoclusters.</title><p>In all merged figures, DNA is blue and tubulin green. (A) Colchicine/hypotonic-treated metaphase chromosomes (c-meta) from control (ctr) cells; metaphase (meta), anaphase (ana) and telophase (telo) from untreated S2 cells. (B) Unreplicated chromosomes and impaired chromosome migration towards the spindle poles in cells treated with <italic>dup</italic> dsRNA (CS1). (C) Precocious sister chromatid separation and defective chromosome segregation observed after RNAi for <italic>dmt</italic> (CS2). Scale bars, 5 µm.</p></caption></fig>", "<fig id=\"pgen-1000126-g006\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g006</object-id><label>Figure 6</label><caption><title>Mitotic phenotypes observed in the CS3 phenocluster.</title><p>In merges, DNA is blue and tubulin green. (A) Top panels, metaphase- and late anaphase-like figures from <italic>U2af50</italic> (splicing factor) RNAi cells. Bottom panels, telophase-like figures from <italic>cid</italic> RNAi cells. Arrowheads point to chromosomes comprised of both sister chromatids. (B) Cyclin B distribution in control (top panels) and in <italic>CG3058</italic> (splicing factor) RNAi cells (bottom panels). The Cyclin B-stained cells correspond to the mitotic figures at their left. Scale bar 5 µm.</p></caption></fig>", "<fig id=\"pgen-1000126-g007\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g007</object-id><label>Figure 7</label><caption><title>Lack of ZW10 streaming and failure in sister chromatid separation observed after RNAi for genes of the CS3 phenocluster.</title><p>In merges, DNA is blue and tubulin green. (A) ZW10 does not stream towards the spindle poles in metaphase and remains associated with kinetochores in telophase after RNAi for <italic>CG6876</italic> (splicing factor). The ZW10 signal is white in the top panels and red in merges below each of these panels. (B) Cenp-C staining (red in merges) shows that ana-telophase-like figures generated by RNAi to the genes of the CS3 group contain chromosomes that comprise both sister kinetochores. Ctr, control; meta, metaphase; ana, anaphase; telo, telophase. Scale bar, 5 µm.</p></caption></fig>", "<fig id=\"pgen-1000126-g008\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g008</object-id><label>Figure 8</label><caption><title>Mitotic phenotypes observed in the CS4 and CS5 phenoclusters.</title><p>In merges, DNA is blue and tubulin green. (A) In RNAi cells of the CS4 phenocluster, sister chromatids do not separate so that the chromosomes remain at the center of the cell, while the spindles elongate and assume morphologies typical of ana/telophase figures. Top panels, anaphase-like and telophase-like spindles from <italic>Cyclin B</italic> RNAi cells. Bottom panels, anaphase-like and telophase-like spindles from <italic>U2A′</italic> (splicing factor) RNAi cells. Note that chromosome arms are parallel to the spindle axis (see text for explanation). (B) Mitotic figures from the CS5 phenocluster. In both <italic>CG3221</italic> and <italic>CG5649</italic> RNAi cells, the two sets of segregating chromosomes remain close to each other and fail to reach the spindle poles (compare with control cells in ##FIG##3##Figure 4A##). Meta, metaphase; ana, anaphase; telo, telophase. Scale bars, 5 µm.</p></caption></fig>", "<fig id=\"pgen-1000126-g009\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g009</object-id><label>Figure 9</label><caption><title>Mitotic phenotypes observed in the CC1 and CC3 phenoclusters.</title><p>In merges, DNA is blue and tubulin green. (A) colchicine/hypotonic-treated metaphase chromosomes (c-meta); metaphase (meta), anaphase (ana) and telophase (telo) observed after RNAi for <italic>Mcm3</italic> (CC1). Note the relative lack of sister chromatid cohesion in the heterochromatic regions of c-metaphase chromosomes, as well as difficulties in chromosome congression and segregation. (B) C-metaphase (c-meta) and mitotic figures in <italic>Top2</italic> (CC3) RNAi cells. Metaphase chromosomes are poorly condensed, leading to problems in chromosome separation during ana/telophase. Scale bars, 5 µm.</p></caption></fig>", "<fig id=\"pgen-1000126-g010\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g010</object-id><label>Figure 10</label><caption><title>Mitotic phenotypes observed in the SA1 phenocluster.</title><p>In merges, DNA is blue and tubulin green. (A) C-metaphase (c-meta) and mitotic figures observed after RNAi for <italic>int6</italic>. The heterochromatic regions of c-metaphase chromosomes are drastically undercondensed; note also the long astral microtubules in the telophase figures. (B) Extremely contracted chromosomes (c-meta) and short spindles in <italic>CG17293</italic> RNAi cells. (C) Examples of short metaphase spindles observed after RNAi for the SA1 genes. (D) pole-to-pole metaphase spindle lengths (mean±SE) observed in the SA1 phenocluster. The average metaphase spindle lengths observed after RNAi for the 17 SA1 genes are all significantly different (with p&lt;0.001 in Student's t-test) from the average spindle length of control (ctr) metaphases. Scale bars, 5 µm.</p></caption></fig>", "<fig id=\"pgen-1000126-g011\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g011</object-id><label>Figure 11</label><caption><title>Mitotic phenotypes observed in the SA2 phenocluster.</title><p>In merges, DNA is blue and tubulin green. (A) Mitotic figures with low microtubule density showing defective chromosome segregation after RNAi for <italic>NippedA</italic>. (B) Disorganized spindles with low microtubule density in <italic>Grip75</italic> RNAi cells; note that the ana/telophase-like spindles contain chromosomes with unseparated sister chromatids. (C) Staining for Cenp-C (red) shows that the chromosomes of the ana/telophase-like figures of <italic>Grip75</italic> RNAi cells comprise both sister kinetochores. Scale bar, 5 µm.</p></caption></fig>", "<fig id=\"pgen-1000126-g012\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000126.g012</object-id><label>Figure 12</label><caption><title>Mitotic phenotypes observed in the SA3 phenocluster.</title><p>In merges, DNA is blue and tubulin green. Spindles with broad, anastral poles in <italic>NiPp1</italic> (A) and CG6937 (B) RNAi cells. Note that these spindles exhibit a normal microtubule density. Scale bar, 5 µm.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s001\"><label>Figure S1</label><caption><p>Precocious sister chromatid separation and defective chromosome segregation after RNAi for <italic>Bub3</italic>. Ctr, control; c-meta, colchicine/hypotonic-treated metaphase chromosomes. Note in addition the elongated and bent ana/telophase spindles. Scale bar, 5 <italic>μ</italic>m.</p><p>(2.10 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s002\"><label>Figure S2</label><caption><p>Lack of sister chromatid separation after RNAi for CS3 genes that encode kinetochore components. Note in addition the elongated and bent ana/telophase-like spindles in <italic>l(1)G023/CG1558</italic> and <italic>CG9938/hec1/Ndc80</italic> RNAi cells. Scale bar, 5 <italic>μ</italic>m.</p><p>(4.78 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s003\"><label>Figure S3</label><caption><p>Lack of sister chromatid separation after RNAi for CS3 genes that encode splicing factors. Similar to other cells in the CS3 phenocluster (##SUPPL##1##Figure S2##), ana/telophase-like spindles are elongated and bent in <italic>CG3605</italic>, <italic>CG5931</italic>, and <italic>CG6015</italic> RNAi cells. Scale bar, 5 <italic>μ</italic>m.</p><p>(4.26 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s004\"><label>Figure S4</label><caption><p>Examples of mitotic figures observed after RNAi for the splicing factor gene <italic>CG10418</italic>. <italic>CG10418</italic> knockdown results in a failure of sister chromatid separation. In some cells with ana/telophase-like spindles, the chromosomes appear to migrate to the poles, while in others the chromosomes remain at the center of the cell. <italic>CG10418</italic> was assigned to the CS3 phenocluster because the first type of cells is more frequent than the second one. Scale bar, 5 µm.</p><p>(2.59 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s005\"><label>Figure S5</label><caption><p>RNAi for genes of the CS3 group results in ana/telophase-like cells that assemble irregular actin-based contractile rings despite the failure of sister chromatid separation. Note that the actin rings form in regions that contain bundled microtubules. Scale bar, 5 µm.</p><p>(2.81 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s006\"><label>Figure S6</label><caption><p>Examples of mitotic cells observed after RNAi for genes of the CS4 phenocluster. The chromosomes remain at the cell equator while the spindle elongates to assume an ana/telophase-like morphology. The chromosomes at the center of the cell often decondense as occurs during normal telophase. Scale bar, 5 <italic>μ</italic>m.</p><p>(5.25 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s007\"><label>Figure S7</label><caption><p>Defects in chromosome segregation observed after RNAi for genes of the CS5 phenocluster. Ctr, control; c-meta, colchicine/hypotonic-treated metaphase chromosomes. Note that in the c-metaphase (c-meta) from <italic>fzy</italic> RNAi cells, chromosomes are extremely condensed and the sister chromatids are separated. However, sister chromatid separation (SCS) is not observed in <italic>fzy</italic> RNAi cells with normally condensed chromosomes. We thus consider SCS as a secondary consequence of the excessive chromosome condensation, rather than a direct effect of gene product depletion. In support of this idea, RNAi for genes of the CS2 phenocluster (##SUPPL##0##Figure S1##) causes sister chromatid separation regardless of the degree of chromosome condensation. Scale bar, 5 µm.</p><p>(5.70 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s008\"><label>Figure S8</label><caption><p>Pole-to-pole spindle lengths (mean±SE) of ana/telophase figures observed in the CS1-CS5 phenoclusters. The ana/telophase spindles observed in all the RNAi experiments included in the graph are significantly longer (p&lt;0.001 in Student's t-test) than control (ctr) spindles.</p><p>(1.01 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s009\"><label>Figure S9</label><caption><p>RNAi for genes of the CC1 phenocluster results both in a lack of sister chromatid cohesion in the heterochromatic regions of the chromosomes and in defective chromosome segregation. Ctr, control; c-meta, colchicine/hypotonic-treated metaphase chromosomes. Scale bar, 5 µm.</p><p>(3.47 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s010\"><label>Figure S10</label><caption><p>Defects in chromosome condensation and segregation observed after RNAi for genes of the CC2 phenocluster. Ctr, control; c-meta, colchicine/hypotonic-treated metaphase chromosomes. Note the extensive chromatin bridges in the ana/telophase figures. Scale bar, 5 µm.</p><p>(3.37 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s011\"><label>Figure S11</label><caption><p>Monopolar spindles, short spindles and defective chromosome segregation observed after RNAi for <italic>msps</italic> and <italic>tho2</italic> (SA1 phenocluster). The telophase-like figures of <italic>msps</italic> RNAi cells display abnormally long astral microtubules. Scale bar, 5 <italic>μ</italic>m.</p><p>(3.48 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s012\"><label>Figure S12</label><caption><p>Extremely short spindles and monopolar spindles observed after RNAi for <italic>eIF-3p66</italic> (translation factor) and <italic>CG4865</italic>. Anaphases are very rare, suggesting that the tiny spindles in these cells are unable to support chromosome segregation. Scale bar, 5 µm.</p><p>(2.75 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s013\"><label>Figure S13</label><caption><p>Disorganized spindles with low microtubule density observed after RNAi for genes of the SA2 phenocluster. The elongated ana/telophase-like spindles contain scattered chromosomes with unseparated sister chromatids. Scale bar, 5 µm.</p><p>(2.06 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s014\"><label>Figure S14</label><caption><p>Anastral and broad spindle poles observed after RNAi for <italic>abnormal spindle (asp)</italic>. Scale bar, 5 µm.</p><p>(1.45 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s015\"><label>Figure S15</label><caption><p>RNAi phenotypes observed in the heterogeneous SA4 group. In most <italic>cdc2</italic> RNAi cells, the chromosomes remain at the center of the cells (as seen after RNAi for the CS4 group genes) and the centrosomes detach from the spindle poles. In <italic>Klp61F</italic> RNAi cells, spindles are either monopolar or monastral bipolar. In cells with monastral bipolar spindles, chromosome segregation does not occur and the chromosomes remain associated with the astral poles. Scale bar, 5 µm.</p><p>(3.93 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s016\"><label>Figure S16</label><caption><p>Abnormal chromosome condensation and multiple mitotic defects observed after RNAi for <italic>Borr</italic> (SC1) and <italic>Myb</italic> (SC2). Ctr, control; c-meta, colchicine/hypotonic-treated metaphase chromosomes. Note that in <italic>Borr</italic> RNAi cells chromosomes are abnormally long and irregularly condensed. In contrast, after RNAi for <italic>Myb</italic>, chromosomes are overcontracted and swollen with no resolution between sister chromatids. Scale bar, 5 µm.</p><p>(3.48 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s017\"><label>Table S1</label><caption><p>Coexpression analysis-based ranking of Drosophila genes.</p><p>(0.21 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s018\"><label>Table S2</label><caption><p>Distribution of 164 known mitotic genes in coexpression lists with cid, glu, eb1, zw10, ida and sti.</p><p>(0.03 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s019\"><label>Table S3</label><caption><p>Individual ranks of 164 mitotic genes in coexpression lists with cid, glu, eb1, zw10, ida and sti.</p><p>(0.06 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s020\"><label>Table S4</label><caption><p>List of the 155 mitotic genes detected in the screen and primers used for dsRNA synthesis.</p><p>(0.05 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s021\"><label>Table S5</label><caption><p>Characterization of the RNAi phenotypes elicited by the genes detected in the screen.</p><p>(0.10 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s022\"><label>Table S6</label><caption><p>Previously known functions of Drosophila mitotic genes and their putative human orthologs.</p><p>(0.23 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s023\"><label>Table S7</label><caption><p>Comparison with the Goshima et al. (2007) screen.</p><p>(0.04 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000126.s024\"><label>Text S1</label><caption><p>Comparison with previous screens.</p><p>(0.05 MB DOC)</p></caption></supplementary-material>" ]

[ "<fn-group><fn fn-type=\"COI-statement\"><p>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p>This work was supported in part by grants from AIRC (Italian Association for Cancer Research) and Italian Telethon to MG, and by a PRIN grant from MUR (Ministero dell'Università e della Ricerca) to MG and FD.</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"pgen.1000126.g001\"/>", "<graphic xlink:href=\"pgen.1000126.g002\"/>", "<graphic xlink:href=\"pgen.1000126.g003\"/>", "<graphic xlink:href=\"pgen.1000126.g004\"/>", "<graphic xlink:href=\"pgen.1000126.g005\"/>", "<graphic xlink:href=\"pgen.1000126.g006\"/>", "<graphic xlink:href=\"pgen.1000126.g007\"/>", "<graphic xlink:href=\"pgen.1000126.g008\"/>", "<graphic xlink:href=\"pgen.1000126.g009\"/>", "<graphic xlink:href=\"pgen.1000126.g010\"/>", "<graphic xlink:href=\"pgen.1000126.g011\"/>", "<graphic xlink:href=\"pgen.1000126.g012\"/>" ]

[ "<media xlink:href=\"pgen.1000126.s001.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s002.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s003.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s004.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s005.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s006.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s007.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s008.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s009.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s010.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s011.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s012.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s013.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s014.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s015.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s016.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s017.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s018.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s019.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s020.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s021.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s022.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s023.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000126.s024.doc\"><caption><p>Click here for additional data file.</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>The innate immune system, based on antimicrobial peptides (AMP) and proteins, provides a first line of defence against invading microbes ##REF##11753073##[1]##–##REF##11807545##[3]##. At present, over 880 different AMPs have been identified in eukaryotes (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.bbcm.univ.trieste.it/tossi/pag5.htm\">www.bbcm.univ.trieste.it/tossi/pag5.htm</ext-link>). During recent years it has become increasingly evident that many AMPs, such as defensins and cathelicidins, are multifunctional, also mediating chemotaxis, apoptosis, and angiogenesis ##REF##12960280##[4]##–##REF##12949495##[6]##. Conversely, molecules previously not considered as AMPs, including proinflammatory and chemotactic chemokines ##REF##11441062##[7]##, neuropeptides ##REF##15703760##[8]##, peptide hormones ##REF##12074933##[9]##,##REF##8204601##[10]##, the anaphylatoxin peptide C3a ##REF##17132627##[11]##,##REF##15550543##[12]##, growth factors ##REF##17454151##[13]## and kininogen-derived peptides ##REF##17093496##[14]##–##REF##16704414##[17]## have recently been found to exert antibacterial activities.</p>", "<p>Histidine-rich glycoprotein (HRG) is a plasma protein which was first isolated in 1972 by Heimburger <italic>et al.</italic>\n##REF##4116336##[18]##,##REF##4116337##[19]##. The protein is present in human plasma at 1.5–2 µM, but the local concentration when HRG is released from activated platelets is likely to be higher ##REF##15748207##[20]##–##REF##15869579##[22]##. It is a type 3 cystatin family protein ##REF##14587292##[23]##, along with α-2-HS-glycoprotein/fetuin-A, fetuin-B and kininogen, and is found in vertebrates as well as in some invertebrates. The structure contains two cystatin-like domains, a central histidine-rich region (HRR) with highly conserved GHHPH tandem repeats flanked by proline-rich regions, and a C-terminal region ##REF##15748207##[20]##. This modular structure of HRG facilitates multiple interactions, involving ligands such as heparin, plasminogen, fibrinogen, thrombospondin, heme, IgG, FcγR, and C1q. Due to its high content of histidine residues (∼13%), which are concentrated to the HRR, HRG can acquire a positive net charge either by incorporation of Zn²⁺, or by protonation of histidine residues at acidic conditions ##REF##15748207##[20]##. In this context it has been proposed that HRG acts as a pH and Zn²⁺ sensor, providing a mechanism for regulating the various activities of HRG ##REF##9488672##[24]##. HRG has recently been ascribed antiangiogenic ##REF##15313924##[25]## effects <italic>in vitro</italic>, as well as antitumor ##REF##14744774##[26]## effects <italic>in vivo</italic>. Recent studies on Hrg^−/− mice furthermore suggest that HRG plays a role as both an anticoagulant and an antifibrinolytic modifier, and may regulate platelet function <italic>in vivo</italic>\n##REF##15869579##[22]##.</p>", "<p>Previous work has also demonstrated that HRG exert direct antibacterial activities <italic>in vitro</italic> which are dependent on Zn²⁺and pH ##REF##17229145##[27]##. However, as many cationic proteins and peptide sequences display antimicrobial properties <italic>in vitro</italic>, the ultimate role(s) of HRG in innate immunity <italic>in vivo</italic> still remained unresolved. During the course of our studies, we observed that HRG had a significant activity against <italic>Candida</italic>. <italic>Candida</italic>, an eukaryote, is present as a commensal at mucosal surfaces and on skin. Although it may cause life-threatening sepsis in immunocompromised individuals it seldom causes invasive disease in immunologically normal individuals ##REF##18079743##[28]##. We therefore speculated that HRG could constitute a natural defence against <italic>Candida</italic> infections. In the present study we show, using a combination of microbiological, biochemical, and biophysical methods, that HRG exerts a potent antifungal activity particularly at low pH, which is mediated via its HRR, and targets ergosterol-rich membrane structures such as those of <italic>Candida</italic>. In mouse infection models, HRG protects against systemic infection by <italic>Candida</italic>, indicating a previously undisclosed antifungal role of HRG in innate immunity.</p>" ]

[ "<title>Materials and Methods</title>", "<title>Materials</title>", "<p>The peptides GHH20 (GHHPHGHHPHGHHPHGHHPH) and histatin 5 (DSHAKRHHGYKRKFHEKHHSHRGPY) were synthesized by Innovagen AB (Lund, Sweden), and were of &gt;95% purity. The purity and molecular weight was confirmed by MALDI-TOF MS analysis (Voyager, Applied Biosystems). 20-mer synthetic peptides (PEP-screen) spanning the sequence of HRG (Table 1) were obtained from Sigma-Genosys (St Louis, MO). Polyclonal rabbit antibodies against GHH20 and TAMRA-labeled GHH20 were from Innovagen AB (Lund, Sweden). HRG was FITC-labeled using the FluoroTag FITC Conjugation Kit (Sigma, St Louis, MO). Human serum and plasma were collected from healthy volunteers. Sterile wound fluids were obtained from surgical drainages after mastectomy. The use of human wound fluid was approved by the Ethics Committee at Lund University (LU 708-01). Seminal plasma was collected at the Fertility Center at Malmö University Hospital, Sweden. The GenBank (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/Genbank/index.html\">http://www.ncbi.nlm.nih.gov/Genbank/index.html</ext-link>) accession number of human histidine-rich glycoprotein is NP_000403.</p>", "<title>Fungal strains</title>", "<p>The fungi <italic>Candida parapsilosis</italic> BD 17837 and <italic>Candida albicans</italic> BD 1060 were clinical isolates. <italic>C. parapsilosis</italic> ATCC 90018, <italic>C. albicans</italic> ATCC 90028, <italic>Candida glabrata</italic> ATCC 90030, and <italic>Candida krusei</italic> ATCC 6258 isolates were from the American Type Culture Collection (ATCC, Rockville, MD).</p>", "<title>Purification of human HRG</title>", "<p>Serum HRG was purified using nickel-nitrilotriacetic acid (Ni-NTA) agarose as described before ##REF##17229145##[27]##. The concentration of the protein was determined using the Bradford method ##REF##942051##[56]##.</p>", "<title>Production and purification of recombinant HRG (rHRG and rHRG1-240)</title>", "<p>Recombinant His-tagged HRGP and truncated version of HRG (HRG1-240), containing amino acids 1-240 was produced and purified as previously described ##REF##14744774##[26]##,##REF##17229145##[27]##.</p>", "<title>Western blot</title>", "<p>Plasma, serum, wound fluids, seminal plasma (1 µl), and platelets (fluid from 1×10³ cells, disrupted by freeze thawing) were electrophoresed on 8% SDS-polyacrylamide (SDS-PAGE) gel or an 16.5% Tris-tricine gel and transferred to a nitrocellulose membrane (Hybond-C, GE Healthcare BioSciences, Little Chalfont, UK) ##REF##5413343##[57]##. The membrane was incubated in 3% skimmed milk in 10 mM Tris, 0.15 M NaCl, pH 7.4 for 1 h at room temperature, followed by incubation for 1 h with rabbit polyclonal antibodies against GHH20 (diluted 1:1000 in the same buffer). The membrane was washed 3 times, and incubated again for 1 h with horseradish peroxidase-conjugated secondary swine anti rabbit antibodies diluted 1:1000 (Dako, Carpinteria, CA). The image was developed using the ECL system (Amersham Biosciences).</p>", "<title>Preparation of fibrin clots</title>", "<p>Human plasma was subjected to a Ni-NTA agarose gel. The eluent (plasma completely depleted of HRG) was collected and used to form clots. Hrg^−/− and C57BL/6 (wild type) mice ##REF##15869579##[22]## were used for preparation of fibrin clots from plasma of the respective animals. Plasma deficient of HRG and normal plasma were incubated with a total concentration of 10 mM Ca²⁺ in eppendorf tubes at 37°C over night. Clots were washed three times and then stored in 10 mM 2-Morpholinoethanesulfonic acid (MES), pH 5.5. Clots (∼0.04g) were used in viable count experiments. To investigate the localization of HRG in fibrin clots, human plasma and HRG-deficient plasma were incubated with 10 µl FITC-labeled HRG (0.4 mg/ml) and then processed as before in the presence of 10 mM Ca²⁺ over night. The clots were then washed in distilled water and mounted on slides using Dako mounting media (Dako).</p>", "<title>Viable count assay</title>", "<p>\n<italic>C. parapsilosis</italic>, <italic>C. albicans</italic>, <italic>C. glabrata and C. krusei</italic> were grown to mid-logarithmic phase in Todd-Hewitt (TH) medium (Becton and Dickinson, Maryland, USA) at 27°C and washed in 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5. For dose-response experiments, purified HRG or GHH20 (0.03–6 µM) were incubated with 1×10⁵\n<italic>C. parapsilosis</italic> ATCC 90018 <italic>or C. albicans</italic> ATCC 90028 for 2 h at 37°C in 10 mM, Tris, pH 7.4 or in 10 mM MES-buffer, pH 5.5, plated on Sabouraud dextrose broth (Becton and Dickinson) agar, and incubated 48 hours at 27°C, whereafter the number of cfu was determined. In order to investigate the antifungal activity of HRG in presence of salt, 6 µM HRG were incubated with 1×10⁵\n<italic>C. parapsilosis</italic> ATCC 90018 for 2 h at 37°C in 10 mM, MES, pH 5.5 containing 0, 25, 50, 100 or 150 mM NaCl, plated and the number of cfu was determined. In kinetic experiments, 0.3 and 3 µM HRG were incubated with <italic>C. parapsilosis</italic> ATCC 90018 for 5, 15, 30, 60, or 120 minutes in 10 mM MES, pH 5.5, plated and the number of cfu was determined. For determination of the effect of HRG on various <italic>Candida</italic> strains, HRG (3 µM) was incubated with <italic>C. parapsilosis</italic> ATCC 90018 or BD 17837, <italic>C. albicans</italic> ATCC 90028 or BD 1060, <italic>C. glabrata</italic> ATCC 90030 or <italic>C. krusei</italic> ATCC 6258 in 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5, plated and number of cfu determined. Truncated and full length recombinant HRG, 0.6 µM rHRG, or rHRG1-240 were incubated with <italic>C. parapsilosis</italic> (1×10⁵) for two hours and then plated and number of cfu determined. To investigate the <italic>in vitro</italic> antifungal activity of HRG, normal or HRG-deficient fibrin clots (∼0.04g) were incubated with <italic>C. parapsilosis</italic> ATCC 90018 for 2 h in 10 mM MES, pH 5.5, plated and number of cfu were determined. For inhibition studies, 0.3 µM HRGP were incubated with <italic>C. parapsilosis</italic> (1×10⁵) in 10 mM MES, pH 5.5, in presence or absence of heparin (50 µg) for two hours and then plated and number of cfu was determined. In all experiments, 100% survival was defined as total survival of fungi in the same buffer and under the same conditions in absence of peptide, protein, or clots. The p-values were determined using Kruskall-Wallis one-way ANOVA analysis.</p>", "<title>Radial diffusion assay</title>", "<p>Radial diffusion assay (RDA) was performed essentially as described earlier ##REF##1901580##[58]##. <italic>C. parapsilosis</italic> ATCC 90018 and <italic>C. albicans</italic> ATCC 90028 were grown to midlogarithmic phase in TH-medium, and then washed with distilled water. 4×10⁶ colony forming units was added to 5 ml of the underlay agarose gel (0.03% (w/v) trypticase soy broth (TSB), 1% (w/v) low electroendosmosis type agarose (Sigma), 0.02% (v/v) Tween 20 (Sigma). The buffers used in the underlay gels were 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5. The underlay gel was poured into an 85-mm Petri dish. After agarose solidification, wells of 4 mm in diameter were punched, and 6 µl of peptide solution was added to each well. Buffers were used as a negative control. Plates were incubated at 28°C for 3 h to allow diffusion of the peptides. The underlay gel was then covered with 5 ml of molten overlay. Antimicrobial activity of a peptide is visualized as a zone of clearance around each well after 18–24 h of incubation at 28°C. Peptides were tested in concentrations of 100 µM.</p>", "<title>Binding of HRG to <italic>Candida</italic>\n</title>", "<p>\n<italic>C. parapsilosis</italic> (1×10⁵ cfu) were incubated with 0.6 µM HRG in 50 µl 10 mM MES, pH 5.5, with or without heparin (50 µg/ml) for 2 h at 37°C, centrifuged and the pellet was washed three times in 10 mM MES, pH 5.5. The pellet and the supernatant were resuspended in SDS sample buffer, electrophoresed (8% SDS-PAGE), and then transferred to a nitrocellulose membrane. Western blotting was performed as above.</p>", "<title>Fluorescence microscopy</title>", "<p>\n<italic>C. parapsilosis</italic> ATCC 90018 fungi were grown in TH medium at 27°C to mid-logarithmic phase. The fungi were washed in 10 mM Tris, pH 7.4, and resuspended in the same buffer. <italic>C. parapsilosis</italic> (2×10⁶/ ml) were incubated with 1 µl of TAMRA-labeled GHH20 (2 mg/ml) in 10 mM MES, pH 5.5, with or without heparin (50 µg/ml), left standing for 5 minutes on ice, and then washed twice in 10 mM Tris, pH 7.4. Fungi were fixed with 4% paraformaldehyde by incubation on ice for 15 minutes and in room temperature for 45 minutes. The fungi were then applied onto Poly-L-lysine coated cover glass and after an incubation time of 30 minutes, finally mounted on slides using Dako mounting media (Dako, Carpinteria, CA). In order to assess permeabilisation, <italic>C. albicans</italic> ATCC 90028 (2×10⁶ cfu) were incubated with HRG or LL-37 (both at 10 µM) in 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5 for 30 minutes at 37°C. Samples were transferred to Poly-L-lysine coated cover glass and incubated for 45 minutes at 37°C, washed and 2 µg of FITC were added in a volume of 200 µl, and incubated for 30 minutes at 30°C, washed and then fixed as above. Samples were visualized using a Nikon Eclipse TE300 inverted fluorescence microscope equipped with a Hamamatsu C4742-95 cooled CCD camera, a Plan Apochromat 100X objective and a high N.A. oil condenser.</p>", "<title>Negative staining and transmission electron microscopy</title>", "<p>\n<italic>C. parapsilosis</italic> ATCC 90018 were grown in TH medium at 37°C to mid-logarithmic phase. The fungi were washed in 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5, and resuspended in the same buffer. HRG or LL-37 (10 µM) was incubated with <italic>C. parapsilosis</italic> (20×10⁶ cfu) for two hours in a total volume of 10 µl in Tris buffer, pH 7.4 or in MES buffer, pH 5.5. Samples of <italic>C. parapsilosis</italic> fungi suspensions were adsorbed onto carbon-coated copper grids for 1 min, washed briefly on two drops of water, and negatively stained on two drops of 0.75 % uranyl formate. The grids were rendered hydrophilic by glow discharge at low pressure in air. Specimens were observed in a Jeol JEM 1230 electron microscope operated at 60 kV accelerating voltage. Images were recorded with a Gatan Multiscan 791 CCD camera.</p>", "<title>Flow cytometry</title>", "<p>\n<italic>C. parapsilosis</italic> ATCC 90018 were grown in TH medium at 27°C to mid-logarithmic phase. The fungi were washed in 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5 and resuspended in the same buffer. <italic>C. parapsilosis</italic> (5×10⁷ in a total volume of 0.5 ml) were incubated with 10 µl of FITC-labeled HRG (0.4 mg/ml) or 10 µl TAMRA-labeled GHH20 (2 mg/ml) in 10 mM Tris, pH 7.4 or in 10 mM MES, pH 5.5, let stand for 5 minutes on ice and then washed in 10 mM Tris, pH 7.4. The cells were fixed with 4% paraformaldehyde by incubation on ice for 15 minutes and in room temperature for 45 minutes. Flow cytometry analysis was performed using a FACS-Calibur flow cytometry equipped with a 15 mW argon laser turned a 488 mm (Becton-Dickinson, Franklin Lakes, NJ). The fungal population was selected by gating with appropriate settings of forward scatter (FSC) and sideward scatter (SSC). The FL1 fluorescence channel (λ_em = 530 nm) was used to record the emitted fluorescence of FITC, and the FL3 fluorescence channel (λ_em = 585 nm) was used to record the emitted fluorescence of Texas red.</p>", "<title>Liposome preparation and leakage assay</title>", "<p>Dry lipid films were prepared by dissolving dioleoylphosphatidylcholine (1,2-dioleoyl-sn-Glycero-3-phoshocholine, &gt;99% purity, Avanti Polar Lipids, Alabaster, AL) (60 mol%) and either ergosterol or cholesterol (both &gt;99% purity, Sigma, St Louis, MO) (40 mol%), and then removing the solvent by evaporation under vacuum overnight. Subsequently, buffer (10 mM Tris, pH 7.4) was added together with 0.1 M carboxyfluorescein (CF) (Sigma, St Louis, MO). After hydration, the lipid mixture was subjected to eight freeze-thaw cycles consisting of freezing in liquid nitrogen and heating to 60°C. Unilamellar liposomes, of about Ø140 nm were generated by multiple extrusions through polycarbonate filters (pore size 100 nm) mounted in a LipoFast miniextruder (Avestin, Ottawa, Canada) at 22°C. Untrapped CF was then removed by two gel filtrations (Sephadex G-50) at 22°C, with Tris buffer as eluent. CF release was determined by monitoring the emitted fluorescence at 520 nm from liposome dispersions (10 mM lipid in 10 mM Tris). An absolute leakage scale was obtained by disrupting the liposomes at the end of the experiment through addition of 0.8 mM Triton X100 (Sigma, St Louis, MO), causing 100% release and dequenching of CF. Although calcein is frequently used for pH-dependent leakage studies, the high charge of this dye has been noted to influence its leakage behaviour in the presence of highly cationic peptides ##REF##17207456##[59]##. Instead, therefore, CF was used as a leakage marker at both pH 6.0 and 7.4, however, avoiding pH-dependent fluorescence effects through neutralization prior to probing the limiting leakage in case of pH 6.0 leakage. Throughout, a SPEX-fluorolog 1650 0.22-m double spectrometer (SPEX Industries, Edison, NJ) was used for the liposome leakage assay. Measurements were performed at 37°C.</p>", "<title>CD spectroscopy</title>", "<p>The CD spectra of the peptides in solution were measured on a Jasco J-810 Spectropolarimeter (Jasco, U.K.). Measurements were performed at 37°C in a 10 mm quartz cuvet under stirring and the effect on protein/peptide secondary structure monitored in the range 200–260 nm. The background value, detected at 250 nm, was subtracted, and signals from the bulk solution were corrected for. The secondary structure was monitored at a concentration of 0.25 µM of HRG in buffer, in the presence of liposomes (lipid concentration 100 µM), and in the presence of mannan from <italic>Saccharomyces cerevisiae</italic> (0.02 wt%; Sigma-Aldrich, St. Luis, USA).</p>", "<title>Fungal growth in plasma</title>", "<p>\n<italic>C. parapsilosis</italic> ATCC 90018 were grown in TH medium at 27°C to mid-logarithmic phase. The fungi were washed in 10 mM MES, pH 5.5 and resuspended in the same buffer. <italic>C. parapsilosis</italic> (2×10⁷) cfu in a total volume of 10 µl was added to 50 µl of human normal plasma or HRG-depleted plasma (eluent from Ni-NTA agarose gel), and incubated for 0, 4, 8 or 18 hours at 27°C and then plated and number of cfu determined.</p>", "<title>Animal experiments</title>", "<p>The original knockout mice 129/B6-<italic>HRG</italic>\n^tm1wja1 were crossed with C57BL/6 mice (Taconic) for 14 generations to obtain uniform genetic background. These HRG-deficient mouse strain was called B6-<italic>HRG</italic>\n^tm1wja1 following ILAR (Institute of Laboratory Animal Resources) rules. Wildtype C57BL/6 control mice and C57BL/6 Hrg^−/− mice (8–12 weeks, 27+/−4g) were bred in the animal facility at Lund University. C57BL/6 Hrg^−/−, lacks the translation start point of exon 1 of the Hrg gene ##REF##15869579##[22]##. Animals were housed under standard conditions of light and temperature and had free access to standard laboratory chow and water. In order to study <italic>Candida</italic> dissemination, <italic>C. albicans</italic> ATCC 90018 were grown to midlogarithmic phase, washed and diluted in PBS, pH 7.4. Two hundred and fifty µl containing 1×10⁹ cfu was injected intraperitoneally into C57BL/6 or C57BL/6 Hrg^−/− mice, divided into weight and sex matched groups. The animals were sacrificed 48 hours post infection, and blood was collected by cardiac puncture. The number of cfu was determined by viable count. In order to study fungal dissemination to target organs, the mice were infected as previously described and three days later the spleen and kidney were harvested on ice.</p>", "<title>Histology</title>", "<p>Representative animals were sacrificed three days post infection and the kidneys were removed into 4% formalin. The tissues were embedded in paraffin, sectioned and stained with Hematoxylin and eosin (H&amp;E) and with Periodic acid-Schiff (PAS).</p>" ]

[ "<title>Results</title>", "<title>Antifungal activity of HRG and binding to <italic>Candida</italic> cells</title>", "<p>In order to assess possible antifungal effects of HRG, we tested the activity of the protein against various <italic>Candida</italic> isolates. HRG was shown to be antifungal against <italic>C. parapsilosis</italic> at normal pH (10 mM Tris, pH 7.4), and the activity was significantly increased in low pH buffer (10 mM MES, pH 5.5) (##FIG##0##Figure 1A##). It is well-known that activities of AMPs and antimicrobial proteins are dependent of the microenvironment. For example, various chemokines, defensins, LL-37 as well as heparin binding protein are partly, or completely, antagonized by high salt conditions or the presence of plasma proteins <italic>in vitro</italic>\n##REF##17229145##[27]##,##REF##11309707##[29]##,##REF##9837875##[30]##. Therefore, the influence of salt was tested. The results showed that HRG partially retained antifungal activity at physiological Cl⁻ levels (0.1 M) but only at low pH (##FIG##0##Figure 1B##). The antifungal activity against <italic>C. parapsilosis</italic> was both time- and dose-dependent (##FIG##0##Figure 1C##). In subsequent experiments various <italic>Candida</italic> strains (<italic>C. parapsilosis</italic>, <italic>C. albicans</italic>, <italic>C. glabrata</italic> and <italic>C. krusei</italic>) were incubated with HRG (at 3 µM) at neutral as well as low pH. ##FIG##0##Figure 1D## demonstrates, in line with the above experiments, that HRG is particularly active at low pH. Thus, <italic>C. parapsilosis</italic>, <italic>C. albicans</italic> and <italic>C. krusei</italic> were all nearly completely killed by HRG at low pH, whereas <italic>C. glabrata</italic> exhibited a partial resistance at this concentration of HRG, the latter in analogy to <italic>C. glabrata</italic> displaying some resistance against histatin 5 ##REF##16034806##[31]##. Next, to investigate the binding of HRG to fungi, <italic>C. papapsilosis</italic> was incubated with HRG at low pH, washed, and analysed by immunoblotting. Since previous results indicated that the HRR of HRG, which binds heparin/heparan sulfate, mediates antibacterial effects ##REF##17229145##[27]##, heparin was added for competition of binding to <italic>Candida</italic>. ##FIG##0##Figure 1E## shows that HRG was able to bind to the fungal cells and that the binding was partially inhibited by an excess of heparin. This finding is compatible with the observation that heparin completely blocks the antifungal effect of HRG (##SUPPL##1##Figure S1##). As demonstrated by flow cytometry, HRG bound to <italic>C. parapsilosis</italic> at neutral pH, and the binding was significantly increased at pH 5.5 (##FIG##0##Figure 1F##), results compatible with the fungal killing assays (##FIG##0##Figure 1A##). In summary, therefore, the results demonstrate that the antifungal actions of HRG were pH-dependent and likely mediated via the heparin-binding region of the protein.</p>", "<title>Membrane-permeabilizing effects of HRG</title>", "<p>Many AMPs kill microbes by membrane lysis, while others may translocate through membranes and subsequently interact with intracellular targets, such as DNA and mitochondria, all eventually resulting in microbial killing ##REF##8514403##[32]##,##REF##14733612##[33]##. Considering the antifungal effects and the binding to <italic>Candida</italic> cells, it was of interest to further study the possible mode of action for HRG on <italic>Candida</italic>. Electron microscopy demonstrated that HRG caused membrane breaks in <italic>Candida</italic> cells and release of cytoplasmic components (##FIG##1##Figure 2A##), effects particularly noted at low pH, where significant extracellular material was detected. The effects were similar to those observed after treatment with the “classical” human AMP LL-37 (##FIG##1##Figure 2A##). These data suggest that HRG acts on fungal membranes, however they do not demonstrate the exact mechanistic events, as secondary metabolic effects on fungi also may trigger death and membrane destabilization. Therefore, the impermeant dye FITC was used to assess permeabilisation. The results showed that HRG indeed was able to permeabilise <italic>Candida</italic> membranes (##FIG##1##Figure 2B##). In line with previous antifungal and binding experiments (see ##FIG##0##Figure 1A and 1F##), the permeabilisation was most apparent at low pH.</p>", "<p>These results were further substantiated by the use of a liposome model to assess membrane permeabilisation. In correspondence with the effects of HRG on <italic>Candida</italic>, HRG caused liposome leakage. Compatible with the pH sensitivity observed for HRG, the molecule preferably disrupted ergosterol-containing liposomes at pH 6.0 when compared with pH 7.4 (##FIG##1##Figure 2C, left panel##). Notably, ergosterol-containing liposomes, mimicking fungal membranes, were more sensitive than cholesterol-containing ones, mimicking mammalian membranes (##FIG##1##Figure 2C, right panel##). These results are in agreement with numerous previous findings on the membrane-stabilizing effects of cholesterol ##REF##15726825##[34]##, as well as the findings that ergosterol induce less membrane stability in phospholipids than cholesterol ##REF##16326903##[35]##. At lower pH, protonation of histidine groups (pKa for the isolated histidine group is approximately 6.5), effectively increases the net charge density of HRG, thus the observed effects are compatible with findings previously reported for histidine-containing consensus peptides and histidine-rich endogenous peptides ##REF##9199465##[36]##,##REF##17005809##[37]##. Also noteworthy is that HRG did not display any major conformational changes either at low pH, or in the presence of fungal mannan (##FIG##1##Figure 2D##) or ergosterol-containing phospholipid liposomes (not shown). Hence, large-scale conformational changes appear not to be critical for the antifungal action of HRG. Taken together, the combination of electron microscopy, FITC-studies, and liposome data demonstrates that HRG acts at least in part through membrane disruption, although it is possible that additional intracellular effects of HRG may also contribute to fungal death. It is also notable that the observed effects were most marked and consistent at low pH. At neutral pH, binding (##FIG##0##Figure 1B##), as well as permeabilization (##FIG##1##Figure 2A and 2B##) was less apparent and these observations reflected the diminished antifungal effects at pH 7.4 (##FIG##0##Figure 1A and 1D##).</p>", "<title>Antifungal regions of HRG</title>", "<p>In order to explore the structure-function relationships of epitopes of HRG, overlapping peptide sequences comprising 20mers (##FIG##2##Figure 3A## and ##SUPPL##0##Table S1##) were synthesized and screened, at both neutral and acidic pH, for antifungal activities against <italic>C. parapsilosis</italic> as well as <italic>C. albicans</italic>. The experiments identified several antifungal regions. In particular peptides no. 20–24 and 26, spanning the HRR, displayed a significant antifungal activity against both <italic>Candida</italic> strains at low pH (##FIG##2##Figure 3B##). There was a clear correlation with net charge (at the respective pH) of the various peptide regions and their observed antifungal activity (##SUPPL##2##Figure S2##). Although intuitively apparent (##FIG##2##Figure 3B##), the analysis furthermore showed that peptides derived from the HRR were (with the exception of the K and R-rich peptide no. 27) characterized by an increase in net charge at low pH (##SUPPL##0##Table S1## and ##SUPPL##2##Figure S2##).</p>", "<p>In order to further study the importance of the HRR we investigated the activity of recombinant HRG (rHRG) and a truncated version (rHRG1-240), lacking the HRR and C-terminal domain. In contrast to full-length rHRG, truncated rHRG (0.6 µM) displayed no activity at pH 5.5 against <italic>Candida</italic> (##FIG##2##Figure 3C##). Taken together, considering the well-known heparin binding capacity of HRR, its pH dependence, as well as the absence of antifungal activity of rHRG1-240, it was logical to focus on the HRR of HRG in the subsequent studies of antifungal activity.</p>", "<p>The HRR contains 12 tandem repeats of five consensus sequences of amino acids, GHHPH ##REF##15748207##[20]##, a motif highly conserved among various vertebrate species ##REF##17229145##[27]##. To examine the activity of this sequence motif further, a 20-mer peptide (GHHPH)₄\n##REF##17655823##[16]##,##REF##17229145##[27]## was chosen for further studies. Similar to intact HRG, GHH20 was antifungal against <italic>C. parapsilosis</italic> and <italic>C. albicans</italic>, particularly at low pH (##FIG##3##Figure 4A##). As demonstrated by FACS analysis, Tetramethyl-6-Carboxyrhodamine (TAMRA)-labeled GHH20 peptide bound to <italic>C. parapsilosis</italic>, and in correspondence with the antifungal data, the binding was stronger at pH 5.5 when compared to neutral pH (##FIG##3##Figure 4B##). As illustrated by fluorescence microscopy, TAMRA-labeled GHH20 showed a significant binding to <italic>Candida</italic> at pH 5.5 (##FIG##3##Figure 4C##). As with the HRG holoprotein, heparin abolished the binding, reflecting the heparin-binding capacity of this region of the HRR ##REF##17229145##[27]##. Also in line with the above experiments on fungi, GHH20 preferably disrupted liposomes at pH 6.0, with no significant activity at pH 7.4 (##FIG##3##Figure 4D##). The GHH20 peptide caused liposome leakage within a few hundred seconds (not shown), which contrasted to the significantly slower HRG-induced liposome leakage (##FIG##1##Figure 2B##), likely a manifestation of the much higher molecular weight of the holoprotein. Again as with intact HRG, CD spectroscopy showed that GHH20 displayed no major conformational changes associated with the histidine protonation at low pH, nor on interaction with phospholipid liposomes or mannan (not shown). Taken together, the GHH20 peptide showed similar characteristics as the holoprotein HRG with respect to activity, binding, and membrane permeabilisation.</p>", "<title>HRG is found in biological fluids and is active in plasma and in fibrin clots</title>", "<p>In order to investigate the functional relevance of the above <italic>in vitro</italic> activities, we first tested the role of HRG against fungi in relevant physiological “settings” <italic>ex vivo</italic>. Initial results showed that HRG was detected in blood fractions (plasma, serum) and in platelets, also in wound fluid from acute wounds, and chronic leg ulcers (##FIG##4##Figure 5A##). The latter wound type is characterized by unregulated and excessive proteinase activity leading to degradation of many plasma proteins ##REF##10954207##[38]##,##REF##8601737##[39]##. However, compared with plasma and serum HRG, the molecule was not fragmented in this chronic wound fluid fraction (##FIG##4##Figure 5A##). The protein was also detected in fibrin clots (##FIG##4##Figure 5A##) but not present in seminal plasma. It is of note that the molecule migrated aberrantly in the used gel systems; relative 55–60 kDa in 8% gels (Tris-Glycine) and 45–50 kDa in 16.5 gels (Tris-Tricine). Identical serum and plasma preparations of HRG were used in the two gel systems, and recombinant HRG showed the same anomalous migration (not shown). In addition to its presence in plasma and other biological fluids, HRG occurs at significant levels in, and binds avidly to, fibrin clots ##REF##3958188##[40]##. Coagulation was initiated in normal and HRG-deficient human plasma in the presence of FITC-labeled HRG (##FIG##4##Figure 5B##). FITC-labeled HRG bound to clots derived from HRG-deficient plasma, and notably, it appeared to be present at clot boundaries, suggesting that it may “coat” the clot surfaces. In clots from normal plasma, no staining was seen, indicative of an inhibition of binding of FITC-HRG by the excess of endogenous HRG (∼150 µg/ml). Clots, physiologically important “barriers”, formed during hemostasis and infection, could thus constitute a unique milieu with high levels of surface-immobilized HRG. Considering the above results we investigated whether the presence of HRG could reduce the growth of <italic>Candida</italic> in plasma. Firstly, the growth of <italic>C. parapsilosis</italic> was investigated in normal human plasma and in plasma depleted of HRG. The results showed that <italic>C. parapsilosis</italic> multiplied significantly faster in HRG-depleted human plasma (##FIG##4##Figure 5C##). Analogous results on fungal growth were observed using plasma from mice deficient in HRG (data not shown). It is of note that these results do not exclude the possibility that other antifungal mechanisms may be involved, such as those dependent of complement activation. Furthermore, although the total protein levels (as determined by the Bradford method) and contents (as assessed by SDS-PAGE on 8% gels, not shown) were the same in depleted plasma (51.0+/−1.2 g/l) when compared with control plasma (51.7+/−3.3 g/l), it cannot be excluded that additional changes of low abundance proteins, induced by passage over Ni-NTA agarose could affect <italic>Candida</italic> growth. Nevertheless, the observation that similar results were obtained with the mice plasmas points at HRG as the main factor responsible for the partial growth inhibition noted. Furthermore, as demonstrated in ##FIG##4##Figure 5D##, fibrin clots derived from plasma of HRG deficient mice were significantly more prone to infection by <italic>C. parapsilosis</italic> than clots from wild-type mice, and similar results were obtained with human plasma depleted of HRG when compared with normal plasma (not shown). The observation that clots devoid of HRG showed detectable, although reduced, antifungal activity (##FIG##4##Figure 5D##) suggest the existence of other yet unidentified factors in clots also mediating fungal killing. Nevertheless, the results indicate that HRG contributes to antifungal activity under physiological conditions.</p>", "<title>\n<italic>In vivo</italic> role of HRG</title>", "<p>To investigate the role of HRG during <italic>Candida</italic> infection <italic>in vivo</italic>, we designed a mouse model of intraperitoneal infection with <italic>C. albicans</italic>. After infection, the body weight of the mice was followed for three days (##FIG##5##Figure 6A##). Hrg^−/− mice showed a significantly increased weight loss at day 1 and 2 (p = 0.02) when compared with wild type mice, and the wild type mice regained their initial weight after three days. Blood samples were collected from the animals 2 days post infection, and the fungal load in blood was determined (##FIG##5##Figure 6B##). A significantly higher amount of <italic>Candida</italic> cells was detected in the blood of Hrg^−/− mice when compared with wild type mice (p = 0.032), indicating that a systemic infection has developed in HRG-deficient mice. In a similar experiment, we determined the ability of the fungi to establish infection in target organs distant from the site of administration. The spleen and kidney were harvested 3 days after initiation of intraperitoneal infection and the fungal load was determined. The results showed significant differences between Hrg^−/− mice and the wild type mice; one animal out of 10 in the control group showed fungal load in the spleens and kidneys compared with 8 out of 10 in the Hrg^−/− group (p = 0.009) (##FIG##5##Figure 6C##). Histopathological examination of the kidney tissues from Hrg^−/− mice showed dense neutrophil infiltrates and notably, <italic>Candida</italic> cells were visualised by PAS staining in the centre of these infiltrates (##FIG##5##Figure 6D##). These results show a striking protective role for HRG against invasive <italic>Candida</italic> infection <italic>in vivo</italic>.</p>" ]

[ "<title>Discussion</title>", "<p>The key findings in our study are the identification of an antifungal activity of HRG <italic>in vivo</italic> together with the characterization of possible epitopes of HRG mediating this effect, as well as mechanistic data on HRG targeting of <italic>Candida</italic> membranes. The results have implications for our understanding of novel antifungal properties of HRG, and demonstrate that HRG constitutes a previously undisclosed natural and antimicrobial defence system.</p>", "<p>From a structural perspective, several lines of evidence indicate that the HRR is, at least to a significant extent, responsible for the HRG interaction with <italic>Candida</italic> membranes. Although the 3D structure of HRG has not yet been determined, modelling studies suggest that the HRR of HRG forms a polyproline (II) helical structure with numerous histidines. At physiological pH, HRG is net negatively charged (pI 6.45). However, due to its high content of histidine residues (∼13%), which are concentrated to the HRR, it can acquire a positive charge by protonation ##REF##15748207##[20]##,##REF##8639676##[41]##, and this in turn likely facilitates the interactions between HRG and <italic>Candida</italic>. These results were substantiated by the finding that a region of HRG containing the motif sequence GHHPH, was antifungal, and that low pH enhanced this activity. The high conservation of this sequence among vertebrates likely reflects its importance for membrane interactions of HRG ##REF##17229145##[27]##. However, as evident in ##FIG##2##Figure 3B##, there are also other antifungal regions in the protein, active irrespective of pH in the interval investigated, an observation compatible with the antifungal activity of HRG detected at neutral pH. It should be pointed out however, that the peptide data do not reflect the complex structure-activity relationships of the holoprotein. Although the CD experiments did not detect any major conformational changes upon interaction with liposomes or polysaccharides, it cannot be ruled out that conformational changes mediated by HRR interactions with intact fungal cells lead to the exposure of additional antimicrobial epitopes in the molecule. Nevertheless, a recombinant and truncated variant of HRG, lacking the histidine-rich and C-terminal domains, was not active against <italic>Candida</italic>, pointing to the HRR as an important, possibly the most important, effector of HRGs antifungal effects.</p>", "<p>Many histidine-rich AMPs are known, among these the clavanins ##REF##9199465##[36]##, histatins, and calprotectin ##REF##8423249##[42]##. We have previously shown that the antibacterial effects <italic>in vitro</italic> of various histidine-rich peptides, both consensus motifs and peptides derived from domain 5 of HMW kininogen ##REF##16704414##[17]## and from HRG ##REF##17229145##[27]## are enhanced at low pH or upon addition of Zn²⁺. Others have reported that the antimicrobial activity of clavanins were substantially increased in low pH as compared with neutral pH ##REF##9199465##[36]##. Furthermore, the antimicrobial effect of histatin 5 is enhanced at low pH ##REF##6724693##[43]##, and histidine-rich variants of magainin, the LAH4-peptides, were recently shown to have increased antibacterial activity in low pH compared to neutral pH ##REF##17005809##[37]##. Taken together, the pH dependent activity of HRG is thus comparable to other histidine-rich proteins and peptides, and provides an additional link between pH sensitive AMPs and HRG. However, contrasting to histatins, which translocate through <italic>Candida</italic> membranes, bind mitochondria, and induce cell death by non-lytic ATP-release ##REF##10383383##[44]##, HRG acts directly on fungal membranes.</p>", "<p>Many AMPs are generated by proteolysis of larger, and non-antimicrobial holoproteins. For example, the cathelicidin LL-37 is released from hCAP18, and other AMPs are proteolytically generated from complement factor C3 and high molecular weight kininogen ##REF##11807545##[3]##, ##REF##17132627##[11]##, ##REF##15550543##[12]##, ##REF##17093496##[14]##–##REF##16704414##[17]##. Considering that intact HRG is antifungal, proteolysis of this molecule does not appear to be needed for activity. It is of note that like HRG, several antimicrobial proteins are antimicrobial <italic>per se</italic>, including bacterial permeability increasing protein, serprocodins such as proteinase 3, elastase and heparin binding protein, as well as lactoferrin ##REF##17229145##[27]##,##REF##11023496##[45]##. However, it is also described that antibacterial proteins, such as bacterial permeability increasing protein and lactoferrin, may give rise to peptides exerting antibacterial activities ##REF##9523110##[46]##,##REF##16261252##[47]##. Likewise, it has been shown that HRG may be degraded by plasmin ##REF##6218829##[48]##, as well as in patients undergoing thrombolytic therapy ##REF##2935971##[49]## and bioactive fragments of HRG are involved in antiangiogenesis ##REF##14744774##[26]##,##REF##8639676##[41]##. Thus, although a major fragmentation of HRG was not observed in this work, e.g., in wound fluid and after binding to fibrin, it is likely that degradation of HRG may occur at sites of high proteolysis and plasmin activity. Indeed, the finding that the HRG-derived peptide GHH20, as well as numerous other other 20mer peptides were antifungal, and as particularly noted for HRR-derived peptides, exhibiting a similar pH dependence as HRG, exemplifies that the holoprotein is not a prerequisite for antifungal action. Clearly, such possibilities need to be addressed in future studies.</p>", "<p>As previously mentioned, HRG is involved in various aspects of angiogenesis, coagulation, and fibrinolysis ##REF##15748207##[20]##, reflecting its interactions with ligands such as heparin, plasminogen, fibrinogen, and thrombospondin. Additionally, it acts as an opsonin by bridging FcyRI receptors on macrophages to DNA on apoptotic cells, stimulating phagocytosis ##REF##12391183##[50]##, and modulates the binding of IgG and immune complexes to FcγRI ##REF##12391183##[50]##. Considering these multiple roles, it is likely that HRG binding to microbial surfaces could induce additional “down-stream” effects, such as modulation of plasminogen activity and phagocytosis. The history of “classic” AMPs have shown that these molecules, initially believed to take part merely in direct microbial killing, have extended their roles into the ability to act as chemokines and to induce chemokine production leading to recruitment of leukocytes, promotion of wound healing, and an ability to modulate adaptive immunity ##REF##16909917##[51]##. Indeed, as interest in the <italic>in vivo</italic> functions of host defence peptides is increasing, it is important to consider the direct antimicrobial and immunomodulatory properties observed. Nevertheless, several findings in this study unequivocally demonstrate that HRG, like many AMPs, acts directly on microbes. Thus, in addition to the antifungal <italic>in vitro</italic> data, the enhanced fungal growth in HRG-deficient plasma, as well as the finding that <italic>Candida</italic> was detected at higher levels in blood of Hrg^−/− animals, indicates a direct antifungal action of the molecule. It is also interesting to note that these HRG deficient animals have also been shown to be more susceptible to <italic>Streptococcus pyogenes</italic> infection (Shannon et al, unpublished results). However, considering both AMP and HRG multifunctionality <italic>in vitro</italic> as well as <italic>in vivo</italic>, it may be envisaged that additional actions, resulting in the observed antifungal effects, will likely be revealed. All of these effects may be dependent on binding of HRG to microbes and subsequent interactions with cells (e.g., neutrophils and macrophages) in different compartments (e.g., skin, internal organs, and blood). In this respect, the pH dependence of HRG is particularly interesting and relevant. It is well known that infection foci, including abscesses, are characterized by low pH levels reaching as low as pH 5, due to increased anaerobic metabolism and lactate production, as well as leukocyte mediated oxidative burst and subsequent acidification ##REF##1651820##[52]##. The capacity of HRG to kill <italic>Candida</italic> at these pH levels and the corresponding increase in salt-resistance at low pH suggest that HRG could target infection foci, resulting in a physiologically relevant concentration and localization of antifungal activity. As previously mentioned, HRG's opsonising activity could hypothetically lead to enhanced phagocytosis. Although it remains to be investigated, such localisation of antifungal activity to endosomal compartments, where acidification could result in enhanced HRG-mediated killing of phagocytosed fungi, could serve as an effective way of eliminating invading <italic>Candida</italic> cells at sites of tissue inflammation without releasing potentially toxic microbial components.</p>", "<p>Again hypothetically, genetic deficiencies of HRG or acquired functional defects could provide interesting clues with respect to functional roles of HRG. In some patients, reduced levels of HRG are associated with a thrombophilic phenotype, indeed compatible with the phenotype observed in Hrg^−/− mice, which had a shorter prothrombin time ##REF##15869579##[22]##. As these patients still have ∼20–50% of normal levels of HRG, the human phenotype of complete absence of HRG remains, however unknown. Although patients with low levels of HRG have not been reported to be more prone to infections, it must be remembered that examples from deficiencies of particular innate immune proteins, e.g., complement and mannose-binding lectin, illustrate that even homozygous deficiency and a complete absence of a particular innate immune molecule may give rise to surprisingly mild symptoms. For example, patients with mannose-binding lectin deficiencies are normally not at risk of developing infections unless compromised by immune suppression or severe disease ##REF##16023210##[53]##. In this context, it is particularly interesting that antibodies against HRG have been detected in patients with antiphospholipid syndrome ##REF##17785336##[54]##, a disease associated with thrombodiathesis and systemic lupus erythematosus. Notably, the latter disease is associated with an increased risk for opportunistic infections, including <italic>Candida</italic>\n##REF##9494681##[55]##. Taken together, and considering the role of HRG in innate immunity, it should be of interest to study potential associations between functional inactivation(s) or deficiencies of HRG as well as genetically determined differences, in relation to the occurrence of infections.</p>", "<p>During the last three decades, research on innate immune molecules has demonstrated the significance of the innate immune system for prevention of invasion by microbes at biological boundaries. Previous studies have emphasized that various molecules, such as “classic” AMPs, complement factors, and cytokines, bridge between innate and adaptive immunity. The present work adds another significant component to this family of molecules, the plasma protein HRG.</p>" ]

[]

[ "<p>Conceived and designed the experiments: V. Rydengård, O. Shannon, M. Mörgelin, M. Malmsten, A. Schmidtchen. Performed the experiments: V. Rydengård, O. Shannon, K. Lundqvist, L. Kacprzyk, A. Chalupka, A. Olsson, M. Mörgelin, M. Malmsten. Analyzed the data: V. Rydengård. Contributed reagents/materials/analysis tools: W. Jahnen-Dechent. Wrote the paper: V. Rydengård, A. Schmidtchen. Critically read the manuscript: W. Jahnen-Dechent. All authors critically revised the manuscript.</p>", "<p>Fungi, such as <italic>Candida</italic> spp., are commonly found on the skin and at mucosal surfaces. Yet, they rarely cause invasive infections in immunocompetent individuals, an observation reflecting the ability of our innate immune system to control potentially invasive microbes found at biological boundaries. Antimicrobial proteins and peptides are becoming increasingly recognized as important effectors of innate immunity. This is illustrated further by the present investigation, demonstrating a novel antifungal role of histidine-rich glycoprotein (HRG), an abundant and multimodular plasma protein. HRG bound to <italic>Candida</italic> cells, and induced breaks in the cell walls of the organisms. Correspondingly, HRG preferentially lysed ergosterol-containing liposomes but not cholesterol-containing ones, indicating a specificity for fungal versus other types of eukaryotic membranes. Both antifungal and membrane-rupturing activities of HRG were enhanced at low pH, and mapped to the histidine-rich region of the protein. <italic>Ex vivo</italic>, HRG-containing plasma as well as fibrin clots exerted antifungal effects. <italic>In vivo</italic>, Hrg^−/− mice were susceptible to infection by <italic>C. albicans</italic>, in contrast to wild-type mice, which were highly resistant to infection. The results demonstrate a key and previously unknown antifungal role of HRG in innate immunity.</p>", "<title>Author Summary</title>", "<p>It has been estimated that humans contain about 1 kg of microbes, an observation that reflects our coexistence with colonizing microbes such as bacteria and fungi. The fungal species <italic>Candida</italic> is present as a commensal at mucosal surfaces and on skin. Although it may cause life-threatening infections, such as sepsis, particularly in immunocompromised individuals, it seldom causes disease in normal individuals. In order to control our microbial flora, humans as well as virtually all life forms are armoured with various proteins and peptides that comprise integral parts of our innate immune system. Here we describe a new component in this system; histidine-rich glycoprotein (HRG), an abundant plasma protein. We show, using a combination of microbiological, biochemical, and biophysical methods, that HRG exerts a potent antifungal activity, which is mediated via a histidine-rich region of the protein, and targets ergosterol-rich membrane structures such as those of <italic>Candida</italic>. HRG killed <italic>Candida</italic> both in plasma as well as when incorporated into fibrin clots. In mouse infection models, HRG was protective against systemic infection by <italic>Candida</italic>, indicating a novel antifungal role of HRG in innate immunity.</p>" ]

[ "<title>Supporting Information</title>" ]

[ "<p>We wish to thank Ms. Lise-Britt Wahlberg and Ms. Maria Baumgarten for expert technical assistance.</p>" ]

[ "<fig id=\"ppat-1000116-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.ppat.1000116.g001</object-id><label>Figure 1</label><caption><title>Antifungal activity and binding of HRG to <italic>Candida</italic>.</title><p>(A) Antifungal activity of HRG. <italic>C. parapsilosis</italic> ATCC 90018 (1×10⁵ cfu) was incubated with purified human HRG at concentrations ranging from 0.03 to 6 µM for 2 hours in 10 mM Tris, pH 7.4 (•) or 10 mM MES, pH 5.5 (○), plated and the number of cfu determined (n = 6). (B) Antifungal effects of HRG in salt. <italic>C. parapsilosis</italic> ATCC 90018 (1×10⁵ cfu) was incubated with 6 µM HRG for 2 hours in 10 mM MES, pH 5.5 containing 0, 25, 50, 100 or 150 mM NaCl, plated and the number of cfu was determined. (C) Killing kinetics. 0.3 or 3 µM HRG were incubated with 1×10⁵ cfu <italic>C. parapsilosis</italic> ATCC 90018 for 0, 5, 15, 30, 60 or 120 minutes in 10 mM Tris, pH 7.4 (•) or 10 mM MES, pH 5.5 (○), plated and the number of cfu determined. (D) Antifungal activity of HRG against different strains of <italic>Candida.</italic> 3 µM HRG were incubated with 1×10⁵ cfu <italic>C. parapsilosis</italic> ATCC 90018 or BD 17837, <italic>C. albicans</italic> ATCC 90028 or BD 1060, <italic>C. glabrata</italic> ATCC 90030 or <italic>C. krusei</italic> ATCC 6258 in 10 mM Tris, pH 7.4 (black bars) or 10 mM MES, pH 5.5 (white bars) for 2 hours, plated and number of cfu determined (n = 6). (E) Binding of HRG to fungi. <italic>C. parapsilosis</italic> (1×10⁵ cfu) was incubated with HRG (0.6 µM) in 10 mM MES, pH 5.5. For inhibition studies, heparin (50 µg/ml) was added. Samples were centrifuged and the pellet and supernatants were extracted and run on 8% SDS-PAGE under reducing conditions. HRG was detected by western and immunoblotting using polyclonal antibodies against GHH20. Purified HRG was used as a positive control (labeled C). (F) Flow cytometry analysis of binding of HRG to fungal membranes. <italic>C. parapsilosis</italic> (5×10⁷ cfu) were incubated with FITC-labeled HRG in 10 mM Tris pH 7.4 or 10 mM MES pH 5.5.</p></caption></fig>", "<fig id=\"ppat-1000116-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.ppat.1000116.g002</object-id><label>Figure 2</label><caption><title>HRG induces membrane permeabilisation of <italic>Candida</italic> cells as well as liposomes.</title><p>(A) Negative staining and electron microscopy analysis of <italic>C. parapsilosis</italic> exposed to HRG. <italic>C. parapsilosis</italic> ATCC 90018 were incubated in the absence of HRG in 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5. These fungi did not exhibit signs of membrane perturbations. In contrast, when treated with 10 µM HRG in 10 mM Tris, pH 7.4 or 10 mM MES, pH 5.5 membrane damage, blebbing and ejection of cytoplasmic components was observed. Fungi treated with 10 µM LL-37 was used as a positive control for membrane damage. The scale bar corresponds to 2 µm. (B) Fungal viability after incubation with HRG and LL-37. <italic>C. albicans</italic> ATCC 90028 were incubated with 10 µM HRG or LL-37 in either 10 mM Tris, pH 7.4 (left panel) or 10 mM MES, pH 5.5 (right panel). The left images in each row are Nomarski Differential Interference Contrast images, whereas the right images show FITC fluorescence of fungi. (C) Effects of 1 µM HRG on liposome permeability. Left panel. Increase in HRG-induced permeabilization of ergosterol-containing liposomes is detected at pH 6.0. Right panel. Increased HRG-induced lysis of (at pH 7.4) ergosterol containing liposomes. (D) CD spectroscopy of HRG under different conditions. CD spectra for 0.25 µM HRG in buffer and in presence of <italic>S. cerevisiae</italic> mannan are presented.</p></caption></fig>", "<fig id=\"ppat-1000116-g003\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.ppat.1000116.g003</object-id><label>Figure 3</label><caption><title>The antifungal activity of the histidine-rich domain of HRG is significantly increased at low pH.</title><p>(A) Sequence of HRG and synthetic peptides used in this study are indicated. (B) Screening of antifungal epitopes of HRG. 20-mer peptides spanning the whole sequence of HRG (for sequences see ##SUPPL##0##Table S1##) were used in radial diffusion assays against <italic>C. parapsilosis</italic> ATCC 90018 and <italic>C. albicans</italic> ATCC 90028 in 10 mM Tris, pH 7.4 or in 10 mM MES, pH 5.5. A 4 mm diameter well was loaded with 6 µl of 100 µM peptide. The clearance zones (mm) were measured after an overnight incubation at 27°C (n = 6). (C) Comparison of the antifungal activity of rHRG and the truncated version rHRG1-240. <italic>C. parapsilosis</italic> ATCC 90018 (1×10⁵ cfu) was incubated with 0.6 µM rHRG or rHRG1-240 in 10 mM Tris, pH 7.4 or in 10 mM MES, pH 5.5. Samples were plated and the number of cfu was determined. Significance was determined using Kruskall-Wallis one-way ANOVA analysis (*** p&lt;0.001, n = 6).</p></caption></fig>", "<fig id=\"ppat-1000116-g004\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.ppat.1000116.g004</object-id><label>Figure 4</label><caption><title>Antifungal activity and fungal binding of GHH20 peptide.</title><p>(A) Antifungal activity of GHH20. <italic>C. parapsilosis</italic> ATCC 90018 <italic>or C. albicans</italic> ATCC 90028 (1×10⁵ cfu) were incubated with GHH20 peptide (0.03 to 6 µM) for 2 hours in 10 mM Tris, pH 7.4 (•) or 10 mM MES, pH 5.5 (○), plated and the number of cfu determined. A representative experiment (of three) is shown. (B) Flow cytometry analysis of binding of GHH20 to fungal membranes. <italic>C. parapsilosis</italic> (5×10⁷ cfu) were incubated with 20 µg TAMRA-labeled GHH20 in 10 mM Tris pH 7.4 or 10 mM MES pH 5.5. (C) Binding of 2 µg TAMRA-labeled GHH20 peptide to <italic>C. parapsilosis</italic> ATCC 90018 and inhibition by an excess of heparin. <italic>C. parapsilosis</italic> were incubated with TAMRA-labeled GHH20 in 10 mM MES (panel 3), pH 5.5 or the same buffer supplemented with heparin (50 µg/ml) (panel 4). The left panel shows Nomarski images (1 and 3), whereas the right panel shows red fluorescence of peptide bound to fungi. (D) The GHH20 peptide permeabilizes ergosterol-containing liposomes preferably at pH 6.0. 1 µM GHH20 was used (n = 6).</p></caption></fig>", "<fig id=\"ppat-1000116-g005\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.ppat.1000116.g005</object-id><label>Figure 5</label><caption><title>Localization and activities of HRG.</title><p>(A) Analysis of HRG in biological fluids. The indicated biological materials were electrophoresed on a 8% gel (Tris-Glycine, non-reducing conditions) (left panel) or on a 16.5% Tris-Tricine gel under reducing conditions (right panel) and transferred to a nitrocellulose membrane. Western blot was performed using polyclonal antibodies directed against the GHH20 epitope of HRG. (B) Localization of HRG in fibrin clots. Human control plasma (panel 4) or plasma depleted of HRG (panel 2) were incubated with FITC-labeled HRG and clots were generated overnight after addition of 10 mM Ca²⁺ at 37°C. The clots were mounted on slides and visualized by fluorescence microscopy. The left side shows Nomarski images (1 and 3), whereas the right part shows fluorescence of HRG associated with the clots. (C) Candida growth in plasma. <italic>C. parapsilosis</italic> (2×10⁷ cfu) was inoculated in human plasma (•) or human HRG-deficient plasma (○) and incubated at 27°C for 0, 4 ,8 or 18 hours and the number of cfu was determined (n = 6). (D) Antifungal activity of HRG <italic>ex vivo.</italic> Mouse control plasma or plasma of Hrg^−/− mice were used to form fibrin clots in the presence of 10 mM Ca²⁺. Clots were incubated with <italic>C. parapsilosis</italic> ATCC 90018 (1×10⁵ cfu) in 10 mM MES, pH 5.5 for 2 hours, plated and the number of cfu determined (p = 0.043, n = 6).</p></caption></fig>", "<fig id=\"ppat-1000116-g006\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.ppat.1000116.g006</object-id><label>Figure 6</label><caption><title>Candida produces severe infection in Hrg^−/− mice.</title><p>(A) Weight loss in mice after infection with <italic>C.albicans</italic>. C57BL/6 (solid line) and C57BL/6 Hrg^−/− (dotted line) mice were infected (i.p.) with 1×10⁹ cfu of <italic>C. albicans</italic>, and the body weight was followed from day 0 to day 3 (n = 6,4 and 2, respectively) (p = 0.02). (B) Fungal dissemination to the bloodstream. C57BL/6 (•) (n = 4) and C57BL/6 Hrg^−/− (○) (n = 5) mice were infected as above and the animals were sacrificed on day 2 and number of cfu in blood was determined (p = 0.032). (C) HRG suppresses fungal dissemination to the spleen and kidney. C57BL/6 (•) and C57BL/6 Hrg^−/− (○) mice were infected as above and the cfu of <italic>C. parapsilosis</italic> in spleen and kidney was determined (p = 0.009, n = 10) (D) In two animals kidneys were collected on day 3, fixed, sectioned and then stained with hematoxylin and eosin (HE) (left panel) or PAS (right panel) (upper section; magnification ×10 and lower section; magnification ×30).</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"ppat.1000116.s001\"><label>Table S1</label><caption><p>Synthetic 20-mer peptides spanning the whole sequence of HRG, used in the screening of antifungal activity in ##FIG##2##Figure 3B##, and relevant descriptive parameters (net charge, activity against <italic>C. albicans</italic> and <italic>C. parapsilosis</italic>, content (%) of the basic amino acids K, R, H, and the acidic D, E.</p><p>(0.07 MB DOC)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"ppat.1000116.s002\"><label>Figure S1</label><caption><p>Correlation between net charge and antifungal activity. 20-mer peptides spanning the whole sequence of HRG (for sequences see ##SUPPL##0##Table S1##) were used in radial diffusion assay against <italic>C. albicans</italic> ATCC 90028 in 10 mM Tris, pH 7.4 (•) or in 10 mM MES, pH 5.5 (○). A 4 mm diameter well was loaded with 6 µl of 100 µM peptide. The clearance zones (mm) were measured after an overnight incubation at 27°C. The equation for the line of regression is y = 0.605x + 1.222 for peptides in pH 7.4 and y = 0.543x + 0.662 for pH 5.5.</p><p>(31.73 MB TIF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"ppat.1000116.s003\"><label>Figure S2</label><caption><p>Inhibition of antifungal activity by heparin. 0.3 µM HRG were incubated with 1×10⁵ cfu <italic>C. parapsilosis</italic> ATCC 90018 in 10 mM MES, pH 5.5 with or without 50 µg heparin, plated and the number of cfu determined (n = 6).</p><p>(1.16 MB PSD)</p></caption></supplementary-material>" ]

[ "<fn-group><fn fn-type=\"COI-statement\"><p>Drs. Schmidtchen and Malmsten have shares in DermaGen AB, a company involved in therapeutical development of antimicrobial peptides. Peptides of HRG are included in patent applications of the company.</p></fn><fn fn-type=\"financial-disclosure\"><p>This research was supported by grants from from the Swedish Research Council (projects 13471 and 2006-4469), the Royal Physiographic Society in Lund, the Söderberg, Welander-Finsen, Crafoord, Österlund, Lundgrens, Lions and Kock Foundations, DermaGen AB, and The Swedish Government Funds for Clinical Research (ALF).</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"ppat.1000116.g001\"/>", "<graphic xlink:href=\"ppat.1000116.g002\"/>", "<graphic xlink:href=\"ppat.1000116.g003\"/>", "<graphic xlink:href=\"ppat.1000116.g004\"/>", "<graphic xlink:href=\"ppat.1000116.g005\"/>", "<graphic xlink:href=\"ppat.1000116.g006\"/>" ]

[ "<media xlink:href=\"ppat.1000116.s001.doc\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"ppat.1000116.s002.tif\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"ppat.1000116.s003.psd\"><caption><p>Click here for additional data file.</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>The first Europeans from the Old World to land in what is now US territory were Columbus' men in 1493. The initial colonization of the region by the Spanish, English, Scots and Irish, French, Dutch, Swedes, Germans, Italians and Portuguese during the 16th and 17th centuries was followed in the 19th and early 20th century by waves of millions of newcomers originating from the northwestern to the southeastern corners of Europe ##UREF##0##[1]##. Thus, the present day European American population is a mosaic of people that represent different levels of admixture between diverse European populations and, to some degree, also with Native American and African American populations.</p>", "<p>The identification of population genetic structure has been discussed at length in recent literature, due to the potential bias it can introduce in association studies, searching for susceptibility genes for common complex disorders ##REF##8091226##[2]##–##REF##16041375##[5]##. Population stratification is a source of confounding in case-control studies, when allele-frequency heterogeneity that is unrelated with the studied phenotype is coupled with disease-risk heterogeneity and biased sampling in cases and controls. Although European populations were initially considered genetically quite homogeneous, it has recently been shown that significant patterns of structure within Europe along a north to south axis do exist and that unidentified population stratification in European derived populations (European Americans) can lead to spurious associations with disease ##REF##16041375##[5]##–##REF##17436249##[8]##.</p>", "<p>As genotyping of thousands of individuals for hundreds of thousands of markers becomes feasible ##REF##17554300##[9]##–##REF##17632545##[14]##, and genome wide association studies in large samples of European American populations become increasingly common ##REF##17728769##[15]##, identifying and correcting for population stratification will undoubtedly play a central part in the quest to unravel the genetic basis of complex traits. The uniform adjustment proposed by the method of genomic control could be too conservative ##REF##11315092##[16]##,##REF##15812171##[17]##, while structured association testing is computationally impractical for very large datasets ##REF##10827107##[18]##. Price et al. ##REF##16862161##[19]## have shown that Principal Components Analysis (PCA), a powerful linear dimensionality reduction technique can be used as a computationally efficient tool to correct for stratification in the setting of genome wide association studies without loss in power.</p>", "<p>Identifying a small set of markers that could be used for inference of population structure and adjustment for stratification is of particular importance in order to reduce genotyping costs in studies seeking to replicate the findings of large-scale genome-wide projects or when pursuing specific loci as candidate susceptibility genes. Most existing metrics to select ancestry informative markers (AIMs) are allele frequency based and demand prior knowledge of the ancestry of the studied individuals. Consequently, measures like <italic>F_st</italic>, <italic>δ</italic> and informativeness for assignment ##REF##7942857##[20]##–##REF##14631557##[25]##, require prior assumptions about individual ancestry and cannot be directly applied to admixed populations, like European Americans, in order to identify a panel of genetic markers that can reproduce the structure of the dataset. European American AIMs had so far been proposed in two recent large studies targeting distinct European populations, used as proxies for European American ancestry ##REF##17044734##[7]##,##REF##17436249##[8]##. Our work here as well as two studies parallel to ours described in ##REF##18208327##[26]##,##REF##18208329##[27]## are the first to attempt the identification of structure informative SNPs through the direct analysis of genomewide datasets of European Americans. All three of these studies are PCA-based. However, here, we directly leverage the power of PCA for the selection of AIMs ##REF##17892327##[28]##, without the need for any intermediate steps, such as assigning individuals to clusters, in order to use allele frequency based metrics.</p>", "<p>We have recently introduced an unsupervised method for the selection of ancestry and structure informative SNPs (PCA-correlated SNPs or PCA-informative SNPs-PCAIMs) ##REF##17892327##[28]##. Our method does not require prior hypotheses or knowledge of individual ancestry and thus is well-suited for selecting AIMs in admixed populations. In this paper, we employ it to analyze a dense, genome-wide dataset (approx. 307,000 SNPs) of more than 1,500 European Americans from two different studies ##REF##17052657##[29]##,##REF##18439552##[30]##. Our main goal is the identification of a small panel of structure informative SNPs in the European American population. The contributions of this paper are three-fold. <italic>First</italic>, from a statistical perspective, we propose a methodology to remove redundancy from any set of genetic markers, an issue that arises with all existing methods (supervised or unsupervised) for the selection of ancestry or structure informative markers, since the “scoring” of the SNPs in all of these methods does not take into account any correlation between them. We reduce the redundancy removal problem to a well-known problem in numerical linear algebra, the so-called Column Subset Selection Problem ##UREF##1##[31]## and we propose an efficient and accurate algorithm that filters out redundant SNPs. <italic>Second</italic>, we demonstrate that as few as 200 SNPs selected with our methodology can be used to very accurately predict the fine structure of European Americans as identified by PCA, and we employ cross-validation experiments to verify the accuracy of our approach. <italic>Third</italic>, we show that our method can be coupled with PCA-based stratification correction tools (such as EIGENSTRAT ##REF##16862161##[19]##) for accurate stratification correction with significant genotyping savings. Using simulated data we experimentally demonstrate that 100–200 PCAIMs can be used to correct for stratification while maintaining power in association studies.</p>" ]

[ "<title>Methods</title>", "<title>Datasets</title>", "<p>We studied two independent European American datasets. The first dataset (CHORI dataset-Children's Hospital Oakland Research Institute), consists of 980 individuals, that were collected as part of two community-based clinical trials evaluating the anti-inflammatory effects of statins. 305 of these samples (part of the CAP study ##REF##16516587##[32]##) were collected from the San Francisco Bay Area and Los Angeles. These individuals all had to report at least 3 grandparents of European or Caucasian background. Another 675 individuals were part of a clinical trial that included a large number of sites across the U.S. (PRINCE study ##REF##11434828##[33]##). These individuals were self-reported white or Caucasian but no additional information was collected about their parents. All 980 individuals were genotyped using the Illumina Infinium 310K array in one laboratory under the same conditions. The second dataset that we studied here (CORIELL dataset), is a publicly available dataset that has been previously described ##REF##17052657##[29]##, and consists of the same SNPs genotyped for 541 samples (data available from the SNP Resource at the NINDS Human Genetics Resource Center DNA and Cell Line Repository (<ext-link ext-link-type=\"uri\" xlink:href=\"http://ccr.coriell.org/ninds/\">http://ccr.coriell.org/ninds/</ext-link>). These are samples from patients with Parkinson's disease and neurologically normal controls, curated at the Coriell institute. Again, genotyping was performed using the Illumina platform (in the laboratory of Drs. Singleton and Hardy (NIA, LNG), Bethesda, MD USA). For all datasets we only considered genotypes for SNPs on autosomal chromosomes in our analysis. Finally, as a third dataset, we also studied the same SNPs using data available from the HapMap database on the HapMap Yoruba (YRI), CEPH European (CEU), Chinese (CHB), and Japanese (JPT) samples ##REF##14685227##[34]##,##REF##16255080##[35]##.</p>", "<title>Preprocessing and Encoding the Data</title>", "<p>The proportion of missing entries in the above datasets was very small (on average less than 0.1%). As a quality control step, we excluded all SNPs with more than 5% missing entries (separately on each of the three datasets). This step further reduced the number of missing entries to less that 0.07% on average. We also excluded from the analysis a small number of SNPs that were not in Hardy-Weinberg equilibrium (HWE). After these preprocessing steps we were left with a total of 307,315 autosomal SNPs that all three datasets had in common.</p>", "<p>In order to simplify and speed up our computations, we filled in the (very small) number of missing entries randomly so that HWE is satisfied for each SNP. The probabilistic filling in was performed separately for each dataset, and separately in each population of the HapMap data. We then transformed the raw data to numeric values, without any loss of information, in order to apply our linear algebraic methods. Consider a dataset of a population <italic>X</italic> consisting of <italic>m</italic> subjects and assume that for each subject <italic>n</italic> biallelic SNPs have been assayed. Thus, we are given a table <italic>T^x</italic>, consisting of <italic>m</italic> rows and <italic>n</italic> columns. Each entry in the table is a pair of bases, ordered alphabetically. We transform this initial data table to an integer matrix <italic>A^x</italic> which consists of <italic>m</italic> rows (one for each subject), and <italic>n</italic> columns (one for each SNP). Each entry of <italic>A^x</italic> will be −1, 0, +1, or empty. Let <italic>B</italic>\n₁ and <italic>B</italic>\n₂ be the bases that appear in the <italic>j</italic>-th SNP (in alphabetical order). If the genotypic information for the <italic>j</italic>-th SNP of the <italic>i</italic>-th individual is <italic>B</italic>\n₁\n<italic>B</italic>\n₁ the (<italic>i,j</italic>) -th entry of <italic>A^x</italic> is set to +1; else if it is <italic>B</italic>\n₁\n<italic>B</italic>\n₂ the (<italic>i,j</italic>)-th entry of <italic>A^x</italic> is set to 0; else if it is <italic>B</italic>\n₂\n<italic>B</italic>\n₂ the (<italic>i,j</italic>)-th entry of <italic>A^x</italic> is set to −1 ##REF##17892327##[28]##,##REF##17151345##[36]##.</p>", "<title>The Singular Value Decomposition and Outlier Removal</title>", "<p>We carefully studied the two European American datasets for outlier individuals. In the CHORI dataset, we identified five pairs of individuals that showed a very high degree of allele sharing and removed these ten subjects from all further analysis. In particular, we determined the proportion of allele sharing between all pairs of individuals for 1000 randomly selected markers, approximately equally spaced throughout the genome, and subtracted it from the proportion of allele sharing expected under a randomly mating population with the same allele frequencies. These five pairs included one pair that had 100% sharing for all 1000 markers (indicating either an identical twin or a duplicate sample) and four others that had significantly higher than expected excess allele sharing, suggesting that they were related.</p>", "<p>We subsequently used Principal Components Analysis and the Singular Value Decomposition to detect outliers. In particular, given <italic>m</italic> subjects and <italic>n</italic> SNPs, let the <italic>m</italic>×<italic>n</italic> matrix <italic>A</italic> denote the subject-SNP matrix encoded as described above. After mean-centering the columns (SNP genotypes) of <italic>A</italic>, the SVD of the matrix returns <italic>m</italic> pairwise orthonormal vectors <italic>uⁱ</italic>, <italic>n</italic> pairwise orthonormal vectors <italic>vⁱ</italic>, and <italic>m</italic> non-negative singular values <italic>σ_i</italic> such that <italic>σ</italic>\n₁≥<italic>σ</italic>\n₂≥…≥<italic>σ_m</italic>≥0. The matrix <italic>A</italic> may be written as a sum of outer products as\n</p>", "<p>Each triplet (<italic>σ_i</italic>,<italic>uⁱ</italic>,<italic>νⁱ</italic>) may be used to form a principal component of <italic>A</italic>. Formally, the <italic>i</italic>-th most significant principal component of a matrix <italic>A</italic> is the rank-one matrix that is equal to . In our setting, the left singular vectors (the <italic>uⁱ</italic> 's) are linear combinations of the columns (SNPs) of <italic>A</italic> and will be called eigenSNPs ##REF##15389393##[37]##. Notice that a principal component is a matrix, whereas an eigenSNP is just a column vector. PCA is a well-known dimensionality reduction technique that, in this case, represents all subjects with respect to a small number of eigenSNPs, corresponding to the top few principal components. All further analysis is then performed on this low-dimensional representation.</p>", "<p>\n##SUPPL##0##Figure S1## shows the plot of the 970 CHORI individuals, the 541 CORIELL individuals, and the HapMap European, African and Asian samples, projected on their top three eigenSNPs (as we shall argue in <xref ref-type=\"sec\" rid=\"s3\">Results</xref> the top eigenSNP is the most informative). This plot illustrates how a few subjects from our European American datasets are “pulled” towards the African and Asian HapMap populations. Based on this analysis, we discarded 12 individuals from the CHORI dataset and 2 individuals from the CORIELL dataset that were far from the vast majority of the European American subjects and seem to have a higher degree of non-European ancestry (##SUPPL##0##Figure S1##). Overall, out of the 1521 subjects in the CHORI and CORIELL datasets, we discarded a total of 24 subjects (ten suspiciously similar subjects and 14 outliers). Thus, we were left with 1497 subjects of European American ancestry, genotyped for 307,315 SNPs.</p>", "<title>Selecting PCAIMs and Removing Redundancy</title>", "<p>In order to select ancestry informative markers, we used the procedure that we described previously ##REF##17892327##[28]##, ##UREF##2##[38]##–##UREF##4##[40]##. This procedure is based on the well-documented fact that Principal Components Analysis reveals population structure. More specifically, a number of studies have verified that retaining the top few eigenSNPs in datasets that contain individuals from a number of different populations, or even admixed populations, efficiently reveals the ancestry of the individuals ##REF##16862161##[19]##, ##REF##17892327##[28]##, ##REF##356262##[41]##–##REF##16848973##[44]##. The PCAIM selection algorithm first determines the number of significant principal components (and thus the number of informative eigenSNPs) in the data, and then assigns a score to each SNP. Higher scores correspond to SNPs that correlate well with all informative eigenSNPs. The algorithm returns the top scoring SNPs, and we have demonstrated that these PCAIMs are very efficient for ancestry prediction ##REF##17892327##[28]##.</p>", "<p>This algorithm does not take any special measures in order to avoid redundancy in the set of identified markers. As we will also discuss later here, redundancy may arise in sets of AIMs selected with any of the existing methods (eg. <italic>δ</italic>, <italic>F_st</italic>, informativeness, PCAIMs). Redundancy in the case of dense sets of SNP markers is due to tight linkage disequilibrium. Given the increased marker density in the genomewide datasets that are becoming available today, this may lead to significant loss in efficiency by selecting highly correlated markers. It is therefore important to add a redundancy removal step after the initial selection of structure informative markers.</p>", "<p>We propose a simple, efficient methodology to deal with this issue. Our methodology is based on reducing the redundancy removal problem to the so-called Column Subset Selection Problem. The latter problem is well studied in the Numerical Linear Algebra literature, and many algorithms, with various accuracy vs speed tradeoffs, have been proposed ##UREF##5##[45]##. More specifically, assume that the top <italic>r</italic>≪<italic>n</italic> highest scoring SNPs are retained as PCAIMs. Thus, we are given a matrix <italic>Ã</italic> that has <italic>m</italic> rows (one for each subject) but only <italic>r</italic> columns (one for each PCAIM). Recall that <italic>n</italic> is the total number of SNPs, and could be in the order of hundreds of thousands, whereas we expect <italic>r</italic> to be in the order of thousands. Our goal is to only retain a small number (say <italic>k</italic>) of columns of <italic>Ã</italic> that are as uncorrelated as possible. A naive way of solving this problem would be to examine all possible choices of sets of <italic>k</italic> SNPs and keep a set that has no pairs of highly correlated SNPs. This is computationally infeasible even for very small values of <italic>k</italic> (say ten) if <italic>r</italic> is even a thousand. Consider the following definition for the C<sc>olumn</sc> S<sc>ubset</sc> S<sc>election</sc> P<sc>roblem</sc> (CSSP):</p>", "<p>\n<bold>Definition 1:</bold>\n<italic>Given an m×r matrix à and a positive integer k, pick k columns (SNPs) of à such that the maximal Pearson correlation coefficient between all </italic>\n\n<italic> pairs of the selected columns (SNPs) is minimized.</italic>\n</p>", "<p>In words, recall that a large (close to one) Pearson correlation coefficient between a pair of SNPs would imply that one of the two SNPs in redundant. Thus, the above problem formulation seeks to minimize the maximal correlation between any pair of selected SNPs, and thus ensure that limited or no redundancy exists. Even though solving the above optimization problem exactly is hard, efficient approximation algorithms exist. For the purposes of this paper, we chose to use an algorithm called <italic>greedy QR</italic>, that was proposed by Golub in ##UREF##1##[31]## and was subsequently analyzed by Gu and Eisenstat in ##UREF##6##[46]##. The algorithm essentially works in <italic>k</italic> iterations; in the first iteration, the first column of <italic>Ã</italic> (the top PCAIM) is picked; in the second iteration, a column of <italic>Ã</italic> is picked that is as uncorrelated with the first column as possible; in the third iteration the chosen column has to be as uncorrelated as possible with the first two columns, etc. When expressed in linear algebraic notation, this iterative procedure boils down to a permuted <italic>QR</italic> decomposition of a matrix, and can be performed efficiently. In particular, an efficient implementation of this algorithm is available in MatLab, and runs in less than one minute when <italic>r</italic> is in the order of thousands and any value of <italic>k</italic> less than <italic>r</italic>.</p>", "<title>Simulated Association Studies</title>", "<p>In order to illustrate the potential of the proposed PCAIMs for the correction of stratification in association studies, we run a large simulated association study that closely followed the simulated association study in Price et al. ##REF##16862161##[19]##. More specifically, Price et al. ##REF##16862161##[19]## demonstrated how EIGENSTRAT (a PCA-based procedure) could efficiently identify population structure and remove stratification from association studies on populations with similar structural characteristics with the European American population.</p>", "<p>To demonstrate the performance of PCAIMs to correct for stratification in association studies on admixed populations with similar characteristics with European American populations, we followed the methods of ##REF##16862161##[19]## to generate an admixed population of 1,000 individuals genotyped on 100,000 SNPs (see ##SUPPL##7##Text S1## for details). Thus, we created a 1,000×100,000 matrix <italic>A</italic> of genotypes. We then estimated the number of significant principal components, both by looking at the singular values, as well as by the permutation test of ##REF##17892327##[28]##. As we will discuss in the <xref ref-type=\"sec\" rid=\"s3\">Results</xref> section, one eigenSNP was deemed significant and was interpreted as ancestry. We then picked panels of PCAIMs from the 100,000 SNPs in order to predict the ancestry of the 1,000 subjects.</p>", "<p>We created large sets of random, stratified, and causal SNPs (100,000 SNPs in each case) following the methods described by Price et al. ##REF##16862161##[19]## (see ##SUPPL##7##Text S1##). We performed ten repetitions, and generated sets of 100,000, since we did not observe any change in the fourth decimal digit of the reported results by increasing the set size to 1,000,000. Affection status for individuals in the admixed population was determined randomly according to an “ancestry risk” parameter <italic>r</italic> as defined previously ##REF##16862161##[19]##. Results are reported for both <italic>r</italic> = 2 and 3.</p>", "<p>Correlation with affection status was determined by taking the Armitage trend statistic of each SNP with the affection status, with the significance threshold set to 10⁻⁴. For comparison purposes we chose the same threshold as in ##REF##16862161##[19]##. Correction for ancestry was first performed using the algorithm of EIGENSTRAT and looking at the top ten eigenSNPs of the full SNP-subject matrix (mean centering was performed). We then performed correction for stratification by looking at the first eigenSNP of the matrix consisting of the panel of selected PCAIMs. Adjustment of genotypes essentially corresponds to “projecting out” the component of each SNP that lies in the subspace spanned by the ancestry prediction. After performing this simple linear algebraic operation on every SNP, the Armitage trend statistic was re-run on the residual of each SNP.</p>" ]

[ "<title>Results</title>", "<title>Population Substructure and Ancestry in European Americans</title>", "<p>We first examined the number of significant principal components in the two European American datasets that we studied (CHORI and CORIELL). ##FIG##0##Figure 1## (panel A) shows the top few singular values of the CORIELL subject-SNP matrix, and ##FIG##0##Figure 1## (panel D) shows the top few singular values of the CHORI subject-SNP matrix. Clearly, there is a significant gap between the first singular value and the remaining ones in both cases. This is a strong indication that the top principal component is the most informative in both datasets and suggests that subsequent principal components may not be of interest. To further validate this finding, we ran the permutation test that we have recently described ##REF##17892327##[28]##. This permutation test essentially measures the ratio of “information” that the <italic>i</italic>-th principal component contains when compared to the amount of structure in a random matrix. When this ratio is sufficiently high, the principal component is deemed as informative. Again, ##FIG##0##Figure 1## (panels B and E), shows that, for both datasets, the first principal component has significantly more structure than a random matrix, whereas the remaining principal components are much less informative and contain less than 20% more information than a purely random matrix.</p>", "<p>The analysis described above suggests that both in the CORIELL and CHORI datasets, individuals of European American ancestry lie along a line, and all the variation is concentrated across the first eigenSNP, which corresponds to the first principal component. Although no information about self-reported ancestry was available for the individuals we studied, we can speculate that this axis of variation corresponds to the well-documented axis of northern to southeastern genetic variation in Europe ##REF##17044734##[7]##, ##REF##17436249##[8]##, ##REF##18208327##[26]##, ##REF##18208329##[27]##, ##REF##356262##[41]##, ##UREF##7##[47]##–##REF##12355353##[50]##. Hence we only retained the top principal component for our European American datasets for all further analysis and we interpreted this principal component as the European American ancestry axis. ##FIG##0##Figure 1## (panels C and F), shows the histogram of the top eigenSNP for individuals in the CORIELL and the CHORI datasets respectively. We would also like to add here a note on the computational efficiency of our methods: our computations are quite efficient and, for example, running PCA on the joint CHORI and CORIELL datasets takes 21 minutes on a standard laptop computer.</p>", "<p>We then compared the structure of the two European American datasets to the structure of the HapMap Yoruba from Ibadan (YRI), CEPH European (CEU), and East Asian populations (CHB and JPT) of the HapMap project. To this end we extracted from the HapMap database genotypes for all SNPs that were also genotyped on our European American samples and computed the top few eigenSNPs of all five populations. ##FIG##1##Figure 2## shows all 1767 individuals (1497 CHORI and CORIELL plus 270 from HapMap) projected on the first, second, and third eigenSNP of the overall subject-SNP matrix. Adding the HapMap data adds two more axes of variation, one for the African subjects, and one for the Asian subjects. The two large European American samples have similar structure with most individual variation lying across one axis. As expected, they overlap with the CEPH European data. Since outliers were removed as part of a preprocessing step (see <xref ref-type=\"sec\" rid=\"s2\">Methods</xref>), no individuals seem to demonstrate high levels of admixture with non-Europeans. CEPH Europeans form a very tight cluster, which does not seem to encompass the full range of variation observed in European Americans. This also becomes apparent in ##SUPPL##1##Figure S2##, which focuses on the CHORI, the CORIELL, and the CEPH European datasets only. The fact that the CEPH European samples essentially represent US residents from Utah with Northern European ancestry, corroborates with this picture. Thus, the position of the CEPH European samples in this analysis seems to mark the end of the axis of variation in our European American datasets, which corresponds to Northern European ancestry.</p>", "<title>Using PCAIMs to Capture European American Population Structure</title>", "<p>We next tested the feasibility of identifying a small subset of SNPs that could be used to reproduce the structure of the European Americans that we analyzed. Using our algorithm ##REF##17892327##[28]## with the number of significant principal components set to one, we selected 100 to 3000 PCAIMs in each dataset in order to predict the ancestry of the European American subjects. As described earlier here, in both European American datasets that we studied, variation lies almost exclusively along the first eigenSNP, which was interpreted as ancestry of the studied individuals.</p>", "<p>In order to evaluate the performance of the PCAIMs that we select, and show that they can be used to preserve the properties of the complete dataset, we computed the first eigenSNP using all available 307,315 SNPs, and compared it to the first eigenSNP using only the selected subset of SNPs. Thus, we essentially predicted the ancestry of each individual by looking at a small subset of SNPs and computing the first eigenSNP of the resulting subject-SNP matrix. ##FIG##2##Figure 3## shows the Pearson correlation coefficients between “true” and predicted ancestry. In the CHORI dataset, about 1,200 PCAIMs are needed in order to reach a correlation coefficient of above 0.9 and 700 are needed in the CORIELL. Random SNPs perform much worse in the CORIELL and as many as 3,000 random SNPs are needed for the correlation coefficient between true and predicted coordinates of the individuals to reach 0.9. On the other hand, in the CHORI dataset, random SNPs perform worse but overall have comparable performance to PCAIMs (correlation coefficient between “true” and predicted ancestry of individuals is approximately 0.9 with 2,000 SNPs). As we will show in the following section this is due to the redundancy in the markers selected as informative and great savings are indeed possible, after application of our redundancy removal algorithm.</p>", "<title>Removing Redundant PCAIMs</title>", "<p>Even though less than 1% (approx. 1,500–2,000) of the total SNPs suffice to predict ancestry in the studied European American datasets with very high accuracy, we still considered this number to be unnecessarily high. This is reinforced by the fact that 2,000–3,000 random SNPs start performing quite well in predicting ancestry. This led us to suspect that the sets of PCAIMs that we were selecting included significant amounts of redundancy. Indeed, we computed all <italic>r</italic>\n² values between all pairs of selected PCAIMs for the CHORI dataset, the CORIELL dataset, as well as the joint CHORI-CORIELL dataset. The results are shown in ##TAB##0##Table 1##. Obviously, a large number of pairs are in high LD, and thus a lot of the selected SNPs are redundant.</p>", "<p>In an effort to further reduce the number of SNPs that are necessary for ancestry prediction in European Americans and increase genotyping savings, we developed an algorithm that minimizes redundancy from the panels of SNPs that are selected with our scoring algorithm. Applying the redundancy removal procedure described in <xref ref-type=\"sec\" rid=\"s2\">Methods</xref>, we extracted panels of non-redundant SNPs from the top 3,000 PCA-correlated SNPs. We varied the size of these panels from 100 to 500 PCA-correlated non-redundant SNPs. As is shown in ##FIG##2##Figures 3## and ##FIG##3##4##, removing redundancy from the selected PCAIMs results in significant savings with as few as 200 SNPs sufficing to accurately predict individual ancestry (with a correlation coefficient above 0.9). Additionally, when we computed all pairwise <italic>r</italic>\n² values in the top 500 non-redundant PCA-correlated SNPs for the CORIELL dataset, the CHORI dataset, and the joint dataset, we observed that there was not a single pair of SNPs with an <italic>r</italic>\n² value above 0.2 and only three pairs in the CORIELL dataset with an <italic>r</italic>\n² value between 0.1 and 0.2. Thus, our algorithm effectively removed redundant SNPs.</p>", "<p>In order to generate a potentially more comprehensive list of structure informative SNPs for European Americans, we also analyzed the two datasets jointly (##FIG##3##Figure 4## and ##SUPPL##2##Figure S3##) and tested the efficiency of selected subsets of PCAIMs. Again PCAIMs, after redundancy removal, prove to be quite powerful and as few as 200 can be used to accurately predict the structure of 1497 individuals. ##SUPPL##3##Figure S4## shows the scores of selected PCAIMs plotted along each autosome.</p>", "<title>Cross-Validation Experiments</title>", "<p>In order to further evaluate our results, we split the CHORI dataset in 50% training set and 50% test set, selected PCAIMs in the training set (with and without redundancy removal) and used these SNPs to predict the ancestry of the individuals in the test set (##FIG##4##Figure 5##). The PCAIMs selected in the training set achieve comparable performance in the test set. We repeated the same experiment in the Coriell dataset, as well as with different split sizes for both datasets (e.g., 80% training, 20% testing) and obtained similar results (data not shown).</p>", "<p>We then cross-validated our results by using the PCAIMs selected in one European American dataset for prediction of structure in the other European American dataset (##FIG##4##Figure 5##). We found that as few as 500 of the top PCAIMs selected in each dataset suffice for the accurate prediction of structure in the other dataset. The actual overlap between the top 500 PCAIMs selected in each sample is relatively small (6.8% or 34 SNPs). Of course this is still highly significant compared to the overlap between two random sets of 500 SNPs selected from approximately 307,315 SNPs, which is 0.16% with a standard deviation of 0.07%. Additionally, some amount of linkage disequilibrium can be observed between the top 500 PCAIMs selected in each of the two datasets. We computed <italic>r</italic>\n² values for all possible pairs and found 44 pairs of SNPs that had <italic>r</italic>\n² of at least 0.1, with an average value of 0.43. These pairs are in addition to the 34 pairs of overlapping SNPs between the two sets. So, it seems that there exist different sets of SNPs that are mildly correlated and yet provide similar information about the structure of the European American population.</p>", "<title>Correcting for Stratification using PCAIMs</title>", "<p>Finally, we examined the extent to which small subsets of PCAIMs can be used for correction of stratification in the setting of an association study. Following the model and parameters used by Price et al. ##REF##16862161##[19]##, we first simulated an admixed population with 1000 members genotyped on 100,000 SNPs, originating from two ancestral populations that are relatively closely related. In particular, the average <italic>F_st</italic> between SNPs in the ancestral populations was set to 10⁻² (see <xref ref-type=\"sec\" rid=\"s2\">Methods</xref> and ##SUPPL##7##Text S1## for details). This gave us the advantage of knowing the “true” ancestry of each simulated individual, while at the same time constructing a simulated population whose structure is quite similar to the structure of our European American datasets. By looking at the singular values associated with the top eigenSNPs of the subject-SNP matrix, as well as by applying our permutation test, one principal component was deemed significant. Thus, in this simulated dataset, again individual variation lies across the first eigenSNP (##SUPPL##4##Figure S5##). In fact, if this eigenSNP is used as a predictor for ancestry, the Pearson correlation coefficient between true and predicted ancestry coefficient over all individuals is 0.9967. As expected, PCAIMs work extremely well for the prediction of ancestry in the simulated data and as few as 100 to 400 PCAIMs are enough to accurately predict the ancestry of each individual (##TAB##1##Table 2##). In fact the Pearson correlation coefficient between true (recall that in this case we know the actual ancestry) and predicted ancestry, calculated by looking at the first eigenSNP of a matrix containing 100, 200, and 400 PCAIMs is 0.9102, 0.9478, and 0.9690 respectively. Using this particular method of constructing simulated SNPs results in mostly uncorrelated SNPs. Consequently, in this synthetic dataset, our redundancy removal algorithm did not improve our results.</p>", "<p>In order to test if small subsets of PCAIMs could be used for correction for stratification, we simulated association studies with sets of 100,000 random, extremely stratified, and truly causal SNPs (see <xref ref-type=\"sec\" rid=\"s2\">Methods</xref> for details) for 10 different datasets. We first replicated the results of Price et al. ##REF##16862161##[19]## in order to correct for stratification using the top 10 principal components computed on all 100,000 SNPs without significant loss in power (##TAB##1##Table 2##). We then selected subsets of 100 to 400 PCAIMs in order to predict the ancestry of all 1,000 individuals. We proceeded to correct for stratification by removing (projecting out) our ancestry prediction from each SNP and then ran the Armitage trend test to the resulting SNPs. (This is essentially the algorithm implemented in EIGENSTRAT.) We measured the percentage of correlations found using the Armitage-trend test in each scenario and report the results before and after stratification correction in ##TAB##1##Table 2##. According to our findings, as few as 100 PCAIMs (instead of 100,000 SNPs) efficiently remove false correlations with disease, while largely maintaining the power of the study.</p>" ]

[ "<title>Discussion</title>", "<p>We have identified small sets of structure informative markers for the European American population through the direct investigation of European American samples and without depending on any assumptions about the ancestry or admixture proportions of the studied individuals. We have analyzed two independent datasets of European Americans, representing a total of almost 1500 individuals genotyped for more than 300,000 SNPs spanning the entire autosomal genome, and we have demonstrated that as few as 200 SNPs (PCAIMs), carefully selected with our methodology, can be used to very accurately predict the genetic structure of European Americans as identified by PCA. The cross-validation experiments that we have performed verify the validity of our approach. Investigating the European American population directly for the selection of structure informative genetic markers results in SNP panels that provide a direct reflection of the complex patterns of sub-structure and admixture in European Americans.</p>", "<p>The analysis of the admixed European American population for the selection of structure informative markers was made possible through the application of the unsupervised method that we have recently introduced for the selection of PCA-correlated SNPs or PCAIMs ##REF##17892327##[28]##. As we have previously described, PCAIMs selection can be carried out without any need for prior knowledge of individual ancestry, and is thus feasible in admixed populations without having to trace the origin of the studied individuals or hypothesize about admixture proportions ##REF##17892327##[28]##. This is not possible when using allele-frequency based methods for the selection of AIMs like <italic>δ</italic>, <italic>F_st</italic> or informativeness for assignment ##REF##7942857##[20]##–##REF##14631557##[25]##.</p>", "<p>An additional important contribution of the present study is the novel algorithm that we developed for the removal of redundancy from a given set of structure informative markers. All existing algorithms for AIM selection (e.g., <italic>δ</italic>, <italic>F_st</italic>, informativeness, as well as PCAIMs), could potentially suffer from selecting a large number of redundant SNPs. For example, consider the simple scenario where a SNP is assigned a high score, and many SNPs are in very high LD with this SNP. Then, they will also be assigned very high scores, and thus will be chosen as AIMs, even though they are clearly redundant. Thus, if the task at hand is to select a minimal set of AIMs (as is the case in our work), a second step is necessary in order to remove redundant AIMs. Given the large number of SNPs (many of which are in LD) in genome-wide scans over the last year, this is certainly a significant concern. Notice for example the fact that, in the datasets we studied, fewer than 10,000 such pairs exist (##TAB##0##Table 1##), and even though this is a proportionally small percentage out of the possible pairs it still significantly increases the number of PCAIMs needed to perfectly capture the structure of the data.</p>", "<p>In order to address this deficiency, we propose an efficient and accurate algorithm that filters out redundant SNPs from the set of PCA-correlated SNPs. The proposed algorithm emerges by reducing the redundancy removal problem to a well-known problem in the numerical linear algebra community, the so-called Column Subset Selection Problem, as defined earlier here. As we have shown here, applying this algorithm significantly increases genotyping savings, reducing the number of SNPs needed for structure identification almost by six-fold. This method for redundancy removal can be applied to any set of SNPs in order to select a minimally correlated subset. We should note that the proposed algorithm does not necessarily return the absolutely optimal solution to the Column Subset Selection problem. Formal mathematical bounds regarding the accuracy of the algorithm do exist, arguing that the selected subset of columns (i.e. SNPs) provides an almost optimal solution ##UREF##6##[46]##. Further discussion on this is perhaps beyond the scope of this paper. Alternatives that take into account LD estimation and physical distance could also be considered. Notice however, that our method is parameter free and achieves effective redundancy removal in a single step.</p>", "<p>The two independent samples of European Americans that we studied show comparable structure, while the CEPH European Americans represent only a small fraction of the entire breadth of variation that we encountered in these large datasets. We are able to faithfully reproduce this fine structure using as few as 200 PCAIMs. We found that the SNPs selected in the first European American dataset we studied could be successfully applied in the second dataset and vice versa; however, the absolute actual overlap was relatively small (although significantly higher than what expected by chance alone) suggesting the possibility that many different such subsets of informative SNPs exist.</p>", "<p>Several other studies have explored intra-European and European American genetic variation. Classic gene frequency ##REF##356262##[41]##,##UREF##7##[47]##, Y-chromosome ##REF##12167671##[48]## or mitochondrial variation ##REF##15309688##[49]##,##REF##12355353##[50]## as well as whole-genome studies ##REF##17044734##[7]##,##REF##17436249##[8]## generally agree on a coarse separation of European populations along a northern to southeastern axis. Seldin et al. ##REF##17044734##[7]## analyzed 5705 SNPs from the ILLUMINA Linkage IV panel to calculate informativeness for assignment, and identified 400 SNPs that could be used in order to broadly cluster the populations they studied to northern and southern Europeans. Bauchet et al. ##REF##17436249##[8]## studied 10,000 SNPs (Affymetrix 10K panel) and about 100 individuals from 12 European populations and concluded that at least 1,200 high Fst SNPs were needed in order to achieve a similar clustering of northern versus southern Europeans. Our results build on these papers, using large datasets of genomewide markers, and an algorithm that can explicitly identify informative markers from admixed populations without knowledge of the ancestral populations. Finally, we demonstrate that our markers are valid across large European American studies. We found almost no overlap between the markers that we identify as ancestry informative and those reported in the above mentioned studies of European populations ##REF##17044734##[7]##,##REF##17436249##[8]## (data not shown). This was to be expected since all three studies analyze different datasets and different populations. Notice, that even between these two previous studies, there is very little overlap between the panels of SNPs reported as ancestry informative.</p>", "<p>Very recently, two studies parallel to ours, used several genomewide sets of markers in European Americans to derive small subsets of European American AIMs ##REF##18208327##[26]##,##REF##18208329##[27]## (see also ##SUPPL##5##Tables S1## and ##SUPPL##6##S2##). An important difference between these studies and ours is the fact that we employed a previously validated algorithm for the selection of AIMs ##REF##17892327##[28]##, that operates directly on raw data without the need for intermediate steps (i.e., artificial assignment of individuals to clusters, depending on candidate genes for local natural selection, etc.). As we have seen here, and as others have also discussed ##REF##18208327##[26]##,##REF##16355252##[51]##, individual variation in the European American population seems to lie along a continuum rather than in distinct clusters. Thus, the method we have used here would be easier to generalize to diverse datasets without access to ancestral populations. Another important difference of our study, is the fact that, as we have also discussed previously here, we have employed a novel, linear algebra based algorithm in order to select the least correlated SNPs as part of our structure informative panel thus increasing the efficiency of our informative SNP sets. In comparison, Price et al. ##REF##18208327##[26]## and Tian et al. ##REF##18208329##[27]## reduced redundancy by applying measures based on physical distance.</p>", "<p>Our results are consistent with the findings of Price et al. ##REF##18208327##[26]## and Tian et al. ##REF##18208329##[27]##, who also demonstrated that the vast amount of inter-individual variation in European Americans lies across a single axis. In concordance with what we have also described here, Tian et al. ##REF##18208329##[27]## mention that the first principal component in their study accounted for greater than five-fold the variance of the second principal component (percentage of total variance according to their analysis is 42.42% for the first principal component, and 8.32% and 6.66% for the second and third respectively ##REF##18208329##[27]##). Both ##REF##18208327##[26]##,##REF##18208329##[27]## analyzed individuals of known ancestry and they could distinguish a cluster comprising of individuals of known Ashkenazi Jewish origin. Price et al. ##REF##18208327##[26]## argue that an additional principal component is needed in order to discern this line of ancestry. However, both of these studies included large subpopulations with known Ashkenazi Jewish ancestry. For example in Price et al. ##REF##18208327##[26]##, in the inflammatory bowel disease (IBD) study, 43% of included individuals self-reported as Ashkenazi Jewish (78% among individuals of known ancestry in this sample). In Tian et al. ##REF##18208329##[27]## 28% of the population analyzed was of known Ashkenazi heritage (for comparison, 2% of the general US population self-reports as Ashkenazi Jewish ##REF##18208327##[26]##,##UREF##8##[52]##,##REF##11411198##[53]##). Thus, the larger Ashkenazi Jewish population in the Price et al. ##REF##18208327##[26]## study likely helped to bring out an additional principal component for this population.</p>", "<p>It is likely that there were Ashkenazi individuals in the datasets that we studied; however, they probably constituted a smaller fraction of the overall population. Analyzing the top two eigenSNPs, corresponding to the top two principal components in our datasets (data not shown), a small cluster of individuals becomes visually apparent. (A similar figure is shown in ##REF##18208327##[26]## for the PD dataset which corresponds to our CORIELL dataset.) As we have no information on individual ancestry, we cannot infer the origin of the individuals in this small cluster. Interestingly, this very small cluster (which might correspond to Ashkenazi individuals in our population), is already reasonably separated from the remaining European Americans along the top eigenSNP, at least in the datasets that we studied. This observation is consistent with Tian et al.'s report ##REF##18208329##[27]##; in the sample they studied, the mean score of the top eigenSNP for individuals of known Ashkenazi Jewish ancestry, lay at one end of the distribution (0.045 for Ashkenazi Jewish individuals, followed by 0.022 for Greeks and 0.015 for Italians). This explains why our permutation test only detects the first principal component as statistically significant: our test removes from the data the amount of information that has already been captured by principal components that were deemed significant.</p>", "<p>The SNP panels proposed by Price et al. ##REF##18208327##[26]## and Tian et al. ##REF##18208329##[27]## perform very well when tested on our samples of European Americans (##SUPPL##5##Table S1##). The Tian et al. ##REF##18208329##[27]## panel of ancestry informative SNPs was selected by calculating <italic>I_n</italic>\n##REF##14631557##[25]## for two discrete clusters; Ashkenazi Jewish individuals (as representatives of southeastern or rather mediterranean European ancestry) and northern Europeans. The SNPs selected by this method perform exceptionally well (comparably to our SNP panel) to recreate the individual ancestry in our analysis (see ##SUPPL##5##Table S1##). This suggests the fact that most of the variation between southeastern and northern European ancestry is captured by the difference between Ashkenazi versus northwestern European ancestry. On the other hand, we could not fully test the SNPs proposed in the second study ##REF##18208327##[26]##, since they had not all been genotyped in our datasets. Price et al. ##REF##18208327##[26]## proposed 300 SNPs as informative for European American ancestry (100 discerning the northern European versus southeastern cluster and 200 differentiating the southeatern versus Ashkenazi Jewish clusters). Out of these SNPs, 141 had also been genotyped in the datasets we studied. However, using these 141 SNPs ##REF##18208327##[26]##, results in a correlation coefficient of 0.75 between true and predicted individual variation in our combined CORIELL and CHORI datasets (##SUPPL##5##Table S1##).</p>", "<p>There is generally little (although far greater than chance) overlap between the lists of structure informative SNPs identified by each of these three studies (see ##SUPPL##6##Table S2##). The greatest overlap is found between the panel we propose here and the 1,441 SNPs proposed by Tian et al. ##REF##18208329##[27]## as distinguishing between northern European and Ashkenazi Jewish ancestry; out of the 1,419 SNPs that were also included in our analysis, 36 were among the 500 top informative SNPs that we selected in the analysis of our combined European American datasets. The overlap between the informative SNPs proposed by Price et al. ##REF##18208327##[26]## and the other two studies is even smaller, partly due to the fact that we could only test 141 out of the 300 proposed SNPs (see ##SUPPL##6##Table S2##). In any case, as we have also suggested earlier here, it is probably not surprising that there exist more than one subsets of SNPs describing European American population structure.</p>", "<p>It is now clear that European derived populations are not homogeneous and recent studies have emphasized the problem of population stratification in genetic association studies which may lead to false positive associations with disease or mask true correlations ##REF##16041375##[5]##,##REF##16862161##[19]##. As association studies of thousands of individuals are starting to become increasingly common ##REF##17554300##[9]##–##REF##17632545##[14]##, population stratification will undoubtedly pose a serious challenge. Various methods have been proposed to tackle the problem ##REF##11315092##[16]##, ##REF##10827107##[18]##, ##REF##16862161##[19]##, ##REF##11119293##[54]##–##REF##17436246##[60]##. Among them, PCA-based stratification correction tools seem particularly attractive, since they are computationally efficient and are not overly conservative. Moreover, such methods do not demand the use of discrete clusters, which as we have discussed earlier here may be an over-simplification, especially in the case of admixed populations.</p>", "<p>We have replicated the analysis of simulated data in ##REF##16862161##[19]## and experimentally demonstrated how our method can complement PCA-based stratification correction methods. Using as few as 100 to 200 PCAIMs, we achieved almost perfect stratification correction with virtually no loss in power. In comparison previous simulation studies ##REF##16862161##[19]## have shown that as many as 5,000 randomly selected SNPs would be needed to reach similar performance, while 20,000 random SNPs were needed in a real dataset ##REF##16862161##[19]##. Comparing the accuracy of ancestry prediction in the simulated and real data we have studied we can extrapolate that as few as 200 SNPs could be enough for stratification correction in real data (reaching a Pearson correlation coefficient above 0.9 between “true” and predicted ancestry across the second eigenvector). While the selection of AIMs for stratification correction may be unnecessary for teams of investigators that undertake an initial genome-wide association study and can afford genotyping of very dense maps of markers, the use of AIMs for stratification correction becomes of critical importance in two-stage study designs, (where replication of initial findings is sought in large independent samples), or studies following the candidate gene approach. In such cases, our methods can greatly facilitate association studies in admixed populations, reducing significantly the genotyping costs needed to ensure correction for stratification.</p>", "<p>We would like to point out that, the sets of European American AIMs that we and others ##REF##17044734##[7]##,##REF##17436249##[8]##,##REF##18208327##[26]##,##REF##18208329##[27]## have identified, are representative of the full genetic structure in the European American population, only to the extent that the samples analyzed in each of these studies are deemed truly representative of the entire European American population. It will be important to study European American population structure with even larger datasets of carefully sampled individuals. Interestingly, in Tian et al. ##REF##18208329##[27]##, the effect of stratification on the case-control study of rheumatoid arthritis was mostly due to a difference in Irish ancestry. This suggests that different European American studies will have to exercise caution in detecting and adjusting for ancestry, since the components/axes that affect ancestry are likely to vary from study to study depending on the phenotype and the region sampled.</p>", "<p>In summary, we are proposing a small set of SNPs that can successfully capture the structure of the European American population samples we studied, as identified by PCA. We identified this minimal set of structure informative SNPs (PCAIMs) by applying a novel redundancy removal algorithm that will undoubtedly increase genotyping savings in many different research scenarios. Lists of the sets of markers that we have identified as well as an implementation of our algorithms are available online at <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.cs.rpi.edu/drinep/EUROAIMs/\">http://www.cs.rpi.edu/drinep/EUROAIMs/</ext-link>. These panels of SNPs will serve as useful tools in the discovery of susceptibility genes for common complex disorders and can spark interesting questions in population genetics regarding the possible role of natural selection in the regions of the genome harboring these polymorphic sites.</p>" ]

[]

[ "<p>Conceived and designed the experiments: PP PD EZ. Performed the experiments: PP PD JL. Analyzed the data: PP PD JL CN EZ. Contributed reagents/materials/analysis tools: PP PD DN JS PR DC RK EZ. Wrote the paper: PP PD EZ.</p>", "<p>Genetic structure in the European American population reflects waves of migration and recent gene flow among different populations. This complex structure can introduce bias in genetic association studies. Using Principal Components Analysis (PCA), we analyze the structure of two independent European American datasets (1,521 individuals–307,315 autosomal SNPs). Individual variation lies across a continuum with some individuals showing high degrees of admixture with non-European populations, as demonstrated through joint analysis with HapMap data. The CEPH Europeans only represent a small fraction of the variation encountered in the larger European American datasets we studied. We interpret the first eigenvector of this data as correlated with ancestry, and we apply an algorithm that we have previously described to select PCA-informative markers (PCAIMs) that can reproduce this structure. Importantly, we develop a novel method that can remove redundancy from the selected SNP panels and show that we can effectively remove correlated markers, thus increasing genotyping savings. Only 150–200 PCAIMs suffice to accurately predict fine structure in European American datasets, as identified by PCA. Simulating association studies, we couple our method with a PCA-based stratification correction tool and demonstrate that a small number of PCAIMs can efficiently remove false correlations with almost no loss in power. The structure informative SNPs that we propose are an important resource for genetic association studies of European Americans. Furthermore, our redundancy removal algorithm can be applied on sets of ancestry informative markers selected with any method in order to select the most uncorrelated SNPs, and significantly decreases genotyping costs.</p>", "<title>Author Summary</title>", "<p>Genetic association studies search to identify disease susceptibility genes through the analysis of genetic markers such as single nucleotide polymorphisms (SNPs) in large numbers of cases and controls. In such settings, the existence of sub-structure in the population under study (i.e. differences in ancestry among cases and controls) may lead to spurious results. It is therefore imperative to control for this possible bias. Such biases may arise for example when studying the European American population, which consists of individuals of diverse ancestry proportions from different European countries and to some degree also from African and Native American populations. Here, we study the genetic sub-structure of the European American population, analyzing 1,521 individuals for over 300,000 SNPs across the entire genome. Applying a powerful method that is based on dimensionality reduction (Principal Components Analysis), we are able to identify 200 SNPs that successfully represent the complete dataset. Importantly, we introduce a novel method that effectively removes redundancy from any set of genetic markers, and may prove extremely useful in a variety of different research scenarios, in order to significantly reduce the cost of a study.</p>" ]

[ "<title>Supporting Information</title>" ]

[]

[ "<fig id=\"pgen-1000114-g001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000114.g001</object-id><label>Figure 1</label><caption><title>Singular values and test of significance of principal components in the CHORI and CORIELL datasets.</title><p>(A) Histogram of the first eigenSNP of the CORIELL dataset. (B) The singular values corresponding to the first up to the tenth eigenSNP of the CORIELL dataset. (C) The results of the permutation test to determine the significance of the principal components of the CORIELL dataset. Higher values on the <italic>y</italic>-axis correspond to principal components containing significantly more structure than a random component would. (D) Histogram of the first eigenSNP of the CHORI dataset. (E) The singular values corresponding to the first up to the tenth eigenSNP of the CHORI dataset. (F) The results of the permutation test to determine the significance of the principal components of the CHORI dataset.</p></caption></fig>", "<fig id=\"pgen-1000114-g002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000114.g002</object-id><label>Figure 2</label><caption><title>Projection of 958 CHORI, 539 CORIELL, and 270 HapMap individuals on their first, second, and third eigenSNPs.</title><p>Notice that the individuals of European American ancestry lie along a line with very little deviations toward the Asian and African populations (14 outliers have been removed from this analysis, as described in <xref ref-type=\"sec\" rid=\"s2\">Methods</xref>).</p></caption></fig>", "<fig id=\"pgen-1000114-g003\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000114.g003</object-id><label>Figure 3</label><caption><title>Non-redundant PCAIMs are very good predictors of ancestry.</title><p>(A) Pearson correlation coefficient between predicted ancestry and “true” ancestry for the 539 subjects in the CORIELL dataset. (B) Pearson correlation coefficient between predicted ancestry and “true” ancestry for the 958 subjects in the CHORI dataset. (For random SNPs, the average over 20 experiments is reported).</p></caption></fig>", "<fig id=\"pgen-1000114-g004\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000114.g004</object-id><label>Figure 4</label><caption><title>PCA extracts meaningful information from genotype data.</title><p>(A) Raster plot of the top 500 PCAIMs for all 1497 subjects in the CHORI and CORIELL datasets, after removing redundant SNPs from the top 3000 PCAIMs using the greedy <italic>QR</italic> method. Red and green denotes heterozygotes while homozygotes are black. Individuals are sorted according to their coordinates in the first eigenSNP. (B) The first eigenSNP of the matrix in (A). This vector corresponds to our prediction of ancestry. (C) The first eigenSNP of the matrix of the CHORI and CORIELL subjects on all 307,315 SNPs. This vector is interpreted as “true” ancestry for the individuals. Notice that the two vectors are highly correlated. (D) Pearson correlation coefficient between predicted ancestry and “true” ancestry for the 1497 subjects of European American ancestry using panels of PCAIMs, non-redundant PCAIMs, and random SNPs. Clearly, non-redundant PCAIMs are very good predictors of ancestry.</p></caption></fig>", "<fig id=\"pgen-1000114-g005\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000114.g005</object-id><label>Figure 5</label><caption><title>Cross-validation of panels of PCAIMs for ancestry prediction in European Americans.</title><p>(A) Pearson correlation coefficient between ancestry prediction of CHORI subjects using SNPs selected in the CORIELL dataset, and “true” ancestry of the CHORI subjects. (B) Pearson correlation coefficient between ancestry prediction of CORIELL subjects using SNPs selected in the CHORI dataset, and “true” ancestry of the CORIELL subjects. (C) Split of the CHORI dataset in 50% training and 50% test set. Pearson correlation coefficient between ancestry prediction of test set subjects using SNPs selected in the training set, and “true” ancestry of the test set subjects. Results are reported over 20 splits.</p></caption></fig>" ]

[ "<table-wrap id=\"pgen-1000114-t001\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000114.t001</object-id><label>Table 1</label><caption><title>Number of pairs of the top 3000 PCAIMs with <italic>r</italic>\n² values above 0.1 in the CORIELL dataset, the CHORI dataset, and the joint CORIELL and CHORI dataset.</title></caption><alternatives><table frame=\"hsides\" rules=\"groups\"><colgroup span=\"1\"><col align=\"left\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/></colgroup><thead><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\"># pairs with</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">CORIELL</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">CHORI</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">All European Americans</td></tr></thead><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.1≤<italic>r</italic>\n²&lt;.2</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1598</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">2246</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">2238</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.2≤<italic>r</italic>\n²&lt;.3</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1095</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1007</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">1060</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.3≤<italic>r</italic>\n²&lt;.4</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">666</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">708</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">699</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.4≤<italic>r</italic>\n²&lt;.5</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">552</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">612</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">622</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.5≤<italic>r</italic>\n²&lt;.6</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">521</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">457</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">490</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.6≤<italic>r</italic>\n²&lt;.7</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">526</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">496</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">452</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.7≤<italic>r</italic>\n²&lt;.8</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">432</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">304</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">309</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.8≤<italic>r</italic>\n²&lt;.9</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">350</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">194</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">181</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">.9≤<italic>r</italic>\n²&lt;1</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">589</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">449</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">460</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">exactly 1</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">115</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">67</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">72</td></tr></tbody></table></alternatives></table-wrap>", "<table-wrap id=\"pgen-1000114-t002\" position=\"float\"><object-id pub-id-type=\"doi\">10.1371/journal.pgen.1000114.t002</object-id><label>Table 2</label><caption><title>Using PCAIMs for stratification correction in conjuction with EIGENSTRAT's algorithm.</title></caption><alternatives><table frame=\"hsides\" rules=\"groups\"><colgroup span=\"1\"><col align=\"left\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/><col align=\"center\" span=\"1\"/></colgroup><tbody><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<bold>Admixed</bold> (<italic>r</italic> = 2)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">No correction</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">400 PCAIMs</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">200 PCAIMs</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">100 PCAIMs</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>Random</italic>\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0002</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>Stratified</italic>\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.1246</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0002</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>Causal</italic>\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.5203</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.4735</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.4716</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.4790</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<bold>Admixed</bold> (<italic>r</italic> = 3)</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">No correction</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">400 PCAIMs</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">200 PCAIMs</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">100 PCAIMs</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>Random</italic>\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0005</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>Stratified</italic>\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.6182</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0001</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.0003</td></tr><tr><td align=\"left\" rowspan=\"1\" colspan=\"1\">\n<italic>Causal</italic>\n</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.5110</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.4141</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.4189</td><td align=\"left\" rowspan=\"1\" colspan=\"1\">0.4340</td></tr></tbody></table></alternatives></table-wrap>" ]

[ "<disp-formula><label>(1)</label></disp-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>" ]

[]

[]

[]

[]

[ "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s001\"><label>Figure S1</label><caption><p>Plot of 970 CHORI, 541 CORIELL, and 270 HapMap subjects on their first, second, and third eigenSNPs. Five CHORI pairs (ten CHORI subjects) were suspiciously similar and were excluded prior to this plot. The red squares represent the 14 individuals (12 CHORI and two CORIELL) that were excluded from further analysis. Notice that these individuals tend to have atypical degrees of Asian and African ancestry.</p><p>(0.09 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s002\"><label>Figure S2</label><caption><p>Plot of 960 CHORI, 539 CORIELL, and 90 CEPH European HapMap subjects on their first eigenSNP. Notice CEPH Europeans form a tight cluster that does not seem to encompass the full variation of European American populations.</p><p>(0.04 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s003\"><label>Figure S3</label><caption><p>Using non-redundant PCAIMs to predict the first eigenSNP in European American datasets. The first eigenSNP of 1497 European Americans (CHORI and CORIELL datasets) analyzing 307,315 SNPs, plotted against the predicted first eigenSNP of each individual with 200 and 300 non-redundant PCAIMs.</p><p>(0.18 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s004\"><label>Figure S4</label><caption><p>PCA scores of 307,315 studied SNPs in the combined CHORI and CORIELL datasets plotted along each autosome. The blue “x” marks the top 3,000 PCAIMs, while the red squares denote the top 500 PCAIMs after redundancy removal. Notice the different scale of the Y axis for each chromosome. (A) Chromosomes 1–5, (B) Chromosomes 6–10, (C) Chromosomes 11–15, (D) Chromosomes 16–20, (E) Chromosomes 21 and 22.</p><p>(0.68 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s005\"><label>Figure S5</label><caption><p>A simulated admixed population of 1000 subjects genotyped on 100,000 SNPs. The admixed population emerges from two ancestral populations with an average F_st of 10⁻², as described in <xref ref-type=\"sec\" rid=\"s2\">Methods</xref>.</p><p>(0.03 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s006\"><label>Table S1</label><caption><p>Performance of published European American AIMs, for population structure prediction in the datasets we studied.</p><p>(0.03 MB DOC)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s007\"><label>Table S2</label><caption><p>Overlap between European American AIMs proposed in studies of European American datasets.</p><p>(0.03 MB PDF)</p></caption></supplementary-material>", "<supplementary-material content-type=\"local-data\" id=\"pgen.1000114.s008\"><label>Text S1</label><caption><p>Supplementary note.</p><p>(0.06 MB PDF)</p></caption></supplementary-material>" ]

[ "<table-wrap-foot><fn id=\"nt101\"><p>There are numerous highly correlated pairs. However, after our redundancy removal step, in the retained 500 PCAIMs there was no pair (in any of the three datasets) with an <italic>r</italic>\n² value above 0.2 and only three pairs in the CORIELL dataset with an <italic>r</italic>\n² value between 0.1 and 0.2.</p></fn></table-wrap-foot>", "<table-wrap-foot><fn id=\"nt102\"><p>The first column shows the proportion of random, stratified, and causal SNPs that are identified as causal using the Armitage's trend test with a cut-off <italic>p</italic>-value of 10⁻⁴. The remaining columns show the respective proportions after stratification correction using 100, 200, and 400 PCAIMs.</p></fn></table-wrap-foot>", "<fn-group><fn fn-type=\"COI-statement\"><p>The authors have declared that no competing interests exist.</p></fn><fn fn-type=\"financial-disclosure\"><p>This work was funded, in part, by a National Science Foundation CAREER award to Petros Drineas, National Institutes of Health U19 AG23122 and K22CA109351, and Department of Defense Breast Cancer Research Program grant BC033051 to Elad Ziv, a Hellenic Endocrine Society Research grant award to Peristera Paschou and the National Institutes of Health grant U01HL069757 to Ronald M. Krauss.</p></fn></fn-group>" ]

[ "<graphic xlink:href=\"pgen.1000114.e001.jpg\" mimetype=\"image\" position=\"float\"/>", "<inline-graphic xlink:href=\"pgen.1000114.e002.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pgen.1000114.e003.jpg\" mimetype=\"image\"/>", "<inline-graphic xlink:href=\"pgen.1000114.e004.jpg\" mimetype=\"image\"/>", "<graphic xlink:href=\"pgen.1000114.g001\"/>", "<graphic xlink:href=\"pgen.1000114.g002\"/>", "<graphic xlink:href=\"pgen.1000114.g003\"/>", "<graphic id=\"pgen-1000114-t001-1\" xlink:href=\"pgen.1000114.t001\"/>", "<graphic xlink:href=\"pgen.1000114.g004\"/>", "<graphic xlink:href=\"pgen.1000114.g005\"/>", "<graphic id=\"pgen-1000114-t002-2\" xlink:href=\"pgen.1000114.t002\"/>", "<inline-graphic xlink:href=\"pgen.1000114.e005.jpg\" mimetype=\"image\"/>" ]

[ "<media xlink:href=\"pgen.1000114.s001.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000114.s002.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000114.s003.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000114.s004.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000114.s005.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000114.s006.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000114.s007.pdf\"><caption><p>Click here for additional data file.</p></caption></media>", "<media xlink:href=\"pgen.1000114.s008.pdf\"><caption><p>Click here for additional data file.</p></caption></media>" ]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Adjuvant use of mitomycin C (MMC) is prevalent in the treatment of various ocular diseases including pterygium [##REF##16546105##1##], glaucoma [##REF##18091183##2##], and refractive surgery [##REF##18047001##3##] to improve the success of clinical therapy. MMC is applied by topical drops, local soaking, or subconjunctival injection during or after pterygium [##REF##11340404##4##] and glaucoma [##REF##2127056##5##] surgeries to inhibit cellular proliferation. For laser refractive surgery, MMC is applied with local soaking to the cornea to modulate corneal keratocyte growth and to decrease the postoperative corneal opacity [##REF##18046999##6##]. When MMC is topically soaked onto the pterygium area, a scleral flap, or the cornea, a certain amount of MMC may penetrate into the corneal endothelium where it causes toxicity to corneal endothelial cells. In clinical observations, a significant loss of these cells is noted after trabeculectomy with adjunctive MMC soaking [##REF##9541440##7##]. Evidence has also presented that a single application of MMC onto the corneal surface can cause dose-dependent corneal edema and endothelial apoptosis in a rabbit model system [##REF##15313301##8##]. Thus, improper use of MMC as an ocular medication may result in damage to the cornea and impair the physiologic function of the corneal endothelium.</p>", "<p>It is known that corneal endothelial cells play a crucial role in maintaining corneal transparency. Corneal clarity requires a net movement of fluid from the corneal stroma to the aqueous humor [##REF##1733238##9##], and the efficiency of this flux depends on the presence of undamaged corneal endothelial cells and adequate cellular density. Any factor that decreases cell density in the endothelium may also reduce the efficiency of the corneal pump function and therefore affect transparency [##REF##7775109##10##].</p>", "<p>We have previously demonstrated that MMC is toxic to corneal endothelial cells in a time-dependent and dose-dependent manner [##REF##10530701##11##]. Exposure of corneal endothelial cells to MMC for a certain dose and incubation time may induce apoptosis. This is a process of natural cell death that is characterized by an extreme heterogeneity of signal transduction pathways that lead to DNA degradation and dysfunctions of the plasma membrane and mitochondria, which are accompanied by a series of degenerative pathways [##REF##9464190##12##,##REF##9477978##13##]. To evaluate the potential chronic toxicity of MMC, cultured porcine corneal endothelial cells were used in the present study to investigate the apoptotic characteristics and mechanisms involved in corneal damage induced by MMC.</p>" ]

[ "<title>Methods</title>", "<title>Materials</title>", "<p>Cell culture materials including trypsin, minimal essential medium (MEM), glutamine, gentamicin, and fetal bovine serum were obtained from Gibco (Grand Island, NY). MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyl tetrazolium bromide) was purchased from Sigma Chemical (St. Louis, MO). The general caspase inhibitor, Z-VAD-FMK; caspase-8 inhibitor, Z-IETD-FMK; caspase-9 inhibitor, Z-LEHD-FMK; and a TUNEL (terminal deoxyribonucleotidyl-transferase-(TdT)-mediated deoxyuridine-5′-triphosphate-digoxigenin (dUTP) nick-end labeling) apoptosis detection kit were purchased from Calbiochem (Bad Soden, Germany). A MitoLight mitochondrial apoptosis detection kit was purchased from Chemicon International Inc. (Temecula, CA). Rabbit antibodies to cytochrome <italic>c</italic> and p21 were purchased from Santa Cruz Biotechnology Inc. (Santa Cruz, CA). Alexa Fluor 488-conjugated goat anti–rabbit antibody; the annexin V– FITC/propidium iodide double staining kit; and mouse anti–human antibodies to Bcl-2 and p53 were purchased from Invitrogen (Carlsbad, CA). Western Blot Chemiluminescence Reagent Plus was purchased from New England Nuclear (PerkinElmer, Boston, MA). Protein assay dye and agents used for electrophoresis were purchased from Bio-Rad (Richmond, CA). Horseradish peroxidase-conjugated sheep anti-mouse IgG and donkey anti–rabbit IgG were obtained from Amersham Pharmacia (Buckinghamshire, England). Mitomycin C (MMC) was purchased from Kyowa (Hakko Kogyo Co., Tokyo, Japan). All other chemicals were obtained from Merck (Darmstadt, Germany).</p>", "<title>Culture of porcine corneal endothelial cells</title>", "<p>Culture of porcine corneal endothelial cells was performed as published previously [##REF##10530701##11##]. Briefly, porcine eyeballs were collected from a local slaughterhouse. Under sterile conditions, the corneal endothelial cells were separated using 0.25% trypsin for 30 min. The action of trypsin was stopped by adding minimal essential medium (MEM) containing 10% fetal bovine serum, 3.8 mM L-glutamine, and 50 μg/ml gentamicin. After centrifugation, the cells were resuspended in a culture flask. The culture was kept in a humidified chamber of 5% CO₂ at 37 °C, and the medium was changed every two to three days.</p>", "<title>Assay of cell viability with MTT</title>", "<p>Cell viability was measured with MTT dye following previously published procedures [##REF##11572464##14##]. The MTT assay is based on the production of purple formazan from a methyl tetrazolium salt by the mitochondrial enzymes of viable cells. Cultured cells at a concentration of 4,000 cells/well were seeded in 96 well culture plates and allowed to form a monolayer for 24 h. The cells were then exposed to 150 μl of serum-free MEM medium containing various concentrations of MMC for 15 h and 24 h. For analysis of apoptotic caspase pathways, cells were pre-incubated with different caspase inhibitors including a general caspase inhibitor (Z-VAD-FMK), a caspase-8 inhibitor (Z-IETD-FMK), and a caspase-9 inhibitor (Z-LEHD-FMK) for 1 h before adding 0.001 mg/ml MMC. After exposure to the drug for 24 h, cells were washed twice with phosphate-buffered saline (PBS) and incubated with 150 μl MTT solution (0.833 mg/ml in PBS) for 4 h at 37 °C. At the end of the incubation period, the MTT solution was carefully aspirated, taking care not to disturb the crystal of purple formazan at the bottom of each well. The formazan reaction product was dissolved by the addition of 150 μl dimethyl sulfoxide (DMSO), and the optical density of the fluid in each well was read at 510 nm in a multi-well spectrophotometer (Titertek Multiscan, Flow Lab, Scotland, UK). Cytotoxicity was calculated based on a significant difference in the optical density between the MMC-treated and control groups.</p>", "<title>Immunofluorescent staining of cytochrome <italic>c</italic> protein in corneal endothelial cells</title>", "<p>Coverslips coated with a confluent monolayer of cells were treated with MMC following previously published procedures [##REF##16251124##15##]. Cells were washed twice with PBS and rinsed with 0 °C acetone for 20 s. Following a thorough PBS washing, rabbit anti-cytochrome <italic>c</italic> antibody (1:50) was applied to cells in a humidified chamber at 37 °C for 1 h. Cells were again washed with PBS and then incubated with Alexa Fluor 488-conjugated goat anti–rabbit antibody (1:200) for 1 h at room temperature. The stained samples were then rinsed and mounted. All slides were examined and photographed with a fluorescence microscope (Leitz Ortholuxix II; Leitz, Wetzlar, Germany).</p>", "<title>Flow cytometry assay of apoptotic changes in plasma membrane and mitochondrial membrane potential</title>", "<p>Apoptosis in the plasma membrane was identified by flow cytometry with annexin V-FITC/propidium iodide (PI) staining. Mitochondrial membrane potential changes were assayed with MitoLight dye. In healthy cells, this dye accumulates in the mitochondria and yields a red fluorescence. In apoptotic cells where mitochondrial membrane potential has been depolarized, the dye aggregates in the cytoplasm and gives off a green fluorescence, allowing discrimination of apoptotic and non-apoptotic cells. In the absence or presence of various concentrations of MMC, cells were incubated with 1 μg/ml annexin V-FITC and PI for 10 min or with 50 μl of pre-diluted MitoLight solution (900 μl of water, 1 μl of MitoLight dye, and 100 μl of 10X incubation buffer) for 15 min following the manufacturer’s instructions. Cells were then analyzed by flow cytometry (Becton, Dickinson Inc., Franklin Lakes, NJ) for a cell count of 10000.</p>", "<title>DNA apoptotic TUNEL staining</title>", "<p>Immunohistochemical evidence for DNA strand breaks was obtained using the TUNEL assay. The cells were plated on coverslips for at least 24 h for attachment and then incubated with various concentrations of MMC at 37 °C. After exposure to MMC, the cells were washed twice with TBS (20 mM Tris-HCl, 140 mM NaCl, pH 7.6). According to the manufacturer’s instruction, the monolayer of cells was fixed for 20 min in 4% formaldehyde at room temperature and then washed three times in TBS. Cell membrane permeability was increased by treating the samples with 20 μg/ml proteinase K solution at room temperature for 5 min. Endogenous peroxidase activity was quenched by immersing the specimens in 2% H₂O₂ at room temperature for 5 min. The samples were rinsed by replacing the hydrogen peroxide solution with labeling buffer and reaction mixture solution. The specimens were then placed in a humidified incubator for 1 h at 37 °C. The labeling reaction was stopped by immersion in stop buffer at room temperature for 5 min. The specimens were subsequently washed in TBS. Then, 100 μl of streptavidin-horseradish peroxidase was applied to the cells for 20 min at room temperature. After washing with TBS, the specimens were immersed in diaminobenzidine (DAB) solution at room temperature for 5–10 min until a satisfactory color reaction was achieved. This assay was validated with the use of control slides, which had been ascertained to contain apoptotic (positive control) and non-apoptotic (negative control) cells.</p>", "<title>Western blot assay of proteins involved in apoptosis</title>", "<p>Bcl-2, p53, and p21 proteins were detected by sodium dodecyl sulfate (SDS)–PAGE following a previously published procedure [##REF##16251124##15##]. Briefly, cells were treated with MMC and washed with 10 ml of buffer A (20 mM of N-2-hydroxyethyl-1-piperazine-N'-2-ethanesulfonic acid [HEPES], 1 mM of ethylene diamine tetraacetic acid [EDTA], 2 μg/ml aprotinin, 2 μg/ml leupeptin, and 1 μg/ml pepstatin A, pH 7.4). Cells were then scraped into the ice-cold buffer A and immediately homogenized. Protein concentrations were determined with Bio-Rad protein assay dye. Protein (10 μg) from each sample was added to the SDS-PAGE sample buffer and heated in a boiling water bath for 5 min. An aliquot was then subjected to 10% SDS–PAGE. The proteins separated by SDS–PAGE were transferred in a Bio-Rad Trans-Blot cell onto nitrocellulose membranes. Transfer was performed at 100 V for 2 h in a buffer containing 25 mM Tris-HCl, 190 mM glycine, 0.01% SDS, and 20% methanol. The blots were blocked at room temperature for 1 h with buffer solution (20 mM Tris-HCI, 137 mM NaCl, pH 7.6) containing 5% nonfat milk and 0.1% Tween-20. Incubation was then performed for 1 h at 37 °C with monoclonal anti-Bcl-2 (1:50), anti-p53 (1:500), or rabbit anti-p21 (1:250) antibodies. The nitrocellulose membrane was washed three times with the same buffer used during the blocking phase and incubated with horseradish peroxidase-conjugated sheep anti-mouse IgG (1:2,000) or donkey anti–rabbit (1:10,000) as the secondary antibody at room temperature for 1 h. After washing, immunocomplexes were visualized by adding Western Blot Chemiluminescence Reagent Plus. The molecular size of the immunoreactive bands was determined by comparing them with a set of molecular weight marker proteins (Bio-Rad). Relative band intensity was then analyzed using the LabWorks software 4.6 from UVP Bioimaging systems (Upland, CA).</p>", "<title>Statistical analysis</title>", "<p>Data were analyzed by one way ANOVA followed by Dunnett’s post hoc analysis. The values were expressed as mean±standard deviation (SD). All data were significantly different at p&lt;0.05.</p>" ]

[ "<title>Results</title>", "<title>Effects of mitomycin C on cell viability and apoptotic caspase pathways</title>", "<p>The MTT assay is used as a marker for cell viability. In the present study, MMC produced toxic effects on corneal endothelial cells. After incubation with MMC for 15 h and 24 h, cell viability was significantly decreased in a time-dependent and dose-dependent manner at concentrations ranging from 0.1, 0.01, and 0.001 mg/ml (##FIG##0##Figure 1A##). To investigate the role of caspase in MMC-induced apoptosis, cells were pretreated with various caspase inhibitors for 1 h then incubated with 0.001 mg/ml MMC for 24 h. After application of caspase inhibitors, the cellular MTT values were significantly increased in comparison with the MMC only group. ##FIG##0##Figure 1B-D## show that MTT values of cells were significantly significantly reduced when comparing the MMC only groups and the control groups. The addition of caspase-8 inhibitor, general caspase inhibitor, and caspase-9 inhibitors (Z-IETD-FMK, Z-VAD-FMK, and Z-LEHD-FMK) at 10^-5 and 10^-6 M reversed the MMC-induced cellular damage.</p>", "<title>Assay of mitochondrial membrane potential change after mitomycin C exposure</title>", "<p>Disruption of the mitochondrial transmembrane potential is one of the earliest intracellular changes induced by apoptosis. Following exposure to 0.00001 and 0.0001 mg/ml MMC for 24 h, the fluorescence of MitoLight dye in cells was obviously changed from the mitochondria to the cytoplasm as detected by the shift of the relative fluorescence intensity. Control cells showed red fluorescence, indicating a normal mitochondrial membrane potential, while cells treated with 0.00001, 0.0001 and 0.001 mg/ml MMC showed green fluorescence (##FIG##1##Figure 2##). The percentage of MitoLight accumulated in the mitochondria was decreased from 100%±3% of the control (red color) to 47%±4% at 0.00001 mg/ml (green color) and 27%±4% at 0.0001 mg/ml (black color). Exposure to 0.001 mg/ml MMC destroyed the mitochondrial membrane potential as shown by the blue color.</p>", "<p>Cytochrome <italic>c</italic> released from the mitochondria into the cytoplasm is characteristic of cellular apoptosis. In control cells, the cytochrome <italic>c</italic> protein was found in the mitochondrial membrane (##FIG##2##Figure 3A##). After exposure to 0.001 and 0.01 mg/ml MMC for 24 h, typical apoptotic cells were recognized that were shrunken in shape and had lost contact with neighboring cells. Exposure to 0.001 mg/ml MMC caused the cytochrome <italic>c</italic> to slightly disperse in the cytoplasm (##FIG##2##Figure 3B##). Apoptotic changes throughout the cytosol were clearly visible following exposure to 0.01 mg/ml MMC for 24 h (##FIG##2##Figure 3C##).</p>", "<title>Apoptotic TUNEL staining</title>", "<p>Characterization of apoptosis was performed by TUNEL staining. In comparison to the negative staining observed in control cells (##FIG##3##Figure 4A##), cells exposed to 0.001 mg/ml MMC for 24 h were seen to contain cellular chromatin condensed and marginalized at the nuclear membrane (##FIG##3##Figure 4B##). After exposure to 0.01 mg/ml MMC for 15 h, almost all cells were seriously damaged and showed typical apoptotic bodies, which contained cytosol, condensed chromatin, and organelles (##FIG##3##Figure 4C##).</p>", "<title>Annexin V-FITC/propidium iodide staining in cellular plasma membrane with flow cytometry</title>", "<p>To identify apoptosis of the plasma membrane, annexin V-FITC/PI double staining in cells was performed by flow cytometry. In the nonapoptotic, viable control cells, the annexin V-FITC staining and PI negative staining were located in the bottom left quadrant of the dots (##FIG##4##Figure 5A##). After exposure of the cells to 0.001 mg/ml MMC for 24 h, a significant number of cells showed annexin V-FITC positive and PI negative staining, which increased the dot numbers in the bottom right quadrant from 7%±2% of control cells to 48%±3% (##FIG##4##Figure 5B##). The cells in this stage of apoptosis were still viable. Following the increase of MMC to 0.01 mg/ml, cells in advanced apoptosis stained positive with annexin V-FITC and PI (upper right quadrant) and were significantly augmented from 12%±3% of control cells to 34%±3% after 15 h (##FIG##4##Figure 5C##) and 49%±4% after 24 h incubation (##FIG##4##Figure 5D##). The population of cells progressed to advanced apoptosis, indicating that the cells were no longer viable.</p>", "<title>Immunoblot analysis of proteins involved in apoptosis</title>", "<p>To investigate the effect of MMC on proteins involved in apoptosis in cultured corneal endothelial cells, we examined the anti-apoptotic protein, Bcl-2, and two apoptotic proteins, p53 and p21, through western blot analysis. The results of three independent experiments demonstrated that MMC significantly decreased the content of Bcl-2 protein and increased the amount of p53 and p21 proteins in a dose-dependent manner (##FIG##5##Figure 6##). Densitometric analysis of Bcl-2 proteins bands showed that the optical density of proteins in control cells, 0.01 mg/ml MMC-treated cells, and 0.001 mg/ml MMC-treated cells was displayed as 336±8 (100%±4% of the control), 241±7 (64%±4% at 0.01 mg/ml MMC), and 292±9 (86%±5%), respectively. In contrast, the optical density for p53 proteins was 168±6 (control), 253±7 (0.01 mg/ml MMC) and 182±6 (0.001 mg/ml MMC). The protein only significantly increased at 0.01 mg/ml (139%±4% of the control). In the presence of 0.01 and 0.001 mg/ml MMC for 24 h, optical density of p21 proteins were 111±7 (control), 415±9 (0.01 mg/ml MMC), and 166±6 (0.001 mg/ml MMC), which corresponded to an increase of 373%±6% for 0.01 mg/ml and 149%±7% for 0.001 mg/ml over control values.</p>" ]

[ "<title>Discussion</title>", "<p>In the present study, the mechanism of MMC-induced apoptosis in corneal endothelial cells was investigated using cultured porcine cells. This work presents the findings that MMC treatment not only decreases cell viability but also induces apoptosis. The effects included turnover of membrane phosphatidylserine, as identified by annexin V-FITC, DNA degradation, as characterized with TUNEL staining, changes in membrane potential, as characterized by the release of cytochrome <italic>c</italic> from mitochondria, and upregulation of p53 and p21 proteins, caspase-8 and caspase-9.</p>", "<p>Application of MMC for the treatment of various ocular diseases has become more and more popular. However, it is not easy to detect obvious damage to corneal endothelial cells unless the side effects of MMC are severe enough to interfere with normal cellular physiologic function. A recent study has reported that a significant loss of corneal endothelial cells was observed after trabeculectomy with adjunctive MMC in glaucoma patients [##REF##9541440##7##]. Evidence has also shown that the corneal endothelial cell count in patients was significantly decreased [##REF##16935583##16##] and that permanent corneal edema was induced after photorefractive keratectomy by topical use of MMC [##REF##15448508##17##]. Some earlier studies also found that the aqueous concentration of MMC linearly increased with increases in the drug’s exposure time and concentration [##REF##17457196##18##] and that a dose-dependent increase in corneal thickness, decrease in corneal clarity, and increase in corneal endothelial apoptosis occurred after intraoperative application of MMC in an animal study [##REF##15313301##8##].</p>", "<p>Many side effects of MMC often occur only after a period of clinical use. For example, dying corneal endothelial cells displayed cell shrinkage and chromatin condensation three weeks after photorefractive keratectomy [##REF##15326110##19##], suggesting a high possibility that cells underwent an apoptosis process induced by MMC. Previous studies have reported that MMC-induced apoptosis in rabbit corneal keratocytes is mediated through the caspase-8 and caspase-9 pathways and the release of cytochrome <italic>c</italic> protein from the mitochondria [##REF##12714623##20##,##REF##16209441##21##]. In other ocular tissues, such as human lens epithelial cells, MMC also induces the upregulation of Bax, p53, and caspase-3 [##REF##11585319##22##], and in human Tenon's capsule fibroblasts, it also increases caspase-3, caspase-9, p53, Fas, FasL, and Bad as well as inducing the release of cytochrome <italic>c</italic> and changes in the mitochondrial membrane potential [##REF##16186332##23##]. However, the mechanism of apoptosis induction by MMC in any tissue, including corneal endothelial cells, is poorly understood at present.</p>", "<p>Based on the dose response of MMC in corneal endothelial cells, we found that disruption of the mitochondrial transmembrane potential was the first apoptotic characteristic detected. The mitochondrial depolarization occurs without necessarily going into the cell death pathway. The down-regulation of Bcl-2 levels, the anti-apoptotic protein, may accelerate damage to the mitochondria. The mitochondria then releases the pro-apoptotic factors such as cytochrome <italic>c</italic> from the inner mitochondrial membrane into the cytosol, which then activates the caspase-9 cascade. Thus, mitochondrial depolarization is an early event marker to more severe apoptotic events.</p>", "<p>Our data illustrated that MMC induced cellular DNA fragmentation and then triggered p53 and p21 expression. MMC damages DNA by cross-linking bases in the same or adjacent strands of DNA, which eventually induces a powerful stimulus such as p53 for apoptosis [##REF##8426740##24##,##REF##9440118##25##]. Then, p53 directly activates the expression of a large panel of genes such as p21, which plays a major role in mediating p53-dependent cell cycle G₁ arrest [##REF##7664346##26##,##REF##10362249##27##].</p>", "<p>Utilization of annexin V-FITC to identify apoptosis in the plasma membrane indicated that early apoptotic cells were clearly stained by annexin V-FITC after 24 h of incubation in 0.001 mg/ml MMC. Following exposure to 0.01 mg/ml MMC for 24 h, about 20% of the cells were still stained by PI and annexin V-FITC. Comparing the apoptotic dose response for the mitochondria and DNA with that for the plasma membrane, the cellular plasma membrane appears to be the most resistant to MMC-induced damage.</p>", "<p>The apoptotic caspases are activated by two signal complexes either in response to the ligation of cell surface death receptors (called extrinsic apoptosis pathways) or in response to signals originating from the mitochondria (called intrinsic apoptosis pathways) [##REF##16186332##23##,##REF##14754601##28##]. Evidence has shown that treatment with MMC increased the expression of Fas and FasL, which belong to the death receptor pathway [##REF##16186332##23##]. Once these receptors are activated, caspase-8 is activated for the execution of apoptosis [##REF##12181741##29##]. Our data indicated that both caspase-8 and caspase-9 inhibitors reversed the MMC-induced damage, suggesting that MMC-induced apoptosis in corneal endothelial cells may occur through both intrinsic mitochondrial-mediated and extrinsic death receptor-mediated caspase activation. However, the MMC-induced extrinsic apoptosis pathway in particular needs to be explored further in corneal endothelial cells.</p>", "<p>In summary, this study has demonstrated that MMC not only results in cytotoxicity in a dose-dependent and time-dependent manner but also induces apoptosis in corneal endothelial cells through activation of intrinsic mitochondrial and extrinsic caspase-8 apoptotic pathways. The appearance of apoptotic characteristics in corneal endothelial cells may extend the chronic toxicity of MMC to cells. Thus, the use of MMC may need to be carefully monitored for adverse changes in corneal endothelial cell viability.</p>" ]

[]

[ "<p>This is an open-access article distributed under the terms of the\n Creative Commons Attribution License, which permits unrestricted use,\n distribution, and reproduction in any medium, provided the original\n work is properly cited.</p>", "<title>Purpose</title>", "<p>Previous studies have indicated that improper use of mitomycin C (MMC) caused cytotoxicity in corneal endothelial cells. The aim of the present study was to investigate whether MMC induces cellular apoptosis in corneal endothelial cells and to determine the mechanism by which this may occur.</p>", "<title>Methods</title>", "<p>Porcine corneal endothelial cells were acquired from primary culture. Cellular damage and caspase pathway were estimated with a MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyl tetrazolium bromide) assay. The apoptotic characteristics were detected by means of flow cytometry, the TUNEL (terminal deoxyribonucleotidyl-transferase-(TdT)-mediated deoxyuridine-5′-triphosphate-digoxigenin (dUTP) nick-end labeling) test, immunofluorescent staining, and western blotting.</p>", "<title>Results</title>", "<p>The results indicated that MMC was toxic to corneal endothelial cells in a time-dependent and dose-dependent manner. Pretreatment with a general caspase inhibitor (Z-VAD-FMK), a caspase-8 inhibitor (Z-IETD-FMK), and a caspase-9 inhibitor (Z-LEHD-FMK) reversed MMC-induced cellular damage. Following exposure to MMC, a change in the mitochondrial membrane potential was positively detected by flow cytometric assay with MitoLight dye while cellular cytochrome <italic>c</italic> that was released from the mitochondria to the cytoplasm was detected by immunofluorescent staining. A positive TUNEL test revealed that cellular DNA apoptosis had occurred following exposure to 0.001 and 0.01 mg/ml MMC for 24 h. Positive annexin V-FITC, and negative propidium iodide (PI) staining indicated that the cellular plasma membrane underwent apoptosis following 0.001 mg/ml MMC exposure for 24 h. Western blot assay demonstrated down-regulation of the Bcl-2 protein and upregulation of the p53 and p21 proteins, which were all involved in apoptosis induced by MMC.</p>", "<title>Conclusions</title>", "<p>These results indicate that mitomycin-induced cellular apoptosis in corneal endothelial cells may be mediated through caspase-8, caspase-9, and the mitochondrial regulated pathways as well as through upregulation of p53-dependent and p21-dependent signal transduction pathways.</p>" ]

[]

[ "<title>Acknowledgments</title>", "<p>This work was supported by the research grant from the Kaohsiung Medical University Hospital (KMUH95–5D02) and the National Health Research Institute of Taiwan (NHRI-EX92, 93, 94, 95–9213SI).</p>" ]

[ "<fig id=\"f1\" fig-type=\"figure\" position=\"float\"><label>Figure 1</label><caption><p>Effects of mitomycin C on cell viability and apoptotic caspase pathways. <bold>A:</bold> Cell viability was measured under exposure to various concentrations of mitomycin C at 15 h and 24 h in cultured porcine corneal endothelial cells. <bold>B</bold>, <bold>C</bold>, and <bold>D</bold>: The mitomycin-induced apoptotic caspase pathways were identified with different caspase inhibitors. The cells were pretreated with a caspase-8 inhibitor (Z-IETD-FMK; <bold>B</bold>), a general caspase inhibitor (Z-VAD-FMK; <bold>C</bold>), or a caspase-9 inhibitor (Z-LEHD-FMK; <bold>D</bold>) for 1 h and then exposed to 0.001 mg/ml mitomycin C for 24 h. Data are presented as means±SD from six replicates and three independent experiments. The asterisk denotes that p&lt;0.05 when comparing the control group in <bold>A</bold> with the MMC only treated group in <bold>B</bold>, <bold>C</bold>, and <bold>D</bold>.</p></caption></fig>", "<fig id=\"f2\" fig-type=\"figure\" position=\"float\"><label>Figure 2</label><caption><p>Flow cytomety assay of mitochondrial membrane potential changes with MitoLight dye in cultured porcine corneal endothelial cells under exposure to 0.00001, 0.0001, and 0.001 mg/ml mitomycin C for 24 h. <bold>A:</bold> The curves represent fluorescent changes in the absence of mitomycin C for 24 h (control, red color) or in the presence of 0.00001 (green color), 0.0001 (black color) and 0.001 (blue color) mg/ml mitomycin C. <bold>B</bold>: It is displayed the ratio of red fluorescence in the mitochondria and green fluorescence in the cytoplasm at different concentrations of mitomycin C with histogram. The asterisk denotes that p&lt;0.05 when comparing the red and green fluorescence with control group. Two other independent experiments produced similar results.</p></caption></fig>", "<fig id=\"f3\" fig-type=\"figure\" position=\"float\"><label>Figure 3</label><caption><p>Immunofluorescent staining of cytochrome <italic>c</italic> in cultured corneal endothelial cells. Cells were incubated either in the absence of mitomycin C for 24 h (control, <bold>A</bold>) or in the presence of 0.001 mg/ml (<bold>B</bold>) and 0.01 mg/ml (<bold>C</bold>) mitomycin C for 24 h. The intensity of fluorescence staining with cytochrome <italic>c</italic> was gradually enhanced from the control cells to the apoptotic cells. Apoptotic changes throughout the cytosol were clearly visible following exposure to 0.01 mg/ml mitomycin C for 24 h. The bar in each panel represents 5 μm. Two other independent experiments produced similar results.</p></caption></fig>", "<fig id=\"f4\" fig-type=\"figure\" position=\"float\"><label>Figure 4</label><caption><p>Apoptotic DNA characteristics of corneal endothelial cells visualized with TUNEL staining. Cells were incubated either in the absence of mitomycin C for 15 h and 24 h (control, <bold>A</bold>) or in the presence of 0.001 mg/ml MMC for 24 h (<bold>B</bold>) and 0.01 mg/ml mitomycin-C for 15 h (<bold>C</bold>). The typical positive DNA apoptosis was clearly stained following exposure to mitomycin C. After exposure of cells to 0.001 mg/ml mitomycin C, cellular chromatin condensed and marginalized at the nuclear membrane. After exposure to 0.01 mg/ml mitomycin C, cells were seriously damaged and formed typical apoptotic bodies containing cytosol, condensed chromatin, and organelles. Each bar represents 3 μM. Two other independent experiments produced similar results.</p></caption></fig>", "<fig id=\"f5\" fig-type=\"figure\" position=\"float\"><label>Figure 5</label><caption><p>Flow cytometry analysis of plasma membranes with annexin V-FITC/PI double staining. Cells were incubated either in the absence of mitomycin C for 15 h and 24 h (control, <bold>A</bold>) or in the presence of 0.001 mg/ml of mitomycin C for 24 h (<bold>B</bold>) and 0.01 mg/ml mitomycin C for 15 h (<bold>C</bold>) and 24 h (<bold>D</bold>). Undamaged cells were stained with negative annexin V-FITC/PI (bottom left quadrant). After incubation with 0.001 mg/ml mitomycin C for 24 h, a significant number of apoptotic cells were stained with positive annexin V-FITC and negative PI (bottom right quadrant). Following the increasing of mitomycin C to 0.01 mg/ml, advanced apoptotic cells stained by positive annexin V-FITC and PI (upper right quadrant) were significantly augmented from 12%±3% of the control to 34%±3% after 15 h and 49%±4% after 24 h incubation. During advanced apoptosis stages, the cells were no longer viable. Data are presented as means±SD from triplicates and three independent experiments. Two other independent experiments produced similar results.</p></caption></fig>", "<fig id=\"f6\" fig-type=\"figure\" position=\"float\"><label>Figure 6</label><caption><p>Western blot assay of proteins involved in apoptosis in corneal endothelial cells. Following incubation of cells either in the absence of mitomycin C for 24 h (labeled as C in the image) or in the presence of 0.01 mg/ml (labeled as 0.01 in the image) and 0.001 (labeled as 0.001 in the image) mg/ml mitomycin C for 24 h, cells were subjected to SDS–PAGE electrophoresis and immunoblotting using antisera against Bcl-2, p53, and p21. Densitometric analysis of protein bands showed the optical density values of various proteins with a detailed description in the Results. The respective control values of three proteins were assumed as 100% response. The optical density of the Bcl-2 protein decreased to 64%±4% at 0.01 mg/ml and 86%±5% at 0.001 mg/ml in comparison with the control protein value of 100%±4%. The p53 protein only significantly increased at 0.01 mg/ml to 139%±4% over the control value. The p21 protein increased to 373%±6% at 0.01 mg/ml and 149%±7% at 0.001 mg/ml in comparison with the control value. Two other independent experiments produced similar results.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"mv-v14-1705-f1\"/>", "<graphic xlink:href=\"mv-v14-1705-f2\"/>", "<graphic xlink:href=\"mv-v14-1705-f3\"/>", "<graphic xlink:href=\"mv-v14-1705-f4\"/>", "<graphic xlink:href=\"mv-v14-1705-f5\"/>", "<graphic xlink:href=\"mv-v14-1705-f6\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Keratins are intermediate filament proteins that form a dense fibrous scaffold within the cytoplasm of epithelial cells. Based on the amino acid sequence, they are classified into type I or type II intermediate filaments, which are expressed in pairs and form obligate heterodimers in a tissue-specific and differentiation-specific manner. The predominant function of these structurally resilient polymeric proteins is to impart mechanical strength to the cells [##REF##12688839##1##]. In addition, accumulating evidence suggests that keratins also have regulatory functions influencing cell size, proliferation, translation control, organelle transport, malignant transformation, and stress responses [##REF##17434482##2##].</p>", "<p>Mutations in keratin genes result in an abnormal fragility of epithelial cells, leading to their detachment, blistering of tissues in response to even mild physical trauma, and impaired keratinization [##REF##12688839##1##,##REF##17476820##3##]. Keratin mutations were detected in several human diseases affecting the epidermis and/or its appendages, e.g., epidermolysis bullosa simplex (a group of heritable skin blistering disorders), keratoderma disorders, and hair and nail defects. They were also found in extracutaneous epithelia such as mucosa and corneal epithelium [##REF##12688839##1##].</p>", "<p>The only known disorder associated with mutation in cornea-specific keratin 3 (<italic>KRT3</italic>) and keratin 12 (<italic>KRT12</italic>) representing type II and I intermediate filaments, respectively, is Meesmann corneal dystrophy (MCD) [##REF##9171831##4##]. As with many other keratin disorders, MCD is inherited as an autosomal dominant trait with variable expression. The majority of mutations were found in <italic>KRT12</italic> and only two in <italic>KRT3</italic> [##REF##9171831##4##, ####REF##10354017##5##, ##REF##16227835##6####16227835##6##]. All of them are located in the central α-helical rod domain responsible for protein heterodimerization and higher order polymerization. They cluster in the highly conserved boundary segments of the rod domain either within its NH₂- (1A subdomain) or COOH- (2B subdomain) terminus [##REF##17653038##7##].</p>", "<p>MCD is characterized by fragility of the anterior corneal epithelium, which may lead to its recurrent erosions. Morphologically, the epithelium is disorganized and thickened with widespread cytoplasmic vacuolation and numerous small, round, keratin aggregate-laden intraepithelial microcysts [##REF##10354017##5##,##REF##17986293##8##]. They appear in childhood and increase in number with age. Although the disease is generally mild, some patients present with symptoms of lacrimation, photophobia, and intermittent diminution of visual acuity [##REF##17986293##8##].</p>", "<p>Here, we present the results of a clinical and molecular study of a previously unreported Polish family with MCD in whom a novel missense mutation in exon 7 of <italic>KRT3</italic> was found to segregate with disease. The E498V mutation affects a highly conserved amino acid [##REF##12064940##9##] at the COOH-terminus of the K3 rod-domain and represents the third mutation to be detected in this gene. Other mutations found in <italic>KRT3</italic> and <italic>KRT12</italic> to date are also reviewed.</p>" ]

[ "<title>Methods</title>", "<p>A three-generation Polish family with four affected individuals was studied. All subjects gave informed consent in accordance with the tenets of the Declaration of Helsinki. A complete ophthalmological check-up including slit-lamp examination and confocal microscopy in vivo by Rostock Cornea Module (RCM) for HRT II (Heidelberg Engineering, Dossenheim, Germany), a preferred laser scanning confocal microscope for corneal epithelium evaluation [##REF##17457193##10##], were performed.</p>", "<p>Genomic DNA was extracted from blood obtained from all available family members (n=7). Control DNA samples came from a repository of anonymous samples (n=100, female:male ratio 1:1) representative of the background population of Central Poland, which has been described previously [##UREF##0##11##]. Ophthalmologic status of these individuals was not known.</p>", "<p>DNA mutation screening was performed by amplifying the entire coding region of <italic>KRT3</italic> and <italic>KRT12</italic> with primers located in the noncoding sequences and designed based on the reference sequences of the respective genes, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/entrez/viewer.fcgi?db=nuccore&amp;id=109148551\">NM_057088</ext-link> and <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/entrez/viewer.fcgi?db=nuccore&amp;id=156602658\">NM_000223</ext-link>. Polymerase chain reaction (PCR) was performed at 94 °C for 2 min followed by 35 cycles at 94 °C for 45 s, 57–63 °C for 90 s, 72 °C for 60 s, and 72 °C for 10 min. Primer sequences and annealing temperatures for each primer set are given in ##TAB##0##Table 1##. PCR products were examined on 1% agarose gels and then sequenced using an ABI PRISM 377 DNA sequencer (Applied Biosystems, Foster City, CA) and BigDye Termiantion cycle sequencing kit v. 3.1 (Applied Biosystems).</p>", "<p>The mutation in exon 7 was verified by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP) analysis using HphI (MBI Fermentas, Vilnius, Lithuania). The PCR reaction was performed with the same primers as used for sequencing (##TAB##0##Table 1##), and the digestion was performed according to the manufacturer’s instructions. The digestion products were separated by electrophoresis in 2% agarose gels and visualized by ethidium bromide staining. In the presence of the E498V mutation, the digestion resulted in the bands of 304 bp and 260 bp whereas in the wild type homozygotes, only a 304 bp band was observed. The PCR-RFLP analysis was used to screen family members and 100 control DNA samples. All family members in whom the E498V mutation was found by PCR-RFLP were also analyzed by direct sequencing.</p>" ]

[ "<title>Results</title>", "<p>In the proband (III-1, 24 years old), the diagnosis of MCD was made on the basis of the typical clinical appearance of the corneal microcysts, which were detected in both eyes during routine eye examination (##FIG##0##Figure 1## and ##FIG##1##Figure 2##). They were also found in three other family members I-2 (71 years old), II-2 (49 years old), and III-2 (23 years old; ##FIG##2##Figure 3B##). Patients I-2, II-2, and III-1 were hyperopic from early childhood. None of the affected family members complained about typical symptoms of MCD.</p>", "<p>The pedigree of the examined family was consistent with an autosomal dominant mode of inheritance (##FIG##2##Figure 3B##). Since most mutations in MCD patients were found in exons 1 and 6 of <italic>KRT12</italic> and the remaining two in exon 7 of <italic>KRT3</italic>, these coding regions were initially sequenced in both the forward and the backward directions in the proband. The analysis revealed the presence of a novel heterozygous 1493A&gt;T transversion in exon 7 of <italic>KRT3</italic> (##FIG##2##Figure 3A##). The mutation predicts a glutamate to valine amino acid change at codon 498 (E498V) within the region of highly conserved 2B rod domain segment in K3. Sequencing of the remaining coding regions of <italic>KRT3</italic> in the proband did not show any alterations.</p>", "<p>The E498V mutation creates a recognition site for the HphI endonuclease. The PCR-RFLP analysis showed that the E498V mutation cosegregated with the MCD phenotype in the studied family members (##FIG##2##Figure 3B,C##). Using this method, DNA samples from 100 subjects from the background population of central Poland were also screened, and no carriers were found.</p>", "<p>Sequencing of <italic>KRT12</italic> exon 1 of from the proband revealed the presence of a common coding homozygous non-synonymous single nucleotide polymorphism (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=11650915\">rs11650915</ext-link>). Additionally, a previously unreported heterozygous substitution of A&gt;C in the 3′ region of <italic>KRT12</italic> (position NT:2741849, chromosome 17:36271079, according to <ext-link ext-link-type=\"uri\" xlink:href=\"http://genewindow.nci.nih.gov/\">Genewindow</ext-link>) was detected. The latter variant did not segregate with the disease. It was found in affected members (III-1 and III-2) as well as unaffected family members (II-3 and III-3).</p>" ]

[ "<title>Discussion</title>", "<p>The E498V mutation in <italic>KRT3</italic> found in the Polish family with MCD affects a strongly conserved residue within the 2B subdomain of the intermediate filament chain [##REF##12064940##9##]. Conservation of an amino acid residue indicates its significance for protein function and low tolerance to replacement [##REF##12064940##9##,##UREF##1##12##]. The E498V mutation predicts a particularly unfavorable substitution of a negatively charged, polar glutamate to an aliphatic and hydrophobic valine. Glutamate is often present in the protein active or binding site. It pairs with positively charged amino acids to create hydrogen bonds, which are important for protein stability. Conversely, valine is preferably present in protein hydrophobic cores. It contains two substituents at its C-beta carbon, which restrict the conformational changes that the main chain can adopt. One of the most pronounced effects of this property is the difficulty of valine to adopt an α-helical conformation [##UREF##1##12##]. Thus, the replacement of glutamate to valine is likely to influence both the physicochemical and structural properties of the α-helical rod domain in K3, leading to the disruption of the cytoskeletal keratin network.</p>", "<p>All mutations in <italic>KRT3</italic> and <italic>KRT12</italic> reported to date affect one or the other terminus of the central α–helical rod domain and all but one (a 27 bp insertion [##REF##15148206##13##]) are missense mutations (##TAB##1##Table 2##, ##FIG##3##Figure 4##). Of note is also the lack of reported mutations in the 1A subdomain of K3. Whether this is only a chance finding resulting from the scarcity of genotyped MCD cases or their incompatibility with normal development, which seems less plausible, remains to be elucidated.</p>", "<p>An interesting and yet unresolved issue in MCD is the occurrence of asymptomatic cases despite the presence of proven <italic>KRT</italic> mutations and morphological findings. An example of such MCD presentation is the family reported in this study. Interestingly, our case and the review of data on the mutations and phenotypes reported so far in MCD (##TAB##1##Table 2##) and other diseases caused by keratin mutations may suggest a framework for understanding genotype/phenotype correlation in MCD.</p>", "<p>Amino acids located in the boundary sequence motifs of the keratin rod domain are highly conserved and particularly important in intermediate filaments assembly as they mediate end-to-end interactions between keratin heterodimers and filament elongation. These regions were found to represent mutational hot spots in MCD as well as in other keratin types [##REF##17476820##3##]. mutations, which affect the boundary sequence motifs, typically exert a dominant-negative effect being highly disruptive to filament assembly and usually associate with the severe disease phenotypes. In contrast, mutations in other parts of keratin genes are compatible with filament assembly, and the disease phenotype is generally milder [##REF##10354017##5##]. Interestingly, all three mutations, which so far have been reported in asymptomatic patients (<italic>KRT12</italic>; V143L, <italic>KRT12</italic>; I426V, and <italic>KRT3</italic>; E498V) or patients with relatively mild symptoms (<italic>KRT12</italic>; I426S), are located innermost relative to other mutations and possibly in less critical regions of the boundary motifs of their respective subdomains (##FIG##3##Figure 4##). The only exception is the <italic>KRT12</italic> 400ins9 mutation, which is not directly comparable to other mutations since it leads to an insertion of nine novel amino acids and is likely to be damaging to filament assembly despite a relatively long distance from the terminus of the 2B subdomain (##FIG##3##Figure 4##). These data suggest that putative missense mutations localized internally in <italic>KRT12</italic> (V143 and I426) or <italic>KRT3</italic> (E498) are also likely to be asymptomatic and thus provide a general framework for genotype/phenotype correlation in MCD.</p>" ]

[]

[ "<p>The first two authors contributed equally to this work.</p>", "<p>This is an open-access article distributed under the terms of the\n Creative Commons Attribution License, which permits unrestricted use,\n distribution, and reproduction in any medium, provided the original\n work is properly cited.</p>", "<title>Purpose</title>", "<p>Juvenile epithelial corneal dystrophy of Meesmann (MCD, OMIM <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/entrez/dispomim.cgi?id=122100\">122100</ext-link>) is a dominantly inherited disorder characterized by fragility of the anterior corneal epithelium and intraepithelial microcyst formation. Although the disease is generally mild and affected individuals are often asymptomatic, some suffer from recurrent erosions leading to lacrimation, photophobia, and deterioration in visual acuity. MCD is caused by mutations in keratin 3 (<italic>KRT3</italic>) or keratin 12 (<italic>KRT12</italic>) genes, which encode cornea-specific cytoskeletal proteins. Seventeen mutations in <italic>KRT12</italic> and two in <italic>KRT3</italic> have been described so far. The purpose of this study was to investigate the genetic background of MCD in a Polish family.</p>", "<title>Methods</title>", "<p>We report on a three-generation family with MCD. Epithelial lesions characteristic for MCD were visualized with slit-lamp examination and confirmed by in vivo confocal microscopy. Using genomic DNA as a template, all coding regions of <italic>KRT3</italic> and <italic>KRT12</italic> were amplified and sequenced. Presence of the mutation was verified with restriction endonuclease digestion.</p>", "<title>Results</title>", "<p>In the proband, direct sequencing of the polymerase chain reaction (PCR) product from amplified coding regions of <italic>KRT3</italic> and <italic>KRT12</italic> revealed a novel 1493A&gt;T heterozygous missense mutation in exon 7 of <italic>KRT3</italic>, which predicts the substitution of glutamic acid for valine at codon 498 (E498V). Using PCR-Restriction Fragment Length Polymorphism (RFLP) analysis, the mutation was demonstrated to segregate with the disease (four affected members, three non-affected) and to be absent in 100 controls from the Polish population, indicating that it is not a common polymorphism.</p>", "<title>Conclusions</title>", "<p>Location of the E498V mutation emphasizes the functional relevance of the highly conserved boundary motifs at the COOH-terminus of the α-helical rod domain in keratin 3 (K3).</p>" ]

[]

[ "<title>Acknowledgments</title>", "<p>This work was supported by Medical University of Warsaw Grants No. 1M15/N/2008 and 2WF/W/09.</p>" ]

[ "<fig id=\"f1\" fig-type=\"figure\" position=\"float\"><label>Figure 1</label><caption><p>Slit-lamp photography of proband (III-1). This image demonstrates the microcystic appearance of the corneal epithelium.</p></caption></fig>", "<fig id=\"f2\" fig-type=\"figure\" position=\"float\"><label>Figure 2</label><caption><p>Confocal microscopy image of proband’s cornea. This image shows the presence of hyperreflective material within the intraepithelial cysts.</p></caption></fig>", "<fig id=\"f3\" fig-type=\"figure\" position=\"float\"><label>Figure 3</label><caption><p>Identification of the heterozygous point mutation E498V in exon 7 of <italic>KRT3</italic> in the Polish MCD family. <bold>A</bold>: DNA sequencing. Electropherograms from bidirectional sequencing of <italic>KRT3</italic> exon 7 in the proband showed a 1493 A&gt;T (G<underline>A</underline>G&gt;G<underline>T</underline>G) heterozygous mutation, predicting the amino aid change E498V. <bold>B</bold>: Pedigree of the studied family. The arrow indicates the proband (III-1). <bold>C</bold>: PCR-RFLP analysis. The <italic>KRT3</italic> E498V mutation creates a recognition site for HphI. The presence of this restriction site is seen to cosegregate with MCD in this family. Upon digestion, the full sized 304 bp product is cut into bands of 260 bp and 44 bp (the latter not visible on the figure). DNA molecular weight markers are shown on the left (lane 1). The heterozygous E498V mutation (lanes 3–6) was detected in all affected family members (I-2, II-2, III-1, and III-2). The homozygous normal allele, represented by the 304 bp band (lane 2, 7, and 8), was found in unaffected family members (II-1, II-3, and III-3).</p></caption></fig>", "<fig id=\"f4\" fig-type=\"figure\" position=\"float\"><label>Figure 4</label><caption><p>Schematic drawing of K3 and K12 structure with assigned positions of the published mutations. Keratins are composed of three main parts, the central α-helical rod domain, which is divided into four subdomains (1A, 1B, 2A, and 2B), and the two non-helical variable domains (V1 and V2) at each end [##REF##17476820##3##]. All three mutations within <italic>KRT3</italic> localize exclusively in the boundary motif of the 2B subdomain. Among the mutations in <italic>KRT12,</italic> 11 were found in the 1A subdomain and six in the 2B subdomain (see also ##TAB##1##Table 2##).</p></caption></fig>" ]

[ "<table-wrap id=\"t1\" position=\"float\"><label>Table 1</label><caption><title>Primers for polymerase chain reaction amplification and sequencing of <italic>KRT3</italic> and <italic>KRT12</italic>, their annealing temperature (Ta), and expected amplicon size.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"50\" span=\"1\"/><col width=\"202\" span=\"1\"/><col width=\"237\" span=\"1\"/><col width=\"37\" span=\"1\"/><col width=\"54\" span=\"1\"/><thead><tr><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Gene/</bold>
<bold>exon</bold></th><th valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Forward primer sequence (5’→ 3’)</bold></th><th valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Reverse primer sequence (5’→ 3’)</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Ta (°C)</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Product size (bp)</bold></th></tr></thead><tbody><tr><td colspan=\"5\" valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\"><italic>KRT3</italic><hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TgCACAggTCTTCATTTCCCATCC<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TCCTCAACCCTggATATCTTCCCA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">887<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AgTgTTgCCTgATgTTgCTTCCTg<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ACCATgCTTggAgAAggAAggTgA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">439<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ATggAgggAgggAAgAgATgAACT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ATTgCTCCAAAggCCTGAACTTgg<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">275<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">gCTCTTTCTTgCTgCAgTTgTggT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">gCACCAgCCTCAAATCTggAAACA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">60<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">238<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 5<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AgTgAACAAgCTCCCTCTgTgTTg<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TgAAACCTCCAgTggATCCCgTAA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">60<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">235<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AAggTTTggTgggTgATgTTggAg<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ATTTgTggAgATACTgCCCTgTgg<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">345<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AATCCATTgCATgTCAggAAgggC<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TATCTGGCCCTTGGCCTATGACTT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">60<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">354<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 8<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TgTTggTgATgTgCTTTgTgACgg<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AAgCCAATCACTTCCCTCTCCTCT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">60<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">228<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 9<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ACAATAACATAgCAgCTggCCTgg<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AATACTCAgAggCCCggAgTgAAA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">756<hr/></td></tr><tr><td colspan=\"5\" valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\"><italic>KRT12</italic><hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1a<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AgTgAACTTTTCAACTgCgA<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TgCCCgAgAgAATACCTAgA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61.5<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">450<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1b<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AggACTgggTgCTggTTAT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CTgCAAgTACAgCTAAATTggA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">62<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">447<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TAgggCTTCAATCTTgTgTgTgTCCC<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TTTATATCAATgAAggCAggACAgTAggAC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61.2<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">200<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CCCTCAACTgCTTTgCACTTggTT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CTCCATACTTgTCCTgACTCCAgA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">58.4<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">289<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 4<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CAgggCCCACgAAAgTCACAAT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">gTTCgCAggCCTTTCTgTgAATgT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">58.3<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">272<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 5<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">ACATTCACAgAAAggCCTgCgAAC<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">TggAAgTCCAAAggATgCTACgTC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">58.4<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">235<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">gCTCgTgCgCAAACAgACgT<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CCCAggCATATCTTTACTAgA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">60<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">490<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AgCCACCTgAACCACCTACTCTAA<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">AgCTATgAggTTACAggCATgAgC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">63<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">469<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 8</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">gCCTACATTAAACAACCAgTgTTgg</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">CAAgCAATCATCTTgCCTCTCAgC</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">61</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">684</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t2\" position=\"float\"><label>Table 2</label><caption><title><italic>KRT3</italic> and <italic>KRT12</italic> genotypes and symptoms in patients with MCD.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"68\" span=\"1\"/><col width=\"264\" span=\"1\"/><col width=\"72\" span=\"1\"/><col width=\"81\" span=\"1\"/><col width=\"1\" span=\"1\"/><col width=\"65\" span=\"1\"/><thead><tr><th valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Gene/exon</bold></th><th valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Nucleotide</bold></th><th valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Protein</bold></th><th valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Ocular symptoms</bold></th><th colspan=\"2\" valign=\"top\" align=\"left\" scope=\"colgroup\" rowspan=\"1\"><bold>Reference</bold></th></tr></thead><tbody><tr><td valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><italic>KRT3</italic><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1493A&gt;T<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">E498V<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">asymptomatic<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\">present study<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1508G&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">R503P<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">foreign body sensation, mild blurred vision<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##16227835##6##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 7<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1525G&gt;A<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">E509K<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##9171831##4##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><italic>KRT12</italic><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"><hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">410T&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">M129T<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##10644419##14##,##REF##16352477##15##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">413A&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Q130P<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">recurrent painful erosions, foreign body sensation, photophobia, lacrimation, blurred vision<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##10781519##16##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">423A&gt;G<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">N133K<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">soreness of both eyes; deterioration in visual acuity<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##12084738##17##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">427A&gt;G<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">R135G<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">photophobia, lacrimation, itching<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##9399908##18##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">428G&gt;T<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">R135I<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">photophobia, lacrimation, itching<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##9399908##18##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">428G&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">R135T<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##9171831##4##,##REF##10644419##14##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">429A&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">R135S<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">post-traumatic recurrent erosion<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##15148206##13##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">433G&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">A137P<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">photophobia<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##12543196##19##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">443T&gt;G<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">L140R<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">photophobia, lacrimation, itching<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##9399908##18##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">451G&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">V143L<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">-<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##9171831##4##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">451G&gt;T<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">V143L<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">asymptomatic<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\"> [##REF##18245975##20##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1222+ATCAGCAACCTGGAGGCACAGCTGCTC<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">400 ins ISNLEAQLL<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\">recurrent erosions, foreign body sensation, photophobia, fluctuating vision, contact lens intolerance<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"> [##REF##15148206##13##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1300A&gt;G<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">I426V<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\">asymptomatic<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"> [##REF##10612503##21##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1301T&gt;G<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">I426S<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\">photophobia<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"> [##REF##16352477##15##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1286A&gt;C<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Y429D<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\">photophobia, lacrimation, itching<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"> [##REF##9399908##18##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1286A&gt;G<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Y429C<hr/></td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\">recurrent erosions, foreign body sensation, photophobia, lacrimation, fluctuation of visual acuity<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"> [##REF##16227835##6##]<hr/></td></tr><tr><td valign=\"top\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">exon 6</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">1289G&gt;C</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">R430P</td><td colspan=\"2\" valign=\"top\" align=\"left\" rowspan=\"1\">symptoms from birth; photophobia, lacrimation, periodic burning, irritation, significant impairment of visual acuity</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\"> [##REF##17653038##7##]</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"mv-v14-1713-f1\"/>", "<graphic xlink:href=\"mv-v14-1713-f2\"/>", "<graphic xlink:href=\"mv-v14-1713-f3\"/>", "<graphic xlink:href=\"mv-v14-1713-f4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Usher syndrome (USH) is an autosomal recessive disorders characterized by sensorineural hearing impairment (HI), retinitis pigmentosa (RP), and variable vestibular dysfunction [##REF##10704190##1##]. It is clinically and genetically heterogeneous, and it is categorized into three clinical subtypes. USH type 1 (USH1) is the most severe form. Patients with USH1 suffer from vestibular dysfunction, delayed motor development, congenital sensorineural HI, and RP starting in early childhood. RP is due to photoreceptor degeneration, which occurs from the periphery of the retina to the macula. Night blindness is the first symptom of RP followed by narrowing of the visual field [##REF##15563868##2##]. Those with USH type II (USH2) have moderate to severe congenital sloping HI, normal vestibular function and a late onset of RP. USH type III (USH3) is characterized by variable RP and vestibular dysfunction combined with progressive HI. There are 11 known loci (USH1B-USH1G, USH2A-USH2D, and USH3), and for nine of them, the corresponding genes have been identified: USH1B/<italic>MYO7A</italic>, USH1C/<italic>USH1C</italic>, USH1D/<italic>CDH23</italic>, USH1F/<italic>PCDH15</italic>, USH1G/<italic>SANS</italic>, USH2A/<italic>USH2A</italic>, USH2C/<italic>VLGR1</italic>, USH2D/<italic>WHRN,</italic> and USH3A/<italic>USH3A</italic> (<ext-link ext-link-type=\"uri\" xlink:href=\"http://webh01.ua.ac.be/hhh/\">Usher homepage</ext-link>). Mutations in USH2 genes can also manifest as atypical USH [##REF##10090909##3##], as nonsyndromic recessive HI [##REF##12833159##4##], or as nonsyndromic recessive RP [##REF##10775529##5##].</p>" ]

[ "<title>Methods</title>", "<title>Family and clinical data</title>", "<p>In this study, we investigated a Tunisian family with USH2. This family originates from centre of Tunisia. Two affected (1 male and 1 female aged 28 and 18 years, respectively) and six healthy family members (2 males and 4 females aged 21-61 years) attended our study. We also recruited 45 controls (22 males and 23 females aged 26-72 years) from different regions of Tunisia. Written informed consent was obtained from both parents, in accordance with the ethics committee of the University Hospital of Sfax. The pedigree was obtained upon interviews with parents (##FIG##0##Figure 1##). Clinical history and physical examinations of family members ruled out the implication of environmental factors in the etiology of HI and RP. Eight family members were subjected to audiologic examination, which consisted of otoscopic exploration and pure-tone audiometry. Testing of the vestibular system was performed by electron stagmography. Ocular examinations included fundus ophthalmoscopy, visual field examination, and Ganzfeld-electroretinogram (ERG). Blood samples were collected from eight family members. Genomic DNA was extracted from whole blood following a standard phenol-chloroform method.</p>", "<title>Microsatellite genotyping and homozygosity mapping</title>", "<p>For each gene and locus responsible for USH (<ext-link ext-link-type=\"uri\" xlink:href=\"http://webh01.ua.ac.be/hhh/\">Usher homepage</ext-link>) at least three microsatellite markers were selected on the basis of their map position (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.genome.ucsc.edu\">UCSC Genome Browser</ext-link>) and heterozygosity coefficient (HE; <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.cephb.fr/en/cephdb/browser.php\">minimal HE of 0.7</ext-link>). Fluorescent dye-labeled microsatellite markers were genotyped for all the participating family members. Furthermore, a genome-wide scan was performed using 400 fluorescent dye-labeled microsatellite markers with an average spacing of approximately 10 cM (Prism Linkage Mapping Set, Applied Biosystems, Foster City, CA). We used the True Allele PCR Premix (Applied Biosystems) for PCR reactions according to the manufacturer’s instructions. Fluorescently labeled alleles were analyzed on an ABI Prism 3100-Avant automated DNA sequencer (Applied Biosystems).</p>", "<p>We used homozygosity mapping to identify autozygous regions in the two affected children. Two-point and multipoint LOD scores were calculated with <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.broad.mit.edu/ftp/distribution/software/genehunter/\">Genehunter software</ext-link> V2.1 version, using 100% penetrance, four alleles with equal allele frequencies, and no phenocopies.</p>", "<title>Mutation analysis and single nucleotide polymorphism genotyping</title>", "<p>Direct sequencing of candidate genes was performed using primers in the intronic regions (##TAB##0##Table 1## and ##TAB##1##Table 2##). Amplified products were directly sequenced using an ABI 3100-Avant automated DNA sequencer and Big Dye Terminator Sequencing V3.1 Kit (Applied Biosystems).</p>", "<p>We also analyzed by direct sequencing three single nucleotide polymorphism (SNP) markers (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=155100\">rs155100</ext-link>, <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=1157595\">rs1157595</ext-link>, and <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=1002207\">rs1002207</ext-link>; <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP\">dbSNP</ext-link>) spanning the ceramide kinase-like gene (<italic>CERKL</italic>) to check their cosegregation with the disease before proceeding to mutation screening. Primers and conditions were previously described [##REF##17279538##6##].</p>" ]

[ "<title>Results</title>", "<title>Family and clinical data</title>", "<p>The pedigree in ##FIG##0##Figure 1## displays a consanguineous Tunisian family segregating USH based on clinical history and audiometric and ophthalmologic tests. Audiometric test showed a moderate sloping bilateral sensorineural HI in USH patients (##FIG##1##Figure 2##). Severity of HI was similar in patients BT188 and BT189. Parents reported that HI was first noted in BT188 when the child was six years old, but observed it in BT189, when the older sibling was ten years old. For patient BT189 two audiograms were made at four-year intervals with no change in the profile (##FIG##1##Figure 2##). The father (60 years) had high frequency HI caused by bilateral presbycusis. No vestibular dysfunction was detected in both patients (BT188 and BT189) using the caloric test, nor was there any history of a delay in the age of walking. Patient BT189 reported having night blindness problem beginning at the age of 13 years. Fundus examination at in BT188 at age 14 and in BT189 at age 24 detected severe retinal degeneration. Visual fields (Goldmann targets III/4e) were significantly reduced to 5° concentric field and temporal island fields in BT189 for both eyes and 5° and 10° respectively in left and right eye in BT188. In BT189 and BT188, the nasal and temporal fields were not preserved, and only central field was maintained (##FIG##2##Figure 3##). The Ganzfeld-ERG recorded in BT189 showed an almost normal response flash visual-evoked potential in both eyes and a significant bilateral global retinal degeneration. Only cone flicker responses of less than 15% of the normal mean were recordable under photopic conditions while all other responses were below noise level (BT189), a typical finding for patients with RP (##FIG##3##Figure 4##). Nystagmus was noted in patient BT188 since her first examination at age 14 years. No other abnormalities were observed in these two patients. Taken together, the clinical signs observed in affected subjects indicate a form of USH2.</p>", "<title>Genome-wide screening and homozygosity mapping</title>", "<p>To localize the causative gene, we performed linkage analysis using polymorphic microsatellite markers bordering all described <ext-link ext-link-type=\"uri\" xlink:href=\"http://webh01.ua.ac.be/hhh/\">USH loci and genes</ext-link>. The USH phenotype segregating in the family was not found to be linked to the published USH loci. ##TAB##2##Table 3## shows statistical evidence for exclusion of <italic>USH2</italic> genes. Therefore, a genome-wide screen using microsatellite markers was performed. Linkage was found with four markers D2S117 (2q32.3), D4S402 (4q26), D15S978 (15q21.1), and D15S117 (15q22.1). Additional markers were genotyped in all three regions to define the critical intervals. The homozygous region in chromosome 2 was delimited by two informative markers <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=1002207\">rs1002207</ext-link> and D2S311; the chromosome 4 region was bordered by D4S1572 and D4S2938; and the chromosome 15 region stretched from D15S1039 to D15S120 (##FIG##0##Figure 1##). Analysis of these markers allowed us to refine the boundary of the critical linkage intervals respectively to 16 Mb, 25.5 Mb, and 55 Mb. In the CH15 region, six polymorphic <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.gdb.org/\">microsatellite markers</ext-link> were found to be noninformative in the family (##FIG##0##Figure 1##). Investigation of the polymorphism of these repeats in the Tunisian population was performed in 40 unrelated individuals from different areas. Our results demonstrated that these microsatellites display a high degree of genetic polymorphism in the general Tunisian population. Microsatellite marker heterozygosity values were estimated using <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.genemapping.cn/util.htm\">HET software</ext-link> version 1.8 and are as follows: 0.61 for D15S993, 0.89 for D15S153, 0.84 for D15S131, 0.86 for D15S205, 0.85 for D15S127 and 0.83 for D15S130. To rule out chromosome 15 aberrations, we performed G banded karyotype analysis on Phytohemagglutinin (PHA)-stimulated blood culture using standard procedures. Chromosome analysis of patient BT189 showed normal karyotype (data not shown).</p>", "<p>We genotyped markers located in the three candidate regions in 40 healthy unrelated Tunisian individuals for more accurate estimation of allele frequencies and to determine the best candidate region. In the first region, we analyzed four markers (D2S148, D2S384, D2S364, and D2S117). In the first region, homozygous alleles were predominantly present in the population and the allele frequencies were 0.35 (D2S148), 0.125 (D2S384), 0.311 (D2S364), and 0.203 (D2S117). For the second region, three markers were analyzed and the frequencies of linked alleles were as follows: 0.025 for D4S2989, 0.122 for D4S402, and 0.125 for D4S2975. In contrast, the homozygous alleles of the chromosome 15 region were not frequent in controls. Allele frequencies of the polymorphic markers D15S992, D15S978, D15S117, and D15S1036 were assumed to be 0.05, 0.05, 0.1, and 0.075, respectively. These results suggest that the disease locus is most probably on chromosome 15. Multipoint LOD scores were calculated for the family data using <ext-link ext-link-type=\"uri\" xlink:href=\"http://linkage.rockefeller.edu/soft/\">Genehunter software</ext-link>. Maximum LOD scores (1.765 at θ=0) were identified for the candidate regions on chromosome 4 between D4S2989 and D4S2975, and chromosome 15 between D15S978 and D15S1036. On chromosome 2, a maximum LOD score of 1.51 was found for D2S117 microsatellite marker.</p>", "<title>Candidate gene screening</title>", "<p>The evaluation of the three homozygous regions revealed a large number of known and hypothetical genes (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.genome.ucsc.edu/\">UCSC Genome Browser</ext-link>). More than 100 candidate genes in these three regions are expressed in the inner ear and in the retina. Although the region on chromosome 2 was not the best candidate locus (the lower LOD score and homozygous alleles of each linked marker are common in Tunisian population), we chose to investigate the <italic>CERKL</italic> gene, encoding a ceramide kinase, as candidate since it has been described to cause nonsyndromic autosomal recessive RP (<italic>RP26</italic>) [##REF##14681825##7##]. The basis of this choice is that mutations in <italic>USH2A</italic> are responsible for USH2 as well as nonsyndromic recessive RP [##REF##10775529##5##]. SNP (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=1157595\">rs1157595</ext-link> and <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=155100\">rs155100</ext-link>) genotyping was compatible with linkage of the <italic>CERKL</italic> gene by cosegregation and homozygosity criteria (##FIG##0##Figure 1##). However, BT188 and BT189 were heterozygous for the <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=1002207\">rs1002207</ext-link> (C/T), which was located at 0.8 Mb from <italic>CERKL</italic> gene. We screened this gene for mutations. Two affected children (BT188 and BT189) were compound heterozygous for two novel variants (##FIG##0##Figure 1##). The first change was a G&gt;A (c.1073+34G&gt;A) transition at position 34 from the donor splicing site of intron 8. The second was a c.242A&gt;C transversion in exon 2, which leads to p.Asp81Ala substitution. Molecular modeling of the N-terminal region showed that the mutation p.Asp81Ala has no structural effect [##REF##10860755##8##, ####REF##8254673##9##, ##REF##8515464##10##, ##REF##15701681##11##, ##REF##7954789##12##, ##REF##7588597##13####7588597##13##]. We detected this variant at heterozygous state in 2 out of 45 Tunisian controls. Taken together, these results exclude this variation to cause any functional defect on the encoded enzyme. Therefore, the locus on chromosome 2 was reduced to 14.7 Mb (##FIG##0##Figure 1##).</p>", "<p>We also screened for mutations in another gene, <italic>MYO1E</italic>, encoding an unconventional myosin and representing a very good candidate on chromosome 15 locus. No nucleotide variant was detected in this gene.</p>" ]

[ "<title>Discussion</title>", "<p>In this paper, we report a consanguineous family of Tunisian origin, composed of two affected children with USH. On the whole, the clinical signs observed in affected subjects from this family were indicative of USH2. USH2 is characterized by moderate to severe HI, and onset of RP in the second decade of life. Vestibular function is not impaired in this subtype. Subtle variations within the USH2 phenotype have been observed in several studies. Liu et al. [##REF##12833159##4##] showed that mutations in the USH2A gene were present at homozygous state not only in typical USH2 patients, but also in USH3-like patients who present with late onset progressive deafness that is occasionally associated with vestibular dysfunction. The p.R334W mutation either causes USH2 or atypical USH [##REF##10738000##14##]. Nystagmus was also described in USH2 patients [##REF##10745043##15##].</p>", "<p>This consanguineous Tunisian family displayed no evidence of linkage to any known USH locus. A genome-wide genotyping was performed and revealed three homozygous regions on chromosomes 2q31.3–33.1, 4q24–28.2, and 15q21–15qter. The highest LOD scores were identified for the regions on chromosome 4 and 15. The determination of population frequencies of the homozygous alleles of each linked marker in these three regions showed that only the homozygous alleles of chromosome 15 were rarely present in 40 control Tunisian individuals. More controls (45) were used to check for the novel variant on <italic>CERKL</italic> gene. On the basis of these results, we believe chromosome 15 locus is the most likely locus for the defective gene. This region colocalizes with an autosomal recessive nonsyndromic HI locus (<italic>DFNB48</italic>) mapped to 15q23-q25.1 in five large Pakistani families [##REF##15711797##16##]. Among interesting candidate genes on chromosome 15 region, one gene, <italic>MYO1E</italic>, was selected. Myosins are motor proteins that hydrolyze ATP and translocate along actin filaments [##REF##11294886##17##]. Indeed, the involvement of unconventional myosins in hereditary HI is well documented [##REF##11577373##18##]. Mutations of myosins IA, IIIA, VI, VIIA, and XVA are associated with HI in humans [##REF##12736868##19##, ####REF##12032315##20##, ##REF##11468689##21##, ##REF##9171833##22##, ##REF##9603735##23####9603735##23##]. Mutations in <italic>MYO7A</italic> have been reported essentially in families with USH1 but also can lead to atypical USH [##REF##11391666##24##]. <italic>MYO1E</italic> is a member of a <italic>Myosin I</italic> isozyme which are essential for hair cells, the sensory cell of inner ear. All eight Myosin I isozymes are expressed in rodent auditory and vestibular epithelia. Three Myosin I isosymes <italic>Myo1b</italic>, <italic>Myo1c</italic>, and <italic>Myo1e</italic>, are expressed at birth in cochlea and vestibular organs. In mouse, <italic>Myo1e</italic> is expressed in hair cell of the auditory and vestibular epithelia. [##REF##12486594##25##]. This isozyme was enriched in the cuticular plate. Myosin Ie may mediate adaptation of mechanoelectrical transduction. All exons and the flanking sequences of the <italic>MYO1E</italic> gene were sequenced in patients and were found to be negative for functional sequence variants.</p>", "<p>As the chromosome 15 interval is large and no more information can be obtained from this family to reduce the size of this locus, other families with USH2, even if small, would be useful to identify the novel gene.</p>" ]

[]

[ "<p>This is an open-access article distributed under the terms of the\n Creative Commons Attribution License, which permits unrestricted use,\n distribution, and reproduction in any medium, provided the original\n work is properly cited.</p>", "<title>Purpose</title>", "<p>Chronic diseases affecting the inner ear and the retina cause severe impairments to our communication systems. In more than half of the cases, Usher syndrome (USH) is the origin of these double defects. Patients with USH type II (USH2) have retinitis pigmentosa (RP) that develops during puberty, moderate to severe hearing impairment with downsloping pure-tone audiogram, and normal vestibular function. Four loci and three genes are known for USH2. In this study, we proposed to localize the gene responsible for USH2 in a consanguineous family of Tunisian origin.</p>", "<title>Methods</title>", "<p>Affected members underwent detailed ocular and audiologic characterization. One Tunisian family with USH2 and 45 healthy controls unrelated to the family were recruited. Two affected and six unaffected family members attended our study. DNA samples of eight family members were genotyped with polymorphic markers. Two-point and multipoint LOD scores were calculated using Genehunter software v2.1. Sequencing was used to investigate candidate genes.</p>", "<title>Results</title>", "<p>Haplotype analysis showed no significant linkage to any known USH gene or locus. A genome-wide screen, using microsatellite markers, was performed, allowing the identification of three homozygous regions in chromosomes 2, 4, and 15. We further confirmed and refined these three regions using microsatellite and single-nucleotide polymorphisms. With recessive mode of inheritance, the highest multipoint LOD score of 1.765 was identified for the candidate regions on chromosomes 4 and 15. The chromosome 15 locus is large (55 Mb), underscoring the limited number of meioses in the consanguineous pedigree. Moreover, the linked, homozygous chromosome 15q alleles, unlike those of the chromosome 2 and 4 loci, are infrequent in the local population. Thus, the data strongly suggest that the novel locus for USH2 is likely to reside on 15q.</p>", "<title>Conclusions</title>", "<p>Our data provide a basis for the localization and the identification of a novel gene implicated in USH2, most likely localized on 15q.</p>" ]

[]

[ "<title>Acknowledgments</title>", "<p>We are indebted to the family members for their invaluable cooperation and for providing the blood samples. We thank Dr. Sandrine Marlin and Dr. Hela Azaiez for helpful suggestions on the manuscript. We also thank Dr. Roser González for comments in preparing the manuscript and for providing primers for SNP analysis. We thank Dr. Fakher Chouaiak for technical help. This research was funded by Ministère de L’Enseignement supérieur, de la Recherche Scientifique et de la Technologie, Tunisia and the European Commission FP6 Integrated Project EUROHEAR, LSHG-CT-20054–512063. These studies were partially supported by the Fonds national de la recherche scientifique (F.N.R.S.) to M.V., a “Maître de recherches du F.N.R.S.”; M.A. is a post-doctoral researcher of the F.N.R.S.</p>" ]

[ "<fig id=\"f1\" fig-type=\"figure\" position=\"float\"><label>Figure 1</label><caption><p>Pedigree, haplotype and statistical data for a Tunisian family segregating Usher type 2 syndrome. <bold>A-C:</bold> In pedigree, the square symbol indicates male, the circle symbol denotes female and black symbols represent affected individuals. Haplotypes for polymorphic markers in three candidate regions on Chromosome 2, 4, and 15 are shown. The disease-linked haplotype is indicated by black bar for markers listed while other haplotypes in gray and white. The critical linkage interval of each candidate region was indicated by box on haplotypes. Analysis of these markers allowed us to refine the boundary of the critical linkage intervals to 14.7 Mb, 25.5 Mb, and 55 Mb respectively. Among interesting candidate genes on chromosome 2 and 15 region two <italic>CERKL</italic> and <italic>MYO1E</italic> were selected for mutation screening. <bold>D-F:</bold> Multipoint lod scores for markers on three candidate regions on Chromosome 2, 4, and 15. Lod scores for the different markers studied were computed using Genehunter software. Maximum lod score of 1.765 was identified for the candidate regions on chromosome 4 between D4S2989 and D4S402 and chromosome 15 between <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/genome/sts/sts.cgi?uid=14202\">D15S978</ext-link> and D15S1036. A maximum lod score of 1.51 was found on chromosome 2 between rs155100 and rs1157595. The following abbreviation was used: Mega base (Mb).</p></caption></fig>", "<fig id=\"f2\" fig-type=\"figure\" position=\"float\"><label>Figure 2</label><caption><p>Serial audiograms of two affected members (BT188 and BT189) shown for the right (R) and left (L) ears separately. Pure tone air-conduction threshold (y-axis) is expressed in decibels (dB). The blue one represents the audiogram from 18-year-old BT188. Both the red and the green represent the audiograms for the patient BT189. The red audiogram was made when he was 24 whereas the second was made at the age of 28. BT188 was not available for audiometric test at the beginning of the study. Audiometric test showed a moderate sloping bilateral sensorineural HI in these two usher patients. The green and the red audiograms for the patient BT189 showed that there is no progression of hearing loss at an interval of four years.</p></caption></fig>", "<fig id=\"f3\" fig-type=\"figure\" position=\"float\"><label>Figure 3</label><caption><p>Visual field test results obtained on the right (RE) and the left eye (LE) of the two patients BT188 and BT189. <bold>A:</bold> Result of measuring the visual fields on BT189 at 28 years of age. <bold>B:</bold> Result of measuring the visual fields on BT188 at the age of 18 years. A series of random lights of different intensities are flashed in the peripheral field of vision of both patients. When they perceive the computer-generated light suddenly appearing in their field of view they press a button to indicate their responses, then we see this spot (Dot see). If the patient is unable to see the light in an appropriate portion of his field of view, then we see on the computer a spot (Dot don’t see) indicating vision loss. Visual field loss was more severe in the older brother BT189. But in both patients, the nasal and temporal fields were not preserved, and only the central field was maintained.</p></caption></fig>", "<fig id=\"f4\" fig-type=\"figure\" position=\"float\"><label>Figure 4</label><caption><p>Ganzfeld-Electroretinogram of the right and left eyes of the patient BT189. The following abbreviations were used: Left eye (LE), right eye (RE), electroretinogram (ERG), visual-evoked potentials (VEP), positive peak (P1 and P2), negative peak (N1 and N2). The ERG and the VEP tests the function of the visual pathway from the retina (ERG) to the occipital cortex (VEP). These tests were conducted by placing a standard ERG device attached to the skin on 2 mm above the orbit. VEPs were recorded simultaneously from electrode attached to the occipital scalp 2 mm above the region on the midsaggital plane. An electrode placed on the fore head provided a ground. The results can be directly related to the part of a visual field that might be defective. This is based on the anatomical relationship of the retinal images and the visual field. After dark adaptation for 30 min, the doctor will place anesthetic drops in the patient's eye and place a contact lens on the surface of the eye. Once the contact lens is in place, a series of blue, red and white lights will be shown to the patient. The VEP is an evoked electrophysiological potential that can be extracted, using signal averaging, from the electroencephalographic activity recorded at the scalp. Both ERG and VEP were differentially amplified band pass filtred (0,1,30 Hz), recorded over 300 ms epochs, and signal average. 2 trials were given. The visual evoked potential to flash stimulation consists of a series of negative and positive waves. The earliest detectable response has a peak latency of approximately 30ms post-stimulus. For the flash VEP, the most robust components are the N2 and P2 peaks. Measurements of P2 amplitude should be made from the positive P2 peak at around 207.3 ms. The ERG recorded in BT189 showed an absence of responses. While the VEP showed a normal responses in both eyes. These traces confirm the evidence of a significant bilateral global retinal degeneration. Only cone flicker responses of less than 15% of the normal mean were recordable under photopic conditions while all other responses were below noise level, a typical finding for patients with retinitis pigmentosa.</p></caption></fig>" ]

[ "<table-wrap id=\"t1\" position=\"float\"><label>Table 1</label><caption><title>Primer pairs used to sequence coding exons of <italic>CERKL</italic>.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"36\" span=\"1\"/><col width=\"216\" span=\"1\"/><col width=\"194\" span=\"1\"/><col width=\"48\" span=\"1\"/><thead><tr><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Exon</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Forward sequence</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Reverse sequence</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>PCR size (bp)</bold></th></tr></thead><tbody><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">GTGCTGGACTGGGTCAGG<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CAAAAGCTCGTGGGTGTAGG<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">490<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CCCCAGTGTCTGTTGTTCCT<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TCAAGGAAACTGGGCTGATT<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">356<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">3<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TGTGTCATTTTAAAGGGAAAGAAA<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TTCCCAAGTTTGCATTAAGGA<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">295<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TTTGCCAGAACAAGTTAAAAAGTG<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TGAACAAGATAGAGCCAAAGTAA<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">273<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">5<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CCCATTGGTTAACTTGTCTGTG<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CACATCAGTCCAACACTTTAGCA<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">295<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">6<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">GGTACATGTGAGCAGTTATGCAC<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TAGTGGGGATGCCAGAAGTC<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">399<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">7<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">AAAAGCAAATGTTAGTTTGAACACA<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">AGAGACAAAGAACCTGCCTTTT<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">249<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">8,9<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">GCTCTCTTATGTTTGCTGTTTTGA<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TCTGATCAATTGTTTGTCAGAATG<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">461<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">10,11<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">GCGCGCGTTATCTGTTTTAT<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CAGTTAATTGGATACCCTGGAAA<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">352<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">12<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CATGGTGATTTATCTATCTTGTCCA<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CAATTCTTGCAGCATCTTTTTC<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">299<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">13<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">CTCAAAGCTATTAAAATGTCAGCA<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">AACCAACTGCCTGCTTTGAT<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">400<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">14</td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">TATTTGGCATTGGCATTGTG</td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\">GGTTTAAAGCATGGCCACAT</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">222</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t2\" position=\"float\"><label>Table 2</label><caption><title>Primer pairs used to sequence coding exons of <italic>MYO1E</italic>.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"29\" span=\"1\"/><col width=\"194\" span=\"1\"/><col width=\"182\" span=\"1\"/><col width=\"66\" span=\"1\"/><thead><tr><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Exon</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Forward sequence</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Reverse sequence</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>PCR size (bp)</bold></th></tr></thead><tbody><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TACGGTTTCCCTGAGGAGTG<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CGCGTCCACCTTCTCCAC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">588<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCTGCACTGCTCTTTCTGCT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AACTCCTGCCTTAGCCTTCC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">395<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">3<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TTGTGAATTCTTGATAACATCTGG<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCAAGAAAAACCATGTCTGCAT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">248<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TAGTGCACGATTCGTTTCCA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CCTGCTTGCTACTCAGACACA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">355<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">5<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GTTTTGTGTGATGGGGGAGA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CCAGTGTCTTTTCTGTGGAAGA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">271<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">6<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GGCCCCTCACCTTAATGC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TATGTGAAAGGCTCCCATTT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">299<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">7<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AGGATGCAGGAGTGACTTCG<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GAAAGAGGCGGACATTTCA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">320<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">8<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TGTGACTGCACAACCCAATC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TGCCACAGAGGACATGTAGA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">440<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">9<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CCCGTGATTGTGCCTTCTAT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CGCACCCAGCCTACTAGTTT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">396<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">10,11<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GTCCTCTGTTTCCTGCAAGC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TTGTTTTTGCATTGCCTAGA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">292<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">12<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AAGGAGTTCACTGCCATGCT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GCCACAATGGCATATGGTTT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">684<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">13<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TGTTCCTTTCCTGTTACCTCTT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCAGAGTTGTCACTTTGCCTGT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">359<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">14<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GCCATGACAGCTTTGGTTTT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AGGAACACACCACCACACC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">299<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">15<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CCCTTCACCCCATCCTCTA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CAGGGGTGCAGTTCCTTACT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">243<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">16<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TGCTTAACGAGCAAATTGTCA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AAGACATGTGCGGACAACTG<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">349<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">17<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCCCTACAGCTTGGAACTGG<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GTACGCTTGAAGTGGGTGAA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">286<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">18<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TTCGAACGCTGGTAAACAGA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CAACATTGATGGCATGAAGC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">398<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">19,20<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCCCGTGTGTGTCATTGTCT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AACGAACACATTCTGATTTGG<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">708<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">21<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CCTCCGAAAGTACTGGGATT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCCTCCTGGCTGTTTGGAAC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">304<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">22<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCATTGTTGTTGGTTTTGTTTG<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GCGATCAAGACCCCTTTTTA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">366<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">23<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">CCCTGCTCCTGGTGTAGATT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">GTGCACATGTTTGCAGCATT<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">374<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">24<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCCACCTGAGAGCTGGAATC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TCCAGATTTAGTGGTCCCAGA<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">250<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">25<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">TTCAAATGCGGAAATTGAGAC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">ATGATGGAGATGGAGCTTGC<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">383<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">26</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AAGGATGGAGCTGGATTTGA</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">AGCAATGTGACTGCATGCTC</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">347</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t3\" position=\"float\"><label>Table 3</label><caption><title>Two point LOD scores calculated for microsatellites bordering all described USH2 gene regions.</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"48\" span=\"1\"/><col width=\"48\" span=\"1\"/><col width=\"48\" span=\"1\"/><col width=\"48\" span=\"1\"/><col width=\"48\" span=\"1\"/><col width=\"48\" span=\"1\"/><col width=\"48\" span=\"1\"/><thead><tr><th rowspan=\"2\" valign=\"bottom\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>Gene</bold></th><th rowspan=\"2\" valign=\"bottom\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>Marker</bold></th><th valign=\"bottom\" colspan=\"5\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>Recombination fraction (q)</bold><hr/></th></tr><tr><th valign=\"bottom\" colspan=\"1\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>0</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>0.1</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>0.2</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>0.3</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>0.4</bold></th></tr></thead><tbody><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><italic>USH2A</italic><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D1S425<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-∞<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.423<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.096<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.01<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.001<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D1S2827<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-2.828<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.024<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.071<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D1S213<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-2.832<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.022<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.07<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.042<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.012<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><italic>VLGR1</italic><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D5S428<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-∞<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.165<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.073<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.034<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.009<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D5S618<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-∞<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.165<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.073<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.034<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.009<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D5S644<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-∞<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.251<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.117<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.051<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.013<hr/></td></tr><tr><td valign=\"bottom\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><italic>WHRN</italic><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D9S1677<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.328<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.206<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.109<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.046<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.013<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D9S1776<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-∞<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.536<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.193<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.067<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.014<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"/><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">D9S1682</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-∞</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.119</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.039</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.018</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">-0.005</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Two-point LOD scores for the different markers studied were computed using Genehunter software. Close linkage to the known USH genes on chromosomes 1, 5, and 9 was excluded with negative 2-point LOD scores.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"mv-v14-1719-f1\"/>", "<graphic xlink:href=\"mv-v14-1719-f2\"/>", "<graphic xlink:href=\"mv-v14-1719-f3\"/>", "<graphic xlink:href=\"mv-v14-1719-f4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Diabetic retinopathy is a progressive neurologic disease that is characterized by neuronal degeneration and extensive vascular changes. However, our knowledge of the mechanisms leading to neuronal cell loss and vascular dysfunction in diabetic retinopathy is still incomplete. We do know cells other than endothelial cells and pericytes are affected by hyperglycemia in diabetes [##REF##12540628##1##,##REF##9710447##2##]. Clinical studies [##REF##6858649##3##, ####REF##6383303##4##, ##REF##1936619##5##, ##REF##1390526##6##, ##REF##8740250##7##, ##REF##11688662##8####11688662##8##] have shown that hyperglycemia causes neural dysfunction in the retina before the onset of diabetic microvasculopathy.</p>", "<p>The development of retinal disease varies among patients with diabetes. Although hypertension may be an important contributor, the precise mechanism by which hypertension can exacerbate diabetic retinopathy remains to be established. Diabetic individuals frequently have concomitant retinopathy and nephropathy, and it has been suggested that similar mechanisms may be involved in these two long-term complications of diabetes [##REF##7489844##9##]. Inhibitors of cyclin-dependent kinases (Cdk) such as p27^Kip1, a negative cell cycle regulator, are involved in the development of diabetic nephropathy, with associated mesangial hypertrophy and extracellular matrix accumulation [##REF##9321907##10##, ####REF##9621281##11##, ##REF##10571777##12##, ##REF##10906152##13####10906152##13##]. In addition, genetic hypertension potentiates the cell-cycle abnormalities induced by renal hyperglycemia [##REF##11978652##14##]. These findings suggest that cell cycle regulators are altered by the diabetic milieu and that such alterations contribute to the pathogenesis of diabetic microvascular complications.</p>", "<p>The early phase of diabetic retinopathy involves microangiopathy characterized by a diffuse increase in vascular permeability and capillary basement membrane thickening [##REF##8603829##15##, ####REF##12032713##16##, ##REF##7607344##17##, ##REF##11508263##18####11508263##18##]. Fibronectin, a component of the basement membrane, is overexpressed in the retina of diabetic adults [##REF##8603829##15##]. Experimental studies have indicated that this accumulation of fibronectin in retinal tissue is simply an epiphenomenon of the diabetic state, but may be operative in sight-threatening diabetic retinopathy. Indeed, the downregulation of fibronectin production in galactose-fed rats partly prevented retinal basement membrane thickening and reduced pericyte and endothelial cell loss [##REF##12716757##19##].</p>", "<p>The developing and postnatal vertebrate retina contains neural progenitor cells that divide, generate neurospheres, and undergo neuronal and glial differentiation [##REF##10411978##20##, ####REF##10720333##21##, ##REF##10753510##22##, ##REF##10753656##23####10753656##23##]. These cells can be identified by their ability to proliferate based on the incorporation of bromodeoxyuridine (BrdU), and by the expression of progenitor markers such as nestin, membrane receptor tyrosine kinase, also designated vascular endothelial growth factor receptor-2 (Flk-1), and paired box gene 6 (Pax6) [##REF##1689217##24##, ####REF##8223275##25##, ##REF##7914735##26##, ##REF##8907167##27####8907167##27##]. p27^Kip1 has recently been implicated in the molecular mechanism that controls the decision of multipotent central nervous system progenitors to withdraw from the cell cycle and to maintain the differentiated state of the postmitotic cell [##REF##10694424##28##]. In the retina, p27^Kip1 is expressed in a pattern coincident with the onset of the differentiation of most retinal cell types, and in vitro the accumulation of p27^Kip1 in retinal cells correlates with cell cycle withdrawal and differentiation, thereby inhibiting progenitor cell proliferation [##REF##10694424##28##].</p>", "<p>These observations prompted us to search for potential progenitor cells in adult rat retina. We also wanted to assess the impact of hypertension and short-term diabetes on the number of retinal BrdU positive cells and to examine the relationship between these cells and the well known abnormalities associated with the early stages of diabetic retinopathy. We have identified a small population of potential progenitor cells characterized by BrdU positivity that colocalize with nestin, protein kinase C-alpha (PKC-α), Flk-1, and glial fibrillary acidic protein (GFAP) antigens. Furthermore, the higher number of 84 proliferating cells in the retina of adult spontaneously hypertensive rats compared with normotensive rats was 85 significantly reduced in diabetic SHR rats. These findings were associated with enhanced expression of p27^Kip1 and with the classic abnormalities of diabetic retina—i.e., fibronectin accumulation, increased expression of retinal vascular endothelial growth factor (VEGF), and blood-retinal barrier breakdown.</p>" ]

[ "<title>Methods</title>", "<p>The experiments were done in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. The protocol for this study complies with the guidelines of the Brazilian College for Animal Experimentation (COBEA) and was approved by the institutional Committee for Ethics in Animal Research (CEEA/IB/UNICAMP, protocol no. 1408–1). This study used spontaneously hypertensive rats (SHR) and their genetically normotensive counterparts, Wistar Kyoto rats (WKY), which were derived from animals supplied by Taconic (Germantown, NY) and bred in our animal facility. The rats were housed at a constant temperature (24 °C) on a 12 h:12 h light-dark cycle with access to food and tap water ad libitum. The rats were weighed on the day before the induction of diabetes and before they were euthanized.</p>", "<p>Experimental diabetes was induced in 4- and 12-week-old hypertensive male SHR and WKY rats, who had been allowed to fast overnight. Each animal was given a single intravenous injection of 50 mg/kg streptozotocin (STZ; Sigma, St. Louis, MO), dissolved in sodium citrate buffer, pH 4.5, within 5 min of its preparation. Control rats received only vehicle (500 μl citrate buffer). Blood glucose levels were measured using an enzymatic colorimetric GOD-PAP assay (Merck, Darmstadt, Germany) 72 h after the injection of STZ or citrate buffer and on the day before euthanasia. Values ≥15 mmol/l were indicative of diabetes. Systolic blood pressure was obtained by tail-cuff plethysmography (Physiograph^® MK-III-S; Narco Bio-System, Houston, TX) in unanesthetized, warmed rats. Three to five determinations were made per rat. Readings were taken on the day before the induction of diabetes and before rats were euthanized. Rats were habituated with the procedure for measurement before taking the blood pressure readings.</p>", "<p>Fifteen days after the induction of diabetes, retinal capillary permeability was assessed, and the rats were euthanized with an overdose (twice the anesthesic dose) of anesthesia. The eye globes and retinas were collected for immunohistochemistry, immunofluorescence, and western blot analyses. We have previously described early inflammatory and oxidative changes in retina from diabetic SHR rats after 20 days of diabetes mellitus [##REF##17886037##29##,##REF##17612969##30##]. The aim of the present work was to examine the alterations in cellular proliferation and apoptosis that precede the changes we have mentioned, and to investigate the relationship between these changes and the well established abnormalities associated with the early stages of diabetic retinopathy.</p>", "<title>Detection of proliferating cells in vivo</title>", "<p>Retinal cell replication was evaluated based on the incorporation of BrdU (Calbiochem, La Jolla, CA), a thymidine analog that incorporates into DNA in the S phase [##REF##11978652##14##,##REF##3528077##31##]. Briefly, the rats were given an intraperitoneal injection of 100 mg/kg BrdU, dissolved in saline, 1 h before they were euthanized. An equal number of control rats received saline alone. The rats were euthanized with an intraperitoneal injection of 30 mg/kg sodium pentobarbital (Hipnol, Fontoveter, Itapira, SP, Brazil). Their eyes were enucleated, and 130 a portion of the gastrointestinal tract (positive control for BrdU staining) were excised. The eye globe and the GI tract were 131 fixed in buffered formalin, and embedded in paraffin-embedded, or placed in OCT 132 cryoprotector (Tissue-Tech, Sakura, CA) and snap frozen in liquid nitrogen for subsequent sectioning in a cryostat. Next, 4 µm-thick consecutive sections were mounted on silane-coated slides for immunofluorescence.</p>", "<title>Immunohistochemistry for BrdU and p27^Kip1</title>", "<p>Detection of BrdU and p27^Kip1 was performed by dewaxing slides of eye sections, rehydrating them, and placing in 2 N HCl at 31 °C for 20 min and in 0.005% trypsin in phosphate-buffered saline (PBS) at 37 °C for 2 min for antigen retrieval. After this pretreatment, the slides were placed in 1% nonfat milk in PBS for 1 h to block nonspecific sites. The sections were then incubated with a 1:50 dilution of mouse antihuman antibodies for BrdU (Dako, Glostrup, Denmark) or a 1:100 dilution of mouse monoclonal anti-p27^Kip1 antibody (Transduction Laboratories, Lexington, KY) for 1.5 h at room temperature. The slides of eye sections were washed in, a biotinylated secondary antimouse IgG antibody (Vector, Burlingame, CA) was applied and allowed to sit for 1 h. Endogenous peroxidase was blocked by incubating the slides in 3% H₂O₂ for 5 min. To detect p27^Kip1, we performed microwave post-fixation of slides immersed in 0.01 M citrate buffer, pH 6.0 using a domestic oven (Panasonic Junior, Sao Paulo, SP, Brazil) operated at 700 W. The slides were incubated with an avidin-biotin complex (ABC) reagent (Vector) for 30 min followed by the addition of diaminobenzidine tetrahydrochloride (DAB; Sigma) as a substrate-chromogen solution. Next, the slides were counterstained with hematoxylin before they were dehydrated and mounted in Entellan (Merck, Darmstadt, Germany). Positive controls for BrdU staining consisted of sections of the gastrointestinal tract of each rat. Negative controls for the reaction consisted of omitting the primary antibody.</p>", "<p>Quantitative analysis of BrdU positivity was done by an observer unaware of the slide identification. The results were expressed as the total number of positive cells counted in eight random retinal sections from the right eye of each rat. For p27^Kip1, the results were expressed as a percentage of positive cells in the ganglion cell and inner nuclear layers, using the following scale: 0 (no positivity), 0.5 (up to 10% positivity), 1 (11%–25% positivity), 1.5 (26%–40% positivity), 2 (41%–53% positivity), 2.5 (54%–66% positivity), 3 (67%–80% positivity), and 4 (&gt;80% positivity) [##REF##17886037##29##]. The intervening distance between retinal sections was approximately 24 μm.</p>", "<title>Immunocolocalization by immunofluorescence</title>", "<p>The slides were fixed with acetone for 3 min at room temperature, after which they were washed with PBS, pH 7.4, and blocked in PBS containing 5% bovine serum albumin (BSA) for 2 h at room temperature. The sections were then incubated with the appropriate primary antibody to identify glial, endothelial, and neuronal cells (1:10 goat polyclonal anti-GFAP antibody; Santa Cruz Biochemical, Santa Cruz, CA; 1:100 mouse monoclonal anti-FLK-1 antibody; Santa Cruz; 1:10 mouse antinestin antibody; BD PharMingen™, Franklin Lakes, NJ; and 1:10 mouse monoclonal anti PKC-α antibody; Abcam, Inc., Cambridge, MA) for 5 h at room temperature. After they were washed with PBS, the sections were incubated with secondary antibody (donkey antigoat IgG-FITC or goat antimouse IgG-FITC) as appropriate for 1 h at room temperature. The slides were then washed with PBS and incubated with 1:10 sheep polyclonal anti-BrdU antibody (Abcam, Inc.) overnight at 4 °C, followed by another wash with PBS and incubation with rabbit polyclonal antibody to 1:1,000 sheep IgG-rhodamine (1,5 mg of IgG; Abcam) for 1 h at room temperature. Finally, the sections were rinsed with PBS then coverslipped with Vectashield antifading medium containing 4',6'-diamino-2-phenylindole (DAPI) to stain nuclei (Vector). The sections were examined under a confocal laser scanning microscope (CLSM, LSM510 Zeiss, Jena, Germany) with appropriate emission filters for FITC and rhodamine. Digital images were captured using specific software (LSM; Zeiss). The negative controls consisted of omitting the primary antibody.</p>", "<title>Terminal deoxynucleotidyl transferase-mediated nick-end labeling</title>", "<p>To determine whether retinal cell apoptosis was influenced by age, diabetes or rat strain, the terminal deoxynucleotidyl transferase (TdT)-mediated dUTP-biotin nick-end labeling (TUNEL) method for detecting DNA breaks in situ [##REF##1400587##32##] was applied to retinal tissue from the same rats used for BrdU immunostaining. [##REF##11978652##14##,##REF##1400587##32##]. Sections (4 μm thick) were deparaffinized, boiled in 0.01 M citric acid, pH 6.0, and incubated with 9.3 µg/ml proteinase K (Boehringer Mannheim, Indianapolis, IN) for 15 min at room temperature. Endogenous peroxidase was quenched, and sections were rinsed in One-Phor-All buffer (Amersham Pharmacia Biotech, Piscataway, NJ), pH 7.2, and incubated with diluted 1:50 TdT (Amersham Pharmacia) and diluted 1:50 biotinylated-dUTP (Gibco, Grand Island, NY) in TdT buffer (100 mM TRIS, 1 mM dithiothreitol, 50% of glycerol, 0.1% sodium azide, 0.01%, Brij®35 [Polyoxyethyleneglycol dodecyl ether], pH 7.2) for 60 min at room temperature. Labeled nuclei were detected with ABC Vectastain (Vector) in PBS and DAB, chloride, and hydrogen peroxide and counterstained with hematoxylin. As a positive control, some slides were treated with 20 Kunitz units/ml DNase (Sigma). The quantitative analysis for TUNEL-positive cells was done by an observer with no knowledge of the studied groups. Results were expressed as the number of positive cells per retinal section in at least eight random retinal sections from the right eye of each animal. The distance between sections was approximately 24 μm.</p>", "<title>Isolation of retina</title>", "<p>The eyes were enucleated, and the retinas were dissected and isolated from the retinal pigmented epithelium. The retinas were lysed directly on ice in 300 µl of a buffer containing 2% SDS and 60 mM Tris-HCl (pH 6.8) supplemented with a cocktail of protease inhibitors (Complete^®, containing irreversible and reversible protease inhibitors such as antipain, aprotinin, bestatin, chymostatin, Ethylenediaminetetraacetic acid (EDTA), leupeptin, pepstatin and phosphoramidon; Boehringer-Mannheim). The lysed retinas were centrifuged. The supernatants were transferred to new tubes, and the protein concentrations were measured by the Bradford method [##REF##942051##33##], using BSA as the standard.</p>", "<title>Western blotting for fibronectin, GFAP, and VEGF</title>", "<p>For western blotting, 30 µg (for GFAP), 50 µg (for fibronectin), and 100 µg (for VEGF) of total retinal protein in 5% glycerol/0.03% bromophenol blue/10 mM dithiothreitol was loaded onto 8%–15% SDS polyacrylamide gels. Molecular weight markers (Rainbow; Amersham Pharmacia) were used as standards. After electrophoresis, proteins were transferred to nitrocellulose membranes (Bio-Rad, Hercules, CA) in transfer buffer consisting of 50 mM Tris-HCl, pH 7.0, 380 mM glycine, 0.1% SDS, and 20% methanol. Nonspecific binding was blocked by incubating the membranes overnight at 4 °C in 5% nonfat milk, or at 4 °C in PBS containing 1% gelatin with 0.1% Tween 20 (Fisher Chemicals, Fairlawn, NJ) for VEGF. The membranes were then incubated for 1 h at room temperature with primary antibodies: 1:1,000 goat monoclonal anti-fibronectin antibody (Calbiochem, San Diego, CA), 1:100 polyclonal goat-GFAP antibody (Santa Cruz, Santa Cruz, CA), and 1:5000 rabbit polyclonal IgG anti-VEGF antibody (Santa Cruz). The blots were subsequently washed in Tris-buffered saline with Tween and incubated with horseradish peroxidase (HRP)-conjugated secondary antibody (New England Biolabs, Beverly, MA). Immunoreactive bands were visualized with the enhanced chemiluminescence method (Super Signal CL-HRP Substrate System; Pierce, Rockford, IL). Exposed films were scanned with a densitometer (Bio-Rad) and analyzed quantitatively with Multi-Analyst Macintosh Software for Image Analysis Systems Bio Rad (Hercules, CA). Western blots were repeated 3–5 times; qualitatively similar results were obtained each time. Equal loading and transfer was ensured by reprobing the membranes for β-actin.</p>", "<title>Quantitative measurement of retinal capillary permeability using Evans blue dye</title>", "<p>Briefly [##REF##11222542##34##], the rats (SHR and WKY) were anesthetized with an intraperitoneal injection of 30 mg/kg sodium pentobarbital, and the left femoral vein and right femoral artery were cannulated with 0.28 mm internal diameter polyethylene tubing (Becton Dickinson, Sparks, MD) filled with heparinized saline (400 units heparin/ml saline). Next, 45 mg/kg Evans blue dye was injected through the femoral vein over a period of 10 s. The blue color of the rats confirmed the uptake and distribution of the dye. Two minutes after the injection of the dye, 0.2 ml of blood was drawn from the femoral artery to determine the initial Evans blue plasma concentration. Subsequently, 0.1 ml of blood was drawn from the femoral artery at 15 min intervals for up to 2 h postinjection to obtain the time-averaged Evans blue plasma concentration. Exactly 2 h after infusion, needle was inserted into the left ventricle and the rats were perfused for 2 minutes at 37°C with 0.05 M, pH 3.5, citrate-buffered paraformaldehyde at a physiological pressure of 120 mm Hg.</p>", "<p>Immediately after perfusion, both eyes were enucleated and bisected at the equator. These animals were euthanized with an injection of overdose of penthabarbitol. The retinas were carefully dissected under an operating microscope and weighed. They were then dried in a Speed-Vac for approximately 5 h. The Evans blue dye was extracted from the tissue by incubating each retina in 120 µl formamide (Sigma) for 18 h at 70 °C followed by centrifugation at 27,000x g for 45 min in a Beckman TLX rotor at 4 °C. The absorbance of 60 μl aliquots of the supernatant was measured spectrophotometrically in triplicate at A_{620 nm} at 5 s intervals. The absorbances were corrected for background readings, and the dye content of the extracts was calculated from a standard curve of Evans blue in formamide. Blood-retinal barrier breakdown was calculated using the following equation:</p>", "<p>(µg of Evans blue)( time-averaged µg of Evans blue)/(retinal wet weight in g)(µl of plasma x h of circulation)</p>", "<p>The results were expressed as µl of plasma x retinal wet wt⁻¹ (g) x h⁻¹.</p>", "<title>Statistical analysis</title>", "<p>The results were expressed as the means±standard deviation or standard error of measurement as indicated. Comparisons between groups were done using ANOVA followed by Fisher’s protected least-significant difference test. All comparisons were done using the StatView statistics software for Macintosh, with a value of p&lt;0.05 indicating significance.</p>" ]

[ "<title>Results</title>", "<p>SHR rats are normotensive when they are 4 weeks old, but are fully hypertensive by the time they are 12 weeks old. We used both age groups because they allowed us to assess the contribution of the genetics of hypertension (4-week-old rats) and the influence of genetics plus hypertension per se (12-week-old rats) on retinal abnormalities. As previously demonstrated by our group and others, the systolic blood pressure (SBP) in 4-week-old SHR was significantly higher than in WKY, although still within the normal range (##TAB##0##Table 1##). The SBP of 12-week-old SHR was significantly higher (p=0.0001) than that of age-matched WKY (##TAB##0##Table 1##). Bodyweight was lower, and blood glucose levels were higher in diabetic rats than in their respective controls at both ages (p=0.0001; ##TAB##0##Table 1##).</p>", "<title>Cells with progenitor characteristics in adult rat retina</title>", "<p>BrdU-positive-stained cells were rare but clearly identifiable in retinal sections of adult rats (##FIG##0##Figure 1A,B##). The proliferating retinal cells of adult rats were characterized further by using specific cell markers, including GFAP, the tyrosine-kinase receptor Flk1, the intermediate filament protein nestin, and PKC-α. BrdU and GFAP colocalized in cells of the ganglion cell layer, suggesting that they may represent glial cells in proliferation (##FIG##1##Figure 2A,B##). Other BrdU-positive cells were also colabeled with nestin (##FIG##1##Figure 2C##), an intermediate filament structural protein expressed in primitive neural tissue. The detection of this protein in the inner nuclear and ganglion cell layers suggested the presence of neural progenitor cells. ##FIG##1##Figure 2D## shows a PKC-α-positive cell in the outer nuclear layer that was also positive for BrdU. In this case, cellular maturation involved a bipolar–amacrine neural cell. The elongated BrdU-positive cells in the inner nuclear layer reacted strongly with antibody against Flk-1 (##FIG##1##Figure 2E##), a surface receptor protein characteristic of endothelial cells in cell cycle progression. These findings indicate that adult rat retina contains different populations of replicating cells with characteristics of glial or endothelial cells. The detection of cells that coexpressed BrdU and nestin or PKC-α suggests that subpopulations of replicating cells may differentiate into neural cell subtypes.</p>", "<title>Diabetes decreases the number of progenitor cells only in retina from 12-week-old SHR</title>", "<p>The total number of BrdU-positive retinal cells in 12-week-old rats was significantly higher in hypertensive SHR than in normotensive WKY (9.5±4.0 versus 1.2±0.4 positive retinal cells; p=0.01). After 15 days of diabetes mellitus, there was a significant decrease in the number of these cells only in SHR (0.2±0.2 positive retinal cells; p=0.007; ##FIG##2##Figure 3##). In contrast, in 4-week-old SHR and WKY rats, the number of proliferating cells was similar in all groups (4.5±0.5 for control WKY versus 2.8±0.4 for diabetic WKY versus 3.4±0.4 for control SHR versus 4.0±0.5 for diabetic SHR BrdU-positive cells (p&gt;0.05).</p>", "<title>TUNEL staining is unaltered by rat strain, animal age, or short-term diabetes</title>", "<p>The number of TUNEL-positive retinal cells in 4-week-old rats (SHR and WKY) was 3.2±2.8 positive cells per retinal section in control WKY rats. This number did not differ significantly from the 2.0±2.7 positive cells per retinal section seen in diabetic WKY rats or the 2.0±2.0 and 0.9±0.8 positive cells per retinal section seen in control and diabetic SHR, respectively. The rates of apoptosis in retinal tissue in 12-week-old rats were also similar among the groups (1.2±1.3 for control WKY versus 1.9±2.7 for diabetic WKY versus 1.6±1.7 control SHR versus 1.1±0.1 for diabetic positive cells/retinal section, p&lt;0.05). Hence, in eight retinal sections, the number of TUNEL-positive cells was 9.6 for control WKY, 15.2 for diabetic WKY, 12.8 for control SHR, and 8.8 for diabetic SHR.</p>", "<p>Since diabetes and hypertension influenced the number of BrdU-positive cells only in 12-week-old rats, all subsequent experiments were done only 12-week-old rats.</p>", "<p>To investigate a potential role for cell cycle regulatory proteins in the reduction in BrdU-positive retinal cells in 12-week-old diabetic SHR, we examined the expression of p27^Kip1. Immunohistochemistry revealed a heterogeneous distribution for p27^Kip1 in the retinal cell layers, with greater positivity in the ganglion cell layer (##FIG##3##Figure 4A##). There was a significant increase in p27^Kip1 protein positivity in this layer in diabetic rats (1.5±1 in control WKY versus 2.5±0.5 in diabetic WKY, and 1.5±1.5 in control SHR versus 2.5±1 in diabetic SHR positive score, p=0.05; ##FIG##3##Figure 4B##). Interestingly, there was a significant increase in p27^Kip1 protein in the inner nuclear layer only in diabetic SHR rats (0.7±0.6 versus 1.6±0.4 positive score for control versus diabetic SHR, p=0.02; ##FIG##3##Figure 4C##). The increase in p27^Kip1 expression in diabetic WKY, was not significant (0.9±0.5 and 1.6±0.5 positive score for normal and diabetic WKY; ##FIG##3##Figure 4C##). The accumulation of p27^Kip1 in retinal tissue is accompanied by increased fibronectin expression in diabetic hypertensive rats.</p>", "<p>To identify factors potentially involved in the pathogenesis of diabetic retinopathy, we examined the retinal expression of fibronectin, an extracellular matrix component associated with functional properties of the inner blood-retinal barrier. Western blot analysis of total retinal lysates revealed a significant increase in fibronectin expression in diabetic SHR compared with control SHR (p=0.04) or diabetic (p=0.03) and control WKY (p=0.009; ##FIG##4##Figure 5##). The enhanced expression of p27^Kip1 and fibronectin reflected a generalized response to hyperglycemia in target organs such as the kidney. This is the first demonstration of an association between increased expression of p27^Kip1 and fibronectin accumulation in retinal tissue, and could contribute to thickening of the retinal capillary basement membrane seen in diabetic retina.</p>", "<title>The expression of retinal GFAP is unaltered in SHR or WKY rats</title>", "<p>The retinal expression of GFAP in total retinal lysates, as evaluated by western blotting, was used as an indicator of glial cell reactivity. This expression did not differ between 345 control SHR (5.2±1.4 arbitrary units ) and WKY rats (4.5±0.5 arbitrary units), nor was it significantly altered by 15 days of diabetes mellitus (4.5±0.5 for diabetic WKY and 4.4±0.8 arbitrary units for diabetic SHR; ##FIG##5##Figure 6##)</p>", "<title>VEGF expression is enhanced in diabetic SHR after 15 days of diabetes mellitus</title>", "<p>There was a significant increase in the retinal expression of VEGF only in diabetic SHR (0.9±0.1 arbitrary units versus 1.1±0.1 arbitrary units for diabetic WKY and 0.8±0.1 arbitrary units for control SHR and 1.3±0.2 arbitrary units for diabetic SHR, p=0.02) (##FIG##6##Figure 7##).</p>", "<title>Blood-retinal barrier breakdown occurs only in diabetic hypertensive rats</title>", "<p>The extent of blood-retinal barrier breakdown, estimated by the extravasation of Evans blue dye, was similar in control and diabetic WKY rats (11.4±1.5 for control WKY and 10.3±1.3 μg plasma x g retinal wet wt^-1 x h^-1 for diabetic WKY; n=5 rats). In contrast, there was a significant increase in retinal capillary permeability in diabetic SHR compared with control SHR (8.6±0.7 for control SHR and 17.4±4.6 for diabetic SHR μg plasma x retinal wet wt^-1xh^-1; p=0.01, n=5 rats; ##FIG##7##Figure 8##).</p>" ]

[ "<title>Discussion</title>", "<p>The results of this study show that the adult rat retina contains a small number of BrdU-positive cells that also express glial, neural, and endothelial progenitor cell markers. The number of these cells was higher in hypertensive rats but was markedly reduced by the concomitant presence of hypertension and diabetes. The greater expression of the Cdk inhibitor, p27^Kip1, probably accounts for the reduction in the number of proliferating cells in diabetic retina. Additionally, diabetic hypertensive rats with a reduction in the number of BrdU-positive cells and enhanced p27^Kip1 expression also displayed classic characteristics of early diabetic retinal disease: enhanced fibronectin and VEGF expression and greater blood-retinal barrier breakdown. These observations—i.e., cell cycle withdrawal and enhanced production of extracellular matrix and VEGF expression—could provide a basis for a new understanding of the mechanisms involved in the pathogenesis of diabetic retinopathy.</p>", "<p>It is known that SHR rats display innate abnormal endothelial function associated with eNOS compensatory activity and increased nitric oxide production [##REF##9622137##35##]. Evidence is accumulating that implicates oxygen free radical formation and N-methyl-D-aspartate (NMDA)-receptor-mediated toxicity in the pathophysiology of ischemic retinal injury [##REF##2923566##36##,##REF##8194364##37##]. These two mechanisms are linked by nitric oxide. NMDA-receptor activation generates nitric oxide, which reacts with superoxide to form toxic species such as peroxynitrite. Neuropeptide Y (NPY) is a 36 amino acid peptide widely present in the central nervous system including the retina. NPY was recently found to stimulate retinal neural cell proliferation mediated through nitric oxide-cyclic guanosine monophosphate (GMP) and extracellular signal-regulated kinases (ERK) 1/2 pathways [##REF##18331583##38##]. These results suggest that the increased retinal cell proliferation observed in SHR rats may be due to activation of NPY through nitric oxide-cyclic GMP and ERK 1/2 pathways. Previous data from other cell types of SHR also demonstrated that mesangial cells have a higher proliferation rate in vitro than control WKY mesangial cells [##REF##9263994##39##], suggesting that the genetics of hypertension may contribute to this phenotype.</p>", "<p>The changes described for diabetic hypertensive SHR were not associated with the toxicity of streptozotocin for two reasons: 1) they occurred in the retina of diabetic hypertensive SHR but not in normotensive diabetic WKY rats; and 2) they were absent in diabetic WKY and SHR rats rendered diabetic at four weeks of age. These changes are also not attributable to differences in metabolic control, because the blood glucose levels were similar in both diabetic groups.</p>", "<p>The presence of neuronal progenitor cells in human retina has recently been demonstrated. It was suggested that these cells may potentially replace neurons and photoreceptors [##REF##15615756##40##]. The BrdU-positive retinal cells seen in the present study were not completely characterized. However, their proliferative capacity and their colocalization with antigens of neural, glial, and endothelial origin suggest multipotent properties—i.e., they may represent progenitor cells.</p>", "<p>Not all dividing cells in the retina are progenitors. Numerous studies have sought to identify neural progenitor cells. The intermediate filament protein nestin has a widespread, early expression in the developing retina and central nervous system and is one of the best-characterized protein markers for immature neural cells [##REF##10842089##41##,##REF##15839245##42##]. The coexpression of nestin with other developmental markers, such as the incorporation of BrdU (such as that seen in the present study) strongly suggests that the cells involved are immature. In addition to nestin, we also observed immunostaining for PKC-α (a bipolar–amacrine cell marker) in BrdU-positive cells in the retina. The coexpression of BrdU and nestin or BrdU and PKC-α was also seen.</p>", "<p>The cyclin kinase inhibitor p27^Kip1 may be involved in reducing the number of proliferating retinal cells. Levine et al. [##REF##10694424##28##] suggested that p27^Kip1 is part of the molecular mechanism that controls the decision of multipotency central nervous system progenitor cells to withdraw from the cell cycle. These authors also proposed that postmitotic Müller glia have a novel, intrinsic requirement for p27^Kip1 to maintain their differentiated state. The heterogeneous distribution of p27^{Kip 1} seen in retinal tissue, with greater expression in the inner nuclear layer only in diabetic SHR, may contribute to the different number of replicating cells in this group. Further clarification of the other mechanisms involved in what prompts a replicating retinal cell to withdraw or continue in the cycle when in the presence of diabetes and hypertension is beyond the scope of this study and requires a new experimental design.</p>", "<p>In the present study, we did not detected an enhanced apoptotic rate in retina from normotensive diabetic WKY and hypertensive diabetic SHR rats, as reported by others. Thus, for example, Gastinger et al. [##REF##16799061##43##] reported a significant increase in the apoptotic rate in retina from diabetic rats, based on the number of TUNEL-positive nuclei seen in whole-mounted retinas from Sprague-Dawley rats after two weeks of streptozotocin-induced diabetes. The discrepancy between our findings and the latter study may be related to differences in rat strain and in the use of whole-mounted retina compared to retinal cross-sections. In our study, the retinas of diabetic=WKY and SHR rats after 12 weeks of diabetes showed a significant increase in the number of TUNEL-positive cells in the diabetic groups, particularly in diabetic SHR rats (p=0,0003; unpublished).</p>", "<p>Increased extracellular matrix protein production leading to structural abnormalities is a hallmark of diabetic microangiopathy and has been demonstrated in all target organs of diabetic complications, including retina, kidney, and heart [##REF##11114100##44##,##REF##8799700##45##]. Enhanced extracellular matrix protein synthesis is instrumental in thickening of the basement membrane [##REF##12032713##16##,##REF##10871206##46##]. Fibronectin is a major extracellular matrix component but its overproduction may decrease the motility and replication of many cells [##REF##10871206##46##]. As revealed in the present study, retinal tissue from diabetic WKY rats and nondiabetic SHR showed enhanced fibronectin expression in total retinal lysates, although this expression was not significantly different from that in control WKY rats.</p>", "<p>The accumulation of fibronectin in retinal tissue is a classic finding encountered early in the pathogenesis of diabetic retinal disease and in the microcirculation in models of hypertension [##REF##9158646##47##, ####REF##16025600##48##, ##REF##12708883##49####12708883##49##]. Several studies have described the accumulation of extracellular matrix in target organs of hypertensive rats. In stroke-prone SHR there is enhanced accumulation of extracellular matrix proteins in the cerebral vasculature [##REF##9158646##47##,##REF##16025600##48##]. Agabiti-Rosei demonstrated that structural and functional changes in the microcirculation during arterial hypertension, e.g., remodeling of the extracellular matrix and accumulation of collagen and fibronectin, were associated with several neurohumoral and hormonal factors [##REF##12708883##49##]. Hence, the similar extracellular protein accumulation seen here between diabetic WKY rats and nondiabetic SHR may be the result of either diabetes or hypertension alone.</p>", "<p>The induction of diabetes in SHR leads to a reduction in renal cell replication with concomitant overexpression of p27^Kip1 [##REF##11978652##14##]. p27^Kip1-deficient mice do not develop the classic features of diabetic nephropathy such as renal hypertrophy, glomerular expression of fibronectin, and albuminuria, all of which are marked in wild-type mice [##REF##12595506##50##]. This finding suggests that modulation of p27^Kip1 function may ameliorate diabetic nephropathy. The effect of a reduction in p27^Kip1 expression on diabetic retinopathy remains to be elucidated.</p>", "<p>In conclusion, the combination of genetic hypertension and experimental diabetes markedly reduced retinal cell proliferation. This reduction was associated with enhanced p27^Kip1, fibronectin, and VEGF retinal expressions and greater blood-retinal barrier breakdown. Additional studies are required to clarify the mechanisms by which these cellular changes contribute to the structural abnormalities associated with the early pathogenesis of diabetic retinopathy.</p>" ]

[]

[ "<p>This is an open-access article distributed under the terms of the\n Creative Commons Attribution License, which permits unrestricted use,\n distribution, and reproduction in any medium, provided the original\n work is properly cited.</p>", "<title>Purpose</title>", "<p>Hyperglycemia and hypertension contribute to the development of diabetic retinopathy, and this may involve alterations in the normal retinal cell cycle. In this work, we examined the influence of diabetes and hypertension on retinal cell replication in vivo and the relationship between these changes and several early markers of diabetic retinopathy.</p>", "<title>Methods</title>", "<p>Diabetes was induced with streptozotocin in 4- and 12-week-old spontaneously hypertensive rats (SHR) and their Wistar Kyoto (WKY) controls. The rats were killed 15 days later. Retinal cells stained with bromodeoxyuridine (BrdU) were seen in rats of both ages.</p>", "<title>Results</title>", "<p>In 12-week-old rats, the number of BrdU-positive retinal cells was higher in SHR than in WKY rats. After 15 days of diabetes mellitus, there was a marked reduction in cell replication only in diabetic SHR (p=0.007). The BrdU-positive cells expressed neural, glial, or vascular progenitor markers. There was greater expression of p27^Kip1 in the ganglion cell layer of both diabetic groups (p=0.05), whereas in the inner nuclear layer there was enhanced expression only in diabetic SHR (p=0.02). There was a marked increase in the retinal expression of fibronectin (p=0.04) and vascular endothelial growth factor (p=0.02) in diabetic SHR that was accompanied by blood-retinal barrier breakdown (p=0.01).</p>", "<title>Discussion</title>", "<p>Concomitant diabetes and hypertension attenuated the proliferation of retinal cells, and it is associated with an increase in p27^Kip1 expression, fibronectin accumulation, and blood-retinal barrier breakdown. The replicative retinal cells displayed characteristics of progenitor cells.</p>" ]

[]

[ "<title>Acknowledgments</title>", "<p>We thank Sérgio Magalhães, Elisa B. M. Peixoto, and Flávia F. Mesquita for technical assistance and Stephen Hyslop for editing the manuscript. This work was supported by Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP; grant nos. 04/00455–9 and 05/58189–5).</p>" ]

[ "<fig id=\"f1\" fig-type=\"figure\" position=\"float\"><label>Figure 1</label><caption><p>Immunohystochemistry for BrDU in retinal slides from studied rats. The BrDU positive cells, although rare, are localized in ganglion cell layer <bold>(A)</bold> and in inner nuclear layer <bold>(B).</bold> The scale bar represents 50 µm.</p></caption></fig>", "<fig id=\"f2\" fig-type=\"figure\" position=\"float\"><label>Figure 2</label><caption><p>Immunofuorescence assay for double labeling of BrDU positive cells in retinal sections against glial, neuron and endothelial markers. In panel <bold>A</bold>, it is shown a retinal section showing two cells labeled for glial fibrillar acidic protein (GFAP) and BrdU in the ganglion cell layer. In <bold>B</bold>, there are the presence of two BrdU-positive cells in outer nuclear layer of the retina co-stained with GFAP antibody. In this slide (<bold>C</bold>), the BrDU positive cell is stained for nestin in the inner nuclear (long arrows) and ganglion cell (short arrows) layers of the retina. For identification of amacrine/bipolar origin of BrDu positive cell in the retina, an immunofluorescence for PKC- α antigen was performed. There is a BrDU positive cell that also stained for PKC- α in the outer nuclear (<bold>D</bold>). In panel E, elongated endothelial cells were identified expressing both Flk-1 and BrdU, localized in theinner nuclear layer of the retina</p></caption></fig>", "<fig id=\"f3\" fig-type=\"figure\" position=\"float\"><label>Figure 3</label><caption><p>Total number of BrdU-positive cells counted in eight random retinal sections from 12-week-old rats. See Methods for further details. The results are expressed as the mean±SEM; asterisk (*) is p=0.02 versus CT- spontaneously hypertensive rats (SHR). The number of BrDU positive cells present in retina from WKY rats is very low and do not change according to experiment condition (presence of experimental diabetes); by contrast in SHR rats, it is observed a higher number of these BrDU cells in retina and the induction of diabetes dramatically reduced this number abbreviations used are as follows: control WKY (CT-WKY), diabetic WKY (DM-WKY), control SHR (CT-SHR), and diabetic SHR (DM-SHR).</p></caption></fig>", "<fig id=\"f4\" fig-type=\"figure\" position=\"float\"><label>Figure 4</label><caption><p>Retinal expression of p27^Kip1 in retina evaluated through immunohystochemistry assay. In (<bold>A</bold>), we can see a representative immunohistochemical staining for p27^Kip1 in retina of control and diabetic Wistar Kyoto (WKY) rats and spontaneously hypertensive rats (SHR). The brownish color represents staining for p27^Kip1, and the blue color represents the staining of retinal tissue with hematoxylin. Original magnification 1000x, the bar in the figure represents a scale of 50µm. The graphs represents the score of positivity of p27^Kip1 in retinal slides expressed by mean±SD in ganglion cell (<bold>B</bold>) and inner nuclear (<bold>C</bold>) layers. Asterisk (*) is p=0.05 versus CT-WKY and CT-SHR, and dagger (†) is p=0.02 versus CT-SHR.</p></caption></fig>", "<fig id=\"f5\" fig-type=\"figure\" position=\"float\"><label>Figure 5</label><caption><p>Western blot assay to access the retinal expression of fibronectin. <bold>A:</bold> The membranes were incubated with antibody against glial fibrillar acidic protein in total retinal lysates. <bold>B:</bold> Band densities (ratio of fibronectin to β actin) expressed in arbitrary densitometric units. The columns are the mean±SD of five independent experiments.</p></caption></fig>", "<fig id=\"f6\" fig-type=\"figure\" position=\"float\"><label>Figure 6</label><caption><p>Retinal expression of glial fibrillar acidic protein assayed by western blot. <bold>A:</bold> The membranes were incubated with antibody against glial fibrillar acidic protein (GFAP) in total retinal lysates. The columns represent the band densities (ratio of GFAP to β-actin) expressed in arbitrary densitometric units. Mean±SD of five independent experiments. <bold>B:</bold> Band densities (ratio of GFAP to β-actin) expressed in arbitrary densitometric units. The used symbols are CT-WKY for control WKY, DM-WKY for diabetic WKY, CT-SHR for control SHR and DM-SHR for diabetic SHR.</p></caption></fig>", "<fig id=\"f7\" fig-type=\"figure\" position=\"float\"><label>Figure 7</label><caption><p>Retinal expression of vascular endothelial growth factor (VEGF) was accessed by western blot assay. In A, the membranes were incubated with antibody against VEGF in total retinal lysates. B: The graph display the band densities (ratio of VEGF to β actin) expressed in arbitrary densitometric units. The columns are the mean±SD of four experiments. The used symbols are CT-WKY for control WKY, DM-WKY for diabetic WKY, CT-SHR for control SHR and DM-SHR for diabetic SHR. Asterisk (*) is p=0.02 versus other groups.</p></caption></fig>", "<fig id=\"f8\" fig-type=\"figure\" position=\"float\"><label>Figure 8</label><caption><p>Retinal capillary permeability assessed by the Evans blue method in control and diabetic spontaneously hypertensive rats (SHR) and Wistar Kyoto (WKY) rats. The permeability was expressed in µl of plasma × g retinal wet wt⁻¹ × h⁻¹ wt. The results are expressed as the mean±SD. The following abbreviations are in effect: control WKY (CT-WKY), diabetic WKY (DM-WKY), control SHR (CT-SHR), and diabetic SHR (DM-SHR).</p></caption></fig>" ]

[ "<table-wrap id=\"t1\" position=\"float\"><label>Table 1</label><caption><title>General characteristics of the studied rats</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"65\" span=\"1\"/><col width=\"65\" span=\"1\"/><col width=\"66\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"62\" span=\"1\"/><col width=\"62\" span=\"1\"/><col width=\"72\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"63\" span=\"1\"/><thead><tr><th rowspan=\"2\" valign=\"top\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>Parameter</bold></th><th colspan=\"4\" valign=\"top\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>Younger rat groups</bold><hr/></th><th colspan=\"4\" valign=\"top\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>Adult rat groups</bold><hr/></th></tr><tr><th valign=\"top\" colspan=\"1\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>CT-WKY</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>DM-WKY</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>CT-SHR</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>DM-SHR</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>CT-WKY</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>DM-WKY</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>CT-SHR</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>DM-SHR</bold></th></tr></thead><tbody><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">N<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">20<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">24<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">25<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">24<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">22<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">26<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">29<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">26<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Final body weight (g)<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">198±21<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">139±23†<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">134±15<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">105±12†<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">406± 54<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">293±43†<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">267±42<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">189±25†<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Systolic blood pressure (mmHg)<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">140 ±7‡<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">142 ±5‡<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">118±13<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">120±13<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">174±14‡<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">174±18‡<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">117±12<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">115±14<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Glycemia (mmol/l)</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">6.9±0.3</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">30.4±2.5§</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">7.1±0.5</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">28.6±4§</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">6.2±1.3</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">25.6±5.2§</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">5.8±1.6</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">25.3±4.2§</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>The following abbreviations are in effect: CT-WKY: control WKY; DM-WKY: diabetic WKY; CT-SHR: control SHR; DM-SHR: diabetic SHR; SBP: systolic blood pressure; * p&lt;0.005 versus respectively WKY groups, ^† p&lt;0.003 versus control groups, ^‡ p=0.0001 versus WKY groups, ^§ p&lt;0.0001 versus control groups.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"mv-v14-1680-f1\"/>", "<graphic xlink:href=\"mv-v14-1680-f2\"/>", "<graphic xlink:href=\"mv-v14-1680-f3\"/>", "<graphic xlink:href=\"mv-v14-1680-f4\"/>", "<graphic xlink:href=\"mv-v14-1680-f5\"/>", "<graphic xlink:href=\"mv-v14-1680-f6\"/>", "<graphic xlink:href=\"mv-v14-1680-f7\"/>", "<graphic xlink:href=\"mv-v14-1680-f8\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Arteriosclerotic retinopathy involves vessel walls by medial layer hypertrophy, hyalinization in the intima, and hyperplasia in the endothelial layer [##REF##9541437##1##]. Arteriosclerotic retinopathy usually occurs as a result of progressive hardening of blood vessels by calcification and loss of elastic tissue. Ophthalmoscopic examination is widely used in clinical practice to examine the presence and the degree of arteriolar retinopathy, and offers a unique noninvasive opportunity to assess the status of systemic arteriosclerosis [##REF##11525792##2##,##REF##15118507##3##]. The retinal arteriole shares similar anatomic and physiologic characteristics with cerebral and coronary microcirculation. Therefore, retinal microvascular disease may also reflect the presence of systemic microvascular disease [##REF##11525792##2##,##REF##16216592##4##]. Factors such as smoking, inflammation, and oxidative stress that are associated with free radical formation have been identified as possible risk factors for both systemic atherosclerosis and arteriolar retinopathy [##REF##16724932##5##,##REF##15145091##6##]. Hyperlipidemia is a powerful risk factor for atherosclerosis and related disorders such as ischemic heart disease, cerebrovascular diseases, and retinal atherosclerosis [##REF##17075215##7##,##REF##15185429##8##]. Although recent studies have shown that lipoprotein(a) [Lp(a)] and total homocysteine (Hcy) may be implicated in the development of systemic arteriosclerosis, little knowledge exists about their role in the pathophysiology of retinal arteriosclerosis [##REF##17110238##9##, ####REF##16601863##10##, ##REF##1438209##11####1438209##11##].</p>", "<p>Lp(a) particles contain the lipid and protein component of low-density lipoprotein (LDL) plus a glycoprotein known as apolipoprotein(a) [apo(a)] [##REF##17378772##12##]. Apo(a) is homologous to plasminogen, and its similarity to plasminogen indicates a prothrombogenic role for Lp(a), whereas the similarity of Lp(a) to LDL suggests a proatherogenic role [##REF##16877880##13##]. Serum Lp(a) concentrations are highly heritable. It is reported that about 90% of determinant factors are genetically dependent on sequences of the apo(a) gene in chromosome 6q26–27 [##REF##9322734##14##,##REF##1386087##15##]. Numerous studies have shown that serum levels of Lp(a) above 30 mg/dl is genetically determined and could be an independent risk factor for atherosclerotic vascular disease [##REF##16585023##16##,##REF##16434808##17##].</p>", "<p>Both clinical and experimental investigations suggest that elevated fasting serum Hcy concentration also is an independent risk factor for atherosclerosis, including coronary artery disease, peripheral vascular disease, cerebrovascular disease, and venous thromboembolism [##REF##8592549##18##,##REF##15307456##19##]. Although the exact implication of increased Hcy on atherosclerosis is not clear, recent studies indicate that 15-30% of patients with premature occlusive vascular disease have moderately elevated Hcy concentrations (higher than 15 μmol/l) [##REF##15307456##19##]. To date, little information exists on the possibility that elevated Hcy might increase the risk of retinal arteriosclerosis.</p>", "<p>We proposed that elevated Lp(a) and Hcy levels synergistically increase the atherogenic process, as the potential biochemical interactions with the two risk factors have already been reported [##REF##1438209##11##,##REF##10669648##20##].</p>", "<p>Although Lp(a) and Hcy levels are higher in systemic atherosclerosis, their relationship with retinal arteriosclerosis still is not clear [##REF##7563456##21##, ####REF##9183227##22##, ##REF##15198961##23####15198961##23##]. The aim of this study was to assess the serum concentration of lipids and lipoproteins as well as investigate the concentrations of Hcy and Lp(a) and their relationship with retinal arteriosclerosis.</p>" ]

[ "<title>Methods</title>", "<p>This case-control study was performed from July, 2005 to May, 2006 in the Retina service, Department of Ophthalmology of Tabriz University of Medical Sciences. All participants were recruited from male patients presenting for routine eye examination at the Tabriz Nikookari Eye Hospital. Females were excluded from the study as it has been hypothesized that there is a close relationship between serum estrogen and Hcy levels [##REF##17877128##24##,##REF##16728383##25##]; in addition, the influence of sex hormones on Hcy concentration is not conclusive [##REF##15536396##26##, ####REF##15750667##27##, ##REF##10512207##28##, ##REF##15114519##29####15114519##29##]. The ethic committee at Tabriz University of Medical Sciences reviewed and approved the present study which is in compliance with the Declaration of Helsinki.</p>", "<p>This study enrolled 80 men with retinal arteriosclerosis and a control group of 54 healthy males. All participants gave informed consent. Each participant was evaluated by a retinal specialist (A.J.). Retinal arteriosclerosis was confirmed by slit-lamp examination with super-field lens and fundus photography (Image net 2000, Topcon TRC50IX, Topcon Corp, Tokyo, Japan). Retinal arteriosclerosis was graded according to the Scheie classification for arteriosclerosis (##TAB##0##Table 1##) [##REF##13007237##30##,##UREF##0##31##]. In addition to female patients, also excluded were patients who were being treated with vitamins, antioxidants, micronutrients supplements, or lipid-lowering drugs; patients who had active systemic infection; smokers; and patients who had a history of alcohol abuse, coronary heart diseases, uncontrolled hypertension (systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg), diabetes mellitus, thyroid, or other metabolic diseases. All participants had normal renal and liver functions as assessed by plasma urea, creatinine, alanine aminotransferase, and aspartate aminotransferase. Blood samples were obtained after an overnight fasting. Serum samples were frozen immediately and stored at -70 °C until needed for analysis. Serum levels of total cholesterol (TC), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C) were determined using commercial reagents with an automated chemical analyzer (Abbott analyzer, Abbott laboratories, Abbott Park, North Chicago, IL). Low-density lipoprotein cholesterol (LDL-C) was calculated by using the Friedewald equation [##REF##4337382##32##]. The serum level of Lp(a) was assayed by commercially available immunoturbidimetric kit (Pars Azmun, Tehran, Iran, Lott No: 85001) using the same analyzer. Serum Hcy concentration was determined by a commercially available enzyme-linked immunoassay (Axis-Shield, Axis Biochemicals ASA, Distributed by IBL, Hamburg, Germany, Cat. No: AX51301).</p>", "<p>SPSS software package version 13 for Windows (SPSS Ins, Chicago, IL) was used to perform statistical analysis. Results were expressed as mean ± standard deviation. Independent <italic>t</italic>-test, Mann–Whitney U, logistic and multiple regression tests, as appropriate, were used to assess significance of differences between the two groups. The correlations between variables were evaluated by Pearson (for parametric data) or Spearman (for nonparametric data) test as appropriate. A p&lt;0.05 was considered significance statistically.</p>" ]

[ "<title>Results</title>", "<p>We compared a group of 80 male patients with retinal arteriosclerosis to 54 male control patients. There were no statistically significant differences between the mean ages of case (64.3±6.8 years) and control (66.7±8.0 years) groups (p&gt;0.05). There was no significant difference in systolic blood pressure between the case (116.90±10.47 mmHg) and control groups (113.55±10.12 mmHg); p&gt;0.05. Also diastolic blood pressure in the case group (69.81±8.50 mmHg) was not statistically different from control group (67.85±8.05 mmHg); p&gt;0.05.</p>", "<p>All controls had degree 0 of retinal arteriosclerosis (by slit-lamp with super-field lens and fundus photography in retina according to Scheie classification in ##TAB##0##Table 1##), while 24 (30%) of patients were affected with degree I, 21 (26.25%) with degree II, 21 (26.25%) with degree III, and 14 (17.5%) with degree IV.</p>", "<p>Lipid profile of patients and control group are shown in ##TAB##1##Table 2##. There were no differences in HDL levels between the two groups, while the mean levels of TG, TC, and LDL were significantly higher in patients than those in control group. Serum level of Hcy was higher in patients (24.2±8.1 μmol/l) than controls (10.5±4.1 μmol/l); p&lt;0.01. Serum levels of Lp(a) in patients (47.9±33.1 mg/dl) was also higher than controls (11.7±7.6 mg/dl); p&lt;0.01.</p>", "<p>Hcy and Lp(a) levels were both significantly correlated with the degree of retinal arteriosclerosis (##FIG##0##Figure 1## and ##FIG##1##Figure 2##). In multivariable analysis, retinal arteriosclerosis was associated with higher levels of Lp(a) (OR: 1.13, 95% CI: 1.07-1.19; P&lt;0.01) and Hcy (OR: 1.52; 95% CI: 1.31-1.77; p&lt;0.001). There was no statistically significant association of retinal arteriosclerosis with age, systolic and diastolic blood pressure, TC, TG, LDL-C, and HDL-C. However, there was a significant direct linear correlation between Lp(a) and Hcy (##FIG##2##Figure 3##).</p>", "<p>Consistent with previous reports [##REF##10669648##20##,##REF##8650712##33##], 30 mg/dl of Lp(a) (likelihood ratio= 69.11, P&lt;0.0001) and 15.5 μmol/l of Hcy (likelihood ratio= 86.93, p&lt;0.0001) were determined as significant cutoff points for arteriosclerosis. Lp(a) and Hcy performance as a predictor of arteriosclerosis was summarized by receiver operating characteristic (ROC) curve (##FIG##3##Figure 4##) and area under the curve (AUC; ##TAB##2##Table 3##). ROC curve was well above the diagonal, indicating good sensitivity and specificity. The AUC indicated a high probability of correctly prediction of arteriosclerosis.</p>" ]

[ "<title>Discussion</title>", "<p>Increased blood pressure, elevated total cholesterol, smoking status, coronary bypass surgery history, and high white blood cell count are important etiologic factors in retinal arteriosclerosis, but there is limited information about a possible role of Lp(a) and Hcy in the pathophysiology of retinal arteriosclerosis [##REF##9541437##1##,##REF##16216592##4##,##REF##15307456##19##]. It is well known that elevated levels of Hcy and Lp(a) are often associated with endothelial dysfunction and enhanced atherosclerosis [##REF##17378772##12##,##REF##14664904##34##]. Various authors have proposed that hyperhomocysteinemia and high serum Lp(a) concentrations play a role in the genesis of systemic atherosclerosis, thrombosis, and related disorders [##REF##1438209##11##,##REF##10669648##20##,##REF##8642472##35##,##REF##16105483##36##].</p>", "<p>The retina is the only place where we can directly observe arteriosclerotic changes noninvasively [##REF##9541437##1##, ####REF##11525792##2##, ##REF##15118507##3####15118507##3##]. Since both cerebral and retinal arteries are peripheral branches of the internal carotid arteries, high Lp(a) and Hcy levels may also be associated with retinal vascular diseases such as retinal arteriosclerosis [##REF##11525792##2##,##REF##16216592##4##,##REF##12824229##37##]. To evaluate the connection between Lp(a) and Hcy levels and retinal arteriosclerosis, we determined serum Lp(a) and Hcy concentrations and lipid and lipoprotein levels. In the present study, patients had significantly higher Hcy and Lp(a) levels. These results are in agreement with those reported by Martin et al., [##REF##11040905##38##] who suggested Hcy might be a risk factor in retinal vascular disease. Present study findings suggest that hyperhomocysteinemia may play an important role in the pathogenesis of retinal arteriosclerosis.</p>", "<p>Lp(a) is a low-density lipoprotein particle in which apolipoprotein B-100 is linked by a single interchain disulfide bridge to a unique glycoprotein apo(a) [##REF##2530631##39##,##REF##1390593##40##]. It has been shown that high plasma Lp(a) levels are closely associated with arterial thrombosis such as myocardial infarction and cerebral infarction [##REF##16585023##16##,##REF##2945294##41##,##REF##11698280##42##]. The role of Lp(a) in retinal vascular changes has been reported in a few cases [##REF##12528289##43##, ####REF##8368186##44##, ##REF##8436260##45##, ##REF##10592864##46####10592864##46##]. Some investigators have proposed that impaired fibrinolysis and atherogenesis induced by Lp(a) may play a role in the pathophysiology of retinal vascular changes, while other investigators have concluded that serum Lp(a) level is not a significant risk factor for atherosclerosis of the retinal arteries [##REF##8368186##44##,##REF##8436260##45##]. Foody et al. [##REF##10669648##20##] found that thiols, such as Hcy, can dissociate apo(a) from Lp(a) complex, leading to the exposure of an additional lysine binding site on apo(a) that can increase the affinity of apo(a) to plasmin-modified fibrin. In the present study, correlations between Lp(a), Hcy, and the degree of retinal arteriosclerosis were also investigated. The data indicate that there is a significant correlation between the degree of retinal arteriosclerosis with both Lp(a) and Hcy levels. Moreover, a significant correlation between Lp(a) and Hcy concentrations was observed. However, because this study only included male participants and the study population was of a low number, further studies are warranted.</p>", "<p>In conclusion, our findings support the idea that Lp(a) and Hcy may play an important role in the development of retinal arteriosclerosis.</p>" ]

[]

[ "<p>This is an open-access article distributed under the terms of the\n Creative Commons Attribution License, which permits unrestricted use,\n distribution, and reproduction in any medium, provided the original\n work is properly cited.</p>", "<title>Purpose</title>", "<p>Elevated levels of lipoprotein(a) [Lp(a)] and homocysteine (Hcy) have been implicated as risk factors for vascular diseases. The study was performed to explore the possible relationship between retinal arteriosclerosis and serum Lp(a) and Hcy levels.</p>", "<title>Methods</title>", "<p>Study subjects consisted of 80 nonsmoking male patients with retinal arteriosclerosis and 54 healthy nonsmoker males as controls. Retinal arteriosclerosis was graded according to the Scheie classification. Serum levels of lipids, lipoproteins, Lp(a), and Hcy were measured by standard methods.</p>", "<title>Results</title>", "<p>The serum level of Hcy was higher in patients (24.2±8.1 μmol/l) than controls (10.5±4.1 μmol/l); p&lt;0.01. Serum levels of Lp(a) in patients (47.9±33.1 mg/dl) was also higher than controls (11.7±7.6 mg/dl); p&lt;0.01. There was a significant direct linear correlation between the degree of retinal arteriosclerosis and Lp(a) level (r=0.61, p&lt;0.01), the degree of retinal arteriosclerosis and Hcy level (r=0.72, p&lt;0.01), and also between Lp(a) and Hcy levels (r=0.67, p&lt;0.01).</p>", "<title>Conclusions</title>", "<p>The association between retinal arteriosclerosis and serum Lp(a) and Hcy levels suggests that Lp(a) as well as Hcy could play a role in the development of retinal arteriosclerosis.</p>" ]

[]

[]

[ "<fig id=\"f1\" fig-type=\"figure\" position=\"float\"><label>Figure 1</label><caption><p>Relationship between Hcy level and the degree of retinal arteriosclerosis. There was a direct linear correlation between Hcy level and the degree of retinal arteriosclerosis in the study population (r=0.72, p&lt;0.01), i.e. patients with higher degree of retinal arteriosclerosis had higher level of Hcy.</p></caption></fig>", "<fig id=\"f2\" fig-type=\"figure\" position=\"float\"><label>Figure 2</label><caption><p>Relationship between Lp(a) level and the degree of retinal arteriosclerosis. There was a direct linear correlation between Lp(a) level and the degree of retinal arteriosclerosis in the study population (r=0.61, p&lt;0.01), i.e. patients with higher degree of retinal arteriosclerosis had higher level of Lp(a)</p></caption></fig>", "<fig id=\"f3\" fig-type=\"figure\" position=\"float\"><label>Figure 3</label><caption><p>Correlation of Hcy level with Lp(a) level. There was a direct linear correlation between Hcy and Lp(a) levels in the study population (r=0.67, p&lt;0.01), i.e., higher level of Hcy was associated with higher level of Lp(a).</p></caption></fig>", "<fig id=\"f4\" fig-type=\"figure\" position=\"float\"><label>Figure 4</label><caption><p>Receiver operating characteristic curves for arteriosclerosis. Shown is the relationship of sensitivity to 1-specificity as plotted for lipoprotein(a) and homocysteine. The area under each receiver operating characteristic (ROC) curve for each parameter indicates its diagnostic accuracy (##TAB##2##Table 3##).</p></caption></fig>" ]

[ "<table-wrap id=\"t1\" position=\"float\"><label>Table 1</label><caption><title>Scheie classification</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"42\" span=\"1\"/><col width=\"318\" span=\"1\"/><thead><tr><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Stage</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Observation</bold></th></tr></thead><tbody><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">0<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Normal<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">1<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">There is broadening of the light reflex from the arteriole, with minimal or no arteriolovenous compression.<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">2<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">Light reflex changes and crossing changes are more prominent.<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">3<hr/></td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">The arterioles have a copper wire appearance, and there is more arteriolovenous compression.<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">4</td><td valign=\"top\" align=\"left\" rowspan=\"1\" colspan=\"1\">The arterioles have a silver wire appearance, and the arteriolovenous crossing changes are most severe.</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t2\" position=\"float\"><label>Table 2</label><caption><title>Lipid profiles of patients with retinal arteriosclerosis and controls</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"95\" span=\"1\"/><col width=\"90\" span=\"1\"/><col width=\"81\" span=\"1\"/><col width=\"63\" span=\"1\"/><thead><tr><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Variables</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Controls (n=54) mean±SD</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Patients (n=80) mean±SD</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>p value</bold></th></tr></thead><tbody><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">TC (mg/dl)<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">159.4±25.2<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">199.8±39.6<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">&lt;0.0001*<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">TG (mg/dl)<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">114.8±35.3<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">152.4±71.8<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.001**<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">HDL-C (mg/dl)<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">37.4±9.5<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">39.7±8.6<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">&gt;0.1*<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">LDL-C (mg/dl)</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">99.5±22.9</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">129.6±37.8</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">&lt;0.0001*</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t3\" position=\"float\"><label>Table 3</label><caption><title>Area under the curve for the receiver operating characteristic curves in ##FIG##3##Figure 4##</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"80\" span=\"1\"/><col width=\"50\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"73\" span=\"1\"/><col width=\"52\" span=\"1\"/><col width=\"63\" span=\"1\"/><thead><tr><th rowspan=\"2\" valign=\"top\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>Test result variables</bold></th><th rowspan=\"2\" valign=\"top\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>Area under curve</bold></th><th rowspan=\"2\" valign=\"top\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>Standard error</bold></th><th rowspan=\"2\" valign=\"top\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>Asymptotic sig.</bold></th><th colspan=\"2\" valign=\"top\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>Asymptotic 95% confidence interval</bold><hr/></th></tr><tr><th valign=\"top\" colspan=\"1\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>Lower bound</bold></th><th valign=\"top\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Upper bound</bold></th></tr></thead><tbody><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Lipoprotein(a)<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.888<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.027<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.000<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.835<hr/></td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.941<hr/></td></tr><tr><td valign=\"top\" align=\"center\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Homocysteine</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.950</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.017</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.000</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.917</td><td valign=\"top\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.983</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Retinal arteriosclerosis was graded by slit-lamp with super-field lens and fundus photography in retina according to Scheie classification as above [##REF##13007237##30##,##UREF##0##31##].</p></table-wrap-foot>", "<table-wrap-foot><p>This table shows the comparison of the lipid profiles between the patients and controls. Except the high-density lipoprotein cholesterol (HDL-C) level, all the total cholesterol (TC), triglyceride (TG), and low-density lipoprotein cholesterol (LDL-C) levels are higher in patients than controls. The asterisk denotes an independent-sample <italic>t</italic>-test, and the double asterisk indicates a Mann-Whitney U test.</p></table-wrap-foot>", "<table-wrap-foot><p>This table shows the values of area under the curve for the receiver operating characteristic curves of lipoprotein (a) and homocysteine in ##FIG##3##Figure 4##.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"mv-v14-1692-f1\"/>", "<graphic xlink:href=\"mv-v14-1692-f2\"/>", "<graphic xlink:href=\"mv-v14-1692-f3\"/>", "<graphic xlink:href=\"mv-v14-1692-f4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Diabetic retinopathy (DR), a leading cause of blindness worldwide, is a microvascular complication of diabetes characterized by increased vascular permeability and hemostatic abnormalities which can eventually lead to vascular occlusion in spite of antidiabetic treatment and result in retinal nonperfusion and neovascularization [##REF##14559961##1##]. DR is associated with disorders in the nitric oxide (NO) pathway, including impaired NO-mediated vasodilation, increased oxidative and nitrative stress, dysregulation of NO synthase isoforms, and endothelial NO synthase uncoupling [##REF##11839570##2##]. Clinically, DR can be classified into two major categories: early and advanced. Early stage of DR, also called nonproliferative diabetic retinopathy (NPDR), is characterized by edema, leakage of fluid, and limited blood flow into the eye. But NPDR has no abnormal neovascularization. The advanced stage of DR, or proliferative diabetic retinopathy (PDR), involves neovascularization and fibrous tissue formation [##REF##13694291##3##].</p>", "<p>Over 30 candidate genes involved in different metabolic mechanisms and functional pathways have been reported to be associated with DR [##REF##12724690##4##,##REF##16868886##5##]. However, only a fraction of them have shown consistent associations with occurrence of DR or its severity in different studies [##REF##12724690##4##,##REF##16868886##5##]. The (CA)_n microsatellite marker at the 5’-end of the aldose reductase (<italic>AR2</italic>) gene is the most frequently reported polymorphism that is associated with DR [##REF##12724690##4##]. Results of our previous study on Chinese DR patients showed that the z-2 allele of the 5′-(CA)_n polymorphism was independently associated with DR in type 2 diabetes [##REF##12882871##6##]. Genetic studies have been conducted for vasoactive and angiogenic factors to explore their contributions to DR, such as endothelial nitric oxide synthase (<italic>eNOS</italic>), lymphotoxin-a (<italic>LTA</italic>), integrin alpha-2 (<italic>ITGA2</italic>), angiotensin converting enzyme (<italic>ACE</italic>), vascular endothelial growth factor (<italic>VEGF</italic>), intercellular adhesion molecule 1 <italic>(ICAM</italic> or <italic>CD45</italic>), β3-adrenergic receptor gene <italic>(ADRB3)</italic>, and endothelin-1 (<italic>EDN1</italic>) [##REF##12724690##4##,##REF##16868886##5##]. However, results for some of these genes, such as <italic>ACE</italic> and <italic>CD45</italic>, have been inconsistent in various ethnic groups and different study populations. Discrepancies among these studies are likely due to variations in case definition, sample sizes, and medical conditions of control subjects. Some studies selected normal individuals as controls instead of diabetic subjects without DR. Thus, the comparisons were only between normal individuals and diabetic individuals, which may lead to identification of genetic variants associated with vulnerability to diabetes mellitus (DM) rather than to DR [##REF##11399938##7##].</p>", "<p>In this study, we focused on a well-defined population with type 2 diabetes. Only patients without DR but with DM for at least 10 years were selected as controls. We hypothesized that genetic variants in vasoactive and angiogenic factors which regulate the retina vasculature might contribute to the development of retinopathy. We chose polymorphisms previously reported to have positive association with DR, but none have been evaluated in a Chinese DM population. The Lys198Asn polymorphism of <italic>EDN1</italic> was also studied because it is associated with hypertension, which is a risk factor of DR [##REF##16097909##8##, ####REF##11593097##9##, ##REF##10334806##10##, ##REF##10460781##11####10460781##11##]. We examined six polymorphisms in four genes, including two vasoactive genes, <italic>EDN1</italic> and <italic>eNOS</italic>, and two hemodynamic blood flow related genes, <italic>LTA</italic> and <italic>ITGA2.</italic> They have been reported to be associated with vascular diseases in different populations.</p>" ]

[ "<title>Methods</title>", "<title>Study subjects</title>", "<p>Unrelated participants with type 2 noninsulin-dependent diabetes mellitus (NIDDM) were recruited from the Eye Clinic of the Prince of Wales Hospital, Hong Kong. Diagnosis of type 2 diabetes was based on WHO criteria [##REF##9686693##12##]. This study was performed in accordance with the ethics standards set by the Declaration of Helsinki. Approval for use of human subjects was obtained from the New Territories East Cluster, Hong Kong, clinical research ethics committee. Informed consent was obtained from the participants after explanation of the nature and possible consequences of the study. DR was diagnosed in a masked manner by independent ophthalmologists by direct ophthalmoscope through dilated pupils. Patients with no signs of DR and with known DM duration &lt;10 years were excluded. Patients without DR but with diabetes duration ≥10 years were designated as control subjects (DM). Diabetic patients with DR were defined as case subjects (DR) either of the NPDR or PDR subtype according to Early Treatment Diabetic Retinopathy Study (ETDRS) criteria [##REF##2062513##13##]. Documented information included age, sex, age at onset of diabetes, duration of diabetes, fast plasma glucose and hemoglobin A₁c (HbA₁c) level, age at onset of DR, family history, and treatment details. A smoker was defined as someone who smoked at least five cigarettes daily for more than one year. Onset of DR was defined as the first ever documentation of the clinical evidence of DR on the chart. Hyperlipidemia was diagnosed on patients with either elevated total cholesterol (&gt;6.2 mM) or elevated triglycerides (&gt;1.7 mM) requiring either lifestyle modification or pharmacological intervention. Renal failure was defined as persistent and irreversible derangement of the renal function test, typically creatinine &gt;80 μM upon consecutive measurements.</p>", "<p>After exclusion of non-DR patients having DM duration of fewer than 10 years, 343 patients were enrolled, consisting of 127 controls and 216 DR patients. Using the allele distributions in non-DR patients, we calculated the statistical power at 80% with a Bonferroni correction significance level of 0.0083, where six comparisons were made, and α, before correction was 0.05. Thus the corrected level, α ÷ 6 is 0.083 (two-sided) to detect an allele odds ratio of at least 2.2 for eNOS or 1.8 for the other five polymorphisms. DR patients were further divided into NPDR (n=144) and PDR (n=72). DR group showed a significantly higher level of HbA₁c (7.35±1.36% versus 6.9±1.91%, p=0.046) and a higher percentage of patients who underwent insulin treatment (11.5% versus 2.4%, p&lt;0.00001) than control group (##TAB##0##Table 1##). The PDR group was slightly younger than NPDR and had earlier onset of diabetes. Diabetic macular edema was more common in PDR than in NPDR.</p>", "<title>Genotyping</title>", "<p>The whole blood specimens (5 ml) from all the patients were collected in EDTA tube and stored at -20 °C for fewer than two months’ storage. Genomic DNA was extracted from whole blood using the Qiamp kit (Qiagen, Hilden, Germany) and stored at −20 °C for fewer than two months before analysis. The polymorphisms of <italic>EDN1</italic> (Lys198Asn or <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=5370\">rs5370</ext-link>) and <italic>LTA</italic> (IVS1–80C&gt;A, IVS1–206G&gt;C, and IVS1–252A&gt;G) were detected by polymerase chain reaction (PCR) and direct DNA sequencing. PCR was performed in a final volume of 25 μl containing 2.5 μl of 10X PCR buffer (Invitrogen^TM Life Technology, Carlsbad, CA), 0.3 μl of each primer (<italic>EDN1</italic>: forward 5’- CTT TTG CCA AAG GGT GAT TT-3’ and reverse 5’- AGG GTG GAG AGT GCA GAG TC-3’; <italic>LTA</italic>: forward 5’- TCC TGC CCC ATC TCC TTG G-3’ and reverse 5’-AGA GAG AGA GAC AGT GAG CGG G-3'), 0.5 μl of 10 mM dNTP, 1 μl of 50 mM MgCl_2, 0.5 U AmpliTaq® DNA Polymerase (Applied Biosystems, Foster City, CA), and 10 ng of DNA. After the initial denaturation at 94 °C for 3 min, 35 cycles were conducted: 94 °C for 30 s, annealing for 30 s (<italic>EDN1</italic>: 54 °C; <italic>LTA</italic>: 58 °C), and 72 °C for 30 s. The final extension lasted for 10 min at 72 °C. Sequencing was performed using a standard protocol on an automated 3130X DNA sequencer (Applied Biosystems) [##REF##10997269##14##]. Sequence data were analyzed on computer (Chromas ver. 2.13; Technelysium Pty Ltd., Tewantin, QLD, Australia) and compared with published gene<strike>s</strike> sequences from <ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ensembl.org/Homo_sapiens/geneview?gene=OTTHUMG00000014266;db=vega\">Ensembl</ext-link>. All rare variants detected were confirmed by bidirectional sequencing.</p>", "<p>Polymorphisms in <italic>eNOS</italic> (894G&gt;T) and <italic>ITGA2</italic> (IVS8–1059T&gt;C) were analyzed by PCR followed by BanII and BgIII restriction analysis respectively [##REF##12436344##15##,##REF##10688808##16##]. To cross-check the genotyping results, we randomly selected one-fourth of the PCR products for direct sequencing. Complete matching of results was obtained.</p>", "<title>Statistical analysis</title>", "<p>Differences in genotype distribution and consistency with Hardy–Weinberg equilibrium were tested by χ² test. Continuous clinical data (age, age at onset of diabetes, duration of diabetes, and HbA₁c) were compared by independent Student <italic>t</italic> test, and categorical clinical data were compared using the χ² test or the Fisher’s exact test. Tukey’s test was used for multiple Post Hoc comparisons, and the Bonferroni method was used for multiple comparison adjustment. To assess the role of gene polymorphisms and search for gene-gene and gene-environmental interactions, we built logistic regression models using various polymorphisms and clinical parameters. Disease status was set as the dependent variable (DR=1; control=0), and gene polymorphisms and environmental factors as independent variables. A stepwise regression approach was used to optimize the analysis. SPSS for Windows, standard version 11.5 (SPSS, Chicago, IL), was used. Statistical significance was defined as p&lt;0.05.</p>" ]

[ "<title>Results</title>", "<p>The distributions of the six polymorphisms of <italic>EDN1</italic>, <italic>eNOS</italic>, <italic>LTA</italic>, and <italic>ITGA2</italic> in both case and control groups were under Hardy–Weinberg equilibrium. A significantly higher frequency of the <italic>EDN1</italic> Asn/Asn genotype was found in controls than in DR patients (11% versus 2.3%, p=0.0002, Bonferroni corrected significance level 0.0083=0.05÷6, ##TAB##1##Table 2##). The Asn allele frequency was also significantly higher in controls than in the DR group (29.5% versus 16.4%, p=0.007, Bonferroni corrected significance level 0.0083). For genotype or allele distributions of <italic>eNOS</italic>, <italic>IGTA2,</italic> and <italic>LTA</italic> polymorphisms, there was no significant difference between DM subjects with or without DR (##TAB##1##Table 2##).</p>", "<p>For the <italic>EDN1</italic> polymorphism, we compared DM controls and DR subtypes. NPDR patients had a higher frequency of Lys/Lys than either DM controls or PDR patients (75.7% versus either 56.9% or 52.0% with p=0.0001 and 0.008, respectively, Bonferroni corrected significance level 0.017 = 0.05 ÷ 3), but PDR did not differ from DM controls (p&gt;Bonferroni corrected significance level 0.017). The Asn allele frequency was also significantly higher in controls than in NPDR (29.5% versus 13.5%; p=0.00005, Bonferroni corrected significance level 0.017). However, allele frequencies did not significantly differ among DR subtypes.</p>", "<p>Multivariable logistic regression analysis showed that <italic>EDN1</italic> Asn/Asn was an independent protective factor for DR after adjustment for age, age at onset of diabetes and insulin therapy. The Odds ratio (OR) was 0.19 with 95% confidence interval (CI) ranging from 0.07 to 0.53 (p=0.002; ##TAB##2##Table 3##). The difference of Lys198Asn genotype distributions between NPDR and PDR was not significant after adjusting for the age at onset of diabetes (p&gt;0.05). But age at onset of diabetes was an independent factor associated with the PDR phenotype (OR=0.94; 95% CI: 0.91–0.97; p=0.00001). We also found no gene-gene or gene-environmental factor interaction in DM samples (p&gt;0.05). The age of DM onset of patients with the Asn/Asn genotype was about six years later than patients with other genotypes (p=0.02; ##TAB##3##Table 4##). Genotype distributions of Lys198Asn were not different between hypertensive and nonhypertensive subjects (##TAB##3##Table 4##).</p>" ]

[ "<title>Discussion</title>", "<p>The vasodilator NO and vasoconstrictor endothelin-1 (ET-1) effectively determine the tone of blood vessels [##REF##11839570##2##,##REF##11756889##17##]. Changes in blood flow contribute to early diabetic microangiopathy in DR [##REF##9315390##18##]. Genetic studies on <italic>eNO</italic>S and risk to DR have shown inconsistent results, and no association between polymorphisms in <italic>EDN1</italic> and DR has been identified [##REF##10450377##19##, ####REF##15890549##20##, ##REF##15333482##21##, ##REF##16636650##22####16636650##22##]. LTA and ITGA2 are important molecules involved in lymphocyte proliferation and platelet adhesion to subendothelial tissues. They are essential for thrombus formation and contribute to tissue ischemia and activation of neovascularization in DR. The BgIII polymorphism of <italic>ITGA2</italic> has been shown to be strongly associated with DR in the Caucasian and Japanese populations [##REF##10688808##16##,##REF##12938014##23##]. There are also ethnic differences in the genetic risk of DR. <italic>LTA</italic> IVS1–252A&gt;G has been found to be associated with DR in Caucasians but not in Japanese [##REF##11399938##7##,##REF##16979413##24##].</p>", "<p>Our study is the first to investigate the <italic>EDN1</italic> Lys198Asn polymorphism in a diabetic population. The Asn/Asn frequency was reported to be 6.7% in the Chinese population (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=5370\">HapMap</ext-link> project). We found that the Asn/Asn genotype was higher in DM controls (11.0%) than in DR subjects (2.3%), suggesting a protective role against DR in general. In addition, Asn/Asn was associated with an older age of onset of diabetes, on average a six-year delay (##TAB##3##Table 4##). Therefore, Asn/Asn might delay DM and protect against progression of DM to NPDR. The Asn/Asn genotype was not associated with hypertension in our DM subjects, indicating that Asn/Asn might not modulate the risk of DR via an effect on hypertension. The frequency of Lys/Lys in PDR (56.9%) was closer to that in DM controls (52.0%; p&gt;0.017) than in NPDR (75.7%; p&lt;0.017). This implies that the pathological mechanisms that lead to NPDR might be different from those in PDR. Lys/Lys might contribute a role in the early breakdown of the blood retinal barrier (as may be the case in NPDR) rather than vascular occlusion and neovascularization (in the case of PDR).</p>", "<p>In this study, we focused on a clinically well defined diabetic population to explore the role of genetic factors in DR progression. Each member of the control group lacked DR but had a ≥10-year history of diabetes. According to the Wisconsin epidemiologic study of diabetic retinopathy (WESDR), the 10-year incidence of DR was almost twice as high as the four year incidence of DR in older onset diabetic subjects (67% versus 34%). In insulin-dependent individuals diagnosed with diabetes before the age of 30 years, the incidence of DR increased from 59% over 4 years to 89% over 10 years, and maintained a similar incidence of 86% over 14 years [##REF##2357354##25##, ####REF##7619101##26##, ##REF##9787347##27####9787347##27##]. Thus, a 10-year duration of DM is a reasonable criterion for selecting controls for a DR genetic study. Diabetes duration and glycemic control are major determinants of the development of DR. Our results also showed that subjects who were younger at DM onset had a higher risk of PDR than did subjects with later onset of DM (OR=0.94; p=0.00001; ##TAB##2##Table 3##).</p>", "<p>ET-1 is implicated in some vascular diseases, including the pathology of DR [##REF##16248775##28##]. <italic>EDN1</italic> has also been identified in an early gestation human eye cDNA library and might be important for eye development [##REF##16368877##29##]. The pathogenic significance of the association of <italic>EDN1</italic> Lys198Asn with DM and DR remains unknown. Some studies suggested that the Asn allele is associated with increased plasma concentrations of ET-1 [##REF##11593097##9##,##REF##10334806##10##]. However, ET-1 levels in vitreous and aqueous from DR patients were only elevated at the advanced stage of DR (PDR), but decreased in the early stage (NPDR) [##REF##11756889##17##,##REF##17001327##30##]. ET-1 is synthesized from a 212-amino-acid precursor protein (preproET-1) through multiple proteolytic steps. In the first step, preproET-1 is cleaved by signal peptidase, resulting in the formation of proET-1. ProET-1 is then cleaved at the paired dibasic amino acids by a furin-like enzyme to give rise to 38-aminoacid big ET-1 or other intermediates. Big ET-1 is subsequently cleaved at Trp73-Val74 by another endopeptidase, endothelin converting enzyme, resulting in the production of mature ET-1 [##REF##2473327##31##]. A functional in vitro expression study of <italic>EDN1</italic> Lys198Asn found that neither ET-1 (bioactive form) nor big ET-1 (intermediate polypeptide) levels in the culture supernatant of the Asn-type transfected cells were significantly changed compared to those of the Lys-type transfected cells in three different cell lines: COS1 cells, 293 cells and human umbilical vein endothelial cells (HUVECs) [##REF##15198485##32##]. Lys198Asn is located near the carboxyl terminal region, which is removed from prepro-ET-1 by the proteolytic action of the furin-like enzyme during the processing of ET-1 [##REF##10334806##10##,##REF##11751711##33##]. This polymorphism may affect the processing of preproET-1 to mature ET-1 rather than modifying the gene expression or the stability of the mRNA. So far there is no data on the functional consequence of the Lys198Asn polymorphism on preproET-1. We hypothesized that the protective effects of 198Asn on diabetes and DR might not exert direct influence on ET-1 production. It could have an impact on the processing of ET-1 or influence linkage disequilibrium with other polymorphisms that affect the production or conformation of the protein. Alterations in ET-1 level or function might affect the response to damage from hyperglycemia and hypoxia in vascular endothelial and other cell types. Another <italic>EDN1</italic> polymorphism, a dinucleotide repeat in the 5′-untranslated region, did not have a major impact on DR [##REF##10450377##19##].</p>", "<p>In conclusion, we identified the Asn/Asn genotype of <italic>EDN1</italic> as a genetic factor for delayed onset of DM and reduced risk of DR in type 2 DM patients. Further studies in other DM types and different ethnic populations should be performed.</p>" ]

[]

[ "<p>This is an open-access article distributed under the terms of the\n Creative Commons Attribution License, which permits unrestricted use,\n distribution, and reproduction in any medium, provided the original\n work is properly cited.</p>", "<title>Purpose</title>", "<p>We tested the hypothesis that genetic variants in vasoactive and angiogenic factors regulating the retina vasculature contribute to the development of diabetic retinopathy (DR).</p>", "<title>Methods</title>", "<p>A case-control study was performed to study the genetic association between DR and polymorphic variants of <italic>EDN1</italic> (Lys198Asn), <italic>LTA</italic> (IVS1–80C&gt;A, IVS1–206G&gt;C, IVS1–252A&gt;G), <italic>eNOS</italic> (Glu298Asp), and <italic>ITGA2</italic> (<italic>BgI II)</italic> in a Chinese population with type 2 diabetes mellitus. A well defined population with type 2 diabetes, consisting of 127 controls and 216 DR patients, was recruited.</p>", "<title>Results</title>", "<p>A higher frequency of the Asn/Asn genotype of <italic>EDN1</italic> was found in individuals with at least 10 years of diabetes and no retinopathy (controls) compared with DR patients with any duration of diabetes (DR: 2.3%; control: 11.0%; p=0.0002). The Asn allele was also more frequent in controls than DR patients (DR: 16.4%; control: 29.5%; p=0.007). Multiple logistic regression analysis showed that the Asn/Asn genotype was the factor most significantly associated with reduced risk of DR (odds ratio=0.19; 95% CI: 0.07-0.53; p=0.002) and with late onset of diabetes (Asn/Asn: 59 years; Lys/Lys + Lys/Asn: 53 years; p=0.02). Moreover, the Lys/Lys genotype was more common among patients with nonproliferative (75.7%) than proliferative DR (56.9%; p=0.008). The distributions of Lys198Asn alleles in hypertension did not differ from normotensive subjects. No associations between DR and polymorphisms of <italic>LTA</italic>, <italic>eNOS</italic>, or <italic>ITGA2</italic> were detected, and there were no detectable gene-gene or gene-environmental interactions among the polymorphisms.</p>", "<title>Conclusions</title>", "<p>The Asn/Asn genotype of <italic>EDN1</italic> was associated with a reduced risk of DR and with delayed onset of type 2 diabetes.</p>" ]

[]

[ "<title>Acknowledgements</title>", "<p>We are grateful to patients for their participation in this study. This study is supported by a block grant of the Chinese University of Hong Kong and the Li Ka Shing Foundation, Ophthalmology research and training fund.</p>" ]

[]

[ "<table-wrap id=\"t1\" position=\"float\"><label>Table 1</label><caption><title>Clinical and metabolic characteristics of patients with type 2 diabetes</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"216\" span=\"1\"/><col width=\"62\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"63\" span=\"1\"/><thead><tr><th valign=\"bottom\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\">  <bold>Characteristic</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>DM (n=127)</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>All DR (n=216)</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>NPDR (n=144)</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>PDR (n=72)</bold></th></tr></thead><tbody><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Age (years)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">68.8±11.2<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">66.8±10.4<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">69.1±9.4<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">62.7±11.0^4,5<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Gender (% male)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">60.6<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">60.6<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">59.4<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">62.5<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Age at onset of DM (years)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">53.8±11.9<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">53.0±10.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">55.4±10.4<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">48.4±10.0^6,7<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Duration of diabetes (years)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">14.7±4.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">13.9±8.3<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">13.7±8.0<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">14.4±8.8<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Age at onset of DR (years)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">None<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">62.1±10.2<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">65.6±9.4<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">56.8±9.9<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Gap between DR and DM onset ages (years) <hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">None<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">8.4±8.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">9.1±7.6<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">7.3±11.1<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Duration of DR (years)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">None<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">4.9±3.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">4.6±3.6<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">5.6±4.1<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">HbA1c (%)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">6.90±0.91<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">7.35±1.39¹<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">7.53±1.56³<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">6.98±0.94<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Fasting plasma glucose (mM)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">8.24±1.98<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">8.26±2.83<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">8.09±2.86<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">8.81±2.89<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Insulin use (% of patients)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">2.4<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">11.5²<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">8.3<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">18.1⁸<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Family history of DM (% of patients)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">17.7<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">25.3<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">22.5<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">29.2<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Hypertension (% of patients)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">68.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">67.1<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">66<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">69.4<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Hyperlipidemia (% of patients)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">26.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">24.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">24.5<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">26.4<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Renal failure (% of patients)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.8<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">4.1<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">3.5<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">5.6<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Current smoking (% of patients)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">9.2<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">11.1<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">11.1<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">11.1<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Diabetic macular edema (% of patients)</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">None</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">31.9</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">26.4</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">43.1⁹</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t2\" position=\"float\"><label>Table 2</label><caption><title>Genotype and allele frequencies in groups with and without retinopathy</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"99\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"45\" span=\"1\"/><col width=\"56\" span=\"1\"/><col width=\"51\" span=\"1\"/><col width=\"27\" span=\"1\"/><col width=\"54\" span=\"1\"/><col width=\"54\" span=\"1\"/><col width=\"54\" span=\"1\"/><thead><tr><th rowspan=\"3\" valign=\"middle\" align=\"center\" scope=\"col\" colspan=\"1\">  <bold>Polymorphism</bold></th><th valign=\"middle\" colspan=\"4\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>Genotype distribution (%)</bold><hr/></th><th valign=\"middle\" colspan=\"4\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>Allele distribution (%)</bold><hr/></th></tr><tr><th rowspan=\"2\" valign=\"middle\" colspan=\"1\" align=\"left\" scope=\"colgroup\"/><th rowspan=\"2\" valign=\"middle\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>DM (n=127)</bold></th><th valign=\"middle\" colspan=\"2\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>DR subtypes</bold><hr/></th><th rowspan=\"2\" valign=\"middle\" align=\"left\" scope=\"col\" colspan=\"1\"/><th rowspan=\"2\" valign=\"middle\" align=\"center\" scope=\"col\" colspan=\"1\"><bold>DM (n=254)</bold></th><th valign=\"middle\" colspan=\"2\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>DR subtypes</bold><hr/></th></tr><tr><th valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>NPDR (n=144)</bold></th><th valign=\"middle\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>PDR (n=72)</bold></th><th valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"colgroup\" rowspan=\"1\"><bold>NPDR (n=288)</bold></th><th valign=\"middle\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>PDR (n=144)</bold></th></tr></thead><tbody><tr><td rowspan=\"6\" valign=\"middle\" align=\"center\" scope=\"row\" colspan=\"1\"><italic>EDN1</italic> Lys198Asn<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">Lys/Lys<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">66 (52.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">109 (75.7)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">41 (56.9)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">Lys<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">179 (70.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">249 (86.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">112 (77.8)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">Lys/Asn<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">47 (37.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">31 (21.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">30 (41.7)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">Asn<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">75 (29.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">39 (13.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">32 (22.2)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">Asn/Asn<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">14 (11.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">4 (2.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">1 (1.4)<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">p value<hr/></td><td valign=\"middle\" colspan=\"3\" align=\"left\" rowspan=\"1\">all DR versus DM: 0.0002<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" colspan=\"3\" align=\"left\" rowspan=\"1\">all DR versus DM: 0.007<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"left\" scope=\"row\" rowspan=\"1\"><hr/></td><td valign=\"middle\" colspan=\"3\" align=\"left\" rowspan=\"1\">NPDR versus DM: 0.0001<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" colspan=\"3\" align=\"left\" rowspan=\"1\">NPDR versus DM: 0.00005<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"left\" scope=\"row\" rowspan=\"1\"><hr/></td><td valign=\"middle\" colspan=\"3\" align=\"left\" rowspan=\"1\">PDR versus NPDR: 0.008<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td rowspan=\"3\" valign=\"middle\" align=\"center\" scope=\"row\" colspan=\"1\"><italic>eNOS</italic> Glu298Asp<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">Glu/Glu<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">98 (77.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">110 (76.9)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">55 (76.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">Glu<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">224 (88.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">249 (87.1)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">127 (88.2)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">Glu/Asp<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">28 (22.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">29 (20.3)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">17 (23.6)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">Asp<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">30 (11.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">   39 (12.9)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">17 (11.8)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">Asp/Asp<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">1 (0.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">4 (2.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">0 (0)<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td rowspan=\"3\" valign=\"middle\" align=\"center\" scope=\"row\" colspan=\"1\"><italic>ITGA2 </italic>BgI II<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">BgI II (-/-)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">65 (51.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">76 (52.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">33 (45.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">-<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">184 (72.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">211(73.3)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">100 (69.4)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">BgI II (-/+)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">54 (42.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">59 (41.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">34 (47.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">+<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">70 (27.6)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">77 (26.7)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">44 (30.6)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">BgI II (+/+)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">8 (6.3)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">9 (6.3)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">5 (6.9 )<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td rowspan=\"3\" valign=\"middle\" align=\"center\" scope=\"row\" colspan=\"1\"><italic>LTA</italic> IVS1-80C&gt;A<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">CC<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">64 (50.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">76 (53.1)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">31 (43.7)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">C<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">174 (68.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">210 (73.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">94 (66.2)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">CA<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">46 (36.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">58 (40.6)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">32 (45.1)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">A<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">80 (31.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">76 (26.6)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">48 (33.8)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">AA<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">17 (13.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">9 (6.3)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">8 (11.3)<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td rowspan=\"3\" valign=\"middle\" align=\"center\" scope=\"row\" colspan=\"1\"><italic>LTA</italic> IVS1-252A&gt;G<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">AA<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">31 (24.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">30 (20.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">19 (26.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">A<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">127 (50.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">134 (46.5)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">71 (49.3)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">GA<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">65 (51.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">74 (51.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">33 (45.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">G<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">127 (50.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">154 (53.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">73 (50.7)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">GG<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">31 (24.4)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">40 (27.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">20 (27.8)<hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td rowspan=\"3\" valign=\"middle\" align=\"center\" scope=\"row\" colspan=\"1\"><italic>LTA </italic>IVS1-206G&gt;C</td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">GG<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">63 (49.6)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">76 (52.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">34 (47.2)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">G<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">173 (68.1)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">210 (72.9)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">101 (70.1)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">GC<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">47 (37.0)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">58 (40.3)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">33 (45.8)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">C<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">81 (31.9)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">78 (27.1)<hr/></td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">43 (29.9)<hr/></td></tr><tr><td valign=\"middle\" colspan=\"1\" align=\"center\" scope=\"row\" rowspan=\"1\">CC</td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">17(13.4)</td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">10 (6.9)</td><td valign=\"middle\" align=\"center\" rowspan=\"1\" colspan=\"1\">5 (6.9)</td><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"/><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"/><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"/><td valign=\"middle\" align=\"left\" rowspan=\"1\" colspan=\"1\"/></tr></tbody></table></table-wrap>", "<table-wrap id=\"t3\" position=\"float\"><label>Table 3</label><caption><title>Odds ratio adjusted by multivariable logistic regression for the association with diabetic retinopathy in patients with type 2 diabetes</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"203\" span=\"1\"/><col width=\"79\" span=\"1\"/><col width=\"93\" span=\"1\"/><thead><tr><th valign=\"bottom\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Factors</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Adjusted OR</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Adjusted p value</bold></th></tr></thead><tbody><tr><td valign=\"bottom\" colspan=\"3\" align=\"left\" scope=\"col\" rowspan=\"1\">DM versus DR<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\"><italic>EDN1</italic>: Asn/Asn versus Lys/Lys+Lys/Asn<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.19 (0.07-0.53)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.002<hr/></td></tr><tr><td valign=\"bottom\" colspan=\"3\" align=\"left\" scope=\"col\" rowspan=\"1\">NPDR versus PDR<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Age at onset of diabetes</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.94 (0.91-0.97)</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.00001</td></tr></tbody></table></table-wrap>", "<table-wrap id=\"t4\" position=\"float\"><label>Table 4</label><caption><title>Relationship between <italic>EDN1</italic>genotype, age of diabetes mellitus onset, and hypertension</title></caption><table frame=\"hsides\" rules=\"groups\"><col width=\"99\" span=\"1\"/><col width=\"72\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"63\" span=\"1\"/><col width=\"54\" span=\"1\"/><thead><tr><th valign=\"bottom\" align=\"left\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Factors</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Lys/Lys</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Lys/Asn</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>Asn/Asn</bold></th><th valign=\"bottom\" align=\"center\" scope=\"col\" rowspan=\"1\" colspan=\"1\"><bold>p value</bold></th></tr></thead><tbody><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Number of patients <hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">216<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">108<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">19<hr/></td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\"><hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Age of diabetes mellitus onset<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">53.1+/-11.1<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">53.0+/-11.6<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">59.0+/-11.6<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.02<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Hypertensive<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">143 (61.9%)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">74 (32.0%)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">14 (6.0%)<hr/></td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">0.79<hr/></td></tr><tr><td valign=\"bottom\" align=\"left\" scope=\"row\" rowspan=\"1\" colspan=\"1\">Nonhypertensive</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">73 (65.2%)</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">34 (30.4%)</td><td valign=\"bottom\" align=\"center\" rowspan=\"1\" colspan=\"1\">5 (4.5%)</td><td valign=\"bottom\" align=\"left\" rowspan=\"1\" colspan=\"1\"/></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>The table showed the comparisons of the clinical and metabolic characteristics among diabetes mellitus controls (DM) and diabetic retinopathy (DR) and DR subtypes. All p values were computed by χ2 or student t-tests. DR group showed a significantly higher level of HbA1c and higher percentage of patients receiving insulin treatment than controls (1: p=0.046; 2: p&lt;0.00001). By Tukey’s test for Post Hoc multiple comparisons (the Bonferroni corrected significance level was 0.025=0.05÷2), the significant difference of HbA1c level was only found between nonproliferative diabetic retinopathy (NPDR) and controls (2: p=0.001) and the significant higher percentage of the patients receiving insulin treatment was only due to proliferative diabetic retinopathy (PDR) group (8: p=0.00009). The PDR group was slightly younger than NPDR (4: p=0.001) and controls (5: p=0.00005) and with earlier onset of diabetes than other two groups (6: p=0.0008; 7: p=0.000003). The presence of diabetic macular edema was evident in PDR compared to NPDR (9: p=0.001).</p></table-wrap-foot>", "<table-wrap-foot><p>The genotype and allele frequency distributions are shown for diabetic retinopathy (DR) and controls (diabetes mellitus without diabetic retinopathy, DM). Only the <italic>endothelin-1 (EDN1)</italic> Lys198Asn genotype and its allele distributions showed a statistically significant difference between DR patients and controls at the Bonferroni corrected significance level of 0.0083=0.05÷6. For comparing DR subtypes and DM distributions of only the <italic>EDN1</italic> Lys198Asn polymorphism, the Bonferroni corrected significance level was 0.017=0.05÷3.</p></table-wrap-foot>", "<table-wrap-foot><p>The relationship between diabetic retinopathy (DR) and its affected factors were tested by multivariable logistic regression analysis. The presence of DR was regarded as the dependent variable, and independent variables included genotype of <italic>EDN1, eNOS, ITGA</italic>, and <italic>LTA</italic>, gender, age, age at onset of diabetes, HbA1c, therapy of insulin, hypertension, and hyperlipidemia. Only items with p values &lt;0.05 were listed. After adjustment for age, age at onset of diabetes and insulin therapy, etc, <italic>EDN1</italic> Asn/Asn was the only independent protective factor for DR and younger onset age of diabetes was a significant risk factor for nonproliferative diabetic retinopathy (NPDR) progressing to proliferative diabetic retinopathy (PDR).</p></table-wrap-foot>", "<table-wrap-foot><p>The onset age of diabetes mellitus (DM) and hypertension frequency were compared among different <italic>EDN1</italic> genotypes in all diabetic patients. The age of DM onset of patients with the Asn/Asn genotype was about six years later than patients with other genotypes (p=0.02) and Lys198Asn genotype distributions were not different between hypertensive and nonhypertensive subjects (p=0.79).</p></table-wrap-foot>" ]

[]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Each year an estimated 1.2 million people die and a further 20–50 million are injured worldwide from road traffic accidents: a major public health problem [##UREF##0##1##]. In Vietnam, road traffic injuries are now the leading cause of fatal and non-fatal injuries [##UREF##1##2##].</p>", "<p>Motorcycle users in Vietnam are most vulnerable to road traffic injuries. Motorcycles account for approximately 95% of the total number of vehicles in Vietnam [##UREF##0##1##]. In 2001, there were an estimated 105 motorcycles per 1,000 population, increasing to 193 by 2005. Such an increase in motorcycle use has had significant effects on the burden of injury from road traffic injuries and the economic costs of treatment and the sequellae of injury. A community-based survey undertaken in 2001 in all eight regions of Vietnam showed that motorcycle users accounted for 51.3% of all non-fatal road traffic injuries, a rate of 734 per 100,000 population [##UREF##1##2##].</p>", "<p>According to the World Health Organization, traumatic brain injury (TBI) is the main cause of fatal and non-fatal injury for motorcycle users [##UREF##0##1##]. In poor countries, economic losses caused by TBI due to expenditure for prolonged treatment, loss of productivity or income due to disability or death commonly tip households into a spiral of poverty [##UREF##2##3##]. No hospital-based or community epidemiological data on TBI in motorcycle users are available in Vietnam. However, it is likely that the burden caused by TBI to the country is significant, given the very low use of motorcycle helmets and the dominance of motorcycles as the main form of transport.</p>", "<p>Mandatory motorcycle helmet use is regarded as the single most effective approach for the prevention of TBI among motorcycle users in both developed and developing countries [##UREF##0##1##]. Wearing a helmet reduces the incidence, severity and mortality rates of TBI in motorcycle accidents, ranging from 20% to 45% reduction of fatal and serious head injury [##REF##12966016##4##]. In Vietnam, a mandatory helmet law was introduced for all roads on 15 December 2007, two years after this study was completed. Prior to this, it was mandatory to wear a helmet only on selected roads, mainly those designated as national roads, but the enforcement of that policy was poor. Nationwide, in 2001, only 7.4% of male and 4.1% of female regular motorcycle users reported using a helmet [##UREF##3##5##]. At the time of writing, compliance with mandatory helmet use appears high, though issues of helmet quality have been raised.</p>", "<p>This study estimates the costs of non-fatal TBI in motorcycle users not wearing helmets in Hanoi, Vietnam, in their first year post-injury. The study examined costs from the perspective of the injured patients and their families. These included direct costs associated with treatment at hospital and at home; indirect costs associated with the loss of productivity; and intangible costs associated with the loss of quality of life. Although the social perspective is considered the most appropriate viewpoint to adopt in economic evaluations [##UREF##4##6##], this was not possible, due to the lack of available data and the currently limited role of health and social insurance in Vietnam.</p>" ]

[ "<title>Methods</title>", "<p>The study was undertaken at VietDuc Hospital, Hanoi, the major trauma centre in North Vietnam. Patients discharged between January 2005 and mid July 2005 with a history of TBI were enrolled in the study, based on the following inclusion criteria: aged 16 years and over; residential address in Hanoi; discharged at least 6 months before the commencement of the study; motorcycle driver or passenger not using a helmet when the accident happened; and no other serious injuries, complications or compounding diseases. Patients were further classified into three levels of TBI severity according to the Glasgow Coma Scale (GCS) at admission: severe (&lt; 9), moderate (9–12) and minor (13–15).</p>", "<title>Cost analysis methods</title>", "<p>Direct and indirect costs were quantified in economic terms. The direct cost method was used to estimate the costs associated with treatment, including household expenditure on all goods and services relating to the medical care of patients. The human capital method was used for the calculation of loss of productivity for the injured and their carers [##REF##9210552##7##, ####UREF##5##8##, ##REF##11928974##9##, ##REF##15860510##10####15860510##10##]. Although valuing the intangible costs of injury is difficult and often contentious [##UREF##0##1##], the health status index method was chosen for the assessment of the health-related quality of life, using years lived with a disability (YLD) as a proxy. The intangible costs, in this case, were <italic>not </italic>estimated in economic terms.</p>", "<p>Structured questionnaires were used to obtain costs associated with the treatment of TBI, and productivity loss. For direct costs, respondents were asked to recall medical and non-medical costs at all health facilities and at home. Where interviews occurred less than one year following injury, projections of costs of home care to one year were estimated, based on current patterns.</p>", "<p>Loss of productivity for patients and caregivers (indirect costs) were quantified in monetary terms using both individual actual income and per capita income for urban areas in Vietnam in 2004 (VND 815,000/month, equivalent to USD 51.5 in 2004). As with direct costs, for patients less than one year post-injury, projections of time off work were based on averages for each severity level. In the severe category, patients who had lost more than the mean number of weeks work at interview were assumed to be incapable of resuming work for the remainder of the year. The opportunity costs for loss of normal activity in students, the elderly or un-paid home-makers were estimated using the national per capita income in 2004 (VND 484,000, equivalent to USD 30.5).</p>", "<p>The European Quality of Life instrument (EQ6D) instrument was translated, back translated and trialed, then used in patient interviews to measure changes in quality of life for discharged patients. This instrument uses six dimensions of health: mobility, pain/discomfort, self-care, anxiety/depression, usual activities and cognition [##UREF##6##11##]. Health status for each patient was represented by a single index with 6 digits. This was converted into a predicted disability weight (DW) under the \"Disability Adjusted Life Years (DALY) form\" using the Dutch Disability regression model ranging from \"zero\" for good health to \"one\" for death [##UREF##6##11##]. The YLD caused by TBI in one year was then calculated using the basic formula applied by The Global Burden of Disease and Injury: YLD = I × DW × L where I is the number of accident cases in the reference period, L is the average duration of disability [##UREF##7##12##]. In this case, the YLD of one patient with TBI was: 1 (case) × the predicted DW × 1 (year) with an assumption that the health state assessed at interview was representative of the patient's health state for one year post-injury.</p>", "<p>Each cost component was calculated by three levels of TBI severity. The ANOVA test was used to compare the variance of the three level averages.</p>" ]

[ "<title>Results</title>", "<title>Demographic characteristics of study population</title>", "<p>Discharge records from VietDuc hospital showed 61 patients met the inclusion criteria. Initial telephone contacts with these 61 patients and their families showed that five patients were deceased, ten were not contactable or had relocated, and three were wearing helmets at the time of the accident. In two cases, motorcyclists were injured as a result of inter-personal violence, rather than motorcycle related incidents. Six patients refused to participate, and the remaining 35 patients were recruited to the study. Four of these had exceptional insurance or other third party financial support. As a result, total treatment costs and lengths of stay were extremely high in comparison with the remaining cases in the same level of severity, and these cases were considered as outliers for the purposes of this study.</p>", "<p>Seventy one percent (22/31) of the study population was male. The mean age for the group was 33.2 years, with almost half (45.2%) between 20 to 29 years. Students accounted for 22.6%, followed by manual labourers in the industry/processing/handicraft sector (16.1%). Three-quarters (74%) were motorcycle drivers at the time of the accident. The GCS based severity of injury was evenly distributed: severe (10), moderate (11) and minor (10).</p>", "<title>Direct costs</title>", "<p>Severity of injury correlated directly with length of stay at health facility and length of medication-use at home respectively: severe (3.2 week and 35.9 weeks); moderate (2 weeks and 17.5 weeks); and minor (2 weeks and 15.3 weeks). Similarly, direct costs, both in hospital and at home, increased with the severity of TBI (Table ##TAB##0##1##).</p>", "<p>Costs at home included medication (including tonics and \"therapeutic\" foods) and rehabilitation in the form of physical therapy to improve health status. The low values for rehabilitation reflect the limited resources available to families, and their limited accessibility. Costs for ongoing home visits by therapists are not financially sustainable in this population. Although home treatment costs rose with severity, they remained substantially less than hospital costs at all levels. The use of rehabilitation services at home or though out-patient attendance was minimal: only four cases reported post-discharge rehabilitation services, accounting for a minor component of overall home costs.</p>", "<title>Indirect costs</title>", "<p>The post-injury period was marked by a diminished ability to work or to conduct normal activities. Sixty percent of patients suffering severe TBI could not resume work or implement their usual daily activities again after 6 months. In the moderate group, twenty percent had persisting disability at this point, though all minor injury patients had returned to normal functionality. Where patients returned to work, it was frequently at lower levels of productivity with commensurate reductions in salary levels. Twenty percent lost their pre-injury role of family primary income earner.</p>", "<p>Eighty-percent of discharged patients in the sample needed support from a caregiver at home after the accident. Where possible, households were strategic in minimizing the loss of household income by selecting caregivers with the lowest earning capacity in the family. In 35.5% of households, care-givers were non-working family members or the very old (home-makers, retired, unemployed or students). For 45.2% of caregivers, their income pre-injury was less than the national per capital income (USD 30.5 in 2004), and in 64% of cases, the selected caregiver had an income less than the capita income for urban areas (USD 53 in 2004). The withdrawal of a child from school to provide care for an injured adult or to work in order to compensate for lost income, represents a substantial opportunity cost, not reflected in the calculations of income foregone. Despite efforts to minimize income lost, opportunity costs for households from providing care were significant, and caregivers were not always available – accounting for the substantial difference between time loss for the injured and their caregivers (Table ##TAB##1##2##).</p>", "<p>Estimates of loss of productivity using the individual's actual income produced average indirect costs that were much higher than the estimates based on per capital income for urban areas (Table ##TAB##2##3##). The advantage of using per capita income to estimate lost productivity, instead of the actual (known) income, is that it eliminates the variation of income evident in small samples. For both estimates, the loss of productivity rose with severity, though using per capita income the estimated losses were more conservative.</p>", "<title>Intangible costs</title>", "<p>Changes in quality of life were measured using the EQ6D instrument, administered to patients (or if unable to respond, to caregivers) in a questionnaire format. Disability again correlated with the severity of injury at admission. Patients with severe TBI were most compromised in their usual activities, with higher levels of anxiety and problems of cognition and mobility. All members of the moderate group faced disruption in their usual activities; with increased pain, anxiety and affected cognition. While none of the minor TBI patients faced difficulty in mobility and self-care, anxiety and pain were persisting problems, with continuing compromise of usual activities and cognition (Figure ##FIG##0##1##).</p>", "<p>The average disability weights for TBI patients were assessed pre- and post-injury at the time of interview. While all patients shared the same disability weight of zero pre-injury, the disparity post injury reflected the level of severity (Figure ##FIG##1##2##). In term of intangible costs, the health related quality of life of the patients in the first year post-injury was reduced, resulting in an average year of life lost due to disability of 0.46 for severe, 0.25 for moderate and 0.15 year for minor TBI.</p>", "<title>Impact of TBI on family economic status</title>", "<p>Eighty four percent of households in the sample faced treatment costs that accounted for more than 40% of the household capacity to pay for health care. The capacity to pay is determined by the remaining income of household after expenditure for basic subsistence needs. For this study, household health care expenditure that accounted for more than 40% of the household capacity to pay was taken to be catastrophic [##UREF##2##3##]. Only 12% of the households could afford to pay the cost associated with the treatment of TBI from household savings. The remaining households had to mobilize money for this payment from two or three sources, such as borrowing from relatives, using accumulated savings and/or selling assets, and resulting in financial stress at least in the medium term. Together with savings, support from relatives seemed to be the principle resource protecting households from catastrophic health expenditure.</p>" ]

[ "<title>Discussion</title>", "<p>This study is the first estimate of the costs of non-fatal TBI in motorcycle users not wearing helmets in Vietnam, in their first year post-injury. Given the limited coverage of health and social insurance in Vietnam, the study focused on out of pocket expenses and foregone earnings for both patients and their families. The cut-off after one year is a limitation of this study and contributes to an underestimation of the true total cost over a lifetime.</p>", "<p>This study shows a large variance in the costs across individuals in the same level of severity, as seen in previous studies internationally and nationally [##REF##9041891##13##, ####REF##10206600##14##, ##REF##11782840##15##, ##UREF##8##16####8##16##], but confirms the significant level of financial burden that TBI imposes on families. It clearly demonstrates the direct correlation between level of severity of injury at admission and subsequent component costs, and the risk of catastrophic health expenditure for affected families.</p>", "<p>As a pilot study using selected cost analysis methods, the study suggests that the use of per capita income to value the loss of productivity of TBI in Vietnam may underestimate indirect costs compared to estimates based on the individual's actual income. This reflects the reality that the majority of victims of motorcycle injuries are males within the economically productive age-group, and likely to be principal income earners for their households. As a result, their average income tends to be higher than the national income per capita. Since the costs of TBI in this study are confined to non-fatal TBI without complex complications, they must be considered as conservative estimates. Strategies such as withdrawing children from school to care for the injured, or to work in order to compensate for lost income have far reaching social consequences. The absence of accessible and affordable long-term rehabilitation is another concern concealed in these conservative estimates.</p>" ]

[ "<title>Conclusion</title>", "<p>This study has shown that all three components costs of TBI were high; the direct cost accounted for the largest proportion, with costs rising with the severity of TBI. The results suggest that the burden of TBI can be catastrophic for families because of high direct costs, significant time off work for patients and caregivers, and impact on health-related quality of life. Further research is warranted to explore the actual social and economic benefits of mandatory helmet use.</p>", "<p>International experience shows that relatively affordable interventions such the implementation of mandatory helmet wearing for motorcycle riders result in the reduction of tangible and intangible costs to individuals, families and society [##UREF##0##1##]. Early unpublished data suggests that this is occurring in Vietnam. With the December 2007 introduction of mandatory helmet use, further research is now required to calculate the benefits of motorcycle helmet use in Vietnam together with research exploring compliance, quality standards and the development appropriate helmets for children. Such research will require larger sample sizes at each level of severity of TBI, covering different provinces and cities, targeting both use and non-use of helmets, and comparing different cost analysis methods. This research, however, already demonstrates a level of cost to individuals and households that is in many cases catastrophic, but which can be reduced through recognized policy interventions.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Road traffic accidents are the leading cause of fatal and non-fatal injuries in Vietnam. The purpose of this study is to estimate the costs, in the first year post-injury, of non-fatal traumatic brain injury (TBI) in motorcycle users not wearing helmets in Hanoi, Vietnam. The costs are calculated from the perspective of the injured patients and their families, and include quantification of direct, indirect and intangible costs, using years lost due to disability as a proxy.</p>", "<title>Methods</title>", "<p>The study was a retrospective cross-sectional study. Data on treatment and rehabilitation costs, employment and support were obtained from patients and their families using a structured questionnaire and The European Quality of Life instrument (EQ6D).</p>", "<title>Results</title>", "<p>Thirty-five patients and their families were interviewed. On average, patients with severe, moderate and minor TBI incurred direct costs at USD 2,365, USD 1,390 and USD 849, with time lost for normal activities averaging 54 weeks, 26 weeks and 17 weeks and years lived with disability (YLD) of 0.46, 0.25 and 0.15 year, respectively.</p>", "<title>Conclusion</title>", "<p>All three component costs of TBI were high; the direct cost accounted for the largest proportion, with costs rising with the severity of TBI. The results suggest that the burden of TBI can be catastrophic for families because of high direct costs, significant time off work for patients and caregivers, and impact on health-related quality of life. Further research is warranted to explore the actual social and economic benefits of mandatory helmet use.</p>" ]

[ "<title>Abbreviations</title>", "<p>DALY: Disability Adjusted Life Years; DW: Disability weight; EQ6D: The European Quality of Life Instrument – 6 Dimensions; GCS: Glasgow Coma Scale; TBI: Traumatic Brain Injury; VND: Vietnamese Dong (currency; USD = 15,850 VND, July 2005); WHO: World Health Organization; YLD: Years lost due to disability.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>HTMH developed the literature review, clarified the research objective, developed instruments, interviewed the subjects, analysed the data, and completed the first draft. TLP developed the literature review, clarified the research objective, developed instruments, interviewed the subjects, analysed the data and assisted with the first draft. TTNV developed the literature review, clarified the research objective, developed instruments, interviewed the subjects, analysed the data and assisted with the first draft. PKN negotiated local permission for the research, assisted in data analysis, and reviewed the draft. CMD and PSH conceived the study, assisted in the study design, instruments development and data analysis, reviewed and edited the draft. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>The researchers would like to acknowledge the University of Queensland and Atlantic Philanthropies for financial support and student scholarships, and the director of VietDuc hospital, Dr Nguyen Tien Quyet and Dr Nguyen Duc Chinh, Vice-Head of Planning Department for their assistance with access to data and patients.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Change in ability to function in the six health dimensions.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Predicted average disability weight under DALY form.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>One-year costs associated with treatment by level of traumatic brain injury <italic>(Unit: USD, 1USD = 15,850 Vietnam dong)</italic></p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"9\"><bold>Level of severity by GCS</bold></td></tr><tr><td/><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\" colspan=\"3\"><bold>Severe</bold></td><td align=\"center\" colspan=\"3\"><bold>Moderate</bold></td><td align=\"center\" colspan=\"3\"><bold>Minor</bold></td></tr><tr><td/><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td></tr></thead><tbody><tr><td align=\"center\">At all health facilities (*)</td><td align=\"center\">1571.3</td><td align=\"center\">285.9</td><td align=\"center\">1313.9</td><td align=\"center\">1060.3</td><td align=\"center\">156.5</td><td align=\"center\">1205</td><td align=\"center\">708.3</td><td align=\"center\">104.7</td><td align=\"center\">789.3</td></tr><tr><td align=\"center\">At home (*)</td><td align=\"center\">793.4</td><td align=\"center\">135.1</td><td align=\"center\">737.3</td><td align=\"center\">329.9</td><td align=\"center\">86.8</td><td align=\"center\">227.4</td><td align=\"center\">140.7</td><td align=\"center\">43.6</td><td align=\"center\">94.6</td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"center\"><bold>Total (*)</bold></td><td align=\"center\"><bold>2364.7</bold></td><td align=\"center\"><bold>336.6</bold></td><td align=\"center\"><bold>2201.3</bold></td><td align=\"center\"><bold>1390.1</bold></td><td align=\"center\"><bold>132.4</bold></td><td align=\"center\"><bold>1457.4</bold></td><td align=\"center\"><bold>849.0</bold></td><td align=\"center\"><bold>110.0</bold></td><td align=\"center\"><bold>861.8</bold></td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Time off work or normal activity by level of severity of traumatic brain injury (Unit: weeks)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Time off work or normal activity</bold></td><td align=\"center\" colspan=\"9\"><bold>Level of severity by GCS</bold></td></tr><tr><td/><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\" colspan=\"3\"><bold>Severe</bold></td><td align=\"center\" colspan=\"3\"><bold>Moderate</bold></td><td align=\"center\" colspan=\"3\"><bold>Minor</bold></td></tr><tr><td/><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td></tr></thead><tbody><tr><td align=\"left\">By patient</td><td align=\"center\">38.4</td><td align=\"center\">4.8</td><td align=\"center\">40.3</td><td align=\"center\">18.9</td><td align=\"center\">5.1</td><td align=\"center\">13</td><td align=\"center\">11.5</td><td align=\"center\">2.2</td><td align=\"center\">13</td></tr><tr><td align=\"left\">BY caregiver</td><td align=\"center\">15.5</td><td align=\"center\">3.1</td><td align=\"center\">11.2</td><td align=\"center\">7.1</td><td align=\"center\">1.4</td><td align=\"center\">7.4</td><td align=\"center\">5.5</td><td align=\"center\">1.4</td><td align=\"center\">4.3</td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"center\"><bold>Total (*)</bold></td><td align=\"center\"><bold>54.0</bold></td><td align=\"center\"><bold>6.9</bold></td><td align=\"center\"><bold>59.5</bold></td><td align=\"center\"><bold>26.0</bold></td><td align=\"center\"><bold>6.2</bold></td><td align=\"center\"><bold>20</bold></td><td align=\"center\"><bold>17.1</bold></td><td align=\"center\"><bold>3.0</bold></td><td align=\"center\"><bold>21</bold></td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Loss of productivity estimates using actual income and per capita income for urban area <italic>(Unit: USD, 1USD = 15,850 Vietnam dong, per capita income for urban area was VND 840,000 per month)</italic></p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Indirect costs for one year</bold></td><td align=\"center\" colspan=\"9\"><bold>Level of severity by GCS</bold></td></tr><tr><td/><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\" colspan=\"3\"><bold>Severe</bold></td><td align=\"center\" colspan=\"3\"><bold>Moderate</bold></td><td align=\"center\" colspan=\"3\"><bold>Minor</bold></td></tr><tr><td/><td colspan=\"9\"><hr/></td></tr><tr><td/><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td><td align=\"center\">Mean</td><td align=\"center\">SE</td><td align=\"center\">Median</td></tr></thead><tbody><tr><td align=\"left\" colspan=\"10\"><bold><italic>Actual income based indirect costs</italic></bold></td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"left\">By patient</td><td align=\"center\">1108.6</td><td align=\"center\">324.2</td><td align=\"center\">833.4</td><td align=\"center\">303.0</td><td align=\"center\">85.1</td><td align=\"center\">164.0</td><td align=\"center\">461.3</td><td align=\"center\">171.9</td><td align=\"center\">200.3</td></tr><tr><td align=\"left\">By caregiver</td><td align=\"center\">303.8</td><td align=\"center\">129.3</td><td align=\"center\">132.9</td><td align=\"center\">102.0</td><td align=\"center\">31.4</td><td align=\"center\">60.9</td><td align=\"center\">77.4</td><td align=\"center\">27.1</td><td align=\"center\">40.9</td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"left\"><bold>Total (*)</bold></td><td align=\"center\"><bold>1412.4</bold></td><td align=\"center\"><bold>366.5</bold></td><td align=\"center\"><bold>1045.3</bold></td><td align=\"center\"><bold>405.0</bold></td><td align=\"center\"><bold>106.2</bold></td><td align=\"center\"><bold>239.0</bold></td><td align=\"center\"><bold>538.5</bold></td><td align=\"center\"><bold>194.0</bold></td><td align=\"center\"><bold>263.5</bold></td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"left\" colspan=\"10\"><bold><italic>Per capita urban income based indirect costs</italic></bold></td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"left\">By patient</td><td align=\"center\">455.8</td><td align=\"center\">56.5</td><td align=\"center\">478</td><td align=\"center\">223.7</td><td align=\"center\">60.6</td><td align=\"center\">154.2</td><td align=\"center\">136.4</td><td align=\"center\">25.7</td><td align=\"center\">154.2</td></tr><tr><td align=\"left\">By caregiver</td><td align=\"center\">183.5</td><td align=\"center\">36.6</td><td align=\"center\">132</td><td align=\"center\">84.0</td><td align=\"center\">17.0</td><td align=\"center\">87.8</td><td align=\"center\">65.4</td><td align=\"center\">17.1</td><td align=\"center\">50.9</td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"left\">Total(*)</td><td align=\"center\">639.4</td><td align=\"center\">82.2</td><td align=\"center\">702.5</td><td align=\"center\">307.8</td><td align=\"center\">73.1</td><td align=\"center\">242.0</td><td align=\"center\">201.8</td><td align=\"center\">35.7</td><td align=\"center\">245.4</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>(*) Anova: p &lt; 0.05</p></table-wrap-foot>", "<table-wrap-foot><p>(*) Anova: p &lt; 0.05</p></table-wrap-foot>", "<table-wrap-foot><p>(*) Anova: p &lt; 0.05</p><p>¹General Statistic Office: The Vietnam Living Standard Survey in 2004 showed that per capita income for urban area in 2004 was VND 840,000 per month; national per capita income was VND 484,000 per month.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1478-7547-6-17-1\"/>", "<graphic xlink:href=\"1478-7547-6-17-2\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Use of complementary and alternative medicine (CAM) among people with cancer has been found to be common and widespread with estimates ranging between 7–64% [##REF##9708945##1##]. Apart from probable differences in CAM utilization in different settings and countries, this variation is partially related to differences in the instrumentalization of CAM definitions [##REF##9708945##1##]. The wide spectrum of therapies often considered as within the CAM domain is indicated for example in the broad definition by the Cochrane Collaboration, which includes \"all such practices and ideas self-defined by their users as preventing or treating illness or promoting health or well-being\" if they are not part of the \"politically dominant health system of a particular society or culture\" at the time [##REF##10480829##2##]. The growing number of CAM modalities subjected to efficacy studies is yet another indication of the breadth of this area. Studies are also called for to investigate potential risks in CAM use, including side-effects and interactions between CAM preparations and BHC treatments.</p>", "<p>Considering the frequency of CAM use and the wide spectrum of therapies included, it is not surprising that studies suggest that there may be important differences among CAM users, e.g. with regard to types of CAM used and how therapies are combined [##REF##17956965##3##]. A number of CAM taxonomies have been proposed to distinguish CAM use by type of therapy. Tataryn [##REF##12614539##4##] suggests categorization according to basic assumptions of health and disease underlying each therapy. Jones [##REF##16296930##5##] on the other hand, argues that it is more clinically relevant to categorize CAM according to primary mode of therapeutic action. In line with Jones' suggestion, the influential N.I.H. National Center of Complementary and Alternative Medicine (NCCAM) in the U.S. [##UREF##0##6##], describes CAM in five categories: <italic>Alternative Medical Systems</italic>; <italic>Mind and Body Interventions</italic>; <italic>Biologically Based Therapies</italic>; <italic>Manipulative and Body Based Therapies; </italic>and <italic>Energy Therapies</italic>, which is further distinguished into the subcategories <italic>Biofield therapies </italic>and <italic>Bioelectromagnetic-based therapies</italic>. Many CAM utilization studies have used this framework when assessing CAM use.</p>", "<p>Several studies have found that cancer patients utilize therapies from all NCCAM categories, often using multiple CAM therapies during their disease trajectory [##REF##9708945##1##,##REF##16270107##7##,##REF##16504038##8##]. In their study of women with breast cancer, Balneaves et al [##REF##16813510##9##] further distinguished \"committed\" CAM use, which they defined as use of a large number of CAM therapies, coupled with extensive time, energy and financial resources spent on this use. With consideration given the dedicated use and the large number of CAM therapies involved, committed CAM use may be particularly relevant to further investigate since it may be associated with elevated risks [##REF##12458989##10##] and unclear benefits. The possibility for interactions makes it important to investigate not only the number of therapies used, but also relationships among them.</p>", "<p>The data presented here provides a complement to existing literature by using detailed unstructured narratives from CAM users as a basis for exploring CAM use. Multiple methodological approaches were used to map and explore patterns of CAM use, including use of therapies across different categories, in a self-selected group hypothesized to represent one type of committed use. This material derives from the Swedish portion of a Nordic collaboration, investigating exceptional experiences around CAM and cancer [##UREF##1##11##], with other reports delving further into the experience of the participants [##REF##17619069##12##].</p>" ]

[ "<title>Methods</title>", "<p>The data presented here are drawn from a larger project exploring different stakeholder perspectives of exceptional CAM experiences, inspired by a similar program by National Cancer Institute (N.C.I.), N.I.H., U.S. [##UREF##2##13##]. We used an inductive approach to allow insight into individuals' perspectives of exceptional experiences in connection to CAM use, with a critical incident design to locate \"extreme\" and \"extraordinary\" accounts [##REF##13177800##14##].</p>", "<title>Data collection</title>", "<p>After approval by the Karolinska Institutet research ethics review board, mass media was used to invite reports about CAM use in connection to experiences perceived as exceptional, in the sense of unexpected or unusual improvement or deterioration of the health of people with cancer. No further predefinition was provided for what was considered exceptional or for what was considered a CAM therapy, in order to explore stakeholders' own conceptualizations.</p>", "<p>Thirty-eight people with cancer were interviewed between April 2004 and November 2005. Twenty-eight participants contacted the researchers actively themselves, while the remaining 10 cases were first reported by a CAM provider after receiving patient consent. We conducted open interviews that generally lasted one to three hours, to encourage participants' accounts of their CAM use and illness experiences. Thirty-six participants consented to the interview being audio-recorded and later transcribed verbatim. Detailed interview notes were taken after consent from two participants who were uncomfortable with audio-recording. All study participants were given fictitious names.</p>", "<title>Descriptive Analysis</title>", "<p>Following principles of manifest content analysis [##REF##14769454##15##,##UREF##3##16##], all CAM mentioned by the participants in interview transcripts, e-mails and letters were identified and coded using the qualitative data analysis program NVivo [##UREF##4##17##]. All therapies defined by participants as utilized as complements or alternatives to BHC treatments in connection to their cancer were first sorted into one of the five NCCAM categories. Fifty accounts did not fit into any of the existing NCCAM categories, despite participants' narrative reports of their therapeutic nature. After further analysis of these accounts, two additional categories were formulated, <italic>Spiritual/health literature </italic>and <italic>Treatment centers combining CAM and BHC </italic>(referred to also as <italic>Treatment centers</italic>). This resulted in seven categories of CAM therapies. A CAM educator external to this project later confirmed categorization. Inter-rater reliability was high, with only two inconsistencies in categorization among the 148 different therapies.</p>", "<p>Demographic and disease characteristics were extracted from the interview data and are presented by minimum (min) and maximum (max) values, with inter-quartile ranges (IQR) shown.</p>", "<title>Explorative statistical analysis</title>", "<p>Principal component analysis was used to explore correlations between usage of CAM categories. New variables (principal components, PCs), were introduced to reduce the dimensionality of the usage pattern while retaining as much as possible of the variation in the original data. PCs are computed as weighted sums of the original variables, where the weights of the original variables are referred to as loadings. By definition, the first PC expresses the greatest amount of variation in the data, the second PC the next largest amount, and so on.</p>", "<p>By applying the weights of the original variables, i.e. the loadings, to the values observed for each participant, we also computed the scores along each PC for these participants [##UREF##5##18##]. Calculation of loadings was based on correlations between therapy counts in the seven CAM categories. Using correlations here corresponds to a standardization of the observed therapy counts to the mean of zero and standard deviation one across all participants. In this manner, all categories have the same weight in calculating the loadings. The use of correlations avoids biasing the PC analysis towards categories comprising more therapy modalities at the expense of categories with less reported modalities [##UREF##5##18##].</p>", "<p>Bootstrap confidence intervals were computed for the loadings of the original variables to allow estimation of standard errors and confidence intervals without strong parametric assumptions [##UREF##6##19##]. The bootstrapped confidence intervals served as guidelines for the selection of the number of PCs to be retained, and for the interpretation of the loadings. No formal inference to a larger underlying population is intended in this exploration.</p>", "<p>Two types of usage patterns indicated by the first two PCs were analyzed. One such usage pattern was calculated based on the magnitude and sign of the contribution of each CAM category to the PCs. A second pattern was calculated based on each user's score on the retained PCs. These calculations resulted in individual usage patterns, displayed in a plot to identify groups of participants with similar usage profiles. To aid the visual impression, an explorative k-means clustering analysis of the scores was performed. The number of clusters was chosen to maximize a measure of average separation between members of different clusters (silhouette width as described in [##UREF##7##20##]).</p>" ]

[ "<title>Results</title>", "<title>Demographic and disease characteristics of participants</title>", "<p>Tables ##TAB##0##1## and ##TAB##1##2## present sample characteristics as reported by the study participants, with resulting gaps in information on occupational status and level of education. At the time of interview, the study sample ranged in age between 36 and 85 years (median = 55, IQR = 48–63 years) and was composed primarily of women, with half the participants living with a partner. The dominance of women is reflected in the diagnostic pattern of the sample, with 24 of the 38 participants reporting primary breast or gynecologic tumors. Fifteen participants reported having metastasized cancer. The time between first cancer diagnosis and time of interview ranged from one to 32 years (median = 5 years, IQR = 1–13 years).</p>", "<title>CAM use in relation to BHC treatment and disease characteristics</title>", "<p>Seventeen participants reported completing BHC treatments according to medical recommendation, six participants did not discuss this issue, and 15 of the 38 participants said they had not completed recommended BHC treatments. In 14 accounts this decision was framed as a choice either against or without BHC medical advice, and in one case as a decision made in conjunction with BHC advice. Participants who reported completing BHC treatment used a median of three different CAM therapies (min = 1, max = 20, Q₁= 2, Q₃= 7) from a median of two different CAM categories (min = 1, max = 7, Q₁= 2, Q₃= 4.5), while participants who did not complete BHC treatments reported using a median of seven different CAM therapies (min = 1, max = 26, Q₁= 4, Q₃= 12) from a median of four CAM categories (min = 1, max = 7, Q₁= 3, Q₃= 5).</p>", "<p>Fifteen participants described having metastasized cancer, while 23 reported local disease. Participants reporting metastasized cancer used a median of seven CAM therapies (min = 3, max = 26, Q₁= 3, Q₃= 12) from a median of five categories (min = 1, max = 6, Q₁= 3, Q₃= 6), whereas those who reported a local cancer used a median of three CAM therapies (min = 1, max = 22, Q₁= 2, Q₃= 7) from a median of two categories (min = 1, max = 7, Q₁= 2, Q₃= 4).</p>", "<title>Description of CAM reports</title>", "<p>The 38 participants diagnosed with cancer described using a total of 274 CAM therapies consisting of 148 different therapeutic modalities (Table ##TAB##2##3##). Between one-26 different CAM therapies were reported as utilized by participants (median = 4, IQR = 1–8). CAM treatments across all categories were used throughout the cancer trajectory. Thirty-two participants reported being in contact with CAM providers, with a median of one CAM provider utilized per participant (min = 1, max = 7, Q₁= 1, Q₃= 2). The remaining six participants reported using only self-care CAM modalities.</p>", "<title>Categorization of therapies according to the NCCAM system</title>", "<p>The category <italic>Biologically-based therapies </italic>was the most commonly described CAM category, reported by 27 of 38 participants. This category comprised 77 different therapies with Iscador^®, an injectable extract of mistletoe, most common (n = 14). <italic>Mind-body interventions </italic>was the second most common CAM category reported by 23 participants, with meditation most frequently reported (n = 9). Painting therapy, mental training and counseling were each reported by several participants. Twenty-one participants reported using <italic>Energy therapies</italic>. The therapies within this category were sorted into the NCCAM's sub-categories; <italic>Biofield therapies </italic>including healing, qi gong and yoga and <italic>Bioelectromagnetic-based therapie</italic>s including magnetic field therapy and laser therapy. Healing, used by 10 participants, was the most common therapeutic modality in this category. The category <italic>Alternative medical systems </italic>was reported by 10 participants and consisted of Antroposophic medicine (n = 6), Homeopathy (n = 3) and Traditional Chinese Medicine (n = 1). The category <italic>Manipulative and body-based therapies </italic>was reported to be used by 12 individuals, with acupuncture, hyperthermia and lymph massage each used by two participants.</p>", "<title>Empirically derived categories</title>", "<p>The empirically-derived category <italic>Spiritual/health literature </italic>consisted predominantly of inspirational literature about CAM and cancer in a broad context, with the book \"Love, Medicine and Miracles\" by Bernie Siegel [##UREF##8##21##] referred to by the largest number of participants (n = 7). In total, 15 participants provided reports categorized under this heading. A second empirically-derived category consisted of seven different <italic>Treatment centers combining CAM and BHC</italic>, which could be placed along a continuum with varying levels of integration between different therapies from use of psychosocial interventions in a BHC setting, to integration of BHC and CAM treatment. Fifteen participants reported such use, most frequently referring to one antroposophic hospital (n = 10), where health care providers are licensed in both BHC and antroposophic medicine. Reports within this category focus on the environment of the centers, as well as encounters with staff and other patients, rather than on specific therapeutic modalities.</p>", "<title>Patterns of CAM use</title>", "<p>We have thus described CAM use in seven categories. Most participants described using therapies from several categories, with a median of three CAM categories (min = 2, max = 7, Q₁= 2, Q₃= 5). The number of specific therapies reported for each category varied widely between CAM categories. The greatest variety of specific therapies used within the same category was found in regard to <italic>Biologically-based therapies</italic>.</p>", "<p>In the explorative PC analysis we found that PC1 to PC3 accounted for 42%, 22% and 14%, respectively, of the underlying data. We interpreted the curve of percentages as steep between PC1 and PC2, and as flattening out after PC2 (as shown in Figure ##FIG##0##1##). This motivated retaining PC1 and PC2 [##UREF##5##18##], which together explained over 63% of the variability of the scaled usage counts.</p>", "<p>As shown in Table ##TAB##3##4##, all loadings for PC1 were positive. We interpret this PC as a weighted average of the number of treatments reported within each category. PC2 has both positive and negative loadings (Table ##TAB##3##4##), with the three categories with loadings significantly different from zero (based on 95% confidence intervals): <italic>Alternative medical systems </italic>and <italic>Treatment centres </italic>significantly negative and <italic>Energy therapies </italic>significantly positive. We interpret PC2 as indicative of a preference for CAM categories along a continuum, indicated by the sign (positive or negative) and size of the loadings.</p>", "<title>Relationship between CAM categories</title>", "<p>Based on the loadings in Table ##TAB##3##4##, a graphic approximation of correlations between CAM categories is presented in Figure ##FIG##1##2##. CAM categories are shown as vectors where a small angle between vectors represents a strong correlation between categories and an orthogonal angle represents independence between categories. Consequently, we find that the seven categories can be grouped into three pairs and one singleton: a) <italic>Energy Therapies </italic>is paired with <italic>Spiritual/Health Literature</italic>, b) <italic>Manipulative and body-based therapies </italic>with <italic>Mind-body interventions</italic>, and c) <italic>Alternative medical systems </italic>with <italic>Treatment centres</italic>. The category <italic>Biologically-based therapies </italic>stands alone, located between pairs a) and b), with approximately equal positive correlations with each. We also find that the category pairs a) and c) are almost orthogonal, suggesting that the use of therapies from these categories is almost uncorrelated.</p>", "<title>Individual user profiles</title>", "<p>The scores of the first two PCs calculated for each user are shown in Figure ##FIG##2##3##, with the x-axis representing PC1 (number of CAM categories used) and y-axis PC2 (CAM category preference). The origin corresponds to the average user profile, i.e. a hypothetical user with average number of therapies per category and average category preference (Table ##TAB##4##5##, column 1). In Figure ##FIG##2##3##, Karolina is the study participant whose usage profile is closest to the hypothetical average user in terms of number of therapies reported, while Sofia and Dinah exemplify extremes in how few therapies and how many therapies are used, respectively. In the same manner, a large positive coordinate for PC2 indicates a stronger preference for treatments at the <italic>Energy therapies </italic>and <italic>Spiritual/Health literature </italic>end of the spectrum, whereas a high negative coordinate indicates a stronger than average preference for treatments from the categories <italic>Alternative medical systems </italic>and <italic>Treatment centers</italic>. We find e.g. that Mary, Ellen, and Karolina are similar in that they use close to the average number of therapies (close to zero on the x-axis) but have different preferences for category type. Karolina shows no particular preference for CAM category (close to zero on the y-axis), while Mary and Ellen fall at the opposite ends of the category preference axis.</p>", "<title>Groups of user profiles</title>", "<p>The user profiles displayed in Figure ##FIG##2##3## are not equally distributed, but fall into several groups. After systematically considering different cluster analysis alternatives, we suggest an interpretation based on four clusters of user patterns, displayed in Figure ##FIG##3##4##. Among these clusters, Cluster A is the largest, containing 63% of all reports, with 24 of the 38 cases (Table ##TAB##4##5##, which also shows the average number of therapies in each category). Cluster A is characterized by a preference towards the <italic>Energy therapies </italic>end of the spectrum, coupled with less than average use of therapies. Cluster B, with 13% of the participants, is characterized by an average number of therapies used with preference towards the categories <italic>Alternative medical systems </italic>and <italic>Treatment centers</italic>. Cluster C, with 18% of participants, is characterized by use of more than the average number of therapies with a preference towards therapies included in the categories <italic>Alternative medical systems </italic>and <italic>Treatment centers</italic>. Cluster C is the most heterogeneous of the four clusters in Figure ##FIG##3##4##, with considerable variation along both axes. Cluster D finally includes only two individuals and is characterized by the use of therapies across all CAM categories, although with a distinct preference for the categories at one end of the spectrum. This preference seems to be driven by the high number of <italic>Biologically-based therapies </italic>used by both individuals.</p>" ]

[ "<title>Discussion and Conclusion</title>", "<p>This study is an attempt to disentangle and further understand some of the variability involved in descriptions of CAM use. Based on users' own accounts, we have categorized CAM use and identified patterns in this use among a group of individuals with reported exceptional experiences in relation to cancer and CAM use. By analyzing users' own descriptions of CAM in relation to the most commonly used predefined taxonomy (i.e. NCCAM categories), this study highlights discrepancies between user and professional conceptualizations of CAM not previously addressed in other utilization studies. Beyond variations in users' reports of type and number of CAM, the explorative statistical analysis indicates some patterns in CAM usage. We suggest that preference for different CAM categories may exist along a spectrum. In this data set, reported use of therapies within the categories <italic>Alternative Medical Systems </italic>and <italic>Treatment Centers </italic>are at one end of this spectrum, while use of therapies within the category <italic>Energy therapies </italic>is at the opposite end of the spectrum. Although these results can not be extrapolated to CAM users in general, the patterns found in this study generate hypotheses worthy of further exploration.</p>", "<title>Use of a large number of CAM</title>", "<p>The descriptive part of this analysis raises the question if use of large numbers of CAM therapies may be a feature of situations in which BHC treatments are no longer a curative option. While it is interesting to note that in contrast to many CAM utilization surveys [##REF##9708945##1##], a large portion of participants in this study reported not completing BHC treatments. While we cannot draw clear conclusions from this exploratory material, this finding calls for further investigation of the relationship between adherence to and/or completion of BHC treatment and CAM use. Moreover, in line with earlier research [##REF##16504038##8##,##REF##11331323##22##], these results suggest a need for further exploration of the relationship between disease stage and CAM use beyond that possible with this data set.</p>", "<title>Discrepancy between users' descriptions of CAM and current classification systems</title>", "<p>As noted above, these findings point to a discrepancy between NCCAM's professionally-derived taxonomy and these users' descriptions of what constitutes CAM, as the five original NCCAM categories did not satisfactorily represent participants' descriptions of the entire CAM spectrum. The categories empirically-derived from participant descriptions, <italic>Spiritual/health literature </italic>and <italic>Treatment centers</italic>, suggest a broader and less technical view of the CAM field. This expanded view is in line with the Cochrane Collaboration's definition [##REF##10480829##2##] that encompasses not only practices, but also their underlying theories and belief systems.</p>", "<p>The data in this study is limited to a brief description of these two additional categories, and further exploration of CAM users' views on therapies within these categories is warranted. Conceptualizing self-help literature as a form of CAM has previously been suggested by Achilles [##UREF##9##23##]. With the increasing information flow in society, it is important to further explore these and other mass-medial influences on patients' treatment choices.</p>", "<p>Since it is necessary to be aware of contextual differences in what is considered CAM by different stakeholders in different societies, we suggest that these results may best serve as a basis for further discussion on appropriate development of CAM categorization to better accommodate users' descriptions of the field rather than as a suggestion for revision of NCCAM's taxomy for global use. Increased understanding of discrepancies between current professional CAM categorization and user descriptions may be crucial to improve communication and collaboration between CAM users and their providers. Moreover, knowledge about such discrepancies may help caregivers and health care organizations to acknowledge patients' views on CAM when, for example, designing integrative cancer care, since these data suggest that the environment and not only the modalities used may have therapeutic relevance.</p>", "<title>Challenges with heterogeneity in CAM use</title>", "<p>While many CAM researchers argue for the importance of BHC professionals being knowledgeable about CAM [##REF##12458989##10##], these results make evident some of the challenges involved. Given the large number and the wide variety of therapies used by even this small number of individuals, it is unreasonable to expect practitioners to be familiar with all possible CAM options. A major challenge lies in how to distill the most essential information about CAM therapies in general, and to find ways for both practitioners and the public to obtain more specific and trustworthy information about specific CAM modalities. Distinction by CAM category may be of value in gaining more specificity about CAM use without excess detail.</p>", "<title>Committed CAM use</title>", "<p>Based on the recruitment strategy employed, the level of initiative demanded for participants to actively contact the researchers, the initial analysis of the qualitative and quantitative data generated, as well as Balneaves et al's discussion of committed CAM use [##REF##16813510##9##], the participants in this study are viewed as representing a degree of commitment beyond that of the average CAM user. The relatively high median number of CAM therapies reported, supports similarities to the study sample in Balneaves et al [##REF##16813510##9##] who were described as committed CAM users. However, our analysis indicates that this form of committed CAM use may still vary greatly both with regard to number and type of therapies used; consequently neither number nor type of CAM therapy is a suitable single measure of commitment. Our data suggest that commitment might be characterized either by the use of a limited number of therapeutic modalities on one hand [##REF##17619069##12##], or by the use of a large number of different therapies [##REF##16813510##9##]. In the latter sense, commitment may refer to a stance in regard to CAM in a broad sense, rather than as a commitment to one or more specific therapeutic modalities. Little is known about the reasons behind the use of a large number of CAM therapies. Such use might be part of an established life style or may perhaps indicate that patients are seeking something not readily found.</p>", "<title>Patterns of CAM use</title>", "<p>While the heterogeneity found in this study is not unique [##REF##16504038##8##,##REF##16813510##9##], it supports a more nuanced view of CAM use. The explorative statistical analysis points to some general trends and patterns in these participants' reported CAM use.</p>", "<p>The indicated relationships between the different categories shown in Figure ##FIG##1##2##, can serve as guidance for further study. The contrasting relationship between the categories <italic>Energy therapies </italic>on one hand and <italic>Alternative Medical Systems </italic>and <italic>Treatment centers </italic>on the other, can be interpreted as either indicating a competitive relationship between categories, or as illustrating different poles of a continuum of what constitutes acceptable and available CAM in the Swedish context. Such a continuum has previously been discussed as indicative of the level of \"alternativeness\" [##REF##9820257##24##], reflecting the diversity of CAM therapies and their differing relationship to BHC. For example, stays at an antroposophic treatment center (category <italic>Treatment centers</italic>) are a reimbursable complement to BHC cancer care in many regions of Sweden, whereas most therapies reported within the category <italic>Energy therapies </italic>are not reimbursable. This suggests that the contrasting relationship between these categories may be a reflection of their accessibility and degree of regulation within government-regulated health care plans. It is also interesting to note that specific anthroposophic therapies in isolation (e.g. category <italic>Biologically-based therapie</italic>s) are not subject to reimbursement, whereas the same therapies are reimburseable when provided within the treatment center. The incorporation of certain CAM therapies within the BHC system, often described as integrative medicine [##REF##11802746##25##], may be one way of assuring patient safety while maintaining patient choice.</p>", "<p>Another possible interpretation of the contrasting relationship among CAM categories is that they may represent a competitive relationship, appealing to similar needs among participants. Since previous studies have found differences in user characteristics depending on type of CAM used, the pattern found may reflect user characteristics beyond those documented in this study. Kelner and Wellman [##REF##9395702##26##] for instance, found that among a sample of 300 people using four types of CAM, users of the least institutionalized therapy had high educational levels and managerial positions to a greater extent than users of more institutionalized therapies. This raises the question if the use of less institutionalized CAM therapies is related to socio-economic factors and/or particular beliefs and attitudes.</p>", "<p>Finally, based on the results from this study showing that individuals use a large number of CAM therapies simultaneously, important questions can be raised about the external validity of studies evaluating the efficacy of single CAM modalities. Results such as these, indicating a discrepancy between user and professional classification of CAM and patterns of CAM use, may be of value in designing intervention studies that better reflect the ways CAM are actually used.</p>" ]

[ "<title>Discussion and Conclusion</title>", "<p>This study is an attempt to disentangle and further understand some of the variability involved in descriptions of CAM use. Based on users' own accounts, we have categorized CAM use and identified patterns in this use among a group of individuals with reported exceptional experiences in relation to cancer and CAM use. By analyzing users' own descriptions of CAM in relation to the most commonly used predefined taxonomy (i.e. NCCAM categories), this study highlights discrepancies between user and professional conceptualizations of CAM not previously addressed in other utilization studies. Beyond variations in users' reports of type and number of CAM, the explorative statistical analysis indicates some patterns in CAM usage. We suggest that preference for different CAM categories may exist along a spectrum. In this data set, reported use of therapies within the categories <italic>Alternative Medical Systems </italic>and <italic>Treatment Centers </italic>are at one end of this spectrum, while use of therapies within the category <italic>Energy therapies </italic>is at the opposite end of the spectrum. Although these results can not be extrapolated to CAM users in general, the patterns found in this study generate hypotheses worthy of further exploration.</p>", "<title>Use of a large number of CAM</title>", "<p>The descriptive part of this analysis raises the question if use of large numbers of CAM therapies may be a feature of situations in which BHC treatments are no longer a curative option. While it is interesting to note that in contrast to many CAM utilization surveys [##REF##9708945##1##], a large portion of participants in this study reported not completing BHC treatments. While we cannot draw clear conclusions from this exploratory material, this finding calls for further investigation of the relationship between adherence to and/or completion of BHC treatment and CAM use. Moreover, in line with earlier research [##REF##16504038##8##,##REF##11331323##22##], these results suggest a need for further exploration of the relationship between disease stage and CAM use beyond that possible with this data set.</p>", "<title>Discrepancy between users' descriptions of CAM and current classification systems</title>", "<p>As noted above, these findings point to a discrepancy between NCCAM's professionally-derived taxonomy and these users' descriptions of what constitutes CAM, as the five original NCCAM categories did not satisfactorily represent participants' descriptions of the entire CAM spectrum. The categories empirically-derived from participant descriptions, <italic>Spiritual/health literature </italic>and <italic>Treatment centers</italic>, suggest a broader and less technical view of the CAM field. This expanded view is in line with the Cochrane Collaboration's definition [##REF##10480829##2##] that encompasses not only practices, but also their underlying theories and belief systems.</p>", "<p>The data in this study is limited to a brief description of these two additional categories, and further exploration of CAM users' views on therapies within these categories is warranted. Conceptualizing self-help literature as a form of CAM has previously been suggested by Achilles [##UREF##9##23##]. With the increasing information flow in society, it is important to further explore these and other mass-medial influences on patients' treatment choices.</p>", "<p>Since it is necessary to be aware of contextual differences in what is considered CAM by different stakeholders in different societies, we suggest that these results may best serve as a basis for further discussion on appropriate development of CAM categorization to better accommodate users' descriptions of the field rather than as a suggestion for revision of NCCAM's taxomy for global use. Increased understanding of discrepancies between current professional CAM categorization and user descriptions may be crucial to improve communication and collaboration between CAM users and their providers. Moreover, knowledge about such discrepancies may help caregivers and health care organizations to acknowledge patients' views on CAM when, for example, designing integrative cancer care, since these data suggest that the environment and not only the modalities used may have therapeutic relevance.</p>", "<title>Challenges with heterogeneity in CAM use</title>", "<p>While many CAM researchers argue for the importance of BHC professionals being knowledgeable about CAM [##REF##12458989##10##], these results make evident some of the challenges involved. Given the large number and the wide variety of therapies used by even this small number of individuals, it is unreasonable to expect practitioners to be familiar with all possible CAM options. A major challenge lies in how to distill the most essential information about CAM therapies in general, and to find ways for both practitioners and the public to obtain more specific and trustworthy information about specific CAM modalities. Distinction by CAM category may be of value in gaining more specificity about CAM use without excess detail.</p>", "<title>Committed CAM use</title>", "<p>Based on the recruitment strategy employed, the level of initiative demanded for participants to actively contact the researchers, the initial analysis of the qualitative and quantitative data generated, as well as Balneaves et al's discussion of committed CAM use [##REF##16813510##9##], the participants in this study are viewed as representing a degree of commitment beyond that of the average CAM user. The relatively high median number of CAM therapies reported, supports similarities to the study sample in Balneaves et al [##REF##16813510##9##] who were described as committed CAM users. However, our analysis indicates that this form of committed CAM use may still vary greatly both with regard to number and type of therapies used; consequently neither number nor type of CAM therapy is a suitable single measure of commitment. Our data suggest that commitment might be characterized either by the use of a limited number of therapeutic modalities on one hand [##REF##17619069##12##], or by the use of a large number of different therapies [##REF##16813510##9##]. In the latter sense, commitment may refer to a stance in regard to CAM in a broad sense, rather than as a commitment to one or more specific therapeutic modalities. Little is known about the reasons behind the use of a large number of CAM therapies. Such use might be part of an established life style or may perhaps indicate that patients are seeking something not readily found.</p>", "<title>Patterns of CAM use</title>", "<p>While the heterogeneity found in this study is not unique [##REF##16504038##8##,##REF##16813510##9##], it supports a more nuanced view of CAM use. The explorative statistical analysis points to some general trends and patterns in these participants' reported CAM use.</p>", "<p>The indicated relationships between the different categories shown in Figure ##FIG##1##2##, can serve as guidance for further study. The contrasting relationship between the categories <italic>Energy therapies </italic>on one hand and <italic>Alternative Medical Systems </italic>and <italic>Treatment centers </italic>on the other, can be interpreted as either indicating a competitive relationship between categories, or as illustrating different poles of a continuum of what constitutes acceptable and available CAM in the Swedish context. Such a continuum has previously been discussed as indicative of the level of \"alternativeness\" [##REF##9820257##24##], reflecting the diversity of CAM therapies and their differing relationship to BHC. For example, stays at an antroposophic treatment center (category <italic>Treatment centers</italic>) are a reimbursable complement to BHC cancer care in many regions of Sweden, whereas most therapies reported within the category <italic>Energy therapies </italic>are not reimbursable. This suggests that the contrasting relationship between these categories may be a reflection of their accessibility and degree of regulation within government-regulated health care plans. It is also interesting to note that specific anthroposophic therapies in isolation (e.g. category <italic>Biologically-based therapie</italic>s) are not subject to reimbursement, whereas the same therapies are reimburseable when provided within the treatment center. The incorporation of certain CAM therapies within the BHC system, often described as integrative medicine [##REF##11802746##25##], may be one way of assuring patient safety while maintaining patient choice.</p>", "<p>Another possible interpretation of the contrasting relationship among CAM categories is that they may represent a competitive relationship, appealing to similar needs among participants. Since previous studies have found differences in user characteristics depending on type of CAM used, the pattern found may reflect user characteristics beyond those documented in this study. Kelner and Wellman [##REF##9395702##26##] for instance, found that among a sample of 300 people using four types of CAM, users of the least institutionalized therapy had high educational levels and managerial positions to a greater extent than users of more institutionalized therapies. This raises the question if the use of less institutionalized CAM therapies is related to socio-economic factors and/or particular beliefs and attitudes.</p>", "<p>Finally, based on the results from this study showing that individuals use a large number of CAM therapies simultaneously, important questions can be raised about the external validity of studies evaluating the efficacy of single CAM modalities. Results such as these, indicating a discrepancy between user and professional classification of CAM and patterns of CAM use, may be of value in designing intervention studies that better reflect the ways CAM are actually used.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>While the use of complementary and alternative medicine (CAM) among cancer patients is common and widespread, levels of commitment to CAM vary. \"Committed\" CAM use is important to investigate, as it may be associated with elevated risks and benefits, and may affect use of biomedically-oriented health care (BHC). Multiple methodological approaches were used to explore and map patterns of CAM use among individuals postulated to be committed users, voluntarily reporting exceptional experiences associated with CAM use after cancer diagnosis.</p>", "<title>Method</title>", "<p>The verbatim transcripts of thirty-eight unstructured interviews were analyzed in two steps. First, manifest content analysis was used to elucidate and map participants' use of CAM, based on the National Center for Complementary Medicine (NCCAM)'s classification system. Second, patterns of CAM use were explored statistically using principal component analysis.</p>", "<title>Findings</title>", "<p>The 38 participants reported using a total of 274 specific CAM (median = 4) consisting of 148 different therapeutic modalities. Most reported therapies could be categorized using the NCCAM taxonomy (n = 224). However, a significant number of CAM therapies were not consistent with this categorization (n = 50); consequently, we introduced two additional categories: <italic>Spiritual/health literature </italic>and <italic>Treatment centers</italic>. The two factors explaining the largest proportion of variation in CAM usage patterns were a) number of CAM modalities used and b) a category preference for <italic>Energy therapies </italic>over the categories <italic>Alternative Medical Systems </italic>and <italic>Treatment centers </italic>or vice versa.</p>", "<title>Discussion</title>", "<p>We found considerable heterogeneity in patterns of CAM use. By analyzing users' own descriptions of CAM in relation to the most commonly used predefined professional taxonomy, this study highlights discrepancies between user and professional conceptualizations of CAM not previously addressed. Beyond variations in users' reports of CAM, our findings indicate some patterns in CAM usage related to number of therapies used and preference for different CAM categories.</p>" ]

[ "<title>Abbreviations</title>", "<p>CAM: Complementary and Alternative Medicine; BHC: Biomedically-oriented health care; NCCAM: National Center for Complementary and Alternative Medicine; IQR: Inter-quartile Ranges; PC: Principal Component</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>The overall study design was conceived by TF, JH and CT. JH and AF conducted the interviews. JH carried out the content analysis with analytic input from AF, TF and CT. AP conducted the statistical analysis, with input from with JH and CT. JH drafted the original manuscript with input from all authors. All authors have participated in revising the manuscript, and have approved the final version.</p>", "<title>Pre-publication history</title>", "<p>The pre-publication history for this paper can be accessed here:</p>", "<p><ext-link ext-link-type=\"uri\" xlink:href=\"http://www.biomedcentral.com/1472-6882/8/48/prepub\"/></p>" ]

[ "<title>Acknowledgements</title>", "<p>We thank all the 38 participants who contributed with their stories. Suzanne Schönström for providing a secondary categorization of the CAM therapies and our colleagues at Unit for studies of integrative care for input during the research process.</p>", "<p>Financial support was provided by the National Research School for Caring Sciences, the Swedish Cancer Society and Cancer and Traffic Injury Fund. CT is supported via funding from the Swedish Research Council.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Scree plot showing the proportion of variance explained by consecutive principal components (PCs)</bold>. Bootstrapped 95% confidence intervals are shown as vertical lines. The dotted horizontal reference line indicates the proportion of variance explained by one of the underlying variables (i.e. category counts).</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>The loadings of the original variables (i.e. number of treatments in each category) for the first two principal components</bold>. The weights of the categories are shown in a scatter plot. See also Table 4.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>User profiles of the study participants</bold>. These are represented as their scores along the first two principal components, labelled by their fictitious name.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>A four cluster grouping of the user profiles</bold>. Circles indicate individual user profiles as in Figure 3, ellipses the clusters. Full circles indicate for each cluster the subject closest to the average user profile within the group. See also Table 5.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Participant characteristics.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Characteristics</td><td align=\"right\">Frequency (n = 38)</td></tr></thead><tbody><tr><td align=\"left\">Age</td><td align=\"right\">Median 55 years (min = 36, max = 85)</td></tr><tr><td colspan=\"2\"><hr/></td></tr><tr><td align=\"left\">≤ 40 years</td><td align=\"right\">3</td></tr><tr><td align=\"left\">41–50 years</td><td align=\"right\">9</td></tr><tr><td align=\"left\">51–60 years</td><td align=\"right\">13</td></tr><tr><td align=\"left\">61–70 years</td><td align=\"right\">6</td></tr><tr><td align=\"left\">&gt;70 years</td><td align=\"right\">5</td></tr><tr><td align=\"left\">Age unknown</td><td align=\"right\">2</td></tr><tr><td align=\"left\">Sex</td><td/></tr><tr><td align=\"left\">Female</td><td align=\"right\">31</td></tr><tr><td align=\"left\">Male</td><td align=\"right\">7</td></tr><tr><td align=\"left\">Marital status</td><td/></tr><tr><td align=\"left\">Married or common-law</td><td align=\"right\">19</td></tr><tr><td align=\"left\">Divorced/Separated/Widowed/Single</td><td align=\"right\">15</td></tr><tr><td align=\"left\">Unknown</td><td align=\"right\">4</td></tr><tr><td align=\"left\">Occupational Status</td><td/></tr><tr><td align=\"left\">Working full-time</td><td align=\"right\">7</td></tr><tr><td align=\"left\">Working part-time</td><td align=\"right\">2</td></tr><tr><td align=\"left\">On sick-leave</td><td align=\"right\">7</td></tr><tr><td align=\"left\">Pension</td><td align=\"right\">9</td></tr><tr><td align=\"left\">Unknown</td><td align=\"right\">13</td></tr><tr><td align=\"left\">Education</td><td/></tr><tr><td align=\"left\">College education</td><td align=\"right\">20</td></tr><tr><td align=\"left\">Elementary school + High School</td><td align=\"right\">4</td></tr><tr><td align=\"left\">Unknown</td><td align=\"right\">14</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Reported disease characteristics.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Reported disease characteristics</td><td align=\"right\">Frequency (n = 38)</td></tr></thead><tbody><tr><td align=\"left\">Breast</td><td align=\"right\">17</td></tr><tr><td align=\"left\">Gynecological (women)</td><td align=\"right\">7</td></tr><tr><td align=\"left\">Stomach, Colon and Rectum</td><td align=\"right\">4</td></tr><tr><td align=\"left\">Lymphatic leukemia</td><td align=\"right\">2</td></tr><tr><td align=\"left\">Lung</td><td align=\"right\">2</td></tr><tr><td align=\"left\">Prostate</td><td align=\"right\">2</td></tr><tr><td align=\"left\">Other sites</td><td align=\"right\">4</td></tr><tr><td align=\"left\">Metastasized disease</td><td align=\"right\">15</td></tr><tr><td align=\"left\">Median time since 1^stcancer diagnosis (years)</td><td align=\"right\">5 years</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>CAM described by participants and sorted into seven CAM categories. </p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>NCCAM CATEGORIES</bold></td><td align=\"right\">Total number of therapies reported</td><td align=\"right\">Number of individuals reporting therapies</td></tr></thead><tbody><tr><td align=\"left\">ALTERNATIVE MEDICAL SYSTEMS<break/>antroposophic medicine (6), homeopathy (3), traditional Chinese medicine</td><td align=\"right\">10</td><td align=\"right\">10</td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td align=\"left\">MIND-BODY INTERVENTIONS<break/>painting (6), music, dance, sculpturing, counselling (6), support groups, mental practice (6), relaxation techniques (2), eurythmy (3), gestalt therapy, bonitology (2), kinesiology, prayer (3), meditation-various types (9), family constellations, visualization (3), rehabilitation program, rosen method body work</td><td align=\"right\">50</td><td align=\"right\">23</td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td align=\"left\">BIOLOGICALLY-BASED THERAPIES<break/>aloe vera (2), angelica, antioxidants (5), apis, ayurvedic preparations, birch ash (2), blutsaft, cayenne pepper, cetraria, chalk, charchole, chinese herbal medicine, cypress, coffee enema, dendrite cell treatment (2), ecomer, edta, enzymes (2), field horsetail, fish oil, garlic (2), geranium, ginger, ginseng, helixor (2), iceland lichen, inhalation mixture-chamomile, peppermint and lemon balm, iscador (14), juniperberry, kan yang, lactase enzyme, lavender, lemon concentrate (2) lemon grass, lemon balm, linseed bandage, lycine, magnesium (2), marjoram, micro-algae, mung bean sprouts (2), new castle virus (2), nouni (2), olibanum, ozone therapy (2) quercetin, pankreon, probion, proline, proteas, radish, raw food diet, rosemary, sage (2), sandal wood, saw palmetto, selen (2), shark liver oil, silica, silymarin, silver, sodium ascorbate, sodiumselen respond selen, sulfur, supergreens, THX, valerian root, vegan diet, vitamin A, vitamin b, vitamin C (3), vitamin D, vitamin e (5), walnut supplements, wheat grass juice, yarrow, zinc (2)</td><td align=\"right\">115</td><td align=\"right\">27</td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td align=\"left\">MANIPULATIVE AND BODY BASED THERAPIES<break/>acupuncture (3), chiropractic care, feldenkreis, fever baths, herbal baths, local and whole body hyperthermia (2), stretching, lymph massage (2), alternative surgical procedure, soft tissue massage</td><td align=\"right\">14</td><td align=\"right\">12</td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td align=\"left\">ENERGY THERAPIES<break/>Biofield therapies: healing (10), qi gong (4), tai chi (2), yoga (3), reflexology, color therapy, homeopathic remedies (gold, arsenic, barium-iodate, viscum/mesenchym comp, conium maculatorn)<break/>Bioelectromagnetic-based therapies: ECT-laser (3), frequency medicine (2), magnetic field therapy (3), plasma lamp therapy</td><td align=\"right\">35</td><td align=\"right\">21</td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td colspan=\"3\"/></tr><tr><td align=\"left\" colspan=\"3\"><bold>EMPIRICALLY DERIVED CATEGORIES</bold></td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td align=\"left\">SPIRITUAL/HEALTH LITERATURE<break/>A Course in Miracles-author unspecified, Bays Brandon – The Journey (2), Chopra Deepak – Perfect Health, Ehdin, Sanna – The Self-Healing Human (4), Gawler Ian – You can conquer cancer, Hamer Gerhard – The New Medicine (3), Pollak Kay – Att välja glädje [only in Swedish], Hayes Louise – no specific book, Alexander Marcus – Kvantmänniskan [in Swedish], Preben Maria – no specific book, Moss Ralph-Cancer &amp; CAM information, Shine Betty – Mind to Mind, Siegel Bernie – Love, Medicine and Miracles (6), Sai Baba – no specific book, Simonton, Carl – Getting Well Again, Stern Bengt – Feeling bad is a good start (3), Walsch Donald – Conversations with God</td><td align=\"right\">31</td><td align=\"right\">15</td></tr><tr><td colspan=\"3\"><hr/></td></tr><tr><td align=\"left\">TREATMENT CENTERS<break/>Centro Antroposophico – Antroposophic center, Spain; Furusjön – Health retreat, Sweden (2); Humlegården – Alternative Clinic for cancer patients, Denmark (2); Lustgården – Rehabilitation unit for cancer patients, Sweden, Mösseberg-rehabilitation for cancer patients, Sweden (2), Vidarkliniken – Antroposophic hospital, Sweden (10), TCM hospital combining TCM and BHC, Germany</td><td align=\"right\">19</td><td align=\"right\">15</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Loadings with bootstrapped 95% confidence intervals for the first two principal components (PC).</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"3\">First PC (42%)</td><td align=\"center\" colspan=\"3\">Second PC (21%)</td></tr><tr><td/><td colspan=\"3\"><hr/></td><td colspan=\"3\"><hr/></td></tr><tr><td align=\"left\">CAM categories</td><td align=\"right\">lci*</td><td align=\"right\">Loading</td><td align=\"right\">uci**</td><td align=\"right\">lci</td><td align=\"right\">Loading</td><td align=\"right\">uci</td></tr></thead><tbody><tr><td align=\"left\">Alternative medical systems</td><td align=\"right\">-0.09</td><td align=\"right\">0.23</td><td align=\"right\">0.41</td><td align=\"right\">-0.71</td><td align=\"right\">-0.64</td><td align=\"right\">-0.26</td></tr><tr><td align=\"left\">Biologically-based therapies</td><td align=\"right\">0.35</td><td align=\"right\">0.44</td><td align=\"right\">0.52</td><td align=\"right\">-0.25</td><td align=\"right\">0.27</td><td align=\"right\">0.41</td></tr><tr><td align=\"left\">Energy therapies</td><td align=\"right\">0.15</td><td align=\"right\">0.38</td><td align=\"right\">0.52</td><td align=\"right\">0.05</td><td align=\"right\">0.49</td><td align=\"right\">0.66</td></tr><tr><td align=\"left\">Manipulative &amp; body-based therapies</td><td align=\"right\">0.42</td><td align=\"right\">0.49</td><td align=\"right\">0.53</td><td align=\"right\">-0.24</td><td align=\"right\">-0.01</td><td align=\"right\">0.30</td></tr><tr><td align=\"left\">Mind-body interventions</td><td align=\"right\">0.25</td><td align=\"right\">0.42</td><td align=\"right\">0.53</td><td align=\"right\">-0.47</td><td align=\"right\">-0.08</td><td align=\"right\">0.37</td></tr><tr><td align=\"left\">Spiritual/Health literature</td><td align=\"right\">-0.14</td><td align=\"right\">0.20</td><td align=\"right\">0.39</td><td align=\"right\">-0.46</td><td align=\"right\">0.29</td><td align=\"right\">0.67</td></tr><tr><td align=\"left\">Treatment centers</td><td align=\"right\">0.06</td><td align=\"right\">0.39</td><td align=\"right\">0.49</td><td align=\"right\">-0.67</td><td align=\"right\">-0.44</td><td align=\"right\">-0.09</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Four clusters of rough user patterns. </p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"right\">Whole set</td><td align=\"right\">Cluster A</td><td align=\"right\">Cluster C</td><td align=\"right\">Cluster B</td><td align=\"right\">Cluster D</td></tr></thead><tbody><tr><td/><td align=\"center\" colspan=\"5\">Number of participants</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td/><td align=\"right\">n = 38</td><td align=\"right\">n = 24</td><td align=\"right\">n = 7</td><td align=\"right\">n = 5</td><td align=\"right\">n = 2</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\">CAM categories</td><td align=\"center\" colspan=\"5\">Average number of treatments per category</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\">Alternative medical systems</td><td align=\"right\">0.3</td><td align=\"right\">0.0</td><td align=\"right\">0.6</td><td align=\"right\">1.0</td><td align=\"right\">0.0</td></tr><tr><td align=\"left\">Biologically-based therapies</td><td align=\"right\">3.0</td><td align=\"right\">1.1</td><td align=\"right\">6.1</td><td align=\"right\">1.0</td><td align=\"right\">20.0</td></tr><tr><td align=\"left\">Energy therapies</td><td align=\"right\">0.9</td><td align=\"right\">0.6</td><td align=\"right\">1.4</td><td align=\"right\">0.2</td><td align=\"right\">5.0</td></tr><tr><td align=\"left\">Manipulative &amp;body-based therapies</td><td align=\"right\">0.4</td><td align=\"right\">0.1</td><td align=\"right\">1.3</td><td align=\"right\">0.2</td><td align=\"right\">1.0</td></tr><tr><td align=\"left\">Mind-body interventions</td><td align=\"right\">1.3</td><td align=\"right\">0.7</td><td align=\"right\">3.0</td><td align=\"right\">1.4</td><td align=\"right\">3.0</td></tr><tr><td align=\"left\">Spiritual/Health literature</td><td align=\"right\">0.8</td><td align=\"right\">0.8</td><td align=\"right\">0.9</td><td align=\"right\">0.2</td><td align=\"right\">2.0</td></tr><tr><td align=\"left\">Treatment centers</td><td align=\"right\">0.5</td><td align=\"right\">0.2</td><td align=\"right\">1.3</td><td align=\"right\">1.0</td><td align=\"right\">0.5</td></tr><tr><td align=\"left\">PC1</td><td align=\"right\">0.0</td><td align=\"right\">-1.0</td><td align=\"right\">2.3</td><td align=\"right\">0.1</td><td align=\"right\">3.7</td></tr><tr><td align=\"left\">PC2</td><td align=\"right\">0.0</td><td align=\"right\">0.4</td><td align=\"right\">-0.7</td><td align=\"right\">-1.9</td><td align=\"right\">3.0</td></tr><tr><td align=\"left\">Representative participant</td><td align=\"right\">Karolina</td><td align=\"right\">Andrea</td><td align=\"right\">Sarah</td><td align=\"right\">Ellen</td><td align=\"right\">Peter, Victor</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>Number of participants reporting a certain CAM is indicated in parentheses (if more than one).</p></table-wrap-foot>", "<table-wrap-foot><p>* lci = lower confidence interval **uci = upper confidence interval</p></table-wrap-foot>", "<table-wrap-foot><p>The table shows number of participants, number of average treatments per category, average value for PCs, and participants with close to average user profile for the whole data set and for the four clusters A-D.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1472-6882-8-48-1\"/>", "<graphic xlink:href=\"1472-6882-8-48-2\"/>", "<graphic xlink:href=\"1472-6882-8-48-3\"/>", "<graphic xlink:href=\"1472-6882-8-48-4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Adolescent idiopathic scoliosis (AIS) is a structural three-dimensional deformity of the spine that occurs at or near the onset of puberty for which no cause can be established. In patients with small curve magnitude in the mean of 10° or so, the male and female prevalence is approximately equal. In curves of larger magnitude, however, there is an overwhelming female predominance in a way that the ratio of females to males with curves measuring 30° or more is 10 to 1 [##REF##17675949##1##,##REF##17141012##2##].</p>", "<p>According to epidemiologic and natural history studies, curve progression is different in male and female patients. Studies conducted by Suh and MacEwen [##REF##3206265##3##], and Karol et al [##REF##8258551##4##] on curve behavior in males verified that scoliotic male patients demonstrated clinically significant curve progression until Risser V. In females, scoliosis beyond Risser IV can be considered as an adult curve; Scoliosis in males, however, can be evaluated as an adult curve only at Risser V.</p>", "<p>Bracing has been shown to effectively prevent curve progression in adolescent girls [##REF##16305271##5##], but it is not always effective for the males [##REF##11547200##6##,##REF##16909249##7##]. Karol reported the result of bracing in 112 boys with AIS. 74% of these boys progressed by more than 6°, which is more than failure rate of bracing in girls. Moreover, the amount of curve correction among male patients in brace is lower compared to girls'. It has also been suggested that the spine is stiffer in males than in females [##REF##11547200##6##].</p>", "<p>Despite the importance of gender difference in curve behavior and the results of brace treatment, there are a limited number of studies comparing the results of surgical treatment between males and females. Thus, the purpose of this study was to compare the radiographic outcome of surgery for AIS between males and females in matched and unmatched groups in regard to age, curve type, and magnitude.</p>" ]

[ "<title>Methods</title>", "<p>Methodologically, a retrospective review of the records of all patients who had been surgically treated for AIS between May 1996 and September 2005 at our hospital was performed at first. 18 patients were treated with Harington instrumentation and 132 patients with modern segmental spinal instrumentation (Cotrel-Dubousset; CD: 24, Diapason: 103, and Universal Spine System; USS: 5 cases). The patients who were treated by only anterior surgery were excluded. Radiographic measurement were performed on standing posteroanterior and lateral radiographs of the total spine (T1-S1) acquired before surgery, at 4 days, 6 weeks, 6 months, 1, and 2 years respectively after surgery and at final follow-up. The Cobb method was used to measure the curve magnitude [##UREF##0##8##].</p>", "<p>At the next step, preoperative coronal curve flexibility measurements from the right and left supine side bending radiographs were acquired. These views were all taken while the patients actively bent laterally (Figure ##FIG##0##1##). In order to calculate the percentage of flexibility, we subtracted the magnitude of the bend Cobb angle from the magnitude of the preoperative upright coronal Cobb angle and then divided it by the preoperative upright coronal Cobb angle calculated the percentage of flexibility. For calculating postoperative percent correction of the coronal curves, we Subtracted the magnitude of the coronal Cobb angle at final follow-up from the preoperative coronal Cobb angle and then divided it by the preoperative Cobb angle calculated postoperative percent correction of the coronal curves. The King classification was used to categorize the curve types [##REF##6654943##9##].</p>", "<p>Our threshold level for doing only PSF (posterior spinal fusion) or combined ASF (anterior spinal fusion) and PSF was a curve magnitude of 70°. We used bending views mostly for determining the fusion levels. In anterior surgery, we released the most rigid segment of the spine and then inserted autogenous (in the thoracic area and from the harvested rib) or allogenous (in the lumbar area) type of cancellous bone graft in the intervertebral spaces without implanting the spine. PSF and instrumentation with or without anterior surgery were conducted in all patients. These combined procedures were done in separate sections with the interval of 5 to 7 days. The details of the types of instrumentation used for our operative technique have been reported previously [##REF##14668498##10##]. The USS was implanted according to the manufacturer's instructions [##UREF##1##11##]. PSF included decortication of the laminae, facet joint cleaning, and use of local bone graft besides an autograft from the posterior iliac crest (Figure ##FIG##1##2##).</p>", "<p>The comparison between male and female patients was done in two stages. At first, all of the males were compared with all of the females. In the second stage, each male patient was matched with a female based on the factors such as age (± 1 year), curve type (according to the King classification), curve magnitude (± 5), and instrumentation used so that the final study series comprised 38 matched pairs.</p>", "<p>Statistical analysis was performed by the x²test or the Mann-Whitney test. P value equal to or below 0.05 was considered statistically significant.</p>" ]

[ "<title>Results</title>", "<p>This analysis includes the data for 150 patients (112 females; 74.7% and 38 males; 25.3%). All patients had clinical and radiographic follow-up of at least 2 years. The mean age of the males at the time of operation was 17.3 ± 2.2 years and that of the females 16.3 ± 2.8 years (p = 0.049; significant). The mean follow-up time was 3.6 years (range 2.3 – 10.2 years).</p>", "<p>The King classification distribution between the two gender is listed in table ##TAB##0##1##. There is similar distribution curve pattern between male and female, with King type III making up 47.4% and 47.3% of all the curve types for males and females, respectively.</p>", "<p>We performed PSF alone on 11 males (28.9%) and 58 females (51.8%). ASF and then PSF with instrumentation were conducted on the rest. Harrington rod was used in 4 (10.5%) male and 14 (12.5%) female patients and segmental spinal instrumentation in other cases.</p>", "<p>At first, in comparing all male patients with all females, the males had greater mean age and primary curve magnitude, but less flexibility and correction percentage that were statistically significant. Loss of correction was comparable between the two groups (table ##TAB##1##2##).</p>", "<p>In the second stage of the study, the 38 male patients were compared with the 38 matched females to determine whether gender difference had an effect on the operative results (table ##TAB##2##3##). In this comparison, flexibility percent was the only index that had a statistically significant difference (the boys had more rigid curves). The correction percentage and loss of correction in boys were less than girls but these were not statistically significant (p = 0.11 and 0.25 respectively). We performed only PSF in 13 and 11 female and male patients in this group respectively. In other cases, ASF and then PSF with instrumentation were conducted. Therefore, the difference in prevalence of the type of surgery in matched group was not statistically significant, (p &gt; 0.05).</p>" ]

[ "<title>Discussion</title>", "<p>In the initial reports of segmental spinal instrumentation in the treatment of AIS, some radiographic distinctions between boys and girls had been reported [##REF##6239334##12##,##REF##4071271##13##]. According to these reports, between 10% and 30% of patients requiring operative intervention for AIS are males [##REF##3206265##3##,##REF##1522092##14##]. We found a similar percentage in our study (i.e., of 150 operative patients in this series, 25.3% were male).</p>", "<p>A comparison of the surgical treatment outcomes in both genders has only been reported from few investigations [##REF##15342764##15##, ####REF##17334288##16##, ##REF##15706345##17####15706345##17##]. We found that preoperative curve pattern between the genders are roughly similar in King classification distribution. However, the older male patients had bigger curve magnitude, and less preoperative primary curve flexibility than female patients. This observation contradicts what has been previously reported by Sucato et al [##REF##15342764##15##]. They found larger primary male curves with similar curve stiffness in their comparison of male-female patients.</p>", "<p>Marks et al. in a study of 547 (449 females and 98 males) patients, found that male AIS patients had more rigid primary curve compared to females but showed a similar degree of postoperative scoliosis correction [##REF##17334288##16##]. They concluded that differences in the preoperative status and perioperative course did not compromise the outcomes of surgical treatment as in all other measures; moreover, the results were comparable between the genders.</p>", "<p>Regardless of the preoperative differences and slight variation in treatment approaches, our study revealed that surgical outcomes are comparable between the genders. Primary curve percent correction and loss of correction over time were not statistically different between the genders.</p>", "<p>According to our knowledge, there are only two matched studies of the surgical treatment of AIS between male and female patients:</p>", "<p>In the first, Helenius and coauthors compared the results of operative treatment of 30 male and female AIS pairs. They finally concluded that the curves in males appear to be more rigid than in females; however, posterior surgery for AIS provides similar short and long-term results in both genders [##REF##15706345##17##].</p>", "<p>The second study that was conducted by Sucato et al. [##REF##15342764##15##], revealed that treatment outcome differences did exist. They reported less correction of the curve in males compared to females. In an attempt to explain this finding, they theorized that perhaps the male patients produce a more powerful supine bend effort, reflected by a greater preoperative flexibility that the surgeon cannot duplicate at the time of surgery. Our findings about preoperative stiffer curves in males yet equal curve correction between the genders after surgery discredit this theory.</p>" ]

[ "<title>Conclusion</title>", "<p>In conclusion, male patients with AIS have the similar curve pattern as that of female patients. Males had more rigid primary curves compared to females but a similar degree of postoperative scoliosis correction. Male AIS patients were older at the time of surgery. These preoperative gender differences, however; did not compromise the radiological outcomes of surgical treatment and the results were comparable between the genders.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Studies on adolescent idiopathic scoliosis have well documented the differences between natural history of male and female patients. There are also differences in responses to nonoperative treatment, but the results of operative treatment in male patients compared with females have not been widely reported. Only few studies had compared the outcomes of operative treatment between male and female patients with different results.</p>", "<title>Methods</title>", "<p>We retrospectively reviewed the outcome of 150 (112 girls and 38 boys) consecutive patients with diagnosis of adolescent idiopathic scoliosis who were managed surgically between May 1996 and September 2005. Next, male radiographic parameters were compared with female ones pre- and postoperatively. Then, a subgroup of 38 matched girls was compared regarding the age, curve type, curve magnitude, and the instrumentation we used.</p>", "<title>Results</title>", "<p>In comparing male patients with unmatched girls, the boys had greater mean age (17.3 ± 2.3 vs. 16.3 ± 2.9; p = 0.049), greater primary curve (71.4 ± 21.3° vs. 62.7 ± 17.5°; p = 0.013), less flexibility (30.1 ± 13.5% vs. 40.3 ± 17.8%; p = 0.01), and less correction percentage (51.3 ± 12.9% vs. 58.8 ± 16.5%; p = 0/013). The loss of correction was comparable between the two groups. In the matched comparison, the flexibility in boys was less than girls (30.1 ± 13.5% vs. 38.1 ± 17.5%; p = 0.027). Also, the boys had a smaller correction percentage compared to the girls, but this finding was not statistically significant.</p>", "<title>Conclusion</title>", "<p>There was similar distribution curve pattern between male and female patients with AIS. Males had more rigid primary curves compared to females but a similar degree of postoperative scoliosis correction. Male AIS patients were older at the time of surgery. These preoperative gender differences, however; did not compromise the radiological outcomes of surgical treatment and the results were comparable between the genders.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>EA, the senior surgeon and has made substantial contributions to conception and design of the manuscript. HB has been involved in drafting the manuscript, participated in the sequence alignment. BaM has made substantial contributions to acquisition of data from literature. FOK holds a spine fellowship. He is a junior surgeon and has had substantial role in preparing and revising the manuscript. BeM is an orthopedic surgeon and very helpful in collecting the data. He has made important critical contributions to manuscript revision in terms of its intellectual content. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>The authors acknowledge the assistance of Mohamadian N in data management and statistical analysis.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>A 15 years old boy presented with AIS.</bold> Above: preoperative standing posteroanterior and lateral views, below: supine left and right bending films.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Postoperative radiographs of the patient on Figure 1. </bold>Posteroanterior and lateral standing views 5 years after surgery. The patient was completely asymptomatic at the latest follow-up visit.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Curve pattern according to King classification</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\">King Classification</td><td align=\"center\">Male (%)</td><td align=\"center\">Female (%)</td></tr></thead><tbody><tr><td align=\"center\">I</td><td align=\"center\">6 (11)</td><td align=\"center\">11 (9.8)</td></tr><tr><td align=\"center\">II</td><td align=\"center\">9 (23.7)</td><td align=\"center\">32 (28.6)</td></tr><tr><td align=\"center\">III</td><td align=\"center\">18 (47.4)</td><td align=\"center\">53 (47.3)</td></tr><tr><td align=\"center\">IV</td><td align=\"center\">2 (5.2)</td><td align=\"center\">9 (8)</td></tr><tr><td align=\"center\">V</td><td align=\"center\">3 (7.9)</td><td align=\"center\">7 (6.3)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>The results of the comparison all males with all females</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Males</bold></td><td align=\"center\"><bold>Females</bold></td><td align=\"center\"><bold>P value</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold><underline>Preoperative</underline></bold></td><td/><td/><td/></tr><tr><td align=\"center\">Age (year)</td><td align=\"center\">17.3 ± 2.3*</td><td align=\"center\">16.3 ± 2.9</td><td align=\"center\">0.049</td></tr><tr><td align=\"center\">Primary curve (°)</td><td align=\"center\">71.4 ± 21.3</td><td align=\"center\">62.7 ± 17.5</td><td align=\"center\">0.013</td></tr><tr><td align=\"center\">Flexibility (%)</td><td align=\"center\">30.1 ± 13.5</td><td align=\"center\">40.3 ± 17.8</td><td align=\"center\">0.01</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\"><bold><underline>Postoperative</underline></bold></td><td/><td/><td/></tr><tr><td align=\"center\">Primary curve (°)</td><td align=\"center\">35.3 ± 16.4</td><td align=\"center\">26.6 ± 15.4</td><td align=\"center\">0.03</td></tr><tr><td align=\"center\">Correction (%)</td><td align=\"center\">51.3 ± 12.9</td><td align=\"center\">58.8 ± 16.6</td><td align=\"center\">0.013</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\"><bold><underline>Final visit</underline></bold></td><td/><td/><td/></tr><tr><td align=\"center\">Primary curve (°)</td><td align=\"center\">37.6 ± 16.3</td><td align=\"center\">29.1 ± 15.6</td><td align=\"center\">0.04</td></tr><tr><td align=\"center\">Loss of correction (°)</td><td align=\"center\">2.3 ± 1.9</td><td align=\"center\">2.6 ± 2.3</td><td align=\"center\">0.528 (NS)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>The results of the comparison between males and matched females</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\"><bold>Males</bold></td><td align=\"center\"><bold>Females</bold></td><td align=\"center\"><bold>P value</bold></td></tr></thead><tbody><tr><td align=\"center\"><bold><underline>Preoperative</underline></bold></td><td/><td/><td/></tr><tr><td align=\"center\">Primary curve (°)</td><td align=\"center\">71.4 ± 21.3*</td><td align=\"center\">70.2 ± 18.5</td><td align=\"center\">0.79 (NS)</td></tr><tr><td align=\"center\">Flexibility (%)</td><td align=\"center\">30.1 ± 13.5</td><td align=\"center\">38.1 ± 17.5</td><td align=\"center\">0.027</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\"><bold><underline>Postoperative</underline></bold></td><td/><td/><td/></tr><tr><td align=\"center\">Primary curve (°)</td><td align=\"center\">35.3 ± 16.3</td><td align=\"center\">31.6 ± 18.8</td><td align=\"center\">0.35 (NS)</td></tr><tr><td align=\"center\">Correction (%)</td><td align=\"center\">51.3 ± 12.9</td><td align=\"center\">57.1 ± 18.2</td><td align=\"center\">0.11 (NS)</td></tr><tr><td colspan=\"4\"><hr/></td></tr><tr><td align=\"center\"><bold><underline>Final visit</underline></bold></td><td/><td/><td/></tr><tr><td align=\"center\">Primary curve (°)</td><td align=\"center\">37.6 ± 16.3</td><td align=\"center\">34.4 ± 18.9</td><td align=\"center\">0.35 (NS)</td></tr><tr><td align=\"center\">Loss of correction (°)</td><td align=\"center\">2.3 ± 1.9</td><td align=\"center\">2.8 ± 2.2</td><td align=\"center\">0.25 (NS)</td></tr></tbody></table></table-wrap>" ]

[ "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>", "<inline-formula></inline-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>*Mean ± Standard Deviation NS: Not Significant</p></table-wrap-foot>", "<table-wrap-foot><p>* Mean ± Standard Deviation NS: Not Significant</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1748-7161-3-12-1\"/>", "<graphic xlink:href=\"1748-7161-3-12-2\"/>", "<inline-graphic xlink:href=\"1748-7161-3-12-i1.gif\"/>", "<inline-graphic xlink:href=\"1748-7161-3-12-i1.gif\"/>", "<inline-graphic xlink:href=\"1748-7161-3-12-i1.gif\"/>", "<inline-graphic xlink:href=\"1748-7161-3-12-i1.gif\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>It has been estimated that pelagic bacteria are extremely abundant, achieving densities of up to 10⁶per ml of seawater, and account for most oceanic biomass and metabolism [##UREF##0##1##]; while numbers of bacteria which are thought to colonize marine snow can reach levels of up to 10⁹per ml [##REF##11734832##2##]. Marine environments, including the subsurface are believed to contain a total of approximately 3.67 × 10³⁰microorganisms [##REF##9618454##3##] and with approximately 71% of the earth's surface of 361 million square kilometers covered by the ocean, this environment represents an enormous pool of potential microbial biodiversity and exploitable biotechnology or \"blue biotechnology\". This untapped potential has resulted in the recent acceleration in interest in the study of marine microorganisms, with the aim of not only providing us with more information on the key role they play in marine food webs and biogeochemical cycling in marine ecosystems, but also in exploiting their ability to produce novel enzymes and metabolites/compounds with potential biotechnological applications. As with terrestrial environments, where more than 99% of bacteria cannot be cultured by conventional means, the same is true for marine environments where the vast majority of these marine microbes have to date not yet been identified, classified or indeed cultured. According to Amann and colleagues as few as 0.001–0.1% of microbes in seawater are currently cultivable [##REF##7535888##4##]. In this respect the recent advances in culture independent techniques to assess microbial diversity and ecology, such as phylogenetic studies based on small ribosomal RNA (rRNA) analysis and metagenomics, which were initially developed for terrestrial based research are now increasingly being employed in marine environments and are proving extremely useful [##REF##16163343##5##,##REF##16880384##6##]. A clear example of this was the large scale metagenome sequencing project which was recently undertaken on oligotrophic seawater samples from the Sargasso Sea [##REF##15001713##7##] and the Global Ocean Sampling (GOS) expedition [##REF##17355171##8##].</p>", "<p>In this review, we highlight the exciting potential that metagenomic based approaches offer us in gaining access to protein-coding genes with biotechnological potential from uncultivable marine microorganisms; thereby allowing us to exploit, to a much greater extent than heretofore, the potential of this vast, and as yet untapped, marine microbial biodiversity resource.</p>", "<title>Metagenomics</title>", "<p>Against the background whereby it is widely believed that more than 99% of bacteria in any given environment cannot be cultured when conventional approaches are employed, metagenomic based approaches have emerged as an attractive option to allow an assessment of the microbial genomes present within these environments [##REF##15922085##9##]. Metagenomics involves the direct cloning of environmental DNA into large clone libraries to facilitate the analysis of the genes and the sequences within these libraries (Figure ##FIG##0##1##) [##REF##15590779##10##,##REF##8550487##11##]. Metagenomics was initially employed to study non-culturable microbiota and focused primarily on providing a better understanding of global microbial ecology in different environmental niches. With the advent of efficient cloning vectors such as bacterial artificial chromosomes (BACs) and cosmids, together with improved DNA isolation techniques and advanced screening methodologies using robotic instrumentation; it is now possible to express large fragments of DNA and subsequently screen large clone libraries for functional activities [##REF##15931168##12##]. Such approaches have been particularly successful in terrestrial environments, where genes involved in antibiotic production, antibiotic resistance and degradative enzymes have been identified among others [##REF##15590779##10##,##REF##17686025##13##,##REF##16963141##14##]. These approaches coupled with additional innovative screening approaches such as Substrate-Induced Gene Expression screening (SIGEX) have facilitated the cloning of catabolic operons potentially involved in benzoate and catechol degradation among others [##REF##15608629##15##]. These functional based screening approaches have also been supplemented with homology-based screens, primarily involving polymerase chain reaction (PCR)-based approaches targeting novel genes with sequences similar to known genes. This has resulted in the cloning of genes such as polyketide synthases [##REF##9393700##16##], alkane hydroxylases [##REF##18087672##17##], cyclomaltodextrinases [##REF##16790023##18##], xylanases [##REF##15920623##19##] and beta-xylanases [##REF##12579381##20##]. Recently novel methods such as pre-amplification inverse-PCR (PAI-PCR) [##REF##18093161##21##] and metagenomic DNA shuffling [##REF##16690231##22##] have been employed to isolate new biocatalysts. PAI-PCR which has been employed to isolate glycosyl hydrolase genes from horse and termite guts, offers the potential to clone genes for which the copy number of target DNA sequences is low, while the shuffling approach, which has been used to construct novel biocatalysts, simulates and accelerates the evolutionary process using molecular biological tools. Homology-based screening approaches are by definition quite limited, given that homologs of existing genes are being targeted and this often results in no novel gene families being detected. The large scale Global Ocean Sampling Project has revealed that despite the large increase in DNA sequence data we have yet to approach saturation for the discovery of new protein folds, implying that there is a large resource of truly novel proteins and enzymes in uncultured marine microbes [##REF##17355171##8##]. It is thus widely believed that functional-based screening holds far more potential for identifying entirely new enzymes with novel biocatalytic activities [##REF##16897828##23##].</p>", "<title>Marine microbes as good sources of novel biocatalysts</title>", "<p>All marine ecosystems are inhabited by microbes, which are both taxonomically diverse and metabolically complex. Marine microbes are both the primary producers of biomass in the oceans, harvesting light and fixing carbon, and the primary recyclers of nutrients. Microbial processes are essential for all the major cycles necessary for the maintenance of ocean life. Marine microbes are also known to be involved in the global cycling of bio-elements such as nitrogen, carbon, oxygen, phosphorous, iron, sulphur and trace elements. The precise contribution of marine microbes to these biogeochemical cycles is unknown. However, because of the versatility of their biochemical capabilities and the vast microbial biomass present in marine ecosystems, they are believed to be the main components responsible for the maintenance of these cycles, which help sustain all living things in these ecosystems. [##REF##17853905##24##].</p>", "<p>The marine environment is extremely diverse and marine microbes are exposed to extremes in pressure, temperature, salinity and nutrient availability. These distinct marine environmental niches are likely to possess highly diverse bacterial communities, possessing potentially unique biochemistry. Enzymes isolated from microbes from such environments are likely to have a range of quite diverse biochemical and physiological characteristics that have allowed the microbial communities to adapt and ultimately thrive in these conditions. For example bacteria which colonize marine snow are known to produce extracellular enzymes, whose function is to degrade proteins and polysaccharides within the snow [##REF##11734832##2##]. Thus the potential exists to exploit the enzymes produced by these marine microbes which are likely to possess unique bio-catalytic activities capable of functioning under extreme conditions.</p>", "<p>Microbes are also known to form symbiotic relationships with various marine invertebrates within diverse marine ecosystems; including species of sponges, corals, squids, sea squirts and tunicates among others. The symbioses between the marine invertebrates and the microbes are thought to offer each partner a number of advantages. For example, in the case of sponges symbiotic bacteria are believed to provide a source of nutrients for the sponge host, process sponge metabolic waste and produce secondary metabolites which play a role in the overall defence mechanisms of the sponge [##REF##17318533##25##]. These sponges are rich sources of biologically active and pharmacologically valuable natural products, with the potential for therapeutic use, and the bacterial symbionts of these sponges are widely believed to be the producers of many of these products [##REF##16598493##26##,##REF##15776313##27##]. Marine sponges are also known to contain large amounts of halogenated organic compounds such as fatty acids and alkaloids and are thus a potential useful source of both halogenases and dehalogenases a group of biotechnologically important enzymes which can be used in the production of pharmaceuticals, herbicides and pesticides [##REF##12839794##28##, ####REF##17574904##29##, ##REF##16544142##30##, ##REF##12738254##31####12738254##31##].</p>", "<p>The potential for the discovery of novel enzymes from marine microbes is illustrated by the fact that quite a diverse range of enzymatic activities have to date been identified from cultured marine microbes (Table ##TAB##0##1##), with the need for novel biocatalysts giving rise to many new enzymes being isolated from the marine environment. These include bacteria isolated from Antarctic seawater and marine sediments among others. Novel enzymes which have recently been identified from marine environments include a non-specific nuclease isolated from a bacteriophage which predates on the marine thermophile <italic>Geobacillus </italic>sp. 6K51. This enzyme has been shown to have no known homology to any previously isolated enzymes and a temperature optimum of 60°C [##REF##18439318##32##]. At the other end of the temperature scale are the cold-adapted enzymes such as the lipase isolated from the γ-proteobacterium, <italic>Pseudoalteromonas haloplanktis </italic>TAC125. This lipase is the first member of a new family of lipases, which share homology to the α/β hydrolases superfamily [##REF##18437283##33##]. Other recently reported enzymes include phospolipases [##REF##18293038##34##], extracelluar amylotic enzymes [##REF##18388462##35##], agarases [##REF##18071641##36##] and endochitinases [##REF##17690022##37##].</p>", "<p>Even environments such as the deep sea floor, where to date only limited numbers of cultivable bacteria have been isolated, mainly obligate sulphate reducing anaerobes and methanogens together with facultative anaerobic heterotrophs such as <italic>Halomonas </italic>and <italic>Psychrobacter</italic>, now appear to be reservoirs for microbes with enzymatic activities with potential biotechnological applications. For example a number of cultivable aerobic microbes have recently been isolated from the deep subseafloor sediments from off-shore the Shimokita peninsula in Japan at a water depth of 1,180 m. These microbes produced a variety of different enzymatic activities including protease, amylase, lipase, chitinase, deoxyribonuclease and phosphatase activities [##REF##18368287##38##].</p>", "<p>However, these novel enzymes have all been isolated from the cultivable fraction of microbes from the diverse range of marine environments which were studied. As we know this cultivable fraction represents only a small proportion of the total bacteria present in these environments. Hence, if the entire potential of these environments is to be explored and ultimately exploited with respect to the presence of novel biocatalysts; then there is a clear need to employ metagenomic or other culture-independent approaches together with robust heterologous expression systems to facilitate such an approach.</p>", "<title>Heterologous expression systems for functional expression of metagenomic libraries from marine environments</title>", "<p>The heterologous expression of novel genes or indeed gene clusters within suitable hosts will be required if new biocatalysts are to be identified from unculturable marine sources. However, significant challenges remain in using functional metagenomic based approaches to identify these enzymes; particularly when extreme environments, such as those present in many marine ecosystems, are being targeted. One of the main problems is that of low biomass yields, which coupled with low cell numbers from these marine environments, can lead to difficulties in obtaining high yields of DNA for subsequent cloning. Thus new methods are required to allow metagenomic library construction from environments with only low bacterial cell-densities. One potential method of overcoming this problem is the use of multiple displacement amplification, which has been successfully employed to assess the microbial diversity of scleractinian coral where environmental considerations require minimal sample sizes [##REF##16817924##39##]. In addition, for functional screening to be successful it requires gene expression and proper folding of the resulting protein in a suitable heterologous host. While it has been predicted that <italic>E. coli </italic>expression systems can be employed to successfully express up to 40% of genes from 32 complete genome sequences of various prokaryotic organisms, this is not always easily achievable [##REF##15305913##40##]. For example insolubility of the target protein remains a major limitation in <italic>E. coli </italic>expression systems, while in some structural genomics projects, up to half of the targets tested failed to fold properly and accumulated as insoluble protein or inclusion bodies [##REF##17992580##41##]. With this in mind a number of alternative bacterial host and expression systems are currently being examined. These include <italic>Streptomyces lividans</italic>, <italic>Pseudomonas putida </italic>and <italic>Rhizobium leguminosarum </italic>[##REF##15066844##42##,##REF##16309390##43##], which should facilitate the construction of metagenomic libraries with different expression capabilities thereby overcoming some of the difficulties being encountered with the <italic>E. coli </italic>system. In addition the possibility exists that functional expression of metagenomic DNA from psychrophilic marine microorganisms in heterologous hosts such as <italic>E. coli </italic>may be sub-optimal when the host is cultured at a higher temperature. The expression of proteins from these psychrophilic microorganisms at higher temperatures may lead to the mis-folding of these proteins and the subsequent formation of inclusion bodies, resulting in loss of function. While this may not always be the case [##REF##17064934##44##], in general psychrophilic microorganisms produce enzymes which have become cold-adapted, but which tend to have a low thermal stability. These problems may potentially be overcome through the use of chaperone-based <italic>E. coli </italic>strains bearing the chaperonin 60 gene (cpn60) and the cochaperonin 10 gene (cpn10) from the psychrophilic bacterium <italic>Oleispira antartica </italic>RB8^T[##REF##14595348##45##]. The use of these strains for metagenomic library construction could facilitate the functional screening of these libraries at temperatures down to 10°C, where increased levels of expression may in fact be observed [##REF##15294778##46##].</p>", "<title>Successes in the marine environment</title>", "<p>Researchers have also begun to employ metagenomic based approaches in an effort to isolate novel compounds from marine environments. Much of the emphasis to date has focused on the identification of gene clusters encoding novel biosynthesis pathways for compounds with potential bio-pharmaceutical applications, from bacterial populations associated with marine invertebrates [##REF##17318533##25##]. The metagenomic approaches being employed are similar to those that have previously been successfully employed in soil, which have resulted in the identification of among others; a biosynthesis gene cluster for the antibiotic violacein [##REF##11418029##47##]; novel <italic>N</italic>-acyltyrosine antibiotics [##REF##12188643##48##]; the novel antibiotic turbomycins [##REF##12200279##49##]; antibiotic compounds related to indirubin [##REF##11321587##50##]; and a family of novel natural products, the terragenines [##REF##10956506##51##]. Successes in the marine environment include identification of the biosynthetic machinery for the cytotoxic peptide patellamide from the cyanobacterial symbiont, <italic>Prochloron</italic>, of a marine didemnid ascidian. Two groups independently identified the biosynthesis genes for patellamide using a DNA sequencing approach and through a functional metagenomics approach. Patellamide biosynthesis was found to proceed though modification of a ribosomally encoded peptide and not via a non-ribosomal peptide synthetases as had been previously assumed [##REF##15988766##52##,##REF##15883371##53##].</p>", "<p>With respect to biocatalysts, a number of novel hydrolytic enzymes have recently been cloned from Antarctic sea water bacterial metagenomic DNA [##REF##18055055##54##], while a novel low-temperature-active lipase has also recently been isolated from a metagenomic library of Baltic Sea marine sediment bacteria. This low-temperature-active lipase gene which displayed 54% amino acid similarity to a <italic>Pseudomonas putida </italic>esterase was successfully heterologously expressed in <italic>E. coli </italic>and subsequently biochemically characterised [##REF##17328767##55##]. This highlights the value of employing metagenomics in marine environments to, in this case, identify novel lipases. Lipases have important applications not only in the detergent industry but also in paper processing, as food additives [##REF##10547694##56##] and in biofuel production through catalysing the conversion of vegetable oil to methylalcohol ester [##REF##12323363##57##]. They also have applications in synthetic organic chemistry, due primarily to their enantio-/stereoselectivity coupled with their ability to retain activity in organic solvents. Thus similar metagenomics approaches in other marine environments may prove fruitful in identifying other novel lipase genes, with biotechnological applications in these areas.</p>", "<p>Another example where metagenomics has been successfully employed to identify novel enzymes from a marine environment is the recent report of the cloning of two alkane hydroxylase genes from a metagenomic library from deep-sea sediment in the Pacific. While this is the first report of the genetic characterisation of an alkane hydroxylase from a deep-sea environment, it is also interesting to note that these two alkane hydroxylase genes were functionally expressed in a <italic>Pseudomonas fluorescens </italic>strain [##REF##18087672##17##]. Identification of these novel proteins may help to increase the range of alkane hydroxylase biocatalytic applications, and again highlights the utility of metagenomics in identifying potential novel biocatalysts.</p>", "<title>Challenges and future directions</title>", "<p>There are several key areas which need to be addressed in order for this exciting area to realise its full potential. The high microbial diversity of these marine environments means that large numbers of clones need to be screened in order to access the full biodiversity of the microbial community and to obtain significant numbers of \"hits\" which can then be taken forward for further analysis. This requirement means that high throughput screens need to be established to identify positive clones in metagenomic libraries. The development of cell-based ultra-high throughput screens would enable more rapid screening of large metagenomic libraries. Flow cytometry based approaches such as the SIGEX approach are leading the way in this area, since millions of clones can be screened in a short time and sorted for further analysis.</p>", "<p>While standard metagenomic approaches allow access to the full biodiversity of microbial populations, the nature of the metagenomic approach means that to access low abundance (&lt; 1%) microbes, extremely large libraries are required. These large libraries contain significant redundancy with respect to the high abundance microbes present within the population. A combination of cell sorting technology (FACS and microfluidics) and whole genome amplification approaches has the potential to significantly improve the ability to access the genetic resources of low-abundance microbes within any given population [##REF##17369337##58##, ####REF##17620602##59##, ##REF##17502618##60####17502618##60##]. These new approaches to access low abundance microbial diversity could be used to make targeted 'metagenomic' libraries of low abundance microbes and increase our ability to access this diversity.</p>", "<p>The functional metagenomic approach, typically used for biocatalyst screening, requires that the desired activity be expressed in a surrogate host, typically <italic>Escherichia coli</italic>. While <italic>E. coli </italic>has proven to be a flexible and useful host for heterologous expression there are a significant proportion of proteins that cannot be expression functionally in this host. In addition, many functional assays rely on the expression of entire metabolic pathways, requiring that promoters be recognised in the heterologous host for the co-ordinated expression of entire sets of genes. The range of surrogate hosts for heterologous expression and their associated vectors needs to be greatly expanded if we are to succeed in expressing DNA from the many diverse phyla that exist in the marine ecosystem, including the abundant Archaea and microeukarya. Novel complementation assays need to be developed whereby the metagenomic DNA complements a mutation in a surrogate host. Such methods would be very high throughput since they would allow the positive selection of desired clones. The ability to screen extremely large libraries also makes the use of easier to construct small insert libraries more practical.</p>", "<p>While metagenomic approaches offer a means to access the total potential of the microbial gene pool, other approaches need to be considered which may offer new avenues of investigation. One such approach may be to explore the metaproteome of the marine microbes (Fig ##FIG##0##1##). In metaproteomics the proteins present in an environmental sample are analysed using high throughput methods such as tandem mass spectroscopy [##REF##16402891##61##], comparison of this data to metagenomic DNA sequence data allows one to infer which genes are actively expressed in any population. Using appropriately designed probes this approach would allow the selection of such active genes for the screening and analysis of metagenomic libraries. Using such an approach in tandem perhaps with the more established metagenomic based approaches would greatly enhance the probability of obtaining novel biocatalysts.</p>" ]

[]

[]

[]

[ "<title>Conclusion</title>", "<p>This review is intended to focus the reader's attention on the potential of exploiting metagenomics, specifically the use of metagenomic libraries constructed from unique marine environments as an approach to successfully exploit the largely \"untapped\" resources within various marine environments. The marine environment is extremely diverse with microbes flourishing in cold polar and warm equatorial waters, in sunlit surface waters, in high-pressure deep sea sediments, in hot acidic water near hydrothermal vents and in association with various vertebrate and invertebrate animals. These diverse ecosystems are potentially very useful sources for novel enzymes with unique properties and great biotechnological potential. Only a small proportion of the bioresources of marine microbiota have thus far been examined and an even smaller proportion has been exploited. Given our present inability to culture the vast majority of microbes from these environments the metagenomic approaches outlined in this review offer the only methodology currently available to access these unique and useful bioresources. There is an ongoing need for a wide range of novel biocatalysts which are required to improve current and develop new, cleaner, industrial production processes, to reduce energy and raw material consumption, and for the generation of renewable biofuels; marine metagenomics coupled with biotechnology has the potential to contribute to all these pressing needs.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<p>Metagenomic based strategies have previously been successfully employed as powerful tools to isolate and identify enzymes with novel biocatalytic activities from the unculturable component of microbial communities from various terrestrial environmental niches. Both sequence based and function based screening approaches have been employed to identify genes encoding novel biocatalytic activities and metabolic pathways from metagenomic libraries. While much of the focus to date has centred on terrestrial based microbial ecosystems, it is clear that the marine environment has enormous microbial biodiversity that remains largely unstudied. Marine microbes are both extremely abundant and diverse; the environments they occupy likewise consist of very diverse niches. As culture-dependent methods have thus far resulted in the isolation of only a tiny percentage of the marine microbiota the application of metagenomic strategies holds great potential to study and exploit the enormous microbial biodiversity which is present within these marine environments.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>ADWD drafted the manuscript. JK and JRM contributed additional content throughout the article. All authors have read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>J.K. is in receipt of a Marie Curie Transfer of Knowledge Host Fellowship (Grant No. MTKD-CT-2006-042062). This project was funded by the Marine Biodiscovery Research award funded by the Irish Government under the National Development Plan (2007–2013).</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Omic based approaches to identify novel biocatalysts from marine ecosystems</bold>.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Enzymes from marine microbes</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\"><bold>Enzyme</bold></td><td align=\"center\"><bold>Producing Organism</bold></td><td align=\"center\"><bold>Habitat</bold></td><td align=\"center\"><bold>Ref</bold></td></tr></thead><tbody><tr><td align=\"center\">Alanine dehydrogenase</td><td align=\"center\">Psychrophilic bacterium strain PA-43</td><td align=\"center\">Sea Urchin</td><td align=\"center\">[##REF##12664266##62##]</td></tr><tr><td align=\"center\">Alcohol dehydrogenase</td><td align=\"center\"><italic>Flavobacterium frigidimaris </italic>KUC-1</td><td align=\"center\">Antarctic seawater</td><td align=\"center\">[##REF##17072683##63##]</td></tr><tr><td align=\"center\">Aminopeptidase</td><td align=\"center\"><italic>Colwellia psychrorythraea </italic>strain 34H</td><td align=\"center\">Marine sediment</td><td align=\"center\">[##REF##15184127##64##]</td></tr><tr><td align=\"center\">α-amylase</td><td align=\"center\"><italic>Nocardiopsis </italic>sp.</td><td align=\"center\">Deep sea sediment</td><td align=\"center\">[##REF##17934774##65##]</td></tr><tr><td align=\"center\">β-Galactosidase</td><td align=\"center\"><italic>Arthrobacter </italic>sp. SB</td><td align=\"center\">Antarctic sea water</td><td align=\"center\">[##REF##16736094##66##]</td></tr><tr><td align=\"center\">β-Galactosidase</td><td align=\"center\"><italic>Guehomyces pullulans</italic></td><td align=\"center\">Antarctic sea water</td><td align=\"center\">[##REF##16338594##67##]</td></tr><tr><td align=\"center\">Catalase</td><td align=\"center\"><italic>Vibrio salmonicida</italic></td><td align=\"center\">Fish microbiota</td><td align=\"center\">[##UREF##1##68##]</td></tr><tr><td align=\"center\">Cellulase</td><td align=\"center\"><italic>Pseudoaltermonas haloplanktis</italic></td><td align=\"center\">Antartic sea water</td><td align=\"center\">[##REF##15287848##69##]</td></tr><tr><td align=\"center\">Cellulase</td><td align=\"center\"><italic>Pseudoaltermonas </italic>sp. DY3</td><td align=\"center\">Deep-sea sediment</td><td align=\"center\">[##REF##16133657##70##]</td></tr><tr><td align=\"center\">Chitinase</td><td align=\"center\"><italic>Arthrobacter </italic>sp. TAD20</td><td align=\"center\">Antarctic ice</td><td align=\"center\">[##REF##11342059##71##]</td></tr><tr><td align=\"center\">Chitinase</td><td align=\"center\"><italic>Rhodothermus marinus</italic></td><td align=\"center\">Marine hot springs</td><td align=\"center\">[##REF##15583965##72##]</td></tr><tr><td align=\"center\">Esterase</td><td align=\"center\"><italic>Vibrio </italic>sp.</td><td align=\"center\">Sea Hare eggs</td><td align=\"center\">[##REF##17712554##73##]</td></tr><tr><td align=\"center\">Epoxide hydrolases</td><td align=\"center\"><italic>Erythrobacter litoralis </italic>HTCC2594</td><td align=\"center\">Seawater</td><td align=\"center\">[##REF##17541582##74##]</td></tr><tr><td align=\"center\">Feruloyl esterase</td><td align=\"center\"><italic>Pseudoaltermonas haloplanktis</italic></td><td align=\"center\">Antartic sea water</td><td align=\"center\">[##REF##18242884##75##]</td></tr><tr><td align=\"center\">β-D-glucosidase</td><td align=\"center\"><italic>Shewanella </italic>sp. G5</td><td align=\"center\"><italic>Munida subrrugosa </italic>(intestine)</td><td align=\"center\">[##REF##18247390##76##]</td></tr><tr><td align=\"center\">Homoserine transsuccinylase</td><td align=\"center\"><italic>Thermotoga maritima</italic></td><td align=\"center\">Marine sediment</td><td align=\"center\">[##REF##16708165##77##]</td></tr><tr><td align=\"center\">Isocitrate dehydrogenase</td><td align=\"center\"><italic>Colwellia psychrerythraea</italic></td><td align=\"center\">Arctic marine sediment</td><td align=\"center\">[##REF##16418792##78##]</td></tr><tr><td align=\"center\">Isocitrate lyase</td><td align=\"center\"><italic>Colwellia psychrerythraea</italic></td><td align=\"center\">Arctic marine sediment</td><td align=\"center\">[##REF##17965824##79##]</td></tr><tr><td align=\"center\">Lipase</td><td align=\"center\"><italic>Pseudoaltermonas haloplanktis </italic>TAC125</td><td align=\"center\">Antartic sea water</td><td align=\"center\">[##REF##18437283##33##]</td></tr><tr><td align=\"center\">Malate dehydrogenase</td><td align=\"center\"><italic>Flavobacterium frigidimar </italic>KUC-1</td><td align=\"center\">Antartic sea water</td><td align=\"center\">[##REF##16306697##80##]</td></tr><tr><td align=\"center\">Quinol oxidase</td><td align=\"center\"><italic>Shewanella </italic>sp. strain DB-172F</td><td align=\"center\">Deep-sea sediment</td><td align=\"center\">[##REF##9672683##81##]</td></tr><tr><td align=\"center\">Pectate lyase</td><td align=\"center\"><italic>Pseudoalteromonas haloplanktis </italic>strain ANT/505</td><td align=\"center\">Antarctic sea ice</td><td align=\"center\">[##REF##11302501##82##]</td></tr><tr><td align=\"center\">Proteases</td><td align=\"center\"><italic>Pseudoalteromonas, Shewanella, Colwellia, Planococcus </italic>species</td><td align=\"center\">Sub-Antarctic sediment</td><td align=\"center\">[##REF##17487446##83##]</td></tr><tr><td align=\"center\">Protease (alkaline)</td><td align=\"center\"><italic>Pseudomonas strain </italic>DY-A</td><td align=\"center\">Deep-sea sediment</td><td align=\"center\">[##REF##12910392##84##]</td></tr><tr><td align=\"center\">Proteases (serine)</td><td align=\"center\">Marine bacterium</td><td align=\"center\">Antartic sea water</td><td align=\"center\">[##REF##18055055##54##]</td></tr><tr><td align=\"center\">Proteases (serine)</td><td align=\"center\"><italic>Aeropyrum pernix </italic>K1</td><td align=\"center\">Coastal solfataric vent</td><td align=\"center\">[##REF##10086839##85##]</td></tr><tr><td align=\"center\">Subtilisin</td><td align=\"center\"><italic>Bacillus </italic>TA 41</td><td align=\"center\">Antartic sea water</td><td align=\"center\">[##REF##8021248##86##]</td></tr><tr><td align=\"center\">Trehalase</td><td align=\"center\"><italic>Rhodothermus marinus</italic></td><td align=\"center\">Marine hot springs</td><td align=\"center\">[##REF##16944251##87##]</td></tr><tr><td align=\"center\">Uracil-DNA Glycosylase</td><td align=\"center\">Marine bacterium strain BMTU3346</td><td align=\"center\">Marine sample</td><td align=\"center\">[##REF##10805566##88##]</td></tr><tr><td align=\"center\">Xylanase</td><td align=\"center\"><italic>Pseudoaltermonas haloplanktis</italic></td><td align=\"center\">Antartic sea water</td><td align=\"center\">[##REF##12089151##89##]</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1475-2859-7-27-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>A daily aspirin delays the progression of occlusive atherosclerotic vascular disease [##UREF##0##1##, ####REF##8298418##2##, ##REF##8312766##3##, ##REF##8054013##4##, ##REF##15494585##5####15494585##5##]. Patients with a history of myocardial infarction who are taking aspirin have a 25% decrease in adverse vascular events while patients with increasing symptoms of angina have a 50% decrease in vaso-occlusive events with aspirin therapy [##REF##11786451##6##, ####REF##1976875##7##, ##REF##6135989##8####6135989##8##]. However, the rapid progression of symptomatic occlusion in some patients prescribed aspirin has led to the notion that these patients are resistant to the effects of aspirin.</p>", "<p>Aspirin resistance has been loosely defined as decreased inhibition of platelets when measured using a platelet function assay or by quantitation of a serum or urinary metabolite of thromboxane B2 and has been described in up 45% of patients [##REF##8425313##9##, ####REF##12911581##10##, ##REF##14988166##11##, ##REF##16386660##12####16386660##12##]. The proportion of patients who fit a particular author's definition of aspirin resistance varies according to the method used to assess aspirin's antiplatelet effect as well as the somewhat arbitrary separation of patients into subsets of aspirin resistant and aspirin sensitive patients [##REF##17131625##13##]. Five studies and two meta-analysis show that patients with, \"aspirin resistance\", have a more rapid progression of their atherosclerotic disease [##REF##12651041##14##, ####REF##11940542##15##, ##REF##8236166##16##, ##REF##17008975##17##, ##REF##17056317##18##, ##REF##18202034##19##, ##REF##17698681##20####17698681##20##].</p>", "<p>Clinically important causes of aspirin resistance are noncompliance and non-aspirin non-steroidal anti-inflammatory drugs (NANSAIDs) interference with aspirin's effect and other hypothesized mechanisms like increased platelet turnover [##REF##11752357##21##, ####UREF##1##22##, ##REF##17319904##23####17319904##23##]. We have previously reported that approximately 9% of MI subjects who were presumed to be aspirin resistant were aspirin resistant because of noncompliance [##REF##15820166##24##]. Compliance was documented as a key factor in explaining decreased platelet inhibition with aspirin in five other reports (Table ##TAB##0##1##) [##REF##16785341##25##, ####REF##16256872##26##, ##REF##16270651##27##, ##REF##14760328##28##, ##REF##7974570##29####7974570##29##]. In order to properly treat a patient who appears to be aspirin resistant it is important for a clinician to have some estimate as to the likely cause.</p>", "<p>We evaluated our aspirin platelet function data to determine what proportion of subjects could be classified as non-compliant and what percentage of subjects could be classified as aspirin resistant from a cause that was independent of compliance [##UREF##2##30##]. Compliance with both off aspirin and 2 hour post aspirin was confirmed with arachidonic acid (AA) light aggregometry. The protocol required patients to stop aspirin and NANSAIDs for 7 days. Compliance was assured by watching the subjects ingest aspirin and by demonstrating diminished post aspirin AA stimulated platelet aggregation. The degree of aspirin induced inhibition of platelet function was assessed using platelet prostaglandin agonist (PPA) stimulated light aggregations measured when subjects were off aspirin and 2 hours after observed aspirin ingestion [##REF##12024110##31##]. The decrease in aspirin induced platelet response was used to calculate a novel measurement of aspirin effect, net aspirin inhibition. The presented data support the thesis that the predominant cause of aspirin resistance is non-compliance [##REF##17208069##32##].</p>" ]

[ "<title>Methods</title>", "<p>Subjects were contacted by phone, and after a detailed explanation of the study were invited to participate. Prior to the study informed consent was obtained. Inclusion criteria for the MI. patients were hospital admission for a myocardial infarction during the period between 1995 and 2000 and having been prescribed aspirin for at least of one month prior to study. Exclusion criteria were: history of aspirin noncompliance; primary care physician determination that the patient may not be withdrawn from aspirin; history of hemorrhagic cerebral vascular accident; coronary arteritis; thrombocytopenia; known hypercoagulable disorders; use of non-aspirin nonsteroidal anti-inflammatory drugs (NANSAIDs) or cyclooxygenase (COX)-2 inhibitors during the seven days prior to blood draw; uremia/dialysis or a creatinine greater than 3.0 mg/dl; and failure to provide written informed consent. In addition, thirty nine normal subjects who had not taken platelet inhibiting drugs for seven days and had no known vascular or renal disease were studied. Their measurements of platelet response were compared with those obtained from the myocardial infarction patients. Of the 350 subjects who met the study criteria, 250 agreed to participate and 230 completed the study. Of the 230 subjects 45 were excluded from the analysis of compliant subjects because their off aspirin aggregation responses to AA were less than 50% of maximal and one patient who admitted to violating the protocol's stipulation not to take NANSAIDs. These subjects were judged not to be compliant with the protocol's instructions to refrain from ingesting aspirin or NANSAIDs for 7 days. The 184 subjects (39 normals and 145 post myocardial infarction) who were compliant with the protocol were studied as known compliant subjects. The total number of compliant and noncompliant post-MI patients was 191. This study was approved by the Institutional Review Boards at McLaren Medical Center, Flint, MI. and at Ingham Regional Medical Center, Lansing, MI. and was carried out according to the principles of the Declaration of Helsinki.</p>", "<title>Light transmittance aggregometry</title>", "<p>Platelet function was measured using light transmittance aggregometry at 2 time points: after stopping aspirin for 7 days (off aspirin); and 2 hours after observed aspirin ingestion (on aspirin). Subjects were instructed to withhold all antiplatelet agents for a period of 7 days. Blood was drawn (off aspirin) for platelet aggregations and the subjects were then instructed to ingest a 325 mg aspirin tablet while the nurse watched. Two hours after the observed ingestion of aspirin (on aspirin) aggregation was assessed for the second time.</p>", "<p>Using a 21 g butterfly a minimum of 5 ml of whole blood was drawn by venipuncture into a separate syringe before an additional 9 ml of whole blood was collected using a 10 ml plastic syringe containing 1 ml of 3.2% sodium citrate.</p>", "<p>Platelet counts were performed using a Coulter AcT, Miami, Fl. Whole blood was centrifuged 200 × g for 10 minutes for platelet-rich plasma and 2000 × g for 15 minutes for platelet poor plasma.</p>", "<p>Light transmittance aggregometry was performed in duplicate using a Helena PACKS4 aggregometer, Beaumont, Tx. The final platelet concentration was adjusted with platelet poor plasma to 150,000/μL. Agreement between the duplicate aggregation curves yielded an intraclass correlation coefficient (ICC) of r = 0.99. (The ICC is a reliability statistic that reflects the extent to which two measurements agree.) To assess the amount of aspirin induced inhibition of platelet function we used the slope of the PPA (30 μM Analytical Control Systems, Fishers, In.) stimulated light aggregometry curve [##REF##12024110##31##]. We defined net aspirin inhibition as the difference between the PPA slopes off and on aspirin. Thus, it was important to know whether the subjects had complied with the protocol's dictum not to take aspirin for 7 days. Light transmittance aggregation with AA (1.0 mM Chronolog, Havertown, Pa.) was used to determine if subjects had refrained from aspirin ingestion for 7 days. AA aggregations were evaluated as percent of maximal aggregation. Normal, non-aspirin exposed, platelets aggregate. While on aspirin platelets exhibit minimal AA aggregation (Figure ##FIG##0##1##) [##REF##12024110##31##]. In this study subjects with less than 50% of maximal AA stimulated platelet aggregation were judged to be aspirin inhibited. Our prior publication showed that the 50% aggregation delimiter visually separated patients into two groups, normal and aspirin inhibited [##REF##15820166##24##].</p>", "<title>Statistical Analysis</title>", "<p>A general linear model was used to evaluate the relationship between the off aspirin PPA slope and the difference between the off and on aspirin PPA slopes. The patient with the largest net aspirin inhibitory response was determined to be an outlier using the Grubbs test for outlying points and was removed from the analysis for normal distribution [##UREF##3##33##]. The Kolmogorov-Smirnov test was used to determine if the data were normally distributed. Statistical significance was set at a p value of 0.05 or less. All statistical analysis was performed by Alpha Biostats, Reno NV. using SPSS software, Chicago, Il.</p>", "<p>The authors had full access to the data and take responsibility for its integrity. All authors have read and agreed to the manuscript as written.</p>" ]

[ "<title>Results</title>", "<title>Compliant Subjects</title>", "<p>The mean age for the 184 known compliant subjects was 63 ± 11 years, with 63% males and 37% females. The mean BMI for myocardial infarction patients was 29.4 with 39 percent having a BMI of more than 31. Thirty nine percent of the coronary artery disease patients had two or more documented myocardial infarctions. The percent of patients with additional risk factors for atherosclerotic vascular disease is presented in Table ##TAB##1##2##.</p>", "<p>All known compliant subjects had a greater than 50% decrease in aggregation response to AA two hours after observed aspirin ingestion. For these compliant subjects the mean off aspirin PPA aggregation curve slope was 57 ± 14. The mean on aspirin PPA aggregation slope was 15 ± 14. The mean difference between the PPA aggregation slopes off and on aspirin represents net aspirin inhibition and was 42 ± 16. Net aspirin inhibition demonstrated a normal distribution curve (Figure ##FIG##1##2##). No difference was observed between the 39 normal subjects and the 145 myocardial infarction patients for PPA aggregation slopes off and on aspirin or for net aspirin inhibition (p = 0.61).</p>", "<p>Decreased aspirin response could be defined using the calculated standard deviation. For example, if a decreased aspirin response is defined as the difference in PPA slope between off and on aspirin of one standard deviation or less, 16, then of the 184 compliant subjects 7 (3%) would be classified as having a decreased aspirin response (Figure ##FIG##2##3##). Of these 7 subjects 5 of 145 (3.4%) had prior myocardial infarctions and 2 of 39 were normal subjects.</p>", "<p>The amount of platelet inhibition by aspirin in compliant subjects was related to the off aspirin response (figure ##FIG##2##3##). As the slope of the off aspirin PPA stimulated aggregation curve increased, net aspirin inhibition increased. Even though net aspirin inhibition was proportional to the off aspirin PPA slope, those subjects with a one standard deviation or less decrease in net aspirin inhibitory response could not be identified by just using their PPA response off aspirin.</p>", "<title>Non-compliant subjects</title>", "<p>The mean age for the 191 post MI patients was 61 ± 13 with 66% males and 34% females. As previously reported 16 post-MI patients evaluated after having been prescribed a daily aspirin for at least one month had normal AA light aggregations. However, 2 hours after observed ingestion of 325 mg of aspirin AA aggregations were blocked demonstrating that non-compliance was the cause of the initial non-response to aspirin [##REF##15820166##24##]. An additional 45 patients were judged to be non-compliant with the protocol's stipulation that they stop taking aspirin for seven days because on the 7^thday of their proscribed abstinence from aspirin their AA aggregations demonstrated aspirin inhibition. One patient violated the protocol's stipulation not to take NANSAIDs. Of the 191 post-MI patients 62 (32%) were either non-compliant with their prescribed daily aspirin or with the protocols dictum of stopping aspirin and NANSAIDS for 7 days (Table ##TAB##2##3##).</p>" ]

[ "<title>Discussion</title>", "<p>After assuring compliance by observing post MI and normal subjects taking their aspirin, we determined that only 7 of 184 (3%) demonstrated decreased platelet inhibition that could be defined as aspirin resistance. There was no difference in net aspirin inhibitory response between normal adults and myocardial infarction patients. This suggests that MI patients who are stable enough to participate in a week-long out patient study have a response to aspirin that is similar to normal subjects.</p>", "<p>We used a novel parameter, net aspirin inhibition, to measure the degree of platelet inhibition produced by aspirin. The differences in PPA aggregrometry slopes off and on aspirin demonstrated a normal distribution suggesting that people possibly classified as aspirin resistant in this study were not a distinct population, but represent the lower portion of the bell shaped curve (Figure ##FIG##1##2##). Because net aspirin inhibition is a continuous variable we thought that designating those subjects with less than a one standard deviation decrease in net aspirin inhibitory response as possibly aspirin resistant seemed reasonable. However, separating aspirin sensitive from aspirin resistant subjects is an arbitrary designation. Perhaps a more clinically useful separation could be derived from a prospective study evaluating net aspirin inhibition as a predictor of future vascular events.</p>", "<p>Net aspirin inhibition was related to the off aspirin response (Figure ##FIG##2##3##). Because aspirin specifically and irreversibly blocks the platelet enzyme cyclooxygenase-1 (COX-1), the variation in observed aspirin inhibition may reflect individual differences in the resting platelet's dependence on activation via the arachidonic acid pathway [##REF##12024110##31##,##REF##10700490##34##,##REF##12413588##35##]. It may be that the subset of subjects with a decreased net aspirin inhibition are those who might benefit from an additional inhibitor of platelet function. This hypothesis needs to be tested in a future study. The information obtained from known compliant subjects confirms that platelet inhibitory response to aspirin is variable, but that a clearly delimited subset of people with a markedly decreased aspirin platelet inhibitory response cannot be defined.</p>", "<p>Current clinical considerations for aspirin resistance include patients who are non-compliant, have NANSAID interference with aspirin's ability to inhibit platelets or have a decreased aspirin response. Taken together with our prior report our post MI population of 191 patients had 16 subjects who were noncompliant while taking their daily aspirin another 46 subjects who were non-compliant with the protocol stipulation that they refrain from aspirin ingestion for 7 days and 1 patient who violated the protocols stipulation not to take NANSAIDs [##REF##15820166##24##]. If the above noncompliant subjects are removed from the analysis only 5 known compliant post MI subjects could be identified as having a decreased aspirin effect that could be labeled as aspirin resistant. Of the 67 post MI subjects with an aberrant aspirin effect on platelets 62 (93%) were because of non-compliance with either their prescribed daily aspirin or with the protocol's direction not to take aspirin or an NANSAID for 7 days (Table ##TAB##1##2##).</p>", "<p>Poor platelet inhibition by aspirin is associated with an increase in rate of occlusive atherosclerotic disease [##REF##12651041##14##, ####REF##11940542##15##, ##REF##8236166##16##, ##REF##17008975##17##, ##REF##17056317##18####17056317##18##]. A meta-analysis confirmed the increased risk of occlusive vascular disease in patients classified as aspirin resistant [##REF##18202034##19##]. In this meta-analysis the risk for vascular disease was not decreased in those patients who were prescribed Plavix to treat their aspirin resistance. Our data supports the thesis that most of the aspirin resistant patients are resistant because of non-compliance. Perhaps Plavix's lack of benefit in aspirin resistant patients is also because of non-compliance.</p>", "<p>Compliance in 1521 myocardial infarction patients was investigated by asking patients to list their medications [##REF##17000940##36##]. Patients who discontinued their daily aspirin one month after their myocardial infarction had a lower survival rate at one year compared to compliant patients 91% vs 97%, p &lt; 0.001 [##REF##17000940##36##]. Strict enforcement of compliance can improve the percentages of patients whose platelets are appropriately inhibited by aspirin [##REF##16270651##27##]. These data suggest that discovering which patient has a problem with compliance could improve the health of patients with CAD.</p>" ]

[ "<title>Conclusion</title>", "<p>Aspirin resistance is an unfortunate descriptor. It suggests an inherited or acquired defect for the ability of aspirin to acetylate platelet COX-1. Our current data interpreted in the context of our prior publication as well as several recent reviews reinforces the importance of patient compliance as a cause of poor inhibition of platelets or aspirin resistance [##REF##15820166##24##,##REF##17180149##37##,##REF##17208069##32##,##REF##17340471##39##]. In view of these findings we suggest that a more accurate nomenclature for patients with poor platelet inhibition by aspirin might rely on the etiology of the poor aspirin response and would accommodate non-compliance, drug interactions and other possible causes.</p>", "<p>Platelet stimulation with AA easily stratifies patients as compliant or noncompliant while PPA stimulated aggregometry allows identification of patients whose platelets demonstrate decreased inhibition with aspirin. For the clinician who is confronted with a patient with increasing CAD symptoms the question of aspirin resistance because of compliance or interference with NANSAIDs or decreased aspirin response represents a problem. If the patient is concurrently taking both aspirin and an NANSAID then the appropriate sequence of these two medication needs to be emphasized [##REF##11752357##21##]. Conversely, if the patient truly has a decreased aspirin response, then an additional anti-platelet agent may be warranted. However, our data suggest that the preponderant cause of poor platelet inhibition with aspirin is non-compliance and that the clinician should be encouraged to work to increase the patient's compliance with the prescribed daily aspirin.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Our previous publication showed that 9% of patients with a history of myocardial infarction MI. could be labeled as aspirin resistant; all of these patients were aspirin resistant because of non-compliance. This report compares the relative frequency of aspirin resistance between known compliant and non-compliance subjects to demonstrate that non-compliance is the predominant cause of aspirin resistance.</p>", "<title>Methods</title>", "<p>The difference in the slopes of the platelet prostaglandin agonist (PPA) light aggregation curves off aspirin and 2 hours after observed aspirin ingestion was defined as net aspirin inhibition.</p>", "<title>Results</title>", "<p>After supposedly refraining from aspirin for 7 days, 46 subjects were judged non-compliant with the protocol. Of the remaining 184 compliant subjects 39 were normals and 145 had a past history of MI. In known compliant subjects there was no difference in net aspirin inhibition between normal and MI subjects. Net aspirin inhibition in known compliant patients was statistically normally distributed. Only 3% of compliant subjects (2 normals and 5 MI) had a net aspirin inhibitory response of less than one standard deviation which could qualify as a conservative designation of aspirin resistance. A maximum of 35% of the 191 post MI subjects could be classified as aspirin resistant and/or non-compliant: 9% aspirin resistant because of non-compliance, 23% non-compliant with the protocol and possibly 3% because of a decreased net aspirin inhibitory response in known compliant patients.</p>", "<title>Conclusion</title>", "<p>Our data supports the thesis that the predominant cause of aspirin resistance is noncompliance.</p>" ]

[ "<title>Abbreviations</title>", "<p>MI: myocardial infarction; NANSAIDs: non-steroidal anti-inflammatory drugs; AA: light aggregometry; PPA: platelet prostaglandin agonist; COX: cyclooxygenase; CAD: coronary arterial disease.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>KAS designed the study, analyzed the data and wrote the manuscript. DES designed the study, edited the manuscript, evaluated the data and performed platelet laboratory evaluations. KB collected and organized data. MR designed the study and analyzed the data. ACDF designed the study and performed data collection and analysis. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>Institution where work was carried out – Ingham Regional Medical Center and McLaren Regional Medical Center</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Note that PPA demonstrates a gradual return of platelet aggregation to normal over 3 days.</bold> This characteristic of PPA stimulated aggregation makes it useful for measuring gradations of aspirin induced platelet inhibition. AA aggregation remains unresponsive for 3 days and returns to normal function between days 3 and 4. AA stimulated platelet aggregations were used to show if aspirin platelet inhibition was present or absent [##REF##12024110##31##].</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>When the single point with the largest aspirin response is removed as a statistical outlier, the net aspirin inhibitory response distribution curve is judged to be normally distributed.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>The seven subjects with less than 1 standard deviation decrease in their on aspirin slopes are depicted by open squares (□), those with a decrease in PPA slope between 1 and 2 standard deviations by solid diamonds (◆) and those with a greater than 3 standard deviation decrease by open circles (○).</bold> A direct relationship is observed between PPA slope off aspirin and the net aspirin inhibitory response. (p &lt; 0.001).</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Demonstration of Aspirin non-Compliance by Repeat Testing</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td/><td align=\"center\" colspan=\"2\">Methods for</td><td/></tr><tr><td/><td/><td colspan=\"2\"><hr/></td><td/></tr><tr><td align=\"center\">N</td><td align=\"center\">% non-Compliant</td><td align=\"center\">ASA Effect</td><td align=\"center\">Repeat Testing for Compliance After</td><td align=\"center\">Reference</td></tr></thead><tbody><tr><td align=\"center\">192</td><td align=\"center\">9.0</td><td align=\"center\">AA Light Aggregometry</td><td align=\"center\">Observed ASA ingestion</td><td align=\"center\">[##REF##15820166##24##]</td></tr><tr><td align=\"center\">212</td><td align=\"center\">14.0</td><td align=\"center\">PFA-100</td><td align=\"center\">Strict reinforcement of compliance</td><td align=\"center\">[##REF##16270651##27##]</td></tr><tr><td align=\"center\">203</td><td align=\"center\">3.4</td><td align=\"center\">Thromboelastography</td><td align=\"center\">Hospitalization</td><td align=\"center\">[##REF##16256872##26##]</td></tr><tr><td align=\"center\">73</td><td align=\"center\">16.0</td><td align=\"center\">Thromboxane B2: plasma</td><td align=\"center\">Admitted to non-compliance</td><td align=\"center\">[##REF##14760328##28##]</td></tr><tr><td align=\"center\">87</td><td align=\"center\">20.0</td><td align=\"center\">Collagen Light Aggregometry</td><td align=\"center\">Admitted to non-compliance</td><td align=\"center\">[##REF##7974570##29##]</td></tr><tr><td align=\"center\">678</td><td align=\"center\">2.0</td><td align=\"center\">AA light Aggregometry</td><td align=\"center\">Ex vivo ASA</td><td align=\"center\">[##REF##16785341##25##]</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>MI Patient Clinical Measures</p></caption><table frame=\"hsides\" rules=\"groups\"><tbody><tr><td align=\"left\">Mean Age</td><td align=\"center\">63 ± 11 years</td></tr><tr><td align=\"left\">Mean BMI</td><td align=\"center\">29.4 ± 5.6</td></tr><tr><td align=\"left\">BMI &gt; 31</td><td align=\"center\">39%</td></tr><tr><td align=\"left\">Two or More Mis</td><td align=\"center\">33%</td></tr><tr><td align=\"left\">Smokers</td><td align=\"center\">39%</td></tr><tr><td align=\"left\">Diabetes</td><td align=\"center\">29%</td></tr><tr><td align=\"left\">Hypertension (≥ 140/≥ 90)</td><td align=\"center\">52%</td></tr><tr><td align=\"left\">Total Cholesterol &gt;200</td><td align=\"center\">28%</td></tr><tr><td align=\"left\">Total Cholesterol &gt;240</td><td align=\"center\">6%</td></tr><tr><td align=\"left\">HDL &lt; 40</td><td align=\"center\">50%</td></tr><tr><td align=\"left\">LDL &gt; 130</td><td align=\"center\">31%</td></tr><tr><td align=\"left\">LDL &gt; 159</td><td align=\"center\">9%</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Post MI Subjects with Aberrant Platelet Response to Aspirin</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"left\"><bold># of subjects</bold></td></tr></thead><tbody><tr><td align=\"left\"><bold>Non-Compliant</bold></td><td/></tr><tr><td align=\"left\"> 1. Prescribed daily aspirin</td><td align=\"left\">16</td></tr><tr><td align=\"left\"> 2. Protocol directive not to take Aspirin for 7 days</td><td align=\"left\">45</td></tr><tr><td align=\"left\"> 3. Protocol directive not to take NANSAIDs for 7 days</td><td align=\"left\">1</td></tr><tr><td/><td/></tr><tr><td align=\"left\"><bold>Protocol Compliant</bold></td><td/></tr><tr><td align=\"left\"> &lt;1 S.D. decrease in Net Aspirin Inhibition</td><td align=\"left\">5</td></tr><tr><td align=\"right\">Total</td><td align=\"left\">67</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>NOTES:</p><p>Clinical measures were not available for the control subjects.</p><p>Not all MI patients have complete clinical measures data.</p></table-wrap-foot>", "<table-wrap-foot><p>93% (62 of 67) of aberrant platelet responses to aspirin were due to non-compliance. Among compliant post MI subjects 3% may be classified as aspirin resistant.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1479-5876-6-46-1\"/>", "<graphic xlink:href=\"1479-5876-6-46-2\"/>", "<graphic xlink:href=\"1479-5876-6-46-3\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Peroxisome proliferator-activated receptor-γ (PPARγ) has been shown to be expressed in many of the cells that play a role in the response to vascular injury and to modulate the actions that are thought to initiate neointimal (NI) growth, including inflammation [##REF##14517165##1##, ####REF##15585194##2##, ##REF##16020748##3##, ##REF##16740417##4####16740417##4##]. Neointimal formation is an important structural change in the vessel wall that leads to restenosis after angioplasty or stenting [##REF##9862273##5##, ####REF##7586270##6##, ##REF##3417980##7##, ##REF##8479518##8####8479518##8##].</p>", "<p>Thiazolidinediones consist of a family of synthetic compounds that acts as high-affinity ligands for PPARγ and were originally developed to facilitate glucose control in patients with type 2 diabetes. In addition, they have a direct impact on vascular cells and reduce circulating factors that are associated with atherosclerosis [##REF##11742860##9##]. In a recent meta-analysis of randomized controlled trials there was evidence that thiazolidinedione therapy in patients undergoing coronary stent implantation may be associated with less in-stent restenosis and repeated revascularization [##REF##17584567##10##, ####REF##17584566##11##, ##REF##17259517##12####17259517##12##]. Three different thiazolidinediones, rosiglitazone, pioglitazone, and troglitazone, have been shown to prevent balloon-injured rat carotid arteries [##REF##11742860##9##]. Rosiglitazone can reduce the NI formation and macrophage content in a mouse injury model [##REF##14517165##1##] and in hypercholesterolemic rabbits [##REF##15585194##2##]. These effects were independent of glycemic control or changes in lipid concentrations [##REF##9370113##13##]. In the present study we analyse the effects of rosiglitazone (RGZ) on neointimal formation administered at different times in hypercholesterolemic rabbits following vascular injury.</p>" ]

[ "<title>Methods</title>", "<title>Animals</title>", "<p>Thirty-nine white adult male rabbits (New Zealand), weighing 2.474 ± 348 Kg, were utilized for this experiment. Animals were handled in compliance with the Guiding Principles in the Care and Use of Animals. Protocol approval was obtained from the Pontifical Catholic University Animal Research Committee. During first 14 days the animals were fed a hypercholesterolemic diet (1% cholesterol-Sigma-Aldrich^®). Subsequently, they were changed to a 0.5% cholesterol diet until sacrifice (42 days). The animals were divided into three groups as follows: control group (CG) 13 rabbits without RGZ; group I, 13 rabbits treated with RGZ from the fifteenth day (after the vascular injury) until sacrifice; and group II, 13 rabbits treated with RGZ during the entire experiment (42 days). Rosiglitazone was administered by oral gavage (3 mg/Kg body weight/day).</p>", "<title>Vascular injury</title>", "<p>The rabbits underwent balloon catheter (20 × 3 mm/5 atm/5 min) injury of the right iliac artery on the fourteenth day of the experiment. Anesthesia was induced with ketamine (Vetanarcol^®-König – 3,5 mg/Kg) and intramuscular xylazine (Coopazine^®-Coopers – 5 mg/Kg). After the procedure the animals received intramuscular analgesics for 3 days (25 mg/day of flunixin – Banamine^®– Schering-Plough) and intramuscular antibiotics for 4 days (100 mg/day of oxitetraciclin – TormicinaP^®-Toruga). The rabbits were sacrificed by a lethal barbiturate dose on day 42 and their aorta and iliac arteries were removed for immunohistochemical and histological analysis.</p>", "<title>Quantitative histopathology</title>", "<p>Histological analysis was performed by an experienced pathologist (LN) unaware of the RGZ treatment. The analyses was done with a microscope attached to the Image Pro-plus^®4.5 Software (Media Cybernetics Inc. Silver Spring, MD. USA). Histomorphometric parameters were performed by calculation of the luminal and intimal layer area, and intima/media layer area ratio (the area of the intimal layer divided by the area of the medial layer) according to the method described by Phillips et al [##REF##14517165##1##]. The quantification of total collagen was made by the Sirius red polarization method [##UREF##0##14##]. Atherosclerotic lesions were analysed and classified according to Virmani et al [##REF##10807742##15##].</p>", "<title>Immunohistochemistry</title>", "<p>Tissue preparation and immunohistological techniques were performed according to the manufacturer's instructions included in the kits (Dako Corporation, Carpinteria, Calif). Sections were stained for macrophage cells using primary monoclonal antibody RAM-11(Dako^®, Carpinteria, CA), and for alpha-actin smooth muscle cells with primary polyclonal antibody HHF-35 (Dako^®, Carpinteria, CA). For quantitative immunocytochemical comparisons of macrophage content or smooth muscle cell content in intimal area, sections were computed and scored in 2 categories based on less than or more than 50% of cells in the balloon injury area.</p>", "<title>Blood chemistry</title>", "<p>Blood samples were obtained on first day of the experiment, immediately before balloon catheter injury, and immediately before sacrifice by cardiac puncture. Clinical laboratory assessment included fasting serum glucose, total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), and triglycerides (TGC). Measurements were done using an automated system (Abbott Architect ci8200; Abbott Laboratories, Abbott Park, Il).</p>", "<title>Statistical analysis</title>", "<p>The calculation of sample size was done based on the study of Wang Zhao-hui, Luo Feng and Liu Xiao-mei [##REF##16499094##16##]. The ratio between the intimal layer and the media layer was considered to be the main variable of interest. In order to detect a minimum difference of 0.15 between groups averages, with a significance level of 5% and power of the test of 80%, the minimum number of animals in each group of the study was defined as 12. Categorical variables were expressed as percentages and continuous variables were expressed as mean ± SD and medians. Data were compared using Anova one-way. The normality of the samples was tested by using Shapiro-Wilk tests. For non-normal samples, the Kruskal-Wallis and Mann-Whitney non parametric tests were used to compare the groups. Fisher's exact test was used for qualitative or categorical variables. Statistical significance was indicated by a value of <italic>p </italic>&lt; 0.05. Analyses were performed using SPSS version 14.0 (SPSS, Inc., Chicago, Illinois).</p>" ]

[ "<title>Results</title>", "<title>Metabolic and lipid profiles</title>", "<p>The rabbit's weight did not differ between groups (data not shown). Baselineglucose, total cholesterol, HDL-cholesterol and triglycerides levels were equal in all groups before initiation of the diet. On day 14, two weeks after feeding, fasting glucose levels were higher in CG and group I. At the time of sacrifice glucose levels did not differ between groups. A graded elevation in TC and TGC levels was observed from the initial phase through the vascular lesion until sacrifice without significant differences between groups. A graded elevation in HDL-C was observed in all three groups. Higher levels of HDL-C were observed in group II versus CG and group I at the time of vascular injury and sacrifice (Table ##TAB##0##1##).</p>", "<title>Histomorphometry</title>", "<p>Intimal area was significantly lower in group II vs. CG (<italic>p </italic>= 0.024) and group I (<italic>p </italic>= 0.006). Luminal layer area was higher in group II vs. CG (<italic>p </italic>&lt; 0.0001) and group I (<italic>p </italic>&lt; 0.0001). There was a significant reduction of 65% and 71% in intima/media layer area ratio (IMR) in group II vs. CG (<italic>p </italic>= 0.021) and vs. group I (<italic>p </italic>= 0.003), respectively. Intima/media layer area ratio was equal between CG and group I. (Table ##TAB##1##2##). (Figures ##FIG##0##1## and ##FIG##1##2##). According to the histological analysis proposed by Virmani et al, none of the criteria from1 trough 9 were found in group II, therefore the comparisons were only made between CG and group I. Neointimal growth, xanthomatous macrophages, proteoglican matrix, the presence and the thickness of fibrous cap, and the presence of calcification did not differ between CG and group I. There was no deposit of collagen into intimal or medial layers in group II, nor were there differences in the extent of collagen deposition between CG and group I. (Table ##TAB##2##3##).</p>", "<title>Immunohistochemistry</title>", "<p>There was no significant difference in macrophage and smooth muscle cell content in the intimal layer between CG and group I (data not show). Group II did not present any intimal cell markers.</p>" ]

[ "<title>Discussion</title>", "<p>Prevention of restenosis after balloon coronary angioplasty or stent implantation with the use of local and systemic therapy is a challenging issue in interventional cardiology [##REF##16740417##4##,##REF##12055567##17##, ####REF##16267248##18##, ##REF##12478229##19##, ##REF##11079654##20####11079654##20##]. Osborne et al [##REF##2913050##21##] showed that a short term model of hypercholesterolemia (two to four weeks) prevents extremely high cholesterol values and formation of advanced atherosclerotic plaques. Nevertheless, the arteries isolated from animals fed a cholesterol-enriched diet developed defects in endothelium-dependent relaxation in both large vessels as well as coronary resistance vessels [##UREF##1##22##]. These effects could be, in part, responsible for the restenosis after balloon angioplasty. Thiazolidinediones have immunomodulatory and antiproliferative effects, independent of their actions in metabolic control and are expressed in most cell types of the vascular wall as in atherosclerotic lesions, where they can affect atherogenic process [##REF##17363721##23##, ####REF##15023560##24##, ##REF##11135615##25##, ##REF##15505001##26##, ##UREF##2##27##, ##REF##16044531##28##, ##REF##16801577##29##, ##REF##11213897##30####11213897##30##]. To investigate the effects of a PPARγ ligand (rosiglitazone) on atherogenesis in an animal model, we used rabbits with six-fold increased cholesterol levels at the time of vascular injury and fourteen-fold increased levels at the time of euthanasia. This animal model was based on previous studies where rabbits develop hypercholesterolemia rapidly after excessive cholesterol feeding [##REF##15585194##2##,##REF##16020748##3##,##UREF##1##22##,##REF##15023560##24##]. The metabolic effects of high cholesterol-containing diet on rabbits were extensively explained in our previous study [##REF##18485218##31##]. Rosiglitazone was used at different times for each group. Group II not only did not present atherosclerotic lesions but also did not show any deposit of collagen or macrophage and smooth muscle cell markers in their intimal layer. The most significant findings were identified in the higher luminal area and the lower intimal area in which rabbits were treated with RGZ before vascular injury. Furthermore, in CG and group I intense reparative response occurred, with exuberant neointimal formation and reduction of luminal area. In addition, immunohistochemical analysis demonstrated a reduced macrophage and smooth muscle cell recruitment into the vascular arterial wall when RGZ was used two weeks before catheter balloon injury. Rosiglitazone did not exert anti-atherosclerotic activity when administered after vascular injury, however, a lesser density of macrophages in the media layer was observed in the animals of group I. We cannot rule out that these effects were due to chance, as our evaluation period was short. These findings suggest a possible protective effect of this drug against neointimal proliferation and remodeling responsible for restenosis after a balloon angioplasty. This is the first study to show the effects of a PPARγ ligand on vascular injury at different times and to document the benefits of pre-treatment with RGZ in hypercholesterolemic rabbits. Nevertheless, this drug has been the focus of extensive discussion in recent publications [##REF##17517853##32##, ####REF##17848659##33##, ##REF##17848653##34##, ##UREF##3##35##, ##REF##17551159##36##, ##REF##18378631##37####18378631##37##]. Nissen and Wolski [##REF##17517853##32##] published a meta-analysis showing a significant increase in the risk of myocardial infarction and an increase in cardiovascular death of borderline significance in patients with diabetes receiving RGZ. Singh et al [##REF##17848659##33##] also published a meta-analysis showing a significantly increased risk of myocardial infarction and heart failure among patients with impaired glucose tolerance or type 2 diabetes using rosiglitazone for at least 12 months, with no significantly increased risk of cardiovascular mortality. Lipscombe et al [##REF##17848653##34##], in a nested case-control analysis of a retrospective cohort study, found that in diabetes patients with an age of 66 years or older, RGZ treatment was associated with an increased risk of congestive heart failure, acute myocardial infarction, and mortality when compared with other combination oral hypoglycemic agent treatments. The mechanism for the apparent increase in myocardial infarction and death from cardiovascular causes associated with RGZ remains uncertain. In the PERISCOPE randomized controlled trial [##REF##18378631##37##], using coronary intravascular ultrasonography, the authors found a significantly lower rate of progression of coronary atherosclerosis in patients treated with pioglitazone when compared with glimiperide. However, it is not possible to extend the positive or negative benefit of one drug to another in the same class. In the next three years, we hope that the final results of the studies RECORD and BARI-2D [##REF##12006382##38##], specifically evaluating cardiovascular effects of RGZ, will provide useful insights.</p>" ]

[ "<title>Conclusion</title>", "<p>The results of our study indicate that when rosiglitazone is administered in hypercholesterolemic rabbits before, but not after, undergoing vascular injury, there is significantly reduced neointimal formation.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Objectives</title>", "<p>To analyse the effects of rosiglitazone administered at different times on neointimal formation in hypercholesterolemic rabbits following vascular injury.</p>", "<title>Methods</title>", "<p>Thirty-nine rabbits on a hypercholesterolemic diet were included. The animals underwent balloon catheter injury to the right iliac artery on day 14. They were divided into three groups as follows: control group, 13 rabbits without rosiglitazone; group I, 13 rabbits treated with rosiglitazone (3 mg/Kg body weight/day) for 28 days after the vascular injury; and group II, 13 rabbits treated with rosiglitazone (3 mg/Kg body weight/day) during all the experiment (42 days). Histological analysis was done by an experienced pathologist who was unaware of the rosiglitazone treatment. Histomorphometric parameters were performed by calculation of the luminal and intimal layer area, and intima/media layer area ratio (the area of the intimal layer divided by the area of the medial layer).</p>", "<title>Results</title>", "<p>Intimal area was significantly lower in group II vs. CG (<italic>p </italic>= 0.024) and group I (<italic>p </italic>= 0.006). Luminal layer area was higher in group II vs. CG (<italic>p </italic>&lt; 0.0001) and group I (<italic>p </italic>&lt; 0.0001). Intima/media layer area ratio was equal between CG and group I. Intima/media layer ratio area was significantly lower in group II vs. control group (<italic>p </italic>&lt; 0.021) and group I (<italic>p </italic>&lt; 0.003). There was a significant reduction of 65% and 71% in intima/media layer area ratio in group II vs. control group and group I, respectively.</p>", "<title>Conclusion</title>", "<p>Pretreatment with rosiglitazone in hypercholesterolemic rabbits submitted to vascular injury significantly reduces neointimal formation.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>AA participated in the study design, ORFN participated in the study design, PRSB oriented in the surgical procedures, CP oriented in the management of the animals, LN made the histological examination, RFKCS oriented in the surgical procedures, LAVB wrote and oriented the manuscript, DBP participated in the study design. All authors read and approved the final manuscript.</p>" ]

[]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Representative histological sections demonstrating neointimal formation – Orcein Staining.</bold><bold>Panel A</bold>: Control group. <bold>Panel B</bold>: Group I. <bold>Panel C</bold>: Group II. NI represents neointima.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Quantification of intima/media layer area ratio; and total intimal layer area.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Metabolic and lipid profiles (mean ± SD)</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td/><td align=\"left\">CG</td><td align=\"left\">Group I</td><td align=\"left\">Group II</td><td align=\"left\">P value</td></tr></thead><tbody><tr><td align=\"left\">Baseline</td><td align=\"left\">TC (mg/dl)</td><td align=\"left\">58.62 ± 25.08</td><td align=\"left\">50.77 ± 18.39</td><td align=\"left\">43.54 ± 14.82</td><td align=\"left\">NS</td></tr><tr><td/><td align=\"left\">HDL-C (mg/dl)</td><td align=\"left\">24.23 ± 5.31</td><td align=\"left\">23.1 ± 6.12</td><td align=\"left\">22.69 ± 6.26</td><td align=\"left\">NS *</td></tr><tr><td/><td align=\"left\">TGC (mg/dl)</td><td align=\"left\">79.62 ± 26.64</td><td align=\"left\">91.15 ± 32.71</td><td align=\"left\">79.85 ± 34.31</td><td align=\"left\">NS *</td></tr><tr><td/><td align=\"left\">Glucose (mg/dl)</td><td align=\"left\">121.38 ± 17.43</td><td align=\"left\">120.77 ± 19.07</td><td align=\"left\">117.92 ± 11.44</td><td align=\"left\">NS</td></tr><tr><td align=\"left\">Vascular injury</td><td align=\"left\">TC (mg/dl)</td><td align=\"left\">524.38 ± 258.94</td><td align=\"left\">431.77 ± 197.38</td><td align=\"left\">318.46 ± 212.86</td><td align=\"left\">NS</td></tr><tr><td/><td align=\"left\">HDL-C (mg/dl)</td><td align=\"left\">32.46 ± 19.66</td><td align=\"left\">28.77 ± 6.15</td><td align=\"left\">50.69 ± 21.91</td><td align=\"left\">NS *</td></tr><tr><td/><td align=\"left\">TGC (mg/dl)</td><td align=\"left\">72.77 ± 34.95</td><td align=\"left\">71.54 ± 46.54</td><td align=\"left\">86.92 ± 56.34</td><td align=\"left\">NS *</td></tr><tr><td/><td align=\"left\">Glucose (mg/dl)</td><td align=\"left\">250.23 ± 93.02</td><td align=\"left\">274.46 ± 58.45</td><td align=\"left\">166.62 ± 38.2</td><td align=\"left\">0.001</td></tr><tr><td align=\"left\">Sacrifice</td><td align=\"left\">TC (mg/dl)</td><td align=\"left\">852.46 ± 308.48</td><td align=\"left\">702.62 ± 261.53</td><td align=\"left\">593.54 ± 219,86</td><td align=\"left\">NS</td></tr><tr><td/><td align=\"left\">HDL-C (mg/dl)</td><td align=\"left\">42.62 ± 38.23</td><td align=\"left\">25.08 ± 13.93</td><td align=\"left\">69.08 ± 19.7</td><td align=\"left\">0,001 *</td></tr><tr><td/><td align=\"left\">TGC (mg/dl)</td><td align=\"left\">126.77 ± 85.66</td><td align=\"left\">398.08 ± 509.76</td><td align=\"left\">277.31 ± 248.14</td><td align=\"left\">NS *</td></tr><tr><td/><td align=\"left\">Glucose (mg/dl)</td><td align=\"left\">212.85 ± 73.71</td><td align=\"left\">210.92 ± 72.99</td><td align=\"left\">233.85 ± 89.65</td><td align=\"left\">NS</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Quantitative histopathological analysis</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Area</td><td align=\"left\">Group</td><td align=\"left\">Mean</td><td align=\"left\">DP</td><td align=\"left\">Minimum</td><td align=\"left\">Maximum</td><td align=\"left\">P</td></tr></thead><tbody><tr><td align=\"left\">Intimal area</td><td align=\"left\">CG</td><td align=\"left\">320340.22</td><td align=\"left\">392880.74</td><td align=\"left\">14720.20</td><td align=\"left\">1512612.11</td><td align=\"left\">0.024</td></tr><tr><td/><td align=\"left\">GI</td><td align=\"left\">282659.14</td><td align=\"left\">346471.14</td><td align=\"left\">14500.40</td><td align=\"left\">1361362.80</td><td align=\"left\">0.006</td></tr><tr><td/><td align=\"left\">GII</td><td align=\"left\">83115.01</td><td align=\"left\">65440.66</td><td align=\"left\">16187.50</td><td align=\"left\">269226.60</td><td/></tr><tr><td align=\"left\">Luminal area</td><td align=\"left\">CG</td><td align=\"left\">458711.01</td><td align=\"left\">363013.82</td><td align=\"left\">1853.54</td><td align=\"left\">1773080.00</td><td align=\"left\">&lt;0.0001</td></tr><tr><td/><td align=\"left\">GI</td><td align=\"left\">556450.31</td><td align=\"left\">330540.15</td><td align=\"left\">3274.59</td><td align=\"left\">1486461.00</td><td align=\"left\">&lt;0.0001</td></tr><tr><td/><td align=\"left\">GII</td><td align=\"left\">861255.24</td><td align=\"left\">303153.71</td><td align=\"left\">222741.70</td><td align=\"left\">1586336.00</td><td/></tr><tr><td align=\"left\">IMR</td><td align=\"left\">CG</td><td align=\"left\">0.50</td><td align=\"left\">0.41</td><td align=\"left\">0.04</td><td align=\"left\">1.13</td><td align=\"left\">0.021</td></tr><tr><td/><td align=\"left\">GI</td><td align=\"left\">0.59</td><td align=\"left\">0.36</td><td align=\"left\">0.08</td><td align=\"left\">1.36</td><td align=\"left\">0.003</td></tr><tr><td/><td align=\"left\">GII</td><td align=\"left\">0.18</td><td align=\"left\">0.14</td><td align=\"left\">0.03</td><td align=\"left\">0.49</td><td/></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Qualitative histopathology between CG and Group I</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"left\">Presence</td><td align=\"left\">CG</td><td align=\"left\">Group I</td><td align=\"left\"><italic>p </italic>value</td></tr></thead><tbody><tr><td align=\"left\">Intimal thickening (%)</td><td align=\"left\">No</td><td align=\"left\">30.76</td><td align=\"left\">15.38</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">69.23</td><td align=\"left\">84.61</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Isolated Xanthomamacrophages (%)</td><td align=\"left\">No</td><td align=\"left\">30.76</td><td align=\"left\">15.38</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">69.23</td><td align=\"left\">84.61</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Agregate Xanthomamacrophages (%)</td><td align=\"left\">No</td><td align=\"left\">38.46</td><td align=\"left\">15.38</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">61.53</td><td align=\"left\">84.61</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Lipid drops of proteoglican matrix (%)</td><td align=\"left\">No</td><td align=\"left\">38.46</td><td align=\"left\">15.38</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">61.53</td><td align=\"left\">84.61</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Lipid lakes of proteoglican matrix (%)</td><td align=\"left\">No</td><td align=\"left\">38.46</td><td align=\"left\">30.76</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">61.53</td><td align=\"left\">69.23</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Thin fibrous cap atheroma (%)</td><td align=\"left\">No</td><td align=\"left\">84.61</td><td align=\"left\">92.30</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">15.38</td><td align=\"left\">7.69</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Calcified nodule (%)</td><td align=\"left\">No</td><td align=\"left\">53.84</td><td align=\"left\">15.38</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">46.15</td><td align=\"left\">84.61</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Calcification (%)</td><td align=\"left\">No</td><td align=\"left\">92.30</td><td align=\"left\">76.92</td><td align=\"left\">&gt; 0.05</td></tr><tr><td/><td align=\"left\">Yes</td><td align=\"left\">7.69</td><td align=\"left\">23.07</td><td align=\"left\">&gt; 0.05</td></tr><tr><td align=\"left\">Collagen Type I (mean ± sd)</td><td/><td align=\"left\">880.9 ± 436.5</td><td align=\"left\">264.5 ± 104.04</td><td align=\"left\">0.29</td></tr><tr><td align=\"left\">Collagen Type III (mean ± sd)</td><td/><td align=\"left\">680.5 ± 267.54</td><td align=\"left\">312.24 ± 98.89</td><td align=\"left\">0.41</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>NS – non significant.</p><p>* Kruskal-Wallis</p></table-wrap-foot>", "<table-wrap-foot><p>IMR represents intima/media area ratio. Area was estimated in square micrometer.</p></table-wrap-foot>", "<table-wrap-foot><p>No represents less than 50% of cells in the balloon injury area. Yes represents more than 50% of cells in the balloon injury area.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1477-9560-6-12-1\"/>", "<graphic xlink:href=\"1477-9560-6-12-2\"/>" ]

[]

{ "acronym": [], "definition": [] }

[]

[]

[]

[ "<title>Discussion</title>", "<p>Spinal manifestations of vertebral artery dissection (VAD) are a rare event.[##REF##16960096##1##] Typical manifestations of VAD include brain ischemia combined with neck pain.[##REF##8073471##2##] In previous reports of patients with spinal manifestations of VAD the prominent sign was spinal ischemic cervical myelopathy.[##REF##8073471##2##, ####REF##9109913##3##, ##REF##15519870##4####15519870##4##] Recently a rare manifestation of VAD with left shoulder weakness and numbness of the left upper limb was reported that was successfully treated by proximal occlusion of the dissected vertebral artery using detachable balloon and Guglielmi detachable coils.[##REF##15227843##5##] Proximal vertebral artery occlusion using an intravascular technique was regarded as a non-invasive and effective option for patients with a cervical radiculopathy due to cervical vertebral artery dissection. Here we report local compression due to expansion of the dissecting aneurysm as a mechanism for sensorimotor radiculopathy with a favorable outcome after anticoagulation and physical therapy. Therefore VAD may be considered as a cause of otherwise unexplained cervical radiculopathic symptoms.</p>" ]

[]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<p>Spinal manifestations of vertebral artery dissection (VAD) are rare events and are typically symptomatic with neck pain and ischemic brain injury. We report a patient presenting with unusual peripheral paresis of the right upper limb due to an intramural hematoma of the right vertebral artery with local compression of C5 and C6 as the cause of cervical radiculopathy. These symptoms completely resolved after anticoagulation and physical therapy.</p>" ]

[ "<title>Case presentation</title>", "<p>A 51-year-old women presented with a two-weeks history of progressive pain and weakness of her right arm. Ten days earlier an orthopaedic specialist had performed a chiropractic maneuver to resolve the pain radiating from the neck to the right upper arm without alleviation of the symptoms. Four days later, the patient had noticed a weakness of elevation and flexion of her right upper arm. A cervical spine MRI scan displayed mild protrusion of the disc between the fifth and sixth cervical vertebrae resulting in a mild compression of the right C6 root and a surgical intervention due to progressive and substantial weakness was considered.</p>", "<p>Examination at admission revealed a distribution of the pain along the dermatomes C5 and C6, a paralysis of the deltoid (0/5) and the biceps brachii muscle (2/5), an atrophy of both muscles, decreased biceps and deltoid reflexes as well as sensory deficits in the right C5 and C6 dermatomes. However, the available MRI scan did not explain the more pronounced affection of the C5 root. MRI angiography then revealed a right vertebral artery dissection with a dissecting aneurysm with maximum expansion at the C5 level. The anterolateral part of the C5 and C6 radixes were compressed, causing cervical radiculopathy at these levels (Fig. ##FIG##0##1##). The MRI scan of the brain displayed no signs of ischemic insult or other pathologies. Due to hypoplasia of the contralateral vertebral artery, a surgical intervention was not taken into account. Thus a conservative approach with oral warfarin administration and a symptomatic treatment for cervical radiculopathy was started.</p>", "<p>After eight weeks of anticoagulation and physical therapy, the weakness of the upper arm markedly improved with an improvement in muscle strength of the deltoid (4/5) and the biceps brachii (5/5) without any sensory deficits. The follow-up cervical MRI scans showed a regularly revascularized right vertebral artery. Four months later, muscle strength of the deltoid completely recovered.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>GT, MW and RK analyzed and interpreted the patient data regarding the neurological symptoms and the management of the patient. GT performed the neurological examination and the follow up examinations. GT and RK wrote the manuscript. WS and UE analyzed and interpreted the MRI. All authors read and approved the final manuscript.</p>", "<title>Consent</title>", "<p>Written informed consent was obtained from the patient for publication of this case report and accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.</p>" ]

[]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>MRI of the cervical spine, axial views (A,B): The T1-weighted fat-saturated axial image (A) shows an eccentric hyperintensity surrounding the right VA at C5 level (arrow) indicating an intramural hematoma.</bold> The T2-weighted image <bold>(B)</bold> indicates compression of spinal root C5 (arrow) by the dissecting aneurysm.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1757-1626-1-139-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Introduction</title>", "<p>Granular cell tumour (GCT) was first described in 1926 by Abrikosoff as a rare myogenic lesion affecting the tongue. Further immunohistochemical tests have subsequently shown this lesion to have probably a perineural or Schwann cell origin. Abrikosoff in 1931 described a similar lesion in the female breast [##UREF##0##1##]. GCT of the breast is relatively uncommon and very easily misdiagnosed for primary breast cancer. We present a case of GCT of the pectoral muscle mimicking breast cancer in a female patient, highlighting the diagnostic challenge and the treatment options in managing patients with GCT.</p>" ]

[]

[]

[ "<title>Discussion</title>", "<p>GCT is a rare usually benign tumour which is most frequently encountered in the tongue, but can occur in a variety of visceral and cutaneous sites. Although there are a limited number of reported cases in the English literature the frequency of GCT is estimated to be 1 per 1000 cases of breast cancer [##REF##2993035##2##]. It is more common in middle age, pre-menopausal, black women. Breast GCT in male patients is extremely rare. The preponderance of these tumours in women, especially premenopausal, has led to the hypothesis that hormones are implicated in the pathogenesis of these tumours in the breast, however, up to date, no oestrogen or progesterone receptors have been found on the tumour cells in any of the cases so far reported in the literature. GCT of the breast arises from interlobular breast stroma and affects predominantly the upper inner quadrants of the breast, in the territory of distribution of the cutaneous sensory branches of the supraclavicular nerve. GCT of the breast are usually benign but malignant cases have been described including 2 cases of GCT and infiltrating breast cancer within the same breast [##REF##9386062##3##,##REF##12033967##4##].</p>", "<p>The histogenesis of GCT remains uncertain, the hypothesis of a neural or neuroectodermal origin is supported by the presence of the S-100 protein, typically expressed by these neoplastic cells, and by the similar ultrastructural features of the tumour cells and the Schwann cells. GCT are macroscopically, solid, firm tumours with a yellowish-white cross sectional surface. Microscopically they are usually characterised by clusters or sheets of polygonal cells with distinct borders and abundant granular eosinophillic cytoplasm. The nuclei are small, central and typically hyperchromatic [Figure ##FIG##2##3##]. Histological features suggestive of malignancy include tumour size &gt; 5 cm, presence of necrosis, cellular and nuclear pleomorphism, increased mitotic activity, high nuclear to cytoplasmic ratio and large nucleoli [##UREF##1##5##].</p>", "<p>GCT can closely resemble primary breast malignancies both clinically, due to their fibrous consistency, and radiologically. They present as hard lump which grow in an infiltrative manner and can involve the skin, causing skin dimpling and tethering, and the underlying muscle causing fixity of the lump to the deeper structures. Misdiagnosis could potentially lead to a far more aggressive treatment then necessary, hence it is important to differentiate GCT from other primary breast tumours. The final diagnosis is generally always achieved through histological examination.</p>", "<p>Mammographic and sonographic appearance of GCT could be misleading. On mammograms GCT could present as round, well circumscribed masses, but also as indistinct densities or speculated masses which resemble primary breast carcinoma. Microcalcifications can sometimes be seen but are usually absent. The ultrasound appearances are also variable and include solid masses with indistinct margins or more benign-appearing, well circumscribed masses. Yang et al [##REF##16615051##6##] has described the sonographic features of a series of 7 GCT of the breast and has interestingly shown that five of these lesions had an echogenic halo or were partially hyperechoic, this may be a result of the infiltrative growth pattern of GCT. In our case, sonographically the lesion was an ill defined hypoechoic mass with mild acoustic shadowing.</p>", "<p>MRI scanning has proved to have additional diagnostic value in the detection of such lesions in the breast. It helps delineating the extent of the disease, the presence of aggressive features and is also valuable for concomitant screening of the controlateral breast. Kohashi et al [##REF##10213961##7##] described a homogenously enhancing mass on T1 weighted imaging, the same mass showed high signal intensity rim on T2 weighted sequence. High T2 signal has been shown to be a sign of benign disease [##REF##10077012##8##]. Features suggestive of malignancy include rim enhancement, speculated margins and irregular ill-defined shape of the breast lesion. Interestingly microcalcifications have been absent in every case of GCT reviewed so far and their presence should point towards a malignancy other then GCT [##REF##17940129##9##].</p>", "<p>For GCT which have been proven to be benign at core biopsy close observation is an acceptable treatment option although wide local excision is regarded as gold standard in the treatment of benign GCT. Local recurrence is associated with incomplete excision hence a complete clearance of the tumours with histologically clear margins is paramount. Axillary sampling or sentinel lymph node biopsy is not indicated [##REF##2993035##2##] in the management of benign GCT as nodal invasion is extremely rare. Malignant GCT should be treated like other malignant breast tumours, however treatment has a poor overall outcome [##REF##1733419##10##].</p>", "<p>Differentiation between benign and malignant breast GCT is almost impossible on the clinical basis, and remains challenging even after accurate histological study of the specimen. High index of suspicion is therefore paramount, especially in presence of a breast mass with associated axillary lymphadenopathy, or if the breast lesion is larger then 4 cm in diameter on radiological imaging. On MRI evidence of infiltration of adjacent tissues and rim enhancement are also regarded as suspicious features.</p>" ]

[ "<title>Conclusion</title>", "<p>In conclusion, as GCT can easily be misdiagnosed as primary breast carcinoma clinically and radiologically, histological analysis of these lesions is needed to achieve a diagnosis. This case highlights the benefits of triple assessment in the management of beast lumps Surgeons should always be aware that GCT of the breast can resemble primary breast tumour in order to avoid performing unnecessary radical surgery in this group of patients.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<p>We describe the case of a 55 year old female who presented with a mass in her right breast. Mammography confirmed a 2 × 2 cm lump, suspicious of malignancy. The lesion was widely resected. Histological examination revealed this to be a benign granular cell tumour.</p>", "<p>Granular cell tumour is a rare tumour that very occasionally presents within the breast. It is possibly of Schwann cell origin. Clinical features and subsequent investigations may be suggestive of breast malignancy. Tumour cells showing positive immunostaining for S-100 and PAS is in keeping with the diagnosis. Wide local resection is the gold standard treatment.</p>" ]

[ "<title>Case presentation</title>", "<p>A 55 year old female presented with a 2 month history of a painless lump in her upper outer quadrant of her right breast. She was on hormone replacement therapy for the previous 4 years. There was no family history of breast cancer and past medical history was unremarkable. Examination revealed a hard non mobile lump in the axillary tail of her right breast, which measured 2 × 2 cm and was highly suspicious of primary breast malignancy. Mammography and ultrasound imaging showed a suspicious lesion in the upper outer quadrant of the right breast which appeared to be attached to the underlying pectoralis major muscle [Figure ##FIG##0##1## and ##FIG##1##2##]. As part of routine triple assessment core biopsies of the lump were taken under ultrasound guidance. Histological examination of the biopsies suggested a granular cell tumour (benign tumour of neural type differentiation/origin). Wide local excision of the tumour was performed. Macroscopically the tumour appeared to be originating from the lateral border of the pectoralis major muscle and not from the breast and was excised with an ellipse of normal muscle tissue.</p>", "<p>Microscopically the specimen consisted of a central area containing cells with finely granular eosinophillic cytoplasm containing central bland nuclei. Immuno-histochemistry staining showed tumour cells were positive for S 100 and weakly cytopositive for PAS, in keeping with a diagnosis of granular cell tumour. The tumour was completely excised. The patient remains well 6 months after surgery.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>WAS and SMY performed the surgery. AP and VL were major contributors in the writing of the whole manuscript. All authors read and approved the final manuscript.</p>", "<title>Consent</title>", "<p>Written informed consent was obtained from the patient for publication of this case report and accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.</p>" ]

[]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Mammographic appearance of granular cell tumour</bold>. Mammogram shows a dense mass deep in the upper outer quadrant of the right breast adjacent to the pectoralis major muscle.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Sonographic appearance of granular cell tumour</bold>. Ultrasound shows an ill defined hypoechoic mass with echogenic haloing. It is difficult to differentiate this mass from a primary breast lesion, based on the sonographic findings.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Photomicrograph of core biopsy specimen</bold>. Typical microscopical features of a granular cell tumour, showing nests of cells with abundant pink granular cytoplasm and small, hyperchromatic nuclei which are centrally located.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1757-1626-1-142-1\"/>", "<graphic xlink:href=\"1757-1626-1-142-2\"/>", "<graphic xlink:href=\"1757-1626-1-142-3\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Most patients with colorectal cancer survive at least five years after diagnosis [##UREF##0##1##], making health-related quality of life (HRQL) an important outcome for these patients. Patient-reported outcomes including HRQL have been used in conjunction with traditional clinical outcomes, such as treatment response rates and disease-free survival, to assess treatment efficacy in randomized clinical trials [##REF##10189130##2##, ####REF##10949778##3##, ##REF##12775739##4####12775739##4##]. HRQL is also used in quality of care research [##REF##9248318##5##] and is a predictor of survival of patients with colorectal cancer [##REF##12091066##6##, ####REF##11736975##7##, ##REF##18421055##8####18421055##8##]. Understanding the characteristics or conditions that predict subsequent HRQL may help clinicians identify patients who are at risk for poor HRQL. Furthermore, if a characteristic or condition is modifiable, an intervention to alter it could lead to improved HRQL.</p>", "<p>Most studies of HRQL of patients with colorectal cancer have been cross-sectional [##REF##7720441##9##, ####REF##9717993##10##, ##REF##14705266##11####14705266##11##]. While cross-sectional analyses are valuable, longitudinal analyses of predictors of HRQL are needed to understand how HRQL changes over time [##REF##14705266##11##]. Previous studies of HRQL that combined data from patients with colorectal cancer and patients with other diagnoses such as breast or lung cancer [##REF##8044158##12##, ####REF##10197951##13##, ##REF##16674321##14####16674321##14##] may have failed to identify relationships involving disease- and treatment-related side-effects specific to patients with colorectal cancer. Furthermore, previous research tends to focus on the HRQL of cancer patients while in treatment [##REF##12505052##15##] or the HRQL of cancer survivors several years post-diagnosis [##REF##12897331##16##, ####REF##12017152##17##, ##REF##11206224##18####11206224##18##]. Less is known about HRQL during the period when patients have completed treatment and are transitioning to the survivorship phase. A final limitation of previous research on predictors of HRQL is that it has not considered whether the variables identified based on statistical criteria are also clinically meaningful. Therefore, our study had the following two objectives: (1) to identify variables that predict HRQL in a prospective, population-based study of patients with colorectal cancer, and (2) to interpret the regression results in the context of clinical importance.</p>" ]

[ "<title>Methods</title>", "<title>Study population</title>", "<p>Men and women aged 40 to 84 diagnosed with invasive (i.e., excluding <italic>in situ</italic>) colorectal cancer from April 1999 through June 2000 were identified from three of the ten regional registries that comprise the statewide population-based California Cancer Registry (CCR): Region 1 (San Jose/Monterey area), Region 3 (Sacramento area), and Region 8 (San Francisco/Oakland area) [##UREF##1##19##].</p>", "<title>Timing of assessments</title>", "<p>This study employed rapid case ascertainment to identify eligible patients within 3 to 6 months of diagnosis. Respondents were first assessed in an initial survey an average of 9.2 months (range 4.6 – 23.0 months; SD 2.6 months) post-diagnosis when most patients would have recovered from their cancer surgery and completed their adjuvant treatment, if any. The primary purpose of the initial survey was to assess the quality of care for colorectal cancer by hospital and patient characteristics. The initial survey, as previously described [##REF##16116149##20##], was completed by 1,079 English- or Spanish-speaking respondents (response rate 72.4%).</p>", "<p>Patients were eligible for participation in the follow-up survey if they completed the initial survey and were English-speaking. A total of 830 English-speaking respondents in the initial survey were invited to participate in the follow-up survey an average of 19.1 months (range 13.0 – 31.8 months; SD 2.7 months) after diagnosis when most patients would be in a relatively stable disease state. The average time between the initial and follow-up surveys was 10.2 months (range 4.7 – 17.4 months; SD 1.8 months). Data for both the initial and follow-up surveys were collected predominantly via telephone by trained interviewers from the California Public Health Institute's Survey Research Group. Some patients were surveyed through a mailed, self-administered questionnaire, particularly patients who were hearing-impaired or not successfully contacted by telephone. Institutional review boards of the California Department of Health Services, Public Health Institute, Harvard Medical School, and the Northern California Cancer Center approved the study protocols for the initial and follow-up surveys. Participant consent was obtained prior to conducting the surveys.</p>", "<title>Surveys</title>", "<p>HRQL was measured in the initial and follow-up surveys with the Functional Assessment of Cancer Therapy-Colorectal (FACT-C) [##REF##10472150##21##]. The FACT-C is a valid and reliable measure of HRQL and includes five subscales: Physical Well-Being (PWB, 7-items), Functional Well-Being (FWB, 7-items), Social/family Well-Being (SWB, 7-items), Emotional Well-Being (EWB, 6-items), and the Colorectal Cancer Subscale (CCS, 7-items), which measures concerns specific to colorectal cancer patients such as appetite and bowel control [##REF##10472150##21##]. The FACT-C was scored as described in the documentation for the instrument, with higher scores indicating better HRQL [##UREF##2##22##]. The Trial Outcome Index (TOI), derived as the sum of the PWB, FWB, and CCS subscales, is a summary measure of physical function and well-being in colorectal cancer patients that is useful as a patient-reported outcome in pharmacologic interventions [##REF##10472150##21##]. The TOI and the remaining two subscales, SWB and EWB, were the HRQL outcomes evaluated in this study.</p>", "<p>Patients' perceived quality of cancer care was measured using 31 items obtained from the Picker Institute of Boston, Massachusetts [##REF##16116149##20##,##REF##9370508##23##,##REF##2966123##24##] that measure problems associated with six aspects of cancer care: psychosocial care, access to cancer care, treatment information, health information, confidence in providers, and coordination of care. Two items measuring patients' perceptions of how well healthcare providers controlled their pain/discomfort and nausea/vomiting were also included. Higher scores indicate greater perceived problems with care.</p>", "<p>Limitations due to each of 14 comorbid conditions were measured in the initial survey and scored as 0 if the condition was not present, 1 if the condition was present but did not limit the patient, or 2 if the condition was present and limited the patient. Scores were summed to create a comorbidity index (range 0 to 28) [##REF##8044158##12##,##REF##4436428##25##]. General health was measured with a single item from the Medical Outcomes Study 36-item Short form (SF-36) health survey [##UREF##3##26##]. Bowel function and overall bowel problems were measured with items from the Prostate Cancer Outcomes Study [##REF##10528021##27##]. Respondents who had a colostomy at the time of the interviews were not asked these bowel-related items, including 131 (12.1%) of 1,079 patients in the initial survey and 54 (9.5%) of 568 patients in the follow-up survey.</p>", "<p>Sociodemographic information collected via survey included race/ethnicity, education, household income, financial difficulty due to cancer, occupational status, marital status, and number of persons living in the household. Gender, age at diagnosis and stage at diagnosis were obtained from the CCR. Neighborhood socioeconomic status (SES) was calculated based on 2000 U.S. Census data using methods described in Yost et al. [##REF##11562110##28##].</p>", "<title>Data analyses</title>", "<title>Non-response bias</title>", "<p>To assess potential non-response bias, characteristics of eligible patients who did not participate in the follow-up survey were compared to those of patients who did using standard statistical tests for continuous and categorical data. All candidate predictor variables for the longitudinal analyses were measured either at the time of diagnosis or during the initial survey.</p>", "<title>Regression models</title>", "<p>Because previous studies of predictors of HRQL are inconclusive [##REF##14705266##11##, ####REF##8044158##12##, ##REF##10197951##13####10197951##13##,##REF##12017152##17##,##REF##12619150##29##, ####REF##10717609##30##, ##REF##12088249##31####12088249##31##], we adopted an exploratory approach to identify variables predictive of HRQL at the follow-up survey. The range of time since diagnosis for the initial survey (4.6 – 23.0 months) overlapped with that for the follow-up survey (13.0 – 31.8 months). Therefore, we created non-overlapping time periods for each survey centered around the means (i.e., 9 months for initial survey and 19 months for the follow-up survey). The time period for the initial survey was restricted from 4 to less than 14 months post-diagnosis and the time period for the follow-up survey was restricted from 14 to less than 24 months post-diagnosis. The regression analyses were restricted to participants who completed their initial and follow-up surveys within these ranges. This resulted in the exclusion of 43 participants who were missing data for time since diagnosis for either the initial or follow-up survey and 29 participants who were outside of the restricted time frames. Following these exclusions, the mean times since diagnosis for the 496 respondents were 8.6 months (range 4.6–13.9, SD 1.7) for the initial survey and 18.8 months (range 14.0–24.0, SD 2.1) for the follow-up survey.</p>", "<p>Separate linear regression analyses [##UREF##4##32##] were conducted for each of the three HRQL outcomes measured at the follow-up survey using the following candidate predictor variables. <bold>Initial HRQL</bold>: TOI, SWB or EWB measured at the initial survey. <bold>Sociodemographic</bold>: age at diagnosis, gender, race/ethnicity, marital status (married/living as married vs. not married), education (high school or less vs. technical school or some college, college or higher), occupational status (working vs. not working), number in household, neighborhood SES (standardized principal component score [##REF##11562110##28##]), financial problems due to cancer, and household income (missing, &lt;$25,000, $25,000–$50,000 vs. $50,000+). Household income was not reported by 58 (10.2%) respondents. Rather than exclude these respondents from the analyses, \"missing\" income was treated as a separate income category. <bold>Cancer/health</bold>: stage at diagnosis (Stage I/II/III vs. Stage IV), general health, colostomy (yes/no), history of radiation therapy (yes/no), history of chemotherapy (yes/no), currently receiving chemotherapy at the time of the initial survey (yes/no), comorbidity index, bowel function, overall bowel problems, family history of colorectal cancer (yes/no), time since diagnosis, and site (colon vs. rectum). <bold>Healthcare</bold>: type of health insurance [Medicare, other/none (e.g., Medicaid, other government-provided, uninsured) vs. commercial (e.g., HMO, PPO, private)], six domains of perceived quality of care, control of pain and discomfort (definitely vs. somewhat/not at all) and control of nausea and vomiting (definitely vs. somewhat/not at all).</p>", "<p>Variables associated with HRQL were selected in two stages. First, follow-up HRQL was regressed on initial HRQL plus one other candidate variable. Candidate variables that had a significant relationship with follow-up HRQL by a liberal (<italic>p </italic>&lt; 0.25) criterion were identified [##REF##10886529##33##]. These variables were then combined into a multivariable model. Backward elimination with a criterion of <italic>p </italic>&lt; 0.05 for retention was used to select a final model. We also conducted forward and stepwise regression to determine whether the same model was identified. Multicollinearity in the final models was assessed with the variance inflation factor (VIF). To facilitate interpretation of our regression results, we report squared semi-partial correlations (<italic>sr</italic>²) in addition to <italic>p</italic>-values. Because the <italic>sr</italic>²expresses the unique variance in the dependent variable explained by a predictor variable, it is a useful measure of the importance of a predictor [##UREF##5##34##].</p>", "<title>Meaningful effects</title>", "<p>We assessed the clinical meaningfulness of each predictor variable [##REF##12064758##35##,##REF##12064759##36##]. For interval variables (e.g., age), we computed the difference in predicted HRQL scores corresponding to a large difference in a predictor variable, defined as a 1 SD difference [##UREF##6##37##], where the SD was based on the data for the 496 respondents evaluated in the regression analyses. For nominal variables (e.g., race/ethnicity), we computed the difference in the predicted HRQL score relative to the reference category. The effect of an explanatory variable on follow-up HRQL was considered meaningful if the corresponding predicted score difference was at least as large as the minimally important difference (MID), which we defined as the smallest difference in HRQL scores that patients perceive as important, and thus might lead a clinician to consider changing the patient's management [##REF##11936935##38##]. MIDs have been determined for the TOI (4–6 points) [##REF##16291468##39##] and the SWB and EWB subscales (2–3 points) [##REF##15804320##40##]. We used the lower bounds of these ranges to identify clinically meaningful effects as an indication of the potential prognostic impact of predictor variables. We also computed the percent of patients whose HRQL improved or declined more than the lower bound of the MID range.</p>" ]

[ "<title>Results</title>", "<title>Sample characteristics</title>", "<p>Of the 830 English-speaking patients invited to participate in the follow-up study, 26 were ineligible because they had died. Follow-up surveys were completed by 568 (70.6%) of the 804 eligible patients. Of the 236 patients who did not participate in the follow-up study, 28 were either hearing or mentally impaired or too ill to participate, 79 refused, and 129 were not successfully contacted.</p>", "<p>Table ##TAB##0##1## summarizes the characteristic at the time of diagnosis or at the time of the initial survey for the 496 participants of the follow-up survey with non-overlapping survey periods who were evaluated in this study. Characteristics for all 568 respondents and the 236 non-respondents of the follow-up survey are also described in Table ##TAB##0##1##. Non-respondents were significantly more likely to be non-white, unmarried, with low income, more financial problems due to cancer, longer time since diagnosis, have more perceived problems with access to care, confidence in providers, and coordination of care, and more perceived problems with control of pain/discomfort. They were significantly less likely to be on chemotherapy at the time of the initial survey. HRQL scores measured in the initial survey were lower for non-respondents than for respondents of the follow-up survey. Cronbach's alpha, a measure of reliability, was acceptable (≥0.70) in both the initial and follow-up surveys, respectively, for the TOI scale (0.90, 0.90) and the SWB (0.74, 0.77) and EWB (0.74, 0.76) subscales. The 95% confidence intervals around the initial scores for SWB (22.9, 23.7) and EWB (20.1, 20.7) for the 568 respondents did not contain the general U.S. population norms of 19.1 and 19.9 for the SWB and EWB, respectively [##REF##15851773##41##], nor did they contain the norms for colon cancer patients of 22.0 and 19.8, respectively [##UREF##2##22##]. This indicates that the respondents had significantly higher scores for these domains than the norms. There are no general or cancer-specific norms for the TOI.</p>", "<title>Clinically meaningful change in HRQL</title>", "<p>The percent of patients with a clinically meaningful decline in EWB scores from the initial to follow-up survey was exactly the same as the percent with a clinically meaningful improvement (Table ##TAB##1##2##). Slightly more patients had a meaningful improvement than decline in TOI scores, but fewer patients had a meaningful improvement than decline in SWB scores. EWB scores were the most stable, with 47.6% of patients experiencing a change less than the MID. The magnitude of the average decline in scores was slightly larger than that of the average improvement for all three HRQL outcomes.</p>", "<title>Prognostic impact on patient-reported HRQL</title>", "<title>Predictors of follow-up TOI scores</title>", "<p>In addition to initial TOI scores, two variables were retained as predictors of follow-up TOI following backward elimination (Table ##TAB##2##3##), accounting for 43.9% of the variance. Forward and stepwise selection identified the same model. No sociodemographic variables were retained. One cancer/health-related variable, general health, was retained. The Treatment Information problem score from the Picker Institute measure was the only healthcare variable retained. Only initial TOI was a clinically meaningful predictor of follow-up TOI, with a 1 SD increase (12.7 points) in initial TOI scores corresponding to a 6.4 point increase in follow-up TOI scores. Initial TOI accounted for the largest proportion of variance, as indicated by the <italic>sr</italic>². The <italic>p</italic>-value, <italic>sr</italic>²and effect on follow-up TOI indicated that general health had a greater prognostic impact than Treatment Information. All VIFs were less than 2.0 indicating there was no multicolinearity among the predictors.</p>", "<title>Predictors of follow-up SWB scores</title>", "<p>Backward selection identified a model with six statistically significant predictors of follow-up SWB explaining a total of 39.3% of the variance (Table ##TAB##3##4##), including initial SWB, three sociodemographic indicators (gender, race/ethnicity, marital status), one cancer/health-related indicator (general health), and one healthcare measure (problems with control of pain/discomfort). Forward and stepwise selection yielded a slightly different model that did not include gender and marital status. We report the model identified using backward selection as it had a larger adjusted R²(39.3 vs. 38.1). Both initial SWB and Hispanic ethnicity were clinically meaningful predictors of follow-up SWB. Hispanic ethnicity also had a larger <italic>sr</italic>²relative to the other sociodemographic, cancer/health and healthcare variables.</p>", "<title>Predictors of follow-up EWB scores</title>", "<p>Predictors of EWB at the follow-up survey included initial EWB, general health and control of nausea/vomiting, explaining 36.5% of the variance (Table ##TAB##4##5##). Backward, forward and stepwise selection all identified the same model. After initial EWB, general health had the largest <italic>sr</italic>², while problems with control of nausea/vomiting had the largest prognostic impact as indicated by the size of the effect on follow-up EWB. Initial EWB was the only meaningful predictor of follow-up EWB.</p>" ]

[ "<title>Discussion</title>", "<p>Exploratory longitudinal analyses were conducted to evaluate the relationship between three HRQL outcomes and sociodemographic, cancer/health, and healthcare variables in a population-based sample of patients with colorectal cancer. General health was the only variable common to all three outcomes, although each model also contained a quality of care variable: Perceived problems with Treatment Information was a predictor of follow-up TOI, perceived problems with control of pain/discomfort predicted follow-up SWB and perceived problems with control of nausea/vomiting was a predictor of follow-up EWB.</p>", "<p>Rather than relying solely on statistical measures such as <italic>p</italic>-values and <italic>sr</italic>²to interpret the results of the regression analyses, we also used clinical meaningfulness of the effect of a predictor variable on the HRQL outcome. For follow-up TOI, both initial TOI and general health were highly statistically significant predictors (<italic>p </italic>&lt; 0.001). The <italic>sr</italic>²for general health was smaller than that for initial TOI, but even with this information, it may still not be intuitive to some clinicians or researchers whether to consider general health as an important predictor. By considering the clinical importance of these variables, we showed that even a large (1 SD) difference in the general health score would not have a clinically meaningful effect on follow-up TOI scores. This information may help clinicians and researchers understand the results regardless of their familiarity with regression modeling or the FACT-C instrument.</p>", "<p>Perceived problems with Treatment Information was also retained in the model for follow-up TOI. A measure of perceived quality of treatment information is not commonly included in studies aimed at identifying predictors of HRQL. That this variable was identified in our study as a significant predictor of HRQL over other commonly evaluated variables such as gender, age, and comorbidities [##REF##8044158##12##,##REF##12017152##17##,##REF##10717609##30##] warrants the addition of perceived quality of treatment information to the list of candidate variables in future research of HRQL predictors.</p>", "<p>Relative to non-Hispanic White participants, Hispanic participants had significantly lower follow-up SWB. Wan et al. [##REF##16674321##14##] also observed significantly lower SWB scores among Hispanic cancer patients relative to White patients. Both initial SWB and Hispanic ethnicity were highly significantly related to follow-up SWB (<italic>p </italic>&lt; 0.001). The <italic>sr</italic>²for Hispanic ethnicity was much smaller than that for initial SWB suggesting that Hispanic ethnicity was the less influential predictor of the two. However, by considering clinical importance, we obtained a slightly different interpretation; that the two variables had comparable clinically meaningful effects on follow-up SWB.</p>", "<p>The relationship between gender and the social domain of HRQL may depend on how the domain is measured. For example, as measured in the present study with the FACT-C, men had worse initial SWB (data not shown) and follow-up SWB than women. Normative data for the SWB in both the general U.S. population and cancer patients also show lower average scores for men than women [##REF##15851773##41##]. However, several studies have evaluated the relationship between gender and social functioning as measured by the EORTC QLQ-C30 in cancer patients and reported no differences [##REF##12088249##31##,##REF##14722043##42##,##REF##16311851##43##].</p>", "<p>Although the association between being married/living as married and having better physical and psychological well-being is well established, the link between marital status and SWB is less clear [##UREF##7##44##]. We found that married/living as married persons had better SWB as measured by the FACT-C, which is predominantly a measure of social support and contains items such as \"I get emotional support from my family\" and \"I feel close to my partner (or the person who is my main support).\" It is reasonable to expect married/living as married persons to answer these questions more favorably.</p>", "<p>Initial EWB was the only meaningful predictor of follow-up EWB. After initial EWB, general health was the strongest predictor based on the <italic>p</italic>-value and <italic>sr</italic>², but problems with control of nausea and vomiting was the strongest predictor based on the effect on follow-up EWB score. Thus, by considering clinical importance, we obtained a slightly different interpretation of the importance of these two predictors.</p>", "<p>The measures of perceived quality of care (problems with Treatment Information, control of pain/discomfort, control of nausea/vomiting) are the only predictors of HRQL in our study that are potentially modifiable. These variables are not typically included as predictors of HRQL in multivariable regression analyses, yet they were more clinically meaningful predictors of the HRQL outcomes than some more commonly evaluated predictors, including gender, marital status and general health. Additional research is needed to better understand the association between factors related to perceived quality of care at the time of cancer treatment and HRQL at some follow-up assessment. Furthermore, as these are potentially modifiable variables, intervention studies could explore methods for improving certain aspects of quality of care to determine whether those changes lead to improved HRQL.</p>", "<p>While only a few of the statistically significant predictors individually met our criteria for clinical importance, in combination they may identify patients at high risk for poor HRQL. For example, unmarried male patients with worse general health and more perceived problems with control of pain/discomfort may have meaningfully lower average follow-up SWB scores (i.e., at least 2 points lower) than married female patients with better general health and fewer perceived problems with control of pain/discomfort even after adjusting for initial SWB. Initial HRQL was consistently the strongest predictor of follow-up HRQL; therefore, clinicians could identify patients at risk for poor future HRQL by routinely assessing HRQL in clinical practice [##REF##12479768##45##].</p>", "<p>A strength of our study is that the respondents were identified through a population-based cancer registry and were therefore representative of English-speaking colorectal cancer patients in California. A potential limitation of our study is possible non-response bias in the follow-up sample. The follow-up respondents and non-respondents differed significantly on a number of variables that may be related to follow-up HRQL, which limits the generalizability of our results. In particular, non-respondents had significantly lower initial HRQL than respondents and therefore likely had lower follow-up HRQL. Nonresponse might also have affected the selection and estimated prognostic impact of predictors of HRQL. Another potential limitation is that there may be other variables predictive of follow-up HRQL that were not considered in this analysis. Variables such as spiritual well-being [##REF##16674321##14##,##REF##8630968##46##], optimism [##REF##11284214##47##], and sexual dysfunction [##REF##9717993##10##] have been shown to be related to HRQL. However, these topics were not measured in the initial survey. The follow-up survey did not measure whether the respondents had experienced a recurrence or whether they were undergoing any adjuvant treatment for either their primary or a recurrent cancer, which could have affected their follow-up HRQL [##REF##12619150##29##,##REF##11219418##48##,##REF##1511396##49##]. Because of inconclusive results in the literature regarding variables that predict HRQL in colorectal patients, we adopted an exploratory approach and evaluated a large number of variables, including variables rarely, if ever, evaluated previously in this population as potential predictors, such as perceived quality of cancer care. Thus, another potential limitation of our study is an inflated type I error. The predictors of HRQL in patients with colorectal cancer may differ from those in patients with other types of cancer. Furthermore, the predictors of HRQL may vary based on how HRQL is measured.</p>" ]

[ "<title>Conclusion</title>", "<p>We identified sociodemographic, clinical, and healthcare variables that predict HRQL as measured by the FACT-C. The most consistent finding was that patients with poor general health and problems with certain domains of perceived quality of cancer care may be at risk for poor HRQL. Other characteristics that might identify at-risk patients were specific to each HRQL outcome and included being male, unmarried or Hispanic. Computing the clinical importance of the effect on the HRQL outcome helped to interpret the impact of specific statistically significant predictors. As the only potentially modifiable variables identified in our study were related to perceived quality of cancer care, future research should evaluate whether interventions aimed at improving these variables enhances subsequent HRQL.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Most studies that have identified variables associated with the health-related quality of life (HRQL) of patients with colorectal cancer have been cross-sectional or included patients with other diagnoses. The objectives of this study were to identify predictors of HRQL in patients with colorectal cancer and interpret the clinical importance of the results.</p>", "<title>Methods</title>", "<p>We conducted a population-based longitudinal study of patients identified through three regions of the California Cancer Registry. Surveys were completed by 568 patients approximately 9 and 19 months post-diagnosis. Three HRQL outcomes from the Functional Assessment of Cancer Therapy – Colorectal (FACT-C) were evaluated: social/family well-being (SWB), emotional well-being (EWB) and the Trial Outcome Index (TOI), which is a colorectal cancer-specific summary measure of physical function and well-being. Sociodemographic, cancer/health, and healthcare variables were assessed in multivariable regression models. We computed the difference in predicted HRQL scores corresponding to a large difference in a predictor variable, defined as a 1 standard deviation difference for interval variables or the difference relative to the reference category for nominal variables. The effect of an explanatory variable on HRQL was considered clinically meaningful if the predicted score difference was at least as large as the minimally important difference.</p>", "<title>Results</title>", "<p>Common predictors of better TOI, SWB and EWB were better general health and factors related to better perceived quality of cancer care. Predictor variables in addition to general health and perceived quality of care were identified only for SWB. Being married/living as married was associated with better SWB, whereas being male or of Hispanic ethnicity was associated with worse SWB. Among the sociodemographic, cancer/health, and healthcare variables evaluated, only Hispanic ethnicity had a clinically meaningful effect on an HRQL outcome.</p>", "<title>Conclusion</title>", "<p>Our findings, particularly the information on the clinical importance of predictor variables, can help clinicians identify patients who may be at risk for poor future HRQL. Potentially modifiable factors were related to perceived quality of cancer care; thus, future research should evaluate whether improving these factors improves HRQL.</p>" ]

[ "<title>List of abbreviations</title>", "<p>HRQL: Health-related quality of life; FACT-C: Functional Assessment of Cancer Therapy-Colorectal; PWB: Physical well-being; EWB: Emotional well-being; SWB: Social/family well-being, FWB: Functional well-being; TOI: Trial outcome index; CCR: California Cancer Registry; SD: Standard deviation; HMO: Health maintenance organization; PPO: Preferred provider organization; VIF: Variance inflation factor.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>KJY, AMZ, JZA, and DWW made substantial contributions to the study concept and design, EAH and AMZ provided guidance on the statistical analysis, KJY conducted the statistical analysis, interpreted the results and drafted the manuscript, EAH, AMZ, JZA, and DWW participated in critical revisions of the manuscript for important intellectual content. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This work was supported by grants from the Surveillance, Epidemiology and End Results Special Study program (NO1-PC-65107), and the Agency for Healthcare Research and Quality and the National Cancer Institute (R01 HS09869, U01 CA93324). The authors would like to thank William Wright for his guidance on study design and the following individuals for assistance with data collection and database management: Mark Allen, Gretchen Agha, Craig Grilley, Scott Riddle, Bonnie Davis, Ann Hitchcock and staff of the California Public Health Institute Survey Research Group. We also thank David Eton for his thoughtful review and comments.</p>" ]

[]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Characteristics of follow-up survey respondents and non-respondents at the time of diagnosis or the initial survey</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Variable Type</bold></td><td align=\"left\"><bold>Variable^a</bold></td><td align=\"center\"><bold>Evaluated</bold><break/><bold>Respondents</bold><break/><bold> (<italic>n </italic>= 496)^b,c</bold></td><td align=\"center\" colspan=\"3\"><bold>Assessment of Response Bias^b</bold></td></tr><tr><td/><td/><td/><td colspan=\"3\"><hr/></td></tr><tr><td/><td/><td/><td align=\"center\"><bold>Follow-up</bold><break/><bold>Survey</bold><break/><bold>Respondents</bold><break/><bold> (<italic>n </italic>= 568)</bold></td><td align=\"center\"><bold>Follow-up</bold><break/><bold>Survey</bold><break/><bold>Non-Respondents</bold><break/><bold> (<italic>n </italic>= 236)</bold></td><td align=\"center\"><bold><italic>p</italic>-value^d</bold></td></tr></thead><tbody><tr><td align=\"left\">Sociodemographic</td><td align=\"left\">Age at diagnosis (years) [mean (SD)]</td><td align=\"center\">66.8 (10.4)</td><td align=\"center\">66.7 (10.5)</td><td align=\"center\">65.0 (12.1)</td><td align=\"center\">0.07</td></tr><tr><td/><td align=\"left\">Male</td><td align=\"center\">48.0</td><td align=\"center\">48.6</td><td align=\"center\">50.0</td><td align=\"center\">0.72</td></tr><tr><td/><td align=\"left\">Race/ethnicity</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> Non-Hispanic White</td><td align=\"center\">80.0</td><td align=\"center\">79.8</td><td align=\"center\">67.8</td><td align=\"center\">&lt;0.001</td></tr><tr><td/><td align=\"left\"> Non-Hispanic Black</td><td align=\"center\">5.7</td><td align=\"center\">5.6</td><td align=\"center\">14.0</td><td/></tr><tr><td/><td align=\"left\"> Hispanic</td><td align=\"center\">7.5</td><td align=\"center\">7.0</td><td align=\"center\">9.3</td><td/></tr><tr><td/><td align=\"left\"> Asian/Other</td><td align=\"center\">6.9</td><td align=\"center\">7.6</td><td align=\"center\">8.9</td><td/></tr><tr><td/><td align=\"left\">Married/Living as married</td><td align=\"center\">65.9</td><td align=\"center\">65.7</td><td align=\"center\">56.6</td><td align=\"center\">0.02</td></tr><tr><td/><td align=\"left\">Education</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> High school or less</td><td align=\"center\">38.1</td><td align=\"center\">38.2</td><td align=\"center\">44.0</td><td align=\"center\">0.30</td></tr><tr><td/><td align=\"left\"> Post high school training/some college</td><td align=\"center\">30.7</td><td align=\"center\">31.1</td><td align=\"center\">27.8</td><td/></tr><tr><td/><td align=\"left\"> College degree or higher</td><td align=\"center\">31.1</td><td align=\"center\">30.7</td><td align=\"center\">28.2</td><td/></tr><tr><td/><td align=\"left\">Working</td><td align=\"center\">28.6</td><td align=\"center\">28.8</td><td align=\"center\">29.7</td><td align=\"center\">0.80</td></tr><tr><td/><td align=\"left\">Household Income</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> Missing</td><td align=\"center\">8.9</td><td align=\"center\">10.2</td><td align=\"center\">11.9</td><td align=\"center\">&lt;0.001</td></tr><tr><td/><td align=\"left\"> Less than $25,000</td><td align=\"center\">22.4</td><td align=\"center\">23.1</td><td align=\"center\">36.0</td><td/></tr><tr><td/><td align=\"left\"> $25,000 to $50,000</td><td align=\"center\">31.1</td><td align=\"center\">30.3</td><td align=\"center\">20.8</td><td/></tr><tr><td/><td align=\"left\"> Over $50,000</td><td align=\"center\">37.7</td><td align=\"center\">36.4</td><td align=\"center\">31.4</td><td/></tr><tr><td/><td align=\"left\">Financial problems due to cancer</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> Not at all</td><td align=\"center\">76.6</td><td align=\"center\">77.2</td><td align=\"center\">66.4</td><td align=\"center\">0.01</td></tr><tr><td/><td align=\"left\"> A little</td><td align=\"center\">10.9</td><td align=\"center\">10.2</td><td align=\"center\">16.4</td><td/></tr><tr><td/><td align=\"left\"> Somewhat</td><td align=\"center\">7.3</td><td align=\"center\">7.1</td><td align=\"center\">11.2</td><td/></tr><tr><td/><td align=\"left\"> A lot</td><td align=\"center\">5.0</td><td align=\"center\">5.5</td><td align=\"center\">6.0</td><td/></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\">Cancer/Health-related</td><td align=\"left\">Late Stage (Stage IV)</td><td align=\"center\">8.3</td><td align=\"center\">7.8</td><td align=\"center\">10.2</td><td align=\"center\">0.26</td></tr><tr><td/><td align=\"left\">General Health</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> Poor</td><td align=\"center\">5.0</td><td align=\"center\">5.3</td><td align=\"center\">5.1</td><td align=\"center\">0.21</td></tr><tr><td/><td align=\"left\"> Fair</td><td align=\"center\">23.2</td><td align=\"center\">22.4</td><td align=\"center\">23.3</td><td/></tr><tr><td/><td align=\"left\"> Good</td><td align=\"center\">38.7</td><td align=\"center\">38.2</td><td align=\"center\">41.1</td><td/></tr><tr><td/><td align=\"left\"> Very Good</td><td align=\"center\">21.2</td><td align=\"center\">22.7</td><td align=\"center\">24.6</td><td/></tr><tr><td/><td align=\"left\"> Excellent</td><td align=\"center\">11.9</td><td align=\"center\">11.4</td><td align=\"center\">5.9</td><td/></tr><tr><td/><td align=\"left\">Colostomy</td><td align=\"center\">14.5</td><td align=\"center\">14.5</td><td align=\"center\">15.7</td><td align=\"center\">0.67</td></tr><tr><td/><td align=\"left\">History of radiation therapy</td><td align=\"center\">15.3</td><td align=\"center\">17.4</td><td align=\"center\">19.4</td><td align=\"center\">0.51</td></tr><tr><td/><td align=\"left\">History of chemotherapy</td><td align=\"center\">51.6</td><td align=\"center\">49.2</td><td align=\"center\">42.0</td><td align=\"center\">0.07</td></tr><tr><td/><td align=\"left\">Receiving chemotherapy at time of initial survey</td><td align=\"center\">30.9</td><td align=\"center\">30.1</td><td align=\"center\">19.5</td><td align=\"center\">0.002</td></tr><tr><td/><td align=\"left\">Comorbidity Index [median (IQR)]</td><td align=\"center\">2 (1–4)</td><td align=\"center\">2 (1–4)</td><td align=\"center\">3 (1–4)</td><td align=\"center\">0.08</td></tr><tr><td/><td align=\"left\">Bowel Function [mean (SD)]</td><td align=\"center\">8.6 (2.3)</td><td align=\"center\">8.6 (2.3)</td><td align=\"center\">8.8 (2.4)</td><td align=\"center\">0.42</td></tr><tr><td/><td align=\"left\">Time since diagnosis [mean (SD)]</td><td align=\"center\">8.6 (1.7)</td><td align=\"center\">9.2 (2.6)</td><td align=\"center\">9.8 (3.1)</td><td align=\"center\">0.02</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\">Healthcare</td><td align=\"left\">Type of Insurance</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> Medicare</td><td align=\"center\">24.2</td><td align=\"center\">24.3</td><td align=\"center\">28.8</td><td align=\"center\">0.40</td></tr><tr><td/><td align=\"left\"> Commercial Insurance</td><td align=\"center\">68.2</td><td align=\"center\">68.0</td><td align=\"center\">64.4</td><td/></tr><tr><td/><td align=\"left\"> Other/none</td><td align=\"center\">7.7</td><td align=\"center\">7.8</td><td align=\"center\">6.8</td><td/></tr><tr><td/><td align=\"left\">Picker problem scores [mean, (SD)]</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> Psychosocial Care</td><td align=\"center\">30.6 (28.9)</td><td align=\"center\">30.7 (28.5)</td><td align=\"center\">32.4 (28.1)</td><td align=\"center\">0.45</td></tr><tr><td/><td align=\"left\"> Access to Care</td><td align=\"center\">10.8 (22.5)</td><td align=\"center\">10.7 (22.4)</td><td align=\"center\">16.7 (26.9)</td><td align=\"center\">0.003</td></tr><tr><td/><td align=\"left\"> Treatment Information</td><td align=\"center\">30.4 (30.7)</td><td align=\"center\">30.5 (30.9)</td><td align=\"center\">33.9 (33.6)</td><td align=\"center\">0.17</td></tr><tr><td/><td align=\"left\"> Health Information</td><td align=\"center\">46.1 (33.6)</td><td align=\"center\">46.6 (33.6)</td><td align=\"center\">49.6 (35.7)</td><td align=\"center\">0.25</td></tr><tr><td/><td align=\"left\"> Confidence in Providers</td><td align=\"center\">11.7 (20.6)</td><td align=\"center\">11.8 (21.4)</td><td align=\"center\">15.5 (26.7)</td><td align=\"center\">0.05</td></tr><tr><td/><td align=\"left\"> Coordination of Care</td><td align=\"center\">19.4 (23.5)</td><td align=\"center\">19.0 (23.5)</td><td align=\"center\">24.4 (27.4)</td><td align=\"center\">0.008</td></tr><tr><td/><td align=\"left\">Control of Nausea/Vomiting (% somewhat/not at all)</td><td align=\"center\">11.3</td><td align=\"center\">10.9</td><td align=\"center\">14.0</td><td align=\"center\">0.22</td></tr><tr><td/><td align=\"left\">Control of Pain (% somewhat/not at all)</td><td align=\"center\">10.3</td><td align=\"center\">10.4</td><td align=\"center\">18.2</td><td align=\"center\">0.002</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\">HRQL</td><td align=\"left\">Initial HRQL scores [mean, (SD)]</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> TOI</td><td align=\"center\">66.6 (12.7)</td><td align=\"center\">66.9 (12.7)</td><td align=\"center\">64.6 (13.7)</td><td align=\"center\">0.02</td></tr><tr><td/><td align=\"left\"> SWB</td><td align=\"center\">23.2 (4.4)</td><td align=\"center\">23.2 (4.6)</td><td align=\"center\">22.0 (5.1)</td><td align=\"center\">0.002</td></tr><tr><td/><td align=\"left\"> EWB</td><td align=\"center\">20.4 (3.7)</td><td align=\"center\">20.4 (3.7)</td><td align=\"center\">19.6 (4.5)</td><td align=\"center\">0.02</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Clinically meaningful change in HRQL* from the initial to follow-up surveys</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Change in HRQL</bold></td><td align=\"center\" colspan=\"2\"><bold>TOI</bold></td><td align=\"center\" colspan=\"2\"><bold>SWB</bold></td><td align=\"center\" colspan=\"2\"><bold>EWB</bold></td></tr><tr><td/><td align=\"center\"><bold>n (%)</bold></td><td align=\"center\"><bold>mean (SD)</bold></td><td align=\"center\"><bold>n (%)</bold></td><td align=\"center\"><bold>mean (SD)</bold></td><td align=\"center\"><bold>n (%)</bold></td><td align=\"center\"><bold>mean (SD)</bold></td></tr></thead><tbody><tr><td align=\"left\">Meaningful Decline</td><td align=\"center\">161 (32.5)</td><td align=\"center\">-11.0 (6.2)</td><td align=\"center\">167 (33.7)</td><td align=\"center\">-5.2 (3.2)</td><td align=\"center\">130 (26.2)</td><td align=\"center\">-4.5 (2.7)</td></tr><tr><td align=\"left\">About the same</td><td align=\"center\">163 (32.9)</td><td align=\"center\">.07 (2.0)</td><td align=\"center\">193 (38.9)</td><td align=\"center\">-.15 (.85)</td><td align=\"center\">236 (47.6)</td><td align=\"center\">.05 (.70)</td></tr><tr><td align=\"left\">Meaningful Improvement</td><td align=\"center\">172 (34.7)</td><td align=\"center\">10.5 (7.3)</td><td align=\"center\">125 (25.2)</td><td align=\"center\">4.2 (2.4)</td><td align=\"center\">130 (26.2)</td><td align=\"center\">3.8 (2.2)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Predictors of follow-up TOI</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Variable Type</bold></td><td align=\"left\"><bold>Variable</bold></td><td align=\"center\"><bold>B (SE)</bold></td><td align=\"center\"><bold><italic>p</italic>-value</bold></td><td align=\"center\"><bold><italic>sr</italic></bold>²</td><td align=\"center\"><bold>Effect on</bold><break/><bold>follow-up</bold><break/><bold> HRQL*</bold></td></tr></thead><tbody><tr><td align=\"left\">HRQL</td><td align=\"left\">Initial TOI</td><td align=\"center\">.50 (.04)</td><td align=\"center\">&lt;.0001</td><td align=\"center\">.179</td><td align=\"center\"><bold>6.39</bold></td></tr><tr><td align=\"left\">Cancer/Health-related</td><td align=\"left\">General health</td><td align=\"center\">1.88 (.48)</td><td align=\"center\">&lt;.0001</td><td align=\"center\">.018</td><td align=\"center\">1.98</td></tr><tr><td align=\"left\">Healthcare</td><td align=\"left\">Treatment information problem score</td><td align=\"center\">-.04 (.01)</td><td align=\"center\">.005</td><td align=\"center\">.009</td><td align=\"center\">-1.18</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">Adjusted R²: 43.9</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Predictors of follow-up SWB</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Variable Type</bold></td><td align=\"left\"><bold>Variable</bold></td><td align=\"center\"><bold>B (SE)</bold></td><td align=\"center\"><bold><italic>p</italic>-value</bold></td><td align=\"center\"><bold><italic>sr</italic></bold>²</td><td align=\"center\"><bold>Effect on</bold><break/><bold>follow-up</bold><break/><bold> HRQL*</bold></td></tr></thead><tbody><tr><td align=\"left\">HRQL</td><td align=\"left\">Initial SWB</td><td align=\"center\">.59 (.04)</td><td align=\"center\">&lt;.0001</td><td align=\"center\">.222</td><td align=\"center\"><bold>2.59</bold></td></tr><tr><td align=\"left\">Sociodemographic</td><td align=\"left\">Male</td><td align=\"center\">-1.08 (.39)</td><td align=\"center\">.006</td><td align=\"center\">.010</td><td align=\"center\">-1.08</td></tr><tr><td/><td align=\"left\">Race/ethnicity (ref = White)</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\"> Black</td><td align=\"center\">-.90 (.78)</td><td align=\"center\">.25</td><td align=\"center\">.002</td><td align=\"center\">-.90</td></tr><tr><td/><td align=\"left\"> Hispanic</td><td align=\"center\">-2.49 (.69)</td><td align=\"center\">.0003</td><td align=\"center\">.016</td><td align=\"center\"><bold>-2.49</bold></td></tr><tr><td/><td align=\"left\"> Asian/other</td><td align=\"center\">-1.15 (.73)</td><td align=\"center\">.11</td><td align=\"center\">.003</td><td align=\"center\">-1.15</td></tr><tr><td/><td align=\"left\">Married/living as married</td><td align=\"center\">1.10 (.41)</td><td align=\"center\">.008</td><td align=\"center\">.009</td><td align=\"center\">1.10</td></tr><tr><td align=\"left\">Cancer/Health-related</td><td align=\"left\">General health</td><td align=\"center\">.48 (.18)</td><td align=\"center\">.009</td><td align=\"center\">.009</td><td align=\"center\">.51</td></tr><tr><td align=\"left\">Healthcare</td><td align=\"left\">Problems with control of pain/discomfort</td><td align=\"center\">-1.43 (.61)</td><td align=\"center\">.02</td><td align=\"center\">.007</td><td align=\"center\">-1.43</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">Adjusted R²: 39.3</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T5\"><label>Table 5</label><caption><p>Predictors of follow-up EWB</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Variable Type</bold></td><td align=\"left\"><bold>Variable</bold></td><td align=\"center\"><bold>B (SE)</bold></td><td align=\"center\"><bold><italic>p</italic>-value</bold></td><td align=\"center\"><bold><italic>sr</italic></bold>²</td><td align=\"center\"><bold>Effect on</bold><break/><bold>follow-up</bold><break/><bold> HRQL*</bold></td></tr></thead><tbody><tr><td align=\"left\">HRQL</td><td align=\"left\">Initial EWB</td><td align=\"center\">.55 (.04)</td><td align=\"center\">&lt;.0001</td><td align=\"center\">.210</td><td align=\"center\"><bold>2.05</bold></td></tr><tr><td align=\"left\">Cancer/Health-related</td><td align=\"left\">General health</td><td align=\"center\">.54 (.15)</td><td align=\"center\">.0004</td><td align=\"center\">.017</td><td align=\"center\">.56</td></tr><tr><td align=\"left\">Healthcare</td><td align=\"left\">Problems with control of nausea/vomiting</td><td align=\"center\">-1.25 (.46)</td><td align=\"center\">.007</td><td align=\"center\">.009</td><td align=\"center\">-1.25</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">Adjusted R²: 36.5</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>HRQL, Health-related Quality of Life; TOI, Trial Outcome Index-Colorectal; SWB, Social/family well-being; EWB, Emotional well-being; SD, standard deviation; IQR, interquartile range.</p><p>^aData for all variables were collected via self report in the initial survey except gender, stage at diagnosis and age, which were reported to the California Cancer Registry at the time of diagnosis.</p><p>^bNumbers represent percentages unless otherwise specified.</p><p>^cParticipants with non-overlapping survey periods who were evaluated in the regression analyses.</p><p>^d<italic>p</italic>-value for response bias at follow-up. Compares follow-up respondents (<italic>n </italic>= 568) and non-respondents (<italic>n </italic>= 236).</p></table-wrap-foot>", "<table-wrap-foot><p>HRQL, Health-related Quality of Life; TOI, Trial Outcome Index-Colorectal; SWB, Social/family Well-being; EWB, Emotional Well-being</p><p>*Clinically meaningful change is at least 4 points for the TOI and at least 2 points for SWB and EWB.</p></table-wrap-foot>", "<table-wrap-foot><p>HRQL, Health-related Quality of Life; TOI, Trial Outcome Index-Colorectal; <bold>B</bold>, regression coefficient; <bold>SE</bold>, standard error; <bold><italic>sr</italic></bold>², squared semi-partial correlation</p><p>*Difference in predicted follow-up TOI score for a 1 SD difference in interval variables or the difference relative to the reference category for nominal variables. Effects with a magnitude of at least 4 points are considered meaningful and are indicated in bold font.</p></table-wrap-foot>", "<table-wrap-foot><p>HRQL, Health-related Quality of Life; SWB, Social/family Well-being; <bold>B</bold>, regression coefficient; <bold>SE</bold>, standard error; <bold><italic>sr</italic></bold>², squared semi-partial correlation</p><p>*Difference in predicted follow-up SWB score for a 1 SD difference in interval variables or the difference relative to the reference category for nominal variables. Effects with a magnitude of at least 2 points are considered meaningful and are indicated in bold font.</p></table-wrap-foot>", "<table-wrap-foot><p>HRQL, Health-related Quality of Life; EWB, Emotional Well-being; <bold>B</bold>, regression coefficient; <bold>SE</bold>, standard error; <bold><italic>sr</italic></bold>², squared semi-partial correlation</p><p>*Difference in predicted follow-up EWB score for a 1 SD difference in interval variables or the difference relative to the reference category for nominal variables. Effects with a magnitude of at least 2 points are considered meaningful and are indicated in bold font.</p></table-wrap-foot>" ]

[]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Patient reported outcome measures (PROMs) are increasingly being used as primary or secondary outcome measures in clinical trials [##REF##14645463##1##,##REF##18031706##2##]. As such, the quality of inferences made from clinical trials is dependent on the PROMs used. This increasingly acknowledged fact has led the United States Food and Drug Administration (FDA) to produce guidelines [##UREF##0##3##,##REF##17307086##4##] that specify minimum criteria for the scientific adequacy of scales in clinical trials. These are likely to be followed by the European Medicines Agency (EMEA) [##UREF##1##5##], and will have implications for all scales used in future clinical trials.</p>", "<p>The Cervical Dystonia Impact Profile (CDIP-58) assesses the health impact of CD [##REF##15534247##6##]. It was developed using new, sophisticated, but not widely known techniques of PROMs construction (Rasch analysis), which are, as of yet, not included in the FDA guidelines. In addition, researchers interested in using the CDIP-58, who may be unfamiliar with new psychometric methods, may find the original paper [##REF##15534247##6##] abstruse and intangible.</p>", "<p>The aim of this study is to provide clinicians with a comprehensive evaluation of the CDIP-58 using a traditional approach to scale evaluation for three reasons: 1) provide traditional psychometric evidence for the CDIP-58 in line with the proposed FDA guidelines; 2) enable researchers and clinicians to make their own judgment of its performance and compare it with existing dystonia scales; and 3) help researchers and clinicians bridge the knowledge gap between old and new reliability and validity testing methods.</p>" ]

[ "<title>Methods</title>", "<title>Setting and Participants</title>", "<p>A random sample of 460 people with CD was recruited from a complete list of 1110 members from the Dystonia Society of Great Britain. The sampling strategy is described elsewhere [##REF##15534247##6##]. A booklet of questionnaires was administered as a postal survey following standard techniques [##UREF##2##7##]. In addition, 140 individuals were randomly selected to receive a second identical battery after 2 weeks to estimate test-retest reproducibility (TRT). This study was reviewed and passed by the local hospital trust research ethics committee.</p>", "<title>Measurement model</title>", "<p>In the traditional psychometric paradigm, a measurement model proposes how items in a measure are grouped into scales, and in turn how scales are scored. This definition of a measurement model is different to that in the Rasch measurement paradigm, which instead views it as a formulation that represents the structure which data should exhibit in order to obtain measurements from the data. The CDIP-58 measurement model groups the 58 items into eight subscales: head and neck symptoms (6 items), pain and discomfort (5 items), upper limb activities (9 items), walking (9 items), sleep (4 items), annoyance (8 items), mood (7 items), and psychosocial functioning (10 items) [##REF##15534247##6##]. We examined whether the model (Figure ##FIG##0##1##) fulfilled fundamental prerequisites for rigorous measurement as defined by traditional psychometric approaches [##REF##8277801##8##,##REF##17115394##9##].</p>", "<title>Data analyses</title>", "<p>CDIP-58 subscale item responses were summed without weighting or standardisation to generate scores [##UREF##3##10##]. Each subscale score was transformed to have a common range of 0 (no impact) to 100 (most impact) [##UREF##4##11##]. Five psychometric properties were examined: data quality, scaling assumptions, targeting, reliability and validity. Table ##TAB##0##1## shows the extent to which the CDIP-58 testing conforms to the draft guidelines proposed by the FDA [##UREF##0##3##,##REF##17307086##4##].</p>", "<title>Data quality</title>", "<p>Data quality concerns the completeness of item- and scale-level data, and was determined by the percentage of missing data for items, and the percentage of computable scale scores [##REF##8277801##8##]. The criterion for acceptable item-level missing data was &lt; 10% [##REF##9672396##12##] and for computable scale scores &gt; 50% completed items [##UREF##5##13##].</p>", "<title>Scaling assumptions</title>", "<p>Three scaling assumptions should be satisfied for scale scores to be generated using the proposed item groups, and Likert's method of summated ratings [##UREF##6##14##,##UREF##7##15##].</p>", "<p>1. Items in each scale should measure a common underlying construct otherwise it is not appropriate to combine them to generate a scale score [##UREF##8##16##]. This was evaluated by examining the correlation between each item and scale score computed from the remaining items in that scale (corrected item-total correlation). The criterion used was corrected item-total correlation ≥ 0.30 [##UREF##9##17##].</p>", "<p>2. Items in each scale should contain a similar proportion of information concerning the construct being measured otherwise they should be given different weights [##UREF##3##10##]. This criterion was determined by examining the equivalence of corrected item-total correlations. The criterion used was corrected item-total correlation ≥ 0.30 [##UREF##9##17##].</p>", "<p>3. Items should be correctly grouped into scales. That is, items should correlate higher with the total score of their own scales (item own-scale correlation) than with the total score of the other scales (item other-scale correlations). The recommended criterion for definite scaling successes are item-own scale correlations (corrected for overlap) exceeding item-other scale correlations by at least two standard errors (2 × 1√n) [##UREF##9##17##]. In situations where this criterion was not reached, we examined the magnitude of differences between item-own and item-other scale correlations. The greater the magnitude of differences between item-own scale and item other-scale correlations, the greater the support for scaling success.</p>", "<title>Targeting</title>", "<p>The targeting of a scale to a sample indicates whether a scale is acceptable as a measure for the sample. It is recommended that: scale scores should span the entire scale range; floor (proportion of the sample at the maximum scale score) and ceiling (proportion of the sample at the minimum scale score) effects should be low (&lt;15%) [##REF##7550178##18##]; and skewness statistics ranging should range from -1 to +1 [##REF##8161978##19##].</p>", "<title>Reliability</title>", "<p>Reliability is the extent to which scale scores are dependable and consistent. Two types were examined. Internal consistency, reported as Cronbach's alpha coefficients, estimates the random error associated with scores from the intercorrelations among the items [##UREF##10##20##]. TRT reproducibility, reported as intraclass correlations coefficients (ICC) on scores produced by a sub sample assessed twice over a 2-week interval, estimates the ability of CDIP-58 subscales to produce stable scores over a given period of time in which the respondents' condition is assumed to have remained the same [##REF##8161978##19##]. We used a two-way random effects model based on absolute agreement as a suitable, conservative estimate of test retest reliability, as this type of ICC accounts for the systematic differences among levels of ratings. This is because the raters used were only a sample of all possible raters. We used a two-way random effects model based on absolute agreement as a suitable conservative estimate test retest reliability, as this type of ICC accounts for the systematic differences among levels of ratings [##UREF##11##21##]. Recommended criteria for adequate reliability are Cronbach's alpha coefficient ≥ 0.80 [##UREF##11##21##], and TRT ICC ≥ 0.80 [##UREF##12##22##].</p>", "<title>Validity</title>", "<p>Validity is the extent to which a scale measures what it intends to measure and is essential for the accurate and meaningful interpretation of scores [##REF##1030700##23##]. Three aspects were tested:</p>", "<p>1. Intercorrelations between CDIP-58 subscales were assessed to examine the extent to which scales measured separate but related constructs [##REF##8277801##8##]. The magnitude of intercorrelations between CDIP-58 subscale scores were predicted to be consistent with expectation about the proximity of the constructs, and were generally expected to be moderate in size (r = 0.30–0.70) [##UREF##13##24##]. In addition, subscale reliabilities should be larger that inter-scale correlations to support that scales measure distinct constructs.</p>", "<p>2. Correlations between CDIP-58 subscales and other scales were examined. Patients were asked to complete three other questionnaires for validity testing: Medical Outcome Study 36-item Short Form Health Survey (SF-36) measures health status in eight multi-item scales (Role Limitations-Emotional, Role Limitations-Physical, Bodily Pain, Vitality, General Health Perceptions, Social Functioning, Physical Functioning, Mental Health) [##REF##1593914##25##]; 28-item version of the General Health Questionnaire (GHQ-28) measures psychological well being in four dimensions (Somatic Symptoms, Anxiety, Social, Depression) [##REF##424481##26##], and Hospital and Anxiety and Depression Scale (HADS) measures mood in two scales (Depression and Anxiety) [##REF##6880820##27##]. A number of hypotheses were made based on the direction, magnitude and pattern of correlations being consistent with expectations based on the proximity of the constructs.</p>", "<p>Ideally for the results of correlations between CDIP-58 subscales and other scales to be fully interpretable the external measures should be reliable and valid. Whereas we have previously examined the psychometric properties of the SF-36 in CD [##REF##17115394##9##], there are no current published articles which have examined the HADS or GHQ-28. Our reasoning for selecting the latter two scales was on the basis of their wide-spread use in neurologic research. Importantly, this is a common limitation of reliability and validity testing and the findings should be interpreted with this borne in mind.</p>", "<p>Criteria were used as guides as to the magnitude of correlations, as opposed to pass/fail benchmarks (high correlation r &gt; 0.70 and moderate correlation r = 0.30–0.70):</p>", "<p>a. The Pain and Discomfort subscale would correlate more highly with the SF-36 bodily pain than with unrelated measures of psychological functioning (SF-36 Mental Health, HADS Anxiety and Depression).</p>", "<p>b. The Upper Limb and Walking subscales would correlate more highly with the SF-36 physical functioning than with unrelated measures of psychological functioning (SF-36 Mental Health, HADS Anxiety and Depression, GHQ-28).</p>", "<p>c. The Annoyance and Mood subscales would correlate more highly with the SF-36 Mental Health than with unrelated measures of physical functioning (e.g. SF-36 physical functioning).</p>", "<p>d. The Annoyance and Mood subscales would correlate moderately with the HADS, GHQ-28 anxiety and depression scales as these reflect aspects of mood.</p>", "<p>e. The Psychosocial Functioning subscale would correlate moderately with the SF-36 social functioning as this reflects an aspect of psychosocial functioning.</p>", "<p>3. Correlations between CDIP-58 subscales and sociodemographic variables (age, sex, and level of education attained) were examined to determine the extent to which they were biased by these variables. We predicted that these correlations would be low &lt; 0.30.</p>" ]

[ "<title>Results</title>", "<title>Sample</title>", "<p>Of the 460 patients who received the CDIP-58, 391 returned completed questionnaires (corrected response rate = 87%). Of the 140 TRT questionnaires, 105 were returned completed (corrected response rate = 75%). The sample included people with a wide range of ages and disease duration (Table ##TAB##1##2##) [##REF##15534247##6##].</p>", "<title>Psychometric properties</title>", "<title>Data quality (Table ##TAB##2##3##)</title>", "<p>Data quality was high. The proportion of item-level missing data was low (≤ 4%). Subscale scores could be computed for at least 96% of the sample.</p>", "<title>Scaling assumptions (Table ##TAB##2##3##)</title>", "<p>Item groupings in each of the eight CDIP-58 subscales passed tests for scaling assumptions:</p>", "<p>1. Corrected item-total correlations for each of the eight CDIP subscales ranged from 0.64–0.93 satisfying the recommended criteria (&gt; 0.30). This supported that items in each subscale of the CDIP-58 measured a common underlying construct.</p>", "<p>2. Corrected item-total correlations &gt; 0.30 indicated that items in each of the subscales contained a similar proportion of information.</p>", "<p>3. Fifty-five of the fifty-eight items correlated higher with their own subscale than other subscales. Forty-seven of these exceeded the criterion (2 × 1√n). This provided some support for the grouping of items in each of the eight subscales. There was less support for three items which correlated higher with other subscales: Head Neck symptoms 'stiffness in your neck' (Pain and Discomfort subscale, r = 0.72), Pain and Discomfort 'tightness in your neck' (Head and Neck symptoms subscale, r = 0.75), and Upper Limb 'getting tired when doing demanding physical activities' (Walking subscale, r = 0.82).</p>", "<title>Targeting (Table ##TAB##2##3##)</title>", "<p>Subscale scores spanned the entire scale range. However, the Walking scale fell just outside of the criterion (ceiling effect = 17%) and the Sleep subscale was found to have a more significant ceiling effect (27%). Despite this, responses were not notably skewed (-0.23 to +0.82). These findings indicate good scale-to-sample targeting, thus supporting total and subscale scores as appropriate for all patients representing the full spectrum of CD impact.</p>", "<title>Reliability (Table ##TAB##2##3##)</title>", "<p>Cronbach's alpha, and test-retest ICCs for all eight CDIP-58 subscales were high (&gt; 0.83), supporting their reliability.</p>", "<title>Validity (Table ##TAB##3##4##)</title>", "<p>1. Intercorrelations between CDIP-58 subscales ranged from 0.44 – 0.84, suggesting the subscales measured related but different constructs. A few of the correlations fell outside of the predicted range of correlations, and were highly correlated (highlighted in Table ##TAB##1##2##). However, the correlations were not unreasonable given the proximity of the constructs in each of the subscales (see Figure ##FIG##0##1##) and scale reliabilities were larger than inter-scale correlations supporting that CDIP-58 subscales measure distinct constructs.</p>", "<p>2. Correlations between CDIP-58 subscales and hypothesised related scales of the SF-36, GHQ and HADS were consistent with predictions (highlighted in Table ##TAB##3##4##). This provides support that the CDIP-58 subscales measure what they intend to measure.</p>", "<p>3. Correlations between CDIP-58 subscales and sociodemographic variables (age, sex, and level of education attained) were in general low (-0.17 to +0.06). This finding suggests that responses to the CDIP-58 were not biased by socio-demographic factors.</p>" ]

[ "<title>Discussion</title>", "<p>The forthcoming FDA guidelines make it increasingly important for researchers and clinicians to be exposed to the science behind PROMs. In this study, the CDIP-58 satisfied traditional psychometric criteria for data quality, scaling assumptions, targeting, reliability and validity. We hope that this, together with our previous work on conceptual model and scale development [##REF##15534247##6##] and assessment of the sensitivity to clinical change of the CDIP-58 following Botulinum Toxin Type A (that found it to be superior to existing CD PROMs [##REF##17190951##28##]), provides an evidence-base for its use in clinical trials, in line with the forthcoming FDA guidelines. As such, the CDIP-58 offers an advance on current PROMs. In addition, our findings are relevant to practicing neurologists, who can use this information to compare the CDIP-58 to existing published CD PROM data, which will help to avoid an ad hoc approach which may negatively impact upon rigorous measurement.</p>", "<p>Three main issues arise from the findings. First, were there any instances where traditional psychometric criteria were not met and how should we interpret these? Second, how can the information provided here be used and what do traditional psychomteric analyses tell us? Third, what is the added value of using Rasch analysis to develop PROMs and in particular, what benefits are gained from the required additional investment in skill level, retraining and software costs?</p>", "<p>Traditional psychometric analyses detected one problem not identified by Rasch analysis. Tests of scaling assumptions were failed by 3 items ('stiffness in the neck', 'tightness in the neck', 'upset'). This means that they correlated similarly with their own subscales and other subscales they were not intended to belong in. There are three reasons why this may be the case. First, the subscales in question were themselves highly correlated. Second, these items may be non-specific indicators of their intended construct. Third, any item can exist conceptually in more than one scale. The clinical implications of this are probably minimal, as the constructs measured by the three subscales in question are anchored by the other items, which in turn performed well psychometrically.</p>", "<p>So, how can the information provided here be used and what do traditional psychomteric analyses tell us? Researchers who are unfamiliar with Rasch analysis can use the information presented here to compare the CDIP-58 to existing published CD PROM data. The caveat is that any inferences made from this paper alone are constrained by the sample and scale limitations inherent to all studies that use traditional psychometric analyses. These include three main points. First, total scores are often analysed as if they were interval measures. However, it has been widely demonstrated that they are not, and therefore, they are not measuring consistently across the range of the scale. Importantly, we do not know the extent to which they are measuring inconsistently across the scale. Second, traditional psychometric analyses rely directly on the items and samples used to estimate them. This means that item properties vary depending on the sample and patient scores in turn depend on the set of items taken. Thus, the reliability and validity estimates of a measure may differ across different patient groups. Third, it is recommended that total scores are only used for group comparison studies and not individual patient measurement, because the confidence intervals around individual patient scores are so wide [##REF##7550178##18##].</p>", "<p>Our study suggests that Rasch analysis can produce a reliable and valid measure as defined by traditional criteria. What then is the added value of using new psychometric methods? First, when scales are successfully developed using Rasch analysis it is possible to transform ordinal level scale scores into interval level measurements [##UREF##14##29##, ####UREF##15##30##, ##UREF##16##31####16##31##]. This improves the accuracy with which we can measure differences between people and clinical change. Second, Rasch analysis enables estimates suitable for individual person measurement. This can help directly inform upon patient monitoring, management and treatment for patients. Third, reliability and validity estimates computed using Rasch analysis are much less sample dependent than those derived from traditional methods. In addition, Rasch measurement methods afford more sophisticated analyses to test theoretically driven concepts and therefore provide empirical evidence for properties such as construct validity. This has important relevance for the generalisability of PROM evaluations. These benefits are further explored in other relevant articles and texts [##REF##14645463##1##,##REF##18031706##2##,##UREF##17##32##, ####REF##12199545##33##, ##REF##14707751##34####14707751##34##].</p>", "<p>We envisage that this article, in conjunction with our previous articles on the Rasch development of the CDIP [##REF##15534247##6##] and our recent review in Lancet Neurology [##REF##18031706##2##] describing traditional and new psychometric techniques, can be used by researchers and clinicians to help bridge the knowledge gap between traditional and modern reliability and validity testing methods. This study has shown that Rasch analysis methods can produce a PROM that stands up to traditional psychometric criteria. A demonstration of this nature is rare. It is much more common that scales developed using traditional methods to be tested <bold><italic>post hoc </italic></bold>using new approaches [##REF##14707753##35##]. Nevertheless, direct comparisons of new and traditional psychometric methods of any nature in the medical literature are sparse, and at best superficial [##REF##9179104##36##,##REF##12952544##37##] In part, this may be due to the fact that these two approaches cannot be compared easily as they use different methods, produce different information, and apply different criteria for success and failure. Both approaches have their supporters and traditional psychometric methods remain the dominant paradigm. However, we believe that state-of-the-art clinical trials and research would benefit from the advantages offered by Rasch analysis.</p>" ]

[ "<title>Conclusion</title>", "<p>This study has shown that new psychometric methods can produce a PROM that stands up to traditional criteria and supports the clinical advantages of Rasch analysis. In addition, the CDIP-58 satisfied traditional reliability and validity criteria further supporting it as a clinically useful measure for use in routine practice, audit and treatment trials.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>The United States Food and Drug Administration (FDA) are currently producing guidelines for the scientific adequacy of patient reported outcome measures (PROMs) in clinical trials, which will have implications for the selection of scales used in future clinical trials. In this study, we examine how the Cervical Dystonia Impact Profile (CDIP-58), a rigorous Rasch measurement developed neurologic PROM, stands up to traditional psychometric criteria for three reasons: 1) provide traditional psychometric evidence for the CDIP-58 in line with proposed FDA guidelines; 2) enable researchers and clinicians to compare it with existing dystonia PROMs; and 3) help researchers and clinicians bridge the knowledge gap between old and new methods of reliability and validity testing.</p>", "<title>Methods</title>", "<p>We evaluated traditional psychometric properties of data quality, scaling assumptions, targeting, reliability and validity in a group of 391 people with CD. The main outcome measures used were the CDIP-58, Medical Outcome Study Short Form-36, the 28-item General Health Questionnaire, and Hospital and Anxiety and Depression Scale.</p>", "<title>Results</title>", "<p>A total of 391 people returned completed questionnaires (corrected response rate 87%). Analyses showed: 1) data quality was high (low missing data ≤ 4%, subscale scores could be computed for &gt; 96% of the sample); 2) item groupings passed tests for scaling assumptions; 3) good targeting (except for the Sleep subscale, ceiling effect = 27%); 4) good reliability (Cronbach's alpha ≥ 0.92, test-retest intraclass correlations ≥ 0.83); and 5) validity was supported.</p>", "<title>Conclusion</title>", "<p>This study has shown that new psychometric methods can produce a PROM that stands up to traditional criteria and supports the clinical advantages of Rasch analysis.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>SC collected, conducted, analysed and interpreted the data and wrote the manuscript. JH conceived and designed the study and contributed to the interpretation of data and writing of the manuscript. TW, RF, KB and AT were involved in guiding the study including design and acquisition of data, and reviewing drafts of this manuscript. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>We wish to thank the people with CD who participated in this study and Mr Alan Leng and Ms Laura Camfield at the Dystonia Society of Great Britain for help with recruitment. This study was supported by a project grant from the Wellcome Trust. During the writing of this paper Dr Hobart benefited by being on secondment to the School of Education, Murdoch University, Perth, Western Australia, and was support by the Royal Society of Medicine (Ellison-Cliffe Travelling Fellowship), the MS Society of Great Britain and Northern Ireland, and the NHS Technology Assessment Programme (but the opinions expressed do not necessarily reflect those of the executive).</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Measurement model of the CDIP-58.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Adapted from table 4 of the FDA draft guidelines for measurement properties reviewed for PRO instruments used in clinical trials</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Measurement Property</bold></td><td align=\"left\"><bold>Test</bold></td><td align=\"center\"><bold>Methods used in testing the CDIP-58</bold></td></tr></thead><tbody><tr><td align=\"left\">Reliability</td><td align=\"left\">Test-retest</td><td align=\"center\">✓</td></tr><tr><td align=\"left\">Internal consistency</td><td align=\"left\">Whether the items in a domain are intercorrelated, as evidenced by an internal consistency statistic (e.g., coefficient alpha)</td><td align=\"center\">✓</td></tr><tr><td align=\"left\">Inter-interviewer reproducibility (for interviewer-administered PROs only)</td><td align=\"left\">Agreement between responses when the PRO is administered by two or more different interviewers</td><td align=\"center\">NA</td></tr><tr><td align=\"left\">Validity</td><td align=\"left\">Content-related</td><td align=\"center\">✓*</td></tr><tr><td align=\"left\">Ability to measure the concept (also known as construct-related validity; can include tests for discriminant, convergent, and known-groups validity)</td><td align=\"left\">Whether relationships among items, domains, and concepts conform to what is predicted by the conceptual framework for the PRO instrument itself and its validation hypotheses.</td><td align=\"center\">✓</td></tr><tr><td align=\"left\">Ability to predict future outcomes (also known as predictive validity)</td><td align=\"left\">Whether future events or status can be predicted by changes in the PRO scores</td><td align=\"center\">✗</td></tr><tr><td align=\"left\">Ability to detect change</td><td align=\"left\">Includes calculations of effect size and standard error of measurement among others</td><td align=\"center\">✓**</td></tr><tr><td align=\"left\">Interpretability</td><td align=\"left\">Smallest difference that is considered clinically important; this can be a specified difference (the minimum important difference (MID)) or, in some cases, any detectable difference. The MID is used as a benchmark to interpret mean score differences between treatment arms in a clinical trial</td><td align=\"center\">✓/✗***</td></tr><tr><td align=\"left\">Responder definition – used to identify responders in clinical trials for analyzing differences in the proportion of responders between treatment arms</td><td align=\"left\">Change in score that would be clear evidence that an individual patient experienced a treatment benefit. Can be based on experience with the measure using a distribution-based approach, a clinical or non-clinical anchor, an empirical rule, or a combination of approaches.</td><td align=\"center\">NA</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Respondent characteristics</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\"><bold>Variable*</bold></td><td align=\"center\"><italic>Sample</italic></td></tr></thead><tbody><tr><td align=\"left\">Number</td><td align=\"center\">391</td></tr><tr><td align=\"left\">Sex</td><td/></tr><tr><td align=\"left\"> Female</td><td align=\"center\">72</td></tr><tr><td align=\"left\">Age</td><td/></tr><tr><td align=\"left\"> Mean (SD)</td><td align=\"center\"><italic>58 (12)</italic></td></tr><tr><td align=\"left\"> Range</td><td align=\"center\"><italic>25–88</italic></td></tr><tr><td align=\"left\">Ethnicity</td><td/></tr><tr><td align=\"left\"> White</td><td align=\"center\"><italic>97</italic></td></tr><tr><td align=\"left\">Years since CD onset</td><td/></tr><tr><td align=\"left\"> Mean (SD)</td><td align=\"center\"><italic>15 (10)</italic></td></tr><tr><td align=\"left\"> Range</td><td align=\"center\"><italic>2 – 50</italic></td></tr><tr><td align=\"left\">Employment status</td><td/></tr><tr><td align=\"left\"> Retired</td><td align=\"center\">35</td></tr><tr><td align=\"left\"> Employed</td><td align=\"center\">29</td></tr><tr><td align=\"left\"> Unable to work due to CD</td><td align=\"center\">20</td></tr><tr><td align=\"left\">Treatment</td><td/></tr><tr><td align=\"left\"> Botulinum Injections</td><td align=\"center\">90</td></tr><tr><td align=\"left\"> Drug therapy</td><td align=\"center\">53</td></tr><tr><td align=\"left\"> Alternative treatment</td><td align=\"center\">0</td></tr><tr><td align=\"left\"> Surgery</td><td align=\"center\">0</td></tr><tr><td align=\"left\">External measures (Mean; SD)</td><td/></tr><tr><td align=\"left\"> SF-36** Bodily Pain</td><td align=\"center\">46 (26)</td></tr><tr><td align=\"left\"> SF-36 Social Functioning</td><td align=\"center\">55 (31)</td></tr><tr><td align=\"left\"> SF-36 Physical Functioning</td><td align=\"center\">57 (29)</td></tr><tr><td align=\"left\"> SF-36 Mental Health</td><td align=\"center\">62 (20)</td></tr><tr><td/><td/></tr><tr><td align=\"left\"> GHQ*** Anxiety</td><td align=\"center\">38 (19)</td></tr><tr><td align=\"left\"> GHQ Depression</td><td align=\"center\">14 (19)</td></tr><tr><td/><td/></tr><tr><td align=\"left\"> HADS*** Anxiety</td><td align=\"center\">43 (24)</td></tr><tr><td align=\"left\"> HADS Depression</td><td align=\"center\">33 (19)</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Data quality, scaling assumptions, targeting, reliability and validity</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"right\"><bold>CDIP Scale</bold></td><td align=\"center\">Head and<break/>Neck<break/>Symptoms<break/> (6 items)</td><td align=\"center\">Pain and <break/>Discomfort<break/> (5 items)</td><td align=\"center\">Upper<break/>Limb<break/>Activities<break/> (9 items)</td><td align=\"center\">Walking<break/> (9 items)</td><td align=\"center\">Sleep<break/> (4 items)</td><td align=\"center\">Annoyance<break/> (8 items)</td><td align=\"center\">Mood<break/> (7 items)</td><td align=\"center\">Psycho-<break/>social<break/>Functioning<break/> (10 items)</td></tr><tr><td align=\"left\"><bold>Psychometric property</bold></td><td/><td/><td/><td/><td/><td/><td/><td/></tr></thead><tbody><tr><td align=\"left\">Data quality</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"> Item missing data (range %)</td><td align=\"center\">1–2</td><td align=\"center\">3–4</td><td align=\"center\">2–4</td><td align=\"center\">3–4</td><td align=\"center\">2</td><td align=\"center\">2–4</td><td align=\"center\">2–4</td><td align=\"center\">1–2</td></tr><tr><td align=\"left\"> Computable scale scores (%)</td><td align=\"center\">98</td><td align=\"center\">97</td><td align=\"center\">98</td><td align=\"center\">97</td><td align=\"center\">98</td><td align=\"center\">98</td><td align=\"center\">96</td><td align=\"center\">99</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Corrected item-total correlations</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"> Mean</td><td align=\"center\">0.76</td><td align=\"center\">0.82</td><td align=\"center\">0.78</td><td align=\"center\">0.87</td><td align=\"center\">0.89</td><td align=\"center\">0.83</td><td align=\"center\">0.78</td><td align=\"center\">0.82</td></tr><tr><td align=\"left\"> Range</td><td align=\"center\">0.67–0.81</td><td align=\"center\">0.70–0.87</td><td align=\"center\">0.64–0.87</td><td align=\"center\">0.82–0.91</td><td align=\"center\">0.84–0.93</td><td align=\"center\">0.79–0.89</td><td align=\"center\">0.68–0.84</td><td align=\"center\">0.70–0.90</td></tr><tr><td align=\"left\">Item-other scale correlations</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"> Range</td><td align=\"center\">0.37–0.72</td><td align=\"center\">0.40–0.75</td><td align=\"center\">0.43–0.72</td><td align=\"center\">0.42–0.75</td><td align=\"center\">0.39–0.52</td><td align=\"center\">0.42–0.73</td><td align=\"center\">0.39–0.82</td><td align=\"center\">0.38–0.69</td></tr><tr><td align=\"left\">Scaling successes (%)</td><td align=\"center\">83</td><td align=\"center\">80</td><td align=\"center\">100</td><td align=\"center\">100</td><td align=\"center\">100</td><td align=\"center\">100</td><td align=\"center\">86</td><td align=\"center\">100</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"> Targeting</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\"> Mean score</td><td align=\"center\">57.6</td><td align=\"center\">53.8</td><td align=\"center\">42.9</td><td align=\"center\">37.1</td><td align=\"center\">33.3</td><td align=\"center\">37.9</td><td align=\"center\">29.7</td><td align=\"center\">49.2</td></tr><tr><td align=\"left\"> Standard deviation</td><td align=\"center\">25.6</td><td align=\"center\">27.9</td><td align=\"center\">27.6</td><td align=\"center\">31.5</td><td align=\"center\">31.5</td><td align=\"center\">26.9</td><td align=\"center\">24.9</td><td align=\"center\">29.5</td></tr><tr><td align=\"left\"> Score range</td><td align=\"center\">0–100</td><td align=\"center\">0–100</td><td align=\"center\">0–100</td><td align=\"center\">0–100</td><td align=\"center\">0–100</td><td align=\"center\">0–100</td><td align=\"center\">0–100</td><td align=\"center\">0–100</td></tr><tr><td align=\"left\"> Floor/ceiling effect (%)</td><td align=\"center\">1/6</td><td align=\"center\">2/7</td><td align=\"center\">7/0</td><td align=\"center\">17/4</td><td align=\"center\">27/7</td><td align=\"center\">7/4</td><td align=\"center\">13/1</td><td align=\"center\">5/4</td></tr><tr><td align=\"left\">  Skewness</td><td align=\"center\">-0.23</td><td align=\"center\">-0.15</td><td align=\"center\">0.11</td><td align=\"center\">0.44</td><td align=\"center\">0.66</td><td align=\"center\">0.51</td><td align=\"center\">0.82</td><td align=\"center\">0.02</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Reliability (n = 377–385)</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">   Cronbach's alpha</td><td align=\"center\">0.92</td><td align=\"center\">0.93</td><td align=\"center\">0.94</td><td align=\"center\">0.97</td><td align=\"center\">0.96</td><td align=\"center\">0.96</td><td align=\"center\">0.95</td><td align=\"center\">0.96</td></tr><tr><td align=\"left\">   TRT (ICC; n = 92–95)</td><td align=\"center\">0.85</td><td align=\"center\">0.83</td><td align=\"center\">0.94</td><td align=\"center\">0.95</td><td align=\"center\">0.86</td><td align=\"center\">0.83</td><td align=\"center\">0.85</td><td align=\"center\">0.89</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Convergent and discriminant construct validity of the CDIP-58</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"left\">Instrument</td><td align=\"left\">Scale/<break/>Dimension/<break/>Variable</td><td/><td/><td/><td/><td/><td/><td/><td/></tr></thead><tbody><tr><td/><td align=\"left\">Validity<break/> (Correlation)</td><td align=\"center\">Head and<break/>Neck<break/>Symptoms<break/> (6 items)</td><td align=\"center\">Pain and<break/>Discomfort<break/> (5 items)</td><td align=\"center\">Upper<break/>Limb<break/>Activities<break/> (9 items)</td><td align=\"center\">Walking<break/> (9 items)</td><td align=\"center\">Sleep<break/> (4 items)</td><td align=\"center\">Annoyance<break/> (8 items)</td><td align=\"center\">Mood<break/> (7 items)</td><td align=\"center\">Psycho-<break/>social <break/>Funct-ioning<break/> (10 items)</td></tr><tr><td colspan=\"10\"><hr/></td></tr><tr><td align=\"left\">CDIP-58</td><td align=\"left\">Head and Neck Symptoms</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td align=\"left\">Pain and Discomfort</td><td align=\"center\"><bold>0.72^a</bold></td><td align=\"center\">-</td><td/><td/><td/><td/><td/><td/></tr><tr><td/><td align=\"left\">Upper Limb Activities</td><td align=\"center\">0.64</td><td align=\"center\">0.53</td><td align=\"center\">-</td><td/><td/><td/><td/><td/></tr><tr><td/><td align=\"left\">Walking</td><td align=\"center\">0.63</td><td align=\"center\">0.65</td><td align=\"center\"><bold>0.79^a</bold></td><td align=\"center\">-</td><td/><td/><td/><td/></tr><tr><td/><td align=\"left\">Sleep</td><td align=\"center\">0.50</td><td align=\"center\">0.53</td><td align=\"center\">0.54</td><td align=\"center\">0.50</td><td align=\"center\">-</td><td/><td/><td/></tr><tr><td/><td align=\"left\">Annoyance</td><td align=\"center\">0.63</td><td align=\"center\">0.55</td><td align=\"center\">0.60</td><td align=\"center\">0.56</td><td align=\"center\">0.52</td><td align=\"center\">-</td><td align=\"center\">-</td><td/></tr><tr><td/><td align=\"left\">Mood</td><td align=\"center\">0.52</td><td align=\"center\">0.48</td><td align=\"center\">0.54</td><td align=\"center\">0.53</td><td align=\"center\">0.49</td><td align=\"center\"><bold>0.84^a</bold></td><td align=\"center\">-</td><td/></tr><tr><td/><td align=\"left\">Psychosocial Functioning</td><td align=\"center\">0.67</td><td align=\"center\">0.51</td><td align=\"center\">0.55</td><td align=\"center\">0.53</td><td align=\"center\">0.44</td><td align=\"center\"><bold>0.73^a</bold></td><td align=\"center\">0.69</td><td align=\"center\">-</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">SF-36*</td><td align=\"left\">Bodily Pain</td><td align=\"center\">-0.60</td><td align=\"center\"><bold>-0.71^b</bold></td><td align=\"center\">-0.72</td><td align=\"center\">-0.62</td><td align=\"center\">-0.59</td><td align=\"center\">-0.54</td><td align=\"center\">-0.51</td><td align=\"center\">-0.48</td></tr><tr><td/><td align=\"left\">Social Functioning</td><td align=\"center\">-0.56</td><td align=\"center\">-0.56</td><td align=\"center\">-0.70</td><td align=\"center\">-0.62</td><td align=\"center\">-0.55</td><td align=\"center\">-0.63</td><td align=\"center\">-0.59</td><td align=\"center\"><bold>-0.69^b</bold></td></tr><tr><td/><td align=\"left\">Physical Functioning</td><td align=\"center\">-0.45</td><td align=\"center\">-0.57</td><td align=\"center\"><bold>-0.80^b</bold></td><td align=\"center\"><bold>-0.78^b</bold></td><td align=\"center\">-0.53</td><td align=\"center\"><bold>-0.43^b</bold></td><td align=\"center\"><bold>-0.43^b</bold></td><td align=\"center\">-0.45</td></tr><tr><td/><td align=\"left\">Mental Health</td><td align=\"center\">-0.43</td><td align=\"center\"><bold>-0.41^b</bold></td><td align=\"center\"><bold>-0.72^b</bold></td><td align=\"center\"><bold>-0.40^b</bold></td><td align=\"center\">-0.43</td><td align=\"center\"><bold>-0.73^b</bold></td><td align=\"center\"><bold>-0.78^b</bold></td><td align=\"center\">-0.60</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">HADS</td><td align=\"left\">Anxiety</td><td align=\"center\">0.34</td><td align=\"center\"><bold>0.39^b</bold></td><td align=\"center\"><bold>0.31^b</bold></td><td align=\"center\"><bold>0.28^b</bold></td><td align=\"center\">0.31</td><td align=\"center\"><bold>0.54^b</bold></td><td align=\"center\"><bold>0.66^b</bold></td><td align=\"center\">0.48</td></tr><tr><td/><td align=\"left\">Depression</td><td align=\"center\">0.41</td><td align=\"center\"><bold>0.45^b</bold></td><td align=\"center\"><bold>0.52^b</bold></td><td align=\"center\"><bold>0.51^b</bold></td><td align=\"center\">0.42</td><td align=\"center\"><bold>0.59^b</bold></td><td align=\"center\"><bold>0.60^b</bold></td><td align=\"center\">0.50</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">GHQ**</td><td align=\"left\">Anxiety</td><td align=\"center\">0.30</td><td align=\"center\"><bold>0.30^b</bold></td><td align=\"center\"><bold>0.34^b</bold></td><td align=\"center\"><bold>0.19^b</bold></td><td align=\"center\">0.22</td><td align=\"center\"><bold>0.44^b</bold></td><td align=\"center\"><bold>0.50^b</bold></td><td align=\"center\">0.30</td></tr><tr><td/><td align=\"left\">Depression</td><td align=\"center\">0.35</td><td align=\"center\"><bold>0.29^b</bold></td><td align=\"center\"><bold>0.31^b</bold></td><td align=\"center\"><bold>0.36^b</bold></td><td align=\"center\">0.18</td><td align=\"center\"><bold>0.53^b</bold></td><td align=\"center\"><bold>0.66^b</bold></td><td align=\"center\">0.43</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td align=\"left\">Demo-graphic</td><td align=\"left\">Age</td><td align=\"center\">-0.05</td><td align=\"center\">-0.03</td><td align=\"center\">-0.02</td><td align=\"center\">0.01</td><td align=\"center\">0.00</td><td align=\"center\">-0.04</td><td align=\"center\">-0.07</td><td align=\"center\">0.00</td></tr><tr><td align=\"left\">Variables</td><td align=\"left\">Sex</td><td align=\"center\">-0.04</td><td align=\"center\">-0.17</td><td align=\"center\">0.02</td><td align=\"center\">-0.07</td><td align=\"center\">0.00</td><td align=\"center\">0.06</td><td align=\"center\">0.04</td><td align=\"center\">0.02</td></tr><tr><td/><td align=\"left\">Education</td><td align=\"center\">-0.04</td><td align=\"center\">-0.05</td><td align=\"center\">-0.01</td><td align=\"center\">-0.07</td><td align=\"center\">-0.02</td><td align=\"center\">-0.03</td><td align=\"center\">-0.03</td><td align=\"center\">-0.08</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>✓ = tested; ✗ = not tested; * Reported in Cano et al 2004 [##REF##15534247##6##]; ** Reported in Cano et al 2006 [##REF##17190951##28##]; *** Although not including MID in our responsiveness paper (Cano et al, 2006 [##REF##17190951##28##]), we include a comparison of relative responsiveness to existing PROs used in CD research in order to increase the interpretability of CDIP-58 change scores against these measures.</p></table-wrap-foot>", "<table-wrap-foot><p>*All values are percentages unless specified otherwise; ** range 0-100, *** converted to 0-100 (original range 0-21)</p></table-wrap-foot>", "<table-wrap-foot><p>*4/8 scales omitted from table as not applicable to analyses; **2/8 scales omitted from table as not applicable to analyses</p><p>^aCorrelations falling outside of the predicted range; ^bCorrelations consistent with predictions</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1477-7525-6-58-1\"/>" ]

[]

{ "acronym": [], "definition": [] }

[]

[]

[]

[]

[ "<title>Conclusion</title>", "<p>This presentation is intended to stimulate thinking about the difficulties in stepping outside of our primary areas of interest, and to offer frames of reference for considering where synergies might be identified and acted upon. Dedicated individuals and organizations who care about equity can find areas of synergy to enhance dynamic social action in support of reproductive health, rights and justice, regardless of their primary objectives. Specific actions, such as inclusion of LAM as an introductory method towards adequate child spacing in breastfeeding, family planning, and child survival programming, are encouraged.</p>", "<p>Our mission here today is to support the reproductive continuum, including health, rights and justice and addressing birth, breastfeeding and family planning. The support of the reproductive continuum is, literally, a matter of life and death. The areas of synergy and action that emerge with transdisciplinary approaches may best support effective and sustainable action inhealth, rights and social justice for mothers, children, and perhaps, for the most naturally synergized unit – the mother/child dyad.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<p>This paper was presented at the symposium on Breastfeeding and Feminism: A Focus on Reproductive Health, Rights and Justice. It underscores the power and potential of synergy between and among organizations and individuals supporting breastfeeding, the mother-child dyad, and reproductive health to increase sustainable breastfeeding support. These concepts were brought together to lay the groundwork for working group discussions of synergy in program and policy actions.</p>" ]

[ "<title>The issue</title>", "<p>Reproduction may be defined as the successful creation of the next generation of individuals of the same species who are also able to reproduce. Given the importance of breastfeeding for ensuring the health and survival of the next generation, its impact on fertility, and its impact on maternal health, it is arguable that breastfeeding is as much a part of the reproductive continuum as are conception, pregnancy, birth and family planning. Many talented individuals from women's collectives, health systems, and policy-making bodies are doing excellent work to support and protect women at various points on the reproductive continuum. However, when any one point on the continuum is addressed, without consideration of the other parts of the continuum, the action cannot yield the most sustainable and effective result. And yet, despite the many commonalities in the philosophies, policies and programs that address maternal survival, birthing practices, family planning, or breastfeeding, there has not as yet been a concerted effort to ensure synergy. Sustainable and ever growing action in these arenas may depend on identification of synergies and building on mutual strengths.</p>", "<p>This article explores reproductive health, rights and justice as an intergenerational continuum, more than as a women's issue or a child issue, alone. As a group of doctors and lawyers, lactation consultants, government health workers, public health specialists and community activists, feminists, scholars, mothers, fathers and children, let us consider together the possibility that each of us might peer outside of the \"silos\" of our own interest areas to create new and increased cross-sectoral synergy in program and policy.</p>", "<p>While there is a general familiarity with terms such as cross-disciplinary and multi-disciplinary, the concept of transdisciplinarity is relatively new. An understanding of the derivation of this term is helpful in understanding its importance for work on reproduction and breastfeeding.</p>", "<p>• <bold>Disciplinary: </bold>Of, or relating to, a specific field of academic study.</p>", "<p>• <bold>Multidisciplinary: </bold>Of, relating to, or making use of several disciplines at once.</p>", "<p>• <bold>Interdisciplinary: </bold>Of, relating to, or involving two or more academic disciplines that are usually considered distinct.</p>", "<p>• <bold>Cross-Disciplinary: </bold>Integrative learning.</p>", "<p>It is important to note that these terms do not necessarily involve initial or ongoing working together throughout all stages of a process. When multiple disciplines do not work together from the start, it can lead to unexpected conflicts and occasionally quite negative outcomes. For example, many disciplines are addressing the international HIV/AIDS issue today, included the disease-specific AIDS community, the Gender Awareness community, and child survival community.</p>", "<p>• The goals of HIV/AIDS disciples include ensuring that every possible case of HIV/AIDS is prevented or treated, and that those living with HIV/AIDS are given all the support that they need. In attempting to prevent every chance of mother-to-child transmission by whatever means possible, and, this group has worked to eliminate breastfeeding by all HIV-positive women.</p>", "<p>• The goals of gender awareness disciples include ensuring that gender issues are considered in all programming, and that rights of all genders are respected. For example, \"Prevention of Mother to Child Transmission (PMTCT)\" terminology is considered gender insensitive, \"blaming the mother.\" Repeated requests to change the term to \"pediatric HIV/AIDS\" or even \"vertical transmission\" have gone unheeded. Another activity considered to be gender insensitive was the decision to concentrate on testing only pregnant women, rather than all women, sending a clear message that infants' health is prioritized above women's health.</p>", "<p>• The goals of child survival disciples include ensuring the reduction of preventable morbidity and mortality, and increasing rates of survival, growth and development. From this perspective, the goal is HIV-free survival, and this is best served by promotion of exclusive breastfeeding and other proven child survival interventions for all, and consideration of HIV/AIDS secondarily as only one of the causes of mortality amongst many.</p>", "<p>Bringing various disciplines together once they are entrenched in their own paradigms often may seem impossible, and later attempts to create synergy too frequently result in entrenchment and less than optimal policy and programme development.</p>", "<title>Suggested responsive actions</title>", "<title>Transdisciplinarity</title>", "<p>Jean Piaget coined the word \"transdisciplinarity\" [##UREF##0##1##]. He called for reaching beyond interdisciplinary coordination, stating \"concerning interdisciplinary discourse, we hope to see a higher level emerge, 'transdisciplinarity', which would not settle for interactions or reciprocities between specializations, but which would internalize such interaction within an overall construct, and break down the walls between disciplines.\" Transdisciplinarity guides problem perceptions and solutions. Alternatively stated, a transdisciplinary approach would require that many disciplines come together not only to address a predefined problem from each one's own perspective, but rather to together define the problem to be addressed.</p>", "<p>If the HIV, gender awareness and child survival groups had taken a transdisciplinary approach initially, they may have found initial commonalities and decided on a different set of goals and messages, such as:</p>", "<p>• Formula use expends dollars needed for treatment or prevention.</p>", "<p>• Where it is not possible to formula feed safely, exclusive breastfeeding is the best option as it could cut transmission by half.</p>", "<p>• Test and treat all women. This may result in decisions not to become pregnant, hence reducing the risk for the next generation, and would help the survival of all family members.</p>", "<p>Transdisciplinarity enhances our competency to address and resolve controversial problems. A decision to employ a transdisciplinary approach might have avoided the programmatic conflicts that remain unresolved in the field. Whether the topic for discussion is HIV, gender or the risks posed by lack of breastfeeding, specialized scientific knowledge is always involved. This knowledge must, however, be contextualized so that it becomes part of problem resolution competencies and can be applied in their conflicts of interest and value. A transdisciplinary approach to messaging would have facilitated working together, and would have supported coordination and synergy of efforts, actions and programs that would have met needs across disciplines.</p>", "<title>Creating paradigm shifts towards mutual visions and synergy</title>", "<p>It is not easy to step outside of our individual disciplinary or special interest boxes. Altering perceptions is difficult, but it may be an essential step in creating collaborative approaches. Visual perception differences are a common part of psychology 101: in Figure ##FIG##0##1##, who sees a vase and who sees two people facing each other?</p>", "<p>It is not always easy to visualize the image differently than it was originally perceived; but once you do see the alternate image, it becomes more difficult to return to your original perspective. These visual images ask the viewer for paradigm shift in how a single image is viewed. Is there any question, then, that we all have to struggle to see each other's viewpoint on complex issues? The analogy between viewpoints on such images, and viewpoints from disciplinary perspectives, is this: once one has seen things from a perspective different than one's own, it is harder to go back to one's original unilateral stance.</p>", "<p>Today, I am asking each of you to come along with me to see if we each can make a paradigm shift on the issues in the reproductive continuum – family planning, pregnancy and birthing and breastfeeding. These are issues that are intimately, biologically, gender linked in women's lives, and yet ones that are generally divided up to be addressed by a variety of different professional disciplines. Despite the impact of child spacing on birthing success, of birthing practices on breastfeeding success, and of breastfeeding on child spacing, we are offered family planning services by a gynecologist, birth attendance by an obstetrician or midwife, and baby care by a pediatrician. Having these \"silos\" of care, each with its own paradigm and priorities, may lead to conflicting messages, and hence, may undermine the search for mutuality in goals, and collaboration.</p>", "<p>Initiatives to support breastfeeding, such as the Baby-friendly Hospital Initiative [##UREF##1##2##] or initiatives to support mother-friendly maternity practices, such as Mother/Baby-friendly Initiative, or even mother support, must actively seek synergy for optimal impact [##UREF##2##3##]. In fact, the revised Baby-friendly Hospital Initiative includes standards and goals for birthing practices, for breastfeeding-friendly communities, and guidance for birth spacing, in addition to reconfirming the original Ten Steps to Successful Breastfeeding [##UREF##3##4##]. Breastfeeding policy and programs would benefit from common messaging and protocols for breastfeeding and mutual support, no matter which discipline is involved. However, this would demand an active decision on the part of healthcare workers and policy-makers to end the \"silo\" approach.</p>", "<p>Existing paradigms can exert an even stronger polarization between family planning and breastfeeding interest groups. This divide is especially unfortunate given the inextricable biological relationships between the two practices. Breastfeeding is a proximate determinant of fertility, and allowing breastfeeding practices to deteriorate will increase fertility rates significantly [##REF##14621252##5##]. One would think, therefore, that breastfeeding would always be a factor of great interest to family planning professionals, whether in program development, in research design, or in policy formation, and that family planning would be essential component for action by breastfeeding supporters in order to ensure an adequate duration of breastfeeding, without a mother suffering the nutrition and health taxing of a concurrent pregnancy. However, examples of active collaboration between family planning programs and breastfeeding support programs are, paradoxically, rare. There is, however, one example of active planning for synergy that has occurred: the research and development of the Lactational Amenorrhea Method (LAM) as an introductory method of family planning [##UREF##4##6##]. For women using LAM, breastfeeding is the physiological basis of a family planning method that indicates when another method must be introduced.</p>", "<p>The development of LAM is a personal story. As a young physician working in the Population Office at USAID, having been mentored by Dr. Cicely Williams, and recognized by La Leche League International recognition for my work on breastfeeding, I had been sensitized both to the importance of breastfeeding and to the potential negative impact of estrogens on breastfeeding. One of my first projects was the development of a combined oral contraceptive (COC) program with the Ministry of Health in Morocco, a country where breastfeeding was the norm; USAID at the time advised introduction of COCs at two weeks postpartum, despite the evidence that this might disrupt breastfeeding. Today, the decision might be different; both family planning and breastfeeding researchers have identified the synergies of these practices in terms of health and spacing outcomes. However, at the time, the blinders were on, and the early introduction of COCs was initiated. This illustration of the need for out-of-the-box thinking inspired my interest in the development of a family planning method that would encourage synergy rather than competition between family planning and breastfeeding advocates.</p>", "<p>The development of LAM began when the same Office sponsored a research session to discuss the issue of the timing of contraception postpartum: if introduced too early during breastfeeding it would be duplicative and possibly detrimental, leading to too early cessation of family planning use, but introduction too late could predispose to an unhealthy, short birth interval. We started funding research on this issue, and I soon left USAID to continue research on this issue at Hopkins. By 1988, when Family Health International and Rockefeller sponsored the first consensus meeting on the issue [##REF##8842819##7##], sufficient data were available from several research centers to identify the three criteria that were later codified as a method algorithm at a meeting held by the Institute for Reproductive Health at Georgetown [##UREF##4##6##] (see Figure ##FIG##1##2##).</p>", "<p>Despite its proven efficacy and acceptability [##REF##9262927##8##,##REF##9262928##9##], LAM is not as yet always listed as a method in texts. Nor do many breastfeeding texts address the need for child spacing for optimal health outcomes. However, WHO's \"Family Planning: A Global Handbook for Providers\" includes LAM as a highly efficacious method, breastfeeding support groups often include LAM as another benefit of exclusive breastfeeding [##UREF##5##10##,##UREF##6##11##]; and USAID is renewing its dedication to this method given the increasing need to integrate and synergize family planning and child survival service delivery [##REF##1348806##12##]. We are clearly beginning to see signs that synergy is possible and happening.</p>", "<p>Two additional \"silos\" where collaboration, cooperation and synergy are beginning to emerge in support of breastfeeding are <italic>Safe Motherhood and Newborn Initiative </italic>and <italic>Child Survival </italic>strategies. Optimal birth intervals help to protect positive birth outcomes for mothers and babies, and improve the nutritional status and odds of survival in the short and long terms [##UREF##7##13##]. And clearly, whatever the child survival strategy employed, whether Accelerated Child Survival and Development, Integrated Management of Childhood Illnesses, or Expanded Programme on Immunization, child survival will benefit from synergy with the number one life-saver as per the Lancet Series on Child Survival: exclusive breastfeeding support [##REF##12853204##14##].</p>", "<title>Policy areas upon which to build synergy and action</title>", "<p>There is no shortage of \"jumping off points\" for a coalition of individuals and organizations committed to healthier, happier mothers and babies. However, four policy \"pillars\" have been defined as a solid base for sustainable change. These pillars include: national/state government commitment, legislation and policy, health worker training and health system support, and family and community support (see Figure ##FIG##2##3##). As we examine a way forward in generating action ideas to build synergy, it may be useful to keep these in mind. Examples of policy actions in support of breastfeeding that demand coalitions of support include:</p>", "<p>• National/governmental Commitment: This may be supported using women's and children's rights as arguments for change;</p>", "<p>• Legislation/policy for maternity protection and paid leave, health insurance coverage, freedom to breastfeed as children need, and protection against aggressive advertising of infant formula. Coalitions including labor unions and business coalitions, such as the National Business Group for Health, could partner with breastfeeding and feminist groups to achieve these policy goals;</p>", "<p>• Health training and services improvement necessitate cooperation and partnership among State Health Departments, health professional associations, accrediting organizations, and academic faculties to ensure that preventive medicine, breastfeeding, and attention to women's equity are included in undergraduate training for all health workers;</p>", "<p>• Policy in support of family/community must include attention to social support for birth spacing and motherhood, as well as the sharing of social marketing and advocacy across sectors. Such policy dictating action would include building with existing socially-oriented NGOs, no matter what their primary social goal is.</p>", "<p>Here in the US, in North Carolina, eight \"Recommended Breastfeeding Action Areas,\" that address the four pillars have already been established in the NC Blueprint for the Protection, Promotion and Support of Breastfeeding [##UREF##8##15##]:</p>", "<p>1. Encouraging the adoption of activities that create breastfeeding-friendly communities;</p>", "<p>2. Creating a breastfeeding-friendly health care system;</p>", "<p>3. Encouraging the adoption of breastfeeding-friendly workplaces;</p>", "<p>4. Assisting childcare facilities in promoting, protecting, and supporting breastfeeding;</p>", "<p>5. Advocating for insurance coverage by all third-party payers for breastfeeding care, services, and equipment when necessary;</p>", "<p>6. Involving media and using social marketing and public education to promote breastfeeding;</p>", "<p>7. Promoting and enforcing new and existing laws, policies and regulations that support and protect breastfeeding; and,</p>", "<p>8. Encouraging research and evaluation on breastfeeding outcomes, trends, quality of care, and best practices.</p>", "<p>Globally there are at least three policies that could serve as a foundation for planning activities that serve as a construct for synergy of breastfeeding and family planning. The first is the Millennium Development Goals for improving maternal and child health, including gender equity and reproductive justice as underlying needs [##UREF##9##16##]. The second is the Partnership for Safe Motherhood and Newborn Health [##UREF##10##17##]. The third policy is the Global Strategy for Infant and Young Child Feeding and the Innocenti Declarations, where advocates could ensure that sufficient attention is paid to family planning, birthing, and breastfeeding [##UREF##11##18##].</p>", "<title>Concepts for planning synergized action steps</title>", "<p>The reason I have emphasized paradigms is that each of us must actively recognize our own paradigms – our entrenched viewpoints – then work to perceive the issue from a new stance. If you will recall the story of my experience in Morocco, the request to synergize family planning and breastfeeding would have necessitated a paradigm shift for all who were working on family planning or child survival. At that time, there was no perceived benefit in seeking synergy between these efforts. Today, with diminished resources for public health, being open to new approaches to creating the much-needed synergy among breastfeeding, family planning and feminism can only serve us well.</p>", "<p>Two additional ways to more easily visualize points of potential synergies are through exploration of the continua, both social and chronological. The socio-ecological model is a visual description of the continua of interactions among the individual and the various strata of society: family, social community, workplace culture, and legislative and other government authorities. This model is useful in identifying interactions among various impacts on health, and in identifying loci for intervention (see Figure ##FIG##3##4##). The construct is conceptually similar to the rights-based model, which recognizes that support must be in place in each stratum in order to actualize the rights of the individual. It is possible to see how various programs may emanate from one stratum, and the logic of creation of synergy rather than conflicting impacts.</p>", "<p>The chronological intergenerational life-cycle approach may also support the development of new integrated and synergized interventions to promote health, rights and justice by accepting that there are sequelae of programs and policies that reach beyond the moment, and are visited upon the future of individuals and into subsequent generations (see Figure ##FIG##4##5##). If we consider this, it is clear that no intervention can allow itself to stand alone, but rather there must be developed and implemented in cooperation among those who have contact at any point to ensure ongoing best and most sustainable outcomes.</p>", "<title>Competing interests</title>", "<p>The author declares that they have no competing interests.</p>" ]

[]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Vase? Or two profiles?</bold>. Our initial perspective may make it difficult to view things from a new perspective, but it is possible.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Lactational amenorrhea method</bold>. This method of family planning evolved from transdisciplinary thinking: How can we address the need to support breastfeeding and child spacing? (figure derived from references in [##REF##1348806##12##]).</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>The four pillars</bold>. Frameworks for Strategic Planning where Breastfeeding and Family Planning and Reproductive Rights could synergize in support of program and policy: The 4 Action Areas (\"pillars\") for Synergy Consideration (figure derived from UNICEF and WHO materials).</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>The socio-ecologic framework</bold>. Comprehensive, multi-level, multi-sectoral protection, promotion and support for breastfeeding. The socio-ecological model can be used to identify points of synergy and intervention; Construct of this model is similar to the rights-based model, both starting with consideration of the child and mother.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p><bold>The intergenerational life-cycle</bold>. The intergenerational life-cycle may help us identify areas where program and policy interventions and be synergized, and rights and justice considered.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1746-4358-3-16-1\"/>", "<graphic xlink:href=\"1746-4358-3-16-2\"/>", "<graphic xlink:href=\"1746-4358-3-16-3\"/>", "<graphic xlink:href=\"1746-4358-3-16-4\"/>", "<graphic xlink:href=\"1746-4358-3-16-5\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Infection of cells by HIV requires a number of host cell factors [##REF##12091904##1##]. Differences in these factors can contribute to variability in infection levels between individuals. Previous studies show that HIV infection levels of cells differ between individuals, with up to a 1000-fold variability in replication for HIV laboratory-adapted strains [##REF##1688473##2##,##REF##1716027##3##] and up to a 40-fold variability for HIV primary isolates [##REF##7983738##4##]. The level of HIV infectivity in cells has been attributed to multiple steps in the HIV replication cycle, including entry and several post-entry events [##REF##15367641##5##]. These studies show that differences in host factors between individuals can impact the efficiency of HIV infection.</p>", "<p>During the HIV replication cycle nascent virions acquire host cell-surface molecules that can retain their biological function. For example, CD54 (ICAM-1) incorporated in the virus membrane has been shown to facilitate attachment to cells independently of CD4 [##REF##9094631##6##]. Additional studies have shown HIV attachment independent of CD4: macrophages can capture HIV through the mannose receptor and syndecans [##REF##12645947##7##,##REF##11533182##8##], dendritic cells through syndecan-3 and a variety of C-type lectin receptors [##REF##18040049##9##, ####REF##10721995##10##, ##REF##12352970##11####12352970##11##], and HeLa cells through cell-surface heparans [##REF##9557643##12##]. Furthermore, we previously showed that the initial attachment of HIV primary isolates to PBMC is independent of CD4 and not attributable to host factors previously suggested to be involved in HIV attachment to cells [##REF##16842879##13##, ####REF##11860673##14##, ##REF##10954556##15####10954556##15##].</p>", "<p>Since previous studies suggested that the initial attachment of HIV to PBMC is attributable to host factors, and that these factors could potentially differ between donor cells, we hypothesized that HIV attachment variability exists between donor cells. Heretofore, it has only been indirectly shown that variability in HIV entry exists between donor cells [##REF##15367641##5##], however, no study has directly examined variability in HIV attachment to donor cells. In this study, we directly tested HIV attachment to donor PBMC and show high levels of variability in HIV binding to donor PBMC that occurs independently of CD4 and virus strain.</p>" ]

[ "<title>Methods</title>", "<title>Cell culture</title>", "<p>PBMC were obtained from healthy donors by Ficoll-Hypaque (Whittaker M.A. Bioproducts, Walkersville, MD) gradient centrifugation of freshly obtained heparanized blood. To activate PBMC, 3 μg/ml phytohemagglutinin (PHA-L; Sigma/Aldrich, St. Louis, MO) was added to cells in RPMI 1640 medium supplemented with 10% heat inactivated fetal bovine serum and 50 μg/ml gentamicin (complete medium). PBMC were cultured for 2 days, washed, and further cultured in complete medium containing 20 U/ml interleukin-2 (obtained through the AIDS Research and Reference Reagent Program [ARRRP], National Institute of Allergy and Infectious Diseases, National Institutes of Health, from Maurice Gately, Hoffman-La Roche, Inc.). All virus strains tested were produced in PHA-stimulated PBMC of healthy donors as previously described [##REF##9657955##19##].</p>", "<title>HIV binding</title>", "<p>PBMC (3 × 10⁶PHA-stimulated or unstimulated) were incubated with 100 μl of virus-culture supernatant, diluted in serum-free RPMI to contain 2000 pg of p24 for 45 min on ice with periodic agitation. After incubation with HIV, 3 ml of ice-cold phosphate buffered saline (PBS) was added to cells. Cells were pelleted by centrifugation and then transferred to a new tube and washed to prevent measurement of HIV binding to tubes. Pelleted cells were lysed with 0.5% Triton X-100, and the amount of virus bound to cells was measured by p24 antigen ELISA (National Institutes of Health AIDS Vaccine Program, Frederick, MD). For experiments with blocking antibody, either anti-CD4 (Clone Sim.4, obtained from American Type Culture Collection, ATCC; IgG₁), an isotype control antibody (Clone G18-145, ATCC; IgG₁), or no antibody was added to tubes at the beginning of the virus/PBMC incubation step.</p>", "<title>HIV infection</title>", "<p>For infection of PBMC, freshly isolated PBMC were cultured in complete medium with 3 μg/ml PHA for 48 h and then washed prior to HIV binding. Cells were incubated with HIV and washed as described above. PBMC were then transferred to 24-well plates (Corning Inc., Corning, NY) and cultured in medium with 20 U/ml IL-2 in a total volume of 1 ml. HIV p24 in culture supernatants was measured by ELISA on days 5 and 7. Cultures were fed on days 5 and 7 by removal of 500 μl and replacement with fresh medium containing IL-2.</p>", "<title>Flow cytometry</title>", "<p>All antibodies (anti-CD4, -CD8, -CD14, -CD20 and isotype control) were FITC-conjugated mouse anti-human IgG (Becton Dickinson, Franklin Lakes, New Jersey). To assess the expression of cell-surface markers, PBMC were incubated with the manufacturer's recommended volume of antibody on ice for 30 min, and then assessed for surface-markers by flow cytometry. For each PBMC donor, the percentage of surface-marker positive cells was plotted against the amount of HIV bound to donor PBMC.</p>" ]

[ "<title>Results</title>", "<p>We assessed the possibility of differences in HIV binding to donor PBMC using the R5 primary isolate HIV_THand freshly isolated PBMC from 19 different healthy donors. HIV binding was variable between donors, with up to a 3.9-fold difference in virus binding between the highest and lowest binding donors (Fig. ##FIG##0##1A##).</p>", "<p>We next hypothesized that the variability in HIV binding would affect infection of PBMC. Donor PBMC from low, medium and high binding donors were treated with phytohemagglutinin (PHA) for 2 days, HIV was bound to PBMC and the cells cultured for up to 7 days. Replication of virus at day 5 post-infection showed a similar trend to virus bound to donor PBMC on day 0 (Fig. ##FIG##0##1B##). A similar trend was observed on day 7 post-infection (data not shown), although higher levels of replication were observed, as anticipated. As a control, virus was bound to PBMC in the presence or absence of PHA stimulation; there was no difference in virus binding between groups (data not shown), showing that PHA does not alter PBMC binding phenotypes.</p>", "<p>While previous studies showed that binding of HIV primary isolates to PBMC was CD4-independent [##REF##10721995##10##, ####REF##12352970##11##, ##REF##9557643##12####9557643##12##], we next sought to confirm this with PBMC donors that bound HIV at differing levels. The CD4-blocking antibody Sim.4 did not significantly affect binding of HIV to PBMC (Fig. ##FIG##0##1C##). regardless of whether the PBMC donor bound relatively high levels of virus (e.g. donor 1) or low levels of virus (e.g. donor 2). To confirm that the Sim.4 antibody could block infection, HIV was bound to PBMC at in the presence of Sim.4, an irrelevant antibody, or without antibody, and then incubated at 37°C to allow infection of PBMC to proceed. HIV infection of PBMC was inhibited with Sim.4, but not the irrelevant antibody (data not shown).</p>", "<p>We next determined whether HIV binding phenotypes of PBMC donors were stable over time. HIV attachment to PBMC from two donors with a high-binding phenotype and two with a low-binding phenotype were examined for virus binding levels once a week over 4 weeks. A stable HIV binding pattern was maintained over 4 weeks for PBMC from both of the donors with a low-binding phenotype and for one of the donors with a high-binding phenotype (Fig. ##FIG##1##2##). However, one donor with a high-binding phenotype changed to a low-binder on the fourth week. This data shows that HIV binding phenotypes are relatively stable over the period of several weeks, but are also able to change.</p>", "<p>Since HIV infection of PBMC is affected by differences in virus strain [##REF##10358771##16##], we also determined whether the relative binding of HIV to high- and low-binding PBMC was affected by the strain of virus. Four different HIV strains – HIV_GP(X4), HIV_NL4-3(X4), HIV_TH(R5), or HIV_SF2(X4) – were incubated with PBMC from one donor with a high-binding phenotype and one with a low-binding phenotype. The high-binding PBMC bound more virus than the low-binding PBMC for all virus strains tested (Fig. ##FIG##2##3##). Specifically, the high-binder PBMC bound 3.1-, 1.7-, 1.5- and 3.9-fold more virus than the low-binder PBMC for HIV_GP, HIV_NL4-3, HIV_TH, and HIV_SF2, respectively. This shows that HIV binding differences of high- and low-binding PBMC donors are preserved across multiple virus strains. Since the HIV binding variability between PBMC donors is not HIV strain-dependent, this suggests a role in cell binding of either host cell-acquired molecules on the HIV membrane, or a highly conserved HIV-encoded envelope protein site.</p>", "<p>PBMC are a heterogeneous population of cells composed primarily of CD4 T cells, CD8 T cells, monocytes, and B cells. To determine whether donor binding variability is caused by higher or lower numbers of specific cell subsets within PBMC, HIV was bound to PBMC from 10 different donors, and flow cytometry was used to assess the percentage of CD4+ T cells (CD4+), CD8+ T cells (CD8+), monocytes (CD14+), and B cells (CD20+), respectively (Fig. ##FIG##3##4##). No correlation was observed between the level of virus binding and the percentage of CD4⁺, CD8⁺, CD14⁺, or CD20⁺cells, suggesting that virus binding variability between PBMC donors is due to differences in HIV attachment factors found on multiple types of cells, as opposed to differences in levels of a particular cell subset.</p>" ]

[ "<title>Discussion</title>", "<p>In this study we found that differences exist in the level of HIV binding to donor PBMC, that binding phenotypes of the donors are stable over four weeks, and that binding phenotypes are consistent across several HIV strains. Importantly, the variability in virus binding to cells correlated with the levels of HIV infection of cells. Ciuffi <italic>et al</italic>. reported that infection of cells from different donors was highly variable, and that 42% of the infection variance between cell donors was at the level of HIV entry [##REF##15367641##5##]. The expression of CCR5 on CD4 T cells correlated with HIV infection levels of these cells [##REF##15367641##5##], but much of this correlation was due to donors that were heterozygous or homozygous for the CCR5Δ32 allele. However, when donors that did not have the mutant allele were excluded from analysis, there remained a greater-than 10-fold difference in infection between individuals and about 50% of this was estimated to be due to virus binding and entry events, suggesting that cell-surface molecules can have large effects on virus infection of cells. While previous studies indicate the initial attachment of HIV to cells occurs independently of CD4 [##REF##10721995##10##, ####REF##12352970##11##, ##REF##9557643##12####9557643##12##] and prior to gp120 interaction with CCR5 [##REF##12873764##17##], we confirm in this study that the variability of HIV binding to donor PBMC is unaffected by the presence of a blocking antibody to CD4, indicating that the differences observed in the initial attachment of HIV to donor PBMC occurs by cell-surface molecules other than CD4.</p>", "<p>Although the mechanism of HIV binding variability to donor PBMC has not been identified, studies performed in our laboratory show that the initial attachment of HIV to PBMC occurs in a Ca²⁺-dependent manner [##REF##16842879##13##], suggesting a potential role for a C-type lectin or other Ca²⁺-dependent molecules in HIV binding to PBMC. Fusion of HIV mediated by HIV gp120/gp41 to permissive cells does not occur until 10 minutes after virus is added to cells at 37°C [##REF##7853478##18##] and does not take place at 20°C, while binding of HIV can occur at 4°C or higher temperatures [##REF##11860673##14##]. Therefore, the binding we observe is likely prior to fusion, and antibody inhibition experiments show it is independent of CD4. Donor binding variability has important implications for the HIV replication cycle, since infection of cells by HIV is greatly impacted by the amount of bound virus during either direct infection of cells or during transfer of virus between cells [##REF##10954556##15##].</p>" ]

[ "<title>Conclusion</title>", "<p>By showing that high and low HIV-binding phenotypes exist, these studies highlight the impact of host-acquired molecules on the initial steps of HIV infection. This study therefore provides added insight into understanding the mechanism of HIV tethering to cells, which could potentially lead to the development of drug strategies that inhibit entry of the virus into cells.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>HIV infection of cells varies greatly between individuals, with multiple steps in the replication cycle potentially contributing to the variability. Although entry and post-entry variability of HIV infection levels in cells has been demonstrated, variability in HIV binding has not been examined. In this study, we examined variability of HIV binding to peripheral blood mononuclear cells (PBMC) from different donors.</p>", "<title>Results</title>", "<p>HIV binding to PBMC varied up to 3.9-fold between individuals and was independent of CD4. Replication of HIV in donor PBMC required CD4 and paralleled virus binding trends of donor PBMC. To assess the stability of virus binding phenotypes over time, HIV was bound to donors with low- and high-binding phenotypes. The binding phenotypes were maintained when tested weekly over a 4-week period for 3 of 4 donors, while one high-binding donor decreased to lower binding on the 4th week. The low- and high-binding phenotypes were also preserved across different HIV strains. Experiments performed to determine if there was an association between HIV binding levels and specific cell subset levels within PBMC showed no correlation, suggesting that HIV binds to multiple cell subsets.</p>", "<title>Conclusion</title>", "<p>These results show that differences exist in HIV binding to donor PBMC. Our data also show that HIV binding to donor PBMC is CD4-independent and can change over time, suggesting that virus binding variability is due to differences in the expression of changeable cell-surface host factors. Taken together, this study highlights the impact of cell-surface factors in HIV binding to, and infection of, PBMC which likely represents an important step in HIV infection <italic>in vivo</italic>.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>JJA and GGO performed all of the experiments. JJA participated in writing of the manuscript. GTS provided overall direction and co-wrote the manuscript. All authors read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This work was supported by National Institutes of Health grant P01HD40539.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>HIV binding to PBMC is variable between donors</bold>. <bold>(A) </bold>Freshly isolated PBMC (3 × 10⁶) were incubated with HIV_TH(2000 pg of p24) in serum-free RPMI-1640 medium for 45 minutes on ice. After incubation with HIV, cells were washed to remove unbound virus. Cells were pelleted, lysed and HIV p24 was measured by ELISA. Shown are values from 19 different healthy PBMC donors. <bold>(B) </bold>HIV_THwas bound to freshly isolated or PHA-stimulated donor PBMC. For virus binding to freshly isolated PBMC, HIV p24 was measured by ELISA as described above. HIV replication of donor PBMC was assessed by p24 ELISA after 5 days of culture. <bold>(C) </bold>HIV_THwas bound to freshly isolated PBMC either in the presence of anti-CD4, isotype control antibody, or in the absence of antibody. Cell-bound virus was measured as described above. Results are the means ± SD of triplicate determinations.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>HIV attachment phenotypes of PBMC donors are stable over time</bold>. Freshly isolated PBMC from two donors with high-binding phenotypes and two with low-binding phenotypes were incubated with 2000 pg of p24 (HIV_TH) in serum-free medium 45 minutes on ice and bound virus assessed as before. Donors that were not available are indicated by ≠.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>HIV binding phenotypes of PBMC donors are maintained across multiple virus strains</bold>. Freshly isolated PBMC were incubated with 2000 pg of p24 (HIV_GP, HIV_NL4-3, HIV_TH, or HIV_SF2) in serum-free medium for 45 minutes on ice. After incubation with HIV, cells were washed to remove unattached virus. Cells and bound virus were lysed and HIV p24 was measured by ELISA. Results are the means ± SD of triplicate determinations.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Donor binding variability does not correlate with a specific cell subset within PBMC</bold>. Donor PBMC were stained with anti-CD4-FITC, anti-CD8-FITC, anti-CD14-FITC, or anti-CD20-FITC and the percentage of each subset measured by flow cytometry. PBMC from donors were also incubated with 2000 pg of p24 (HIV_TH), and cell-bound virus was measured by ELISA as before. HIV binding variability is plotted on the y-axis and percentage marker positive cell subsets is plotted on the x-axis. r²represents the goodness of fit value. Data represent means ± SD of triplicate determinations.</p></caption></fig>" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1743-422X-5-95-1\"/>", "<graphic xlink:href=\"1743-422X-5-95-2\"/>", "<graphic xlink:href=\"1743-422X-5-95-3\"/>", "<graphic xlink:href=\"1743-422X-5-95-4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Smallpox epidemiology has the longest and richest history in investigating the mechanisms of spread and in evaluating the effectiveness of vaccination [##UREF##0##1##,##UREF##1##2##]. Modern epidemiological methods have developed in parallel with smallpox control practice, and consequently, the disease had already been eradicated before statistical and epidemiological techniques for analyzing infectious disease outbreaks had sufficiently matured.</p>", "<p>Although the world is free from smallpox, researchers continue revisiting smallpox epidemiology and virology with recent techniques. In the aftermath of the 9-11-2001 attack, the awareness of the threat of bioterrorism has grown significantly [##REF##10367824##3##]. Mathematical models and computer simulations have been developed to design and optimize public health measures against re-introduced variola virus, the causative agent of smallpox [##REF##15922002##4##, ####REF##17085740##5##, ##REF##12851224##6##, ##REF##14562094##7##, ##REF##18284358##8##, ##REF##15200816##9##, ##REF##14979585##10##, ##REF##16899385##11##, ##REF##11747722##12##, ##REF##15162977##13##, ##REF##16894173##14##, ##REF##15764191##15##, ##REF##12118122##16##, ##REF##12434061##17####12434061##17##]. These models are based on different epidemiological assumptions of smallpox. For example, assumptions about the number of secondary transmissions before onset of illness had not been not carefully validated in earlier mathematical modelling studies [##REF##12118122##16##,##REF##12434061##17##]. Accordingly, the policy implications of these models differed widely, and thus the necessity arose to capture the basic mechanisms of smallpox transmission precisely [##REF##12851224##6##,##REF##12851223##18##]. To date, it has been demonstrated that transmission dynamics and intervention strategies cannot be modelled without sufficiently quantifying the detailed intrinsic mechanisms by using observed data [##REF##12851224##6##,##REF##16894142##19##,##REF##15071187##20##]. Because of the global eradication, we have had to maximize the use of historical data to estimate nearly all biological and epidemiological parameters that are needed to optimize interventions [##UREF##2##21##].</p>", "<p>This review article has two purposes. The first purpose is to summarize the issues that have been clarified in recent mathematical and statistical studies and to discuss the relevant policy implications. The second purpose is to specify what important aspects of smallpox epidemiology remain unknown and to suggest how these could be addressed by analyzing historical records. In the following section, we first give a technical overview of the use of historical data and then present some examples of quantification. Subsequently, we summarize the basic concept and interpretation of the transmission potential and the resulting implications for vaccination strategies. The paper concludes with a summary of the findings, emphasizing the importance of systematically analyzing historical datasets.</p>" ]

[]

[]

[]

[ "<title>Conclusion</title>", "<p>This article has reviewed quantifications of the transmission and spread of smallpox using historical data. Although historical data are limited and we cannot answer all questions regarding smallpox epidemiology, many publications are available from previous efforts. However, a systematic listing of surveillance data and/or outbreak reports irrespective of language (e.g. see [##REF##16836830##57##]) still remains a future task. It is essential that historians, smallpox specialists and epidemiologists interact more.</p>", "<p>Since the eradication, smallpox deaths have disappeared from the world [##REF##6252467##119##], and hope has arisen that we will succeed in eradicating other infectious diseases. Owing to the conceived threat of bioterrorism, researchers nevertheless have to continue working on smallpox, and we have entered yet another round of discussing the pros and cons of smallpox vaccination. The current debates of preparedness issues are far more complex than mass vaccination, and newer vaccination strategies complicate the balance between individual and community benefits [##REF##12920181##120##]. Once other infectious diseases have been eradicated, we will see similar discussions arise, but before this becomes the case, it is important to make sure that systematically collected data are aggregated and stored for posterity.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Quantification of the transmission dynamics of smallpox is crucial for optimizing intervention strategies in the event of a bioterrorist attack. This article reviews basic methods and findings in mathematical and statistical studies of smallpox which estimate key transmission parameters from historical data.</p>", "<title>Main findings</title>", "<p>First, critically important aspects in extracting key information from historical data are briefly summarized. We mention different sources of heterogeneity and potential pitfalls in utilizing historical records. Second, we discuss how smallpox spreads in the absence of interventions and how the optimal timing of quarantine and isolation measures can be determined. Case studies demonstrate the following. (1) The upper confidence limit of the 99th percentile of the incubation period is 22.2 days, suggesting that quarantine should last 23 days. (2) The highest frequency (61.8%) of secondary transmissions occurs 3–5 days after onset of fever so that infected individuals should be isolated before the appearance of rash. (3) The U-shaped age-specific case fatality implies a vulnerability of infants and elderly among non-immune individuals. Estimates of the transmission potential are subsequently reviewed, followed by an assessment of vaccination effects and of the expected effectiveness of interventions.</p>", "<title>Conclusion</title>", "<p>Current debates on bio-terrorism preparedness indicate that public health decision making must account for the complex interplay and balance between vaccination strategies and other public health measures (e.g. case isolation and contact tracing) taking into account the frequency of adverse events to vaccination. In this review, we summarize what has already been clarified and point out needs to analyze previous smallpox outbreaks systematically.</p>" ]

[ "<title>Review</title>", "<title>Extracting key information from historical data</title>", "<p>Although historical data have frequently been revisited using modern statistical techniques to identify epidemiological determinants of smallpox, many key issues remain unknown in spite of great efforts. To clarify important aspects of smallpox epidemiology, it remains necessary to maximize the use of historical data. To understand their usefulness and to avoid common pitfalls, we briefly discuss technical issues in utilizing historical publications.</p>", "<title>Issues to consider when looking at historical smallpox data</title>", "<p>In the following, we list key points to be remembered whenever we statistically extract information from the historical literature. As we may not be able to find all the answers to the following questions in a single historical data set, we may have to combine different data sets or to merge in information from laboratory experiments.</p>", "<title>(A) Were all cases caused by variola virus?</title>", "<p>As cases could not be confirmed virologically before the middle of the 20th century, it is crucial to know on what observations historical diagnoses were based. It was not uncommon to misdiagnose chickenpox cases as smallpox [##UREF##3##22##,##UREF##4##23##]. In the older literature, it sometimes even remains unclear which kind of \"plague\" was being described [##REF##15081502##24##,##REF##15879045##25##]. Ascertainment of diagnostic methods is one of the biggest challenges in utilizing historical outbreak data.</p>", "<title>(B) Clinical documentations and time-varying medical trends</title>", "<p>Similarly, clinical classifications of smallpox have been revised over time [##UREF##0##1##,##UREF##5##26##, ####UREF##6##27##, ##UREF##7##28####7##28##]. The definition of severe smallpox has varied greatly even in the 20th century. Vaccines have continuously been improved [##UREF##8##29##], and we still do not even know from where the vaccinia virus emerged and when it started to be used as a smallpox vaccine [##REF##198484##30##]. It is necessary to identify and to select the most useful sources of literature, in order to understand which classification in a given publication was adopted and which type of vaccine was most likely to have been used in the population described.</p>", "<title>(C) Pathogenicity and virulence of the variola virus</title>", "<p>Classically, smallpox was classified into two different types according to the observed case fatality. The traditional form of smallpox, referred to as variola major, was believed to have a case fatality of 20% or more. A milder form of the disease, variola minor, with a case fatality of 1% or less was first reported in the late 19th century in South Africa, then it was observed in European countries and finally in Brazil [##UREF##0##1##,##REF##1010665##31##, ####UREF##9##32##, ##REF##13365528##33####13365528##33##]. Variola minor accounted for the majority of cases in the early 20th century in the United States, where it remained the only form of smallpox from the 1930s until its eradication [##UREF##10##34##]. The epidemiology of variola minor and its interplay with variola major have only partly been clarified [##REF##6296226##35##,##REF##3004200##36##]. There are clear genetic differences between variola major and minor, supporting the taxonomic distinction; recently, the virulence of variola virus has also been attributed to detailed genomic information [##REF##16873609##37##,##REF##17901212##38##]. However, if case fatality was a major criterion in determining the virulence of variola virus, the outcome of these laboratory studies may have been distorted by the vaccination history of cases and maybe also by other factors. Epidemiological clarification of this point still remains an open question.</p>", "<title>(D) Definition of the reported events</title>", "<p>When extracting information on the incubation and infectious periods (or similar parameters describing the epidemiological characteristics), it is crucial to know how the time of infection (which cannot be observed directly) and the onset of disease were defined. There were two traditional ways to define the onset of smallpox: onset of fever or appearance of rash. If the period from onset to recovery is documented, it is important furthermore to identify what \"recovery\" stands for (e.g. recovery from pyrexia or solidification/disappearance of rash).</p>", "<title>Extracting data from historical publications</title>", "<p>The foregoing list does not cover all common pitfalls. Tackling historical data requires not only statistical techniques, but also understanding of the social history and the background of the cases. Moreover, as noted above, we often have to draw conclusions with implications for public health decision-making using combined data from different sources. Identifying the most useful and important data and addressing key questions are major parts of an essential process to shed light on the mechanisms of transmission and spread of smallpox. In the next two sections, we review studies on parameter estimation that use historical records and predominantly originate from our previous studies. The following case studies were conducted, carefully accounting for common technical problems as listed above when looking at the historical data.</p>", "<title>Intrinsic transmission process of smallpox</title>", "<p>To understand the spread of smallpox, it is essential to know the intrinsic transmission process, i.e. after what time symptoms appear, when secondary transmission occurs, and how severe the disease will be. Although basic, such knowledge of the intrinsic transmission process already allows us to assess whether public health interventions in the event of a bioterrorist attack can contain smallpox by means of mass vaccination or by a combination of contact tracing, quarantine and isolation [##REF##12851224##6##,##REF##16894142##19##]. As practical examples, here we briefly discuss basic methodologies and recent findings concerning the incubation period, the infectious period and the case fatality.</p>", "<title>Incubation period of smallpox</title>", "<p>The incubation period is defined as the time from infection to onset of disease [##REF##17466070##39##]. Usually, symptoms of smallpox appear 10–14 days after infection [##REF##5007552##40##]. The knowledge of the incubation period distribution enables us to determine the appropriate length of quarantine [##REF##16237660##41##]. 'Quarantine' here refers to physical separation of healthy individuals who were exposed to cases. In the practice of outbreak investigations, the time of exposure is sometimes determined by contact tracing. Historically, the suggested length of smallpox quarantine tended to be 14–16 days, based on professional experience and an accumulation of epidemiological data, but not on an explicit statistical analysis of the incubation period distribution.</p>", "<p>Restricting the movement of exposed individuals for longer than the maximum incubation period ensures the effectiveness of quarantine measures. Unfortunately, the length of the incubation period requires knowing the exact time of infection, and thus can only be determined for cases who were exposed for a very short period of time. In addition, the maximum observed incubation period clearly depends on the sample size: the larger the sample size, the more likely we are to find cases whose incubation periods exceed the previously known maximum. The number of smallpox cases with well-known incubation period (e.g. documented patients who had been exposed for a single day only) is limited in historical records. The problem of stating a maximum incubation period can be circumvented by fitting a statistical distribution to the data. This distribution allows a time point to be determined beyond which the onset of further cases becomes extremely unlikely (e.g. the time after infection until which 99% of the patients develop symptoms). If the incubation period follows a lognormal distribution with mean, <italic>μ</italic>, and standard deviation, <italic>σ </italic>(of the variable's logarithm), the probability density of observing an incubation period of length <italic>t</italic>_iis given by</p>", "<p></p>", "<p>We can estimate the parameters <italic>μ </italic>and <italic>σ </italic>from a dataset of <italic>n </italic>known incubation times <italic>t</italic>_iby maximizing the likelihood function</p>", "<p></p>", "<p>Figure ##FIG##0##1## shows the frequency distribution of the incubation period of smallpox, which was estimated from 131 cases of smallpox who were exposed only for a single day [##UREF##11##42##]. The mean incubation period is 12.5 days (SD 2.2 days). The 99th percentile is 18.6 days with a 95% confidence interval (CI) ranging from 16.8 to 22.2 days. This indicates that a quarantine of 23 days ensures that more than 99% of infected individuals will develop symptoms before being released.</p>", "<title>Infectious period of smallpox</title>", "<p>The infectious period has traditionally been defined as the period in which pathogens are discharged [##UREF##12##43##]. It presently refers to the period in which infected individuals are capable of generating secondary cases. Knowledge of the infectious period allows us to determine for how long known cases need to be isolated and what should be the latest time point after exposure at which newly infected individuals should be in isolation.</p>", "<p>One approach to addressing this issue is to quantify how the pathogen load changes over time using the most sensitive microbiological techniques (e.g. polymerase chain reaction), but such observations are usually limited to the period after onset of symptoms. Several attempts have been made to measure the distribution of the virus-positive period of smallpox cases [##REF##13887636##44##,##REF##4359678##45##], but sample sizes were small and only very few samples could be obtained during the early stage of illness. Moreover, linking \"virus-positive\" results to the probability of causing secondary transmission is difficult without further information (e.g. frequency, mode and degree of contact).</p>", "<p>Another way of addressing this complicated issue is to determine the frequency of secondary transmission relative to disease-age, i.e. the time since onset of fever [##REF##17156499##46##]. An estimate of the relative infectiousness is obtained by analyzing historical data in which it is known who acquired infection from whom. The known transmission network permits serial intervals to be extracted, i.e. the times from symptom onset in a primary case to symptom onset in a secondary case [##REF##14630599##47##, ####REF##16684896##48##, ##REF##16790838##49####16790838##49##]. Given the length of the serial interval <italic>s </italic>and the corresponding length of the incubation period <italic>f</italic>, the disease-age <italic>l </italic>from onset of symptoms to secondary transmission satisfies</p>", "<p></p>", "<p>Considering the statistical distributions for each length results in a convolution equation:</p>", "<p></p>", "<p>The frequency <italic>l</italic>(<italic>t</italic>-<italic>τ</italic>) of secondary transmission relative to disease-age can be back-calculated by extracting the serial interval distribution <italic>s</italic>(<italic>t</italic>) from a known transmission network, and by using the incubation period distribution <italic>f</italic>(<italic>τ</italic>) given above. If we have information on the length <italic>t</italic>_iof the serial interval for <italic>n </italic>cases, the likelihood function is given by</p>", "<p></p>", "<p>The parameters that describe the frequency of secondary transmission relative to disease-age can be estimated by maximizing this function. A similar method has been applied to estimate the number of HIV-infections from AIDS incidence [##UREF##13##50##].</p>", "<p>Figure ##FIG##1##2## shows the back-calculated infectiousness of smallpox relative to disease-age [##REF##17156499##46##]. In the following text, day 0 represents the onset of fever. Before onset of fever (i.e. between day -5 and day -1) altogether only 2.7% of all transmissions occurred. Between day 0 and day 2 (i.e. in the prodromal period before the onset of rash) a total of 21.1% of all transmissions occurred. The daily frequency of passing on the infection was highest between day 3 and day 5, yielding a total of 61.8% of all transmissions. These estimates help determine the latest time by which cases should be in isolation. If each primary case infects on average six individuals, and if the effectiveness of isolation is 100%, the isolation of a primary case before the onset of rash reduces the expected number of victims to 6 × (0.027 + 0.211) = 1.428. Optimal isolation could, therefore, substantially reduce the number of secondary cases, and the outbreak could quickly be brought under control by additional countermeasures (e.g. contact tracing [##REF##12851224##6##,##REF##12496354##51##]).</p>", "<title>Case fatality</title>", "<p>Case fatality is the proportion of deaths among those developing the disease. It is particularly important to understand the case fatality of smallpox in order to estimate the magnitude of the disaster in the event of a bioterrorist attack. It may also be of practical importance to predict the burden of hospital admissions and burials in such an event. The case fatality of smallpox was systematically reviewed during the Eradication Programme [##UREF##14##52##], showing extremely high crude estimates of 26% and 36% in East-Pakistan and Madras, respectively, but suggesting a wide geographical heterogeneity. Recent studies attributed part of this heterogeneity to viral genomic differences [##REF##16873609##37##,##REF##17901212##38##], but many of the previously published mathematical models simply assumed an overall estimate of 30% for unvaccinated cases.</p>", "<p>Various factors influence case fatality, most importantly previous vaccination history (which will be discussed in the Section on public health interventions) and the age at infection. Following a previous study by Dietz and Heesterbeek [##REF##12387913##53##], we assume the following parametric model for the age-specific case fatality of smallpox:</p>", "<p></p>", "<p>where <italic>α</italic>, <italic>β</italic>, <italic>γ </italic>and <italic>δ </italic>are parameters that need to be estimated. If we have a dataset with <italic>M</italic>_ideaths and <italic>N</italic>_isurvivors of age <italic>a</italic>_i, the likelihood function is</p>", "<p></p>", "<p>where <italic>a</italic>_iis a mid-point of age group <italic>i</italic>. Figure ##FIG##2##3## shows age-specific case fatality estimates of unvaccinated cases in Verona, Italy, from 1810–38 and Sheffield, UK, from 1887–88, respectively [##UREF##15##54##,##UREF##16##55##]. The age-specific case fatality of smallpox can be depicted as a U-shaped curve that peaks in infancy and in old age. Smallpox case fatality also depends on other biological factors of the host such as pregnancy [##REF##13973041##56##], which increases the case fatality from 12.7% (estimate for non-pregnant healthy adults; 95% CI: 11.2–14.3) to 34.3% (95% CI: 31.4–37.1) [##REF##16836830##57##]. Underlying diseases (e.g. cancer, diabetes mellitus, HIV infection and medical immunosuppression for transplantation) could further increase the case fatality.</p>", "<p>Above, we have presented the three most important components of the intrinsic transmission process. Each of them plays a key role in determining the optimal intervention strategy. We showed some basic applications of utilizing likelihood functions [##UREF##17##58##], but various other statistical approaches have been taken which were motivated by similar epidemiological interests. These include the applications of non-linear models [##UREF##18##59##] and of Bayesian techniques [##UREF##19##60##].</p>", "<title>Transmission potential</title>", "<p>In addition to the epidemiological parameters that characterize the natural history of smallpox, we have to know the most important summary measure of transmission, the basic reproduction number, <italic>R</italic>₀, in order to design and optimize public health interventions. <italic>R</italic>₀is defined as the average number of secondary cases arising from a single index case in a fully susceptible population in the absence of interventions [##UREF##20##61##,##REF##8261248##62##]. Here, we discuss the concept of <italic>R</italic>₀and its estimation, starting with its historical development. Then we use the basic reproduction number to assess the eradication threshold of smallpox by means of mass vaccination.</p>", "<title>R₀and vaccination</title>", "<p>Smallpox is the disease with the longest history in theoretical modelling. During the 18th century, the famous mathematician Daniel Bernoulli modelled the spread of smallpox and assessed the effectiveness of the variolation practice (variolation was the precursor of vaccination, consisting of the inoculation of variola virus) [##REF##12387913##53##,##REF##15334536##63##]. Moreover, the earliest formulation and calculation of <italic>R</italic>₀may have been stimulated by smallpox [##REF##16875892##64##]. The earliest concept of <italic>R</italic>₀and the relevant insights into the effectiveness of smallpox vaccination are revisited in the following.</p>", "<p>Figure ##FIG##3##4a## shows the result of the simple mathematical model developed by Theophil Lotz in the late 19th century [##REF##16875892##64##]. If a single primary case generates on average <italic>R</italic>₀= 2 secondary cases, and if we ignore for the sake of simplicity the depletion of susceptible individuals, the number of cases grows geometrically. If there are <italic>a </italic>index cases in generation 0, the expected numbers of cases in generations 1, 2, 3, ..., <italic>n </italic>will be</p>", "<p></p>", "<p>respectively. Following Lotz's example of <italic>R</italic>₀= 2, and assuming a single index case (<italic>a </italic>= 1), we expect 2, 4, 8, ..., 2ⁿcases in the subsequent generations. Although the model ignores variations in the number of secondary transmissions (which are deemed particularly important for directly transmitted diseases [##REF##16292310##65##]), the process described reasonably captures the essential dynamics of transmission during the early stages of an epidemic.</p>", "<p>We now move on to describe various attempts to estimate <italic>R</italic>₀, summarized in Table ##TAB##0##1## together with the underlying key assumptions. Whereas an analysis of long-term temporal dynamics using a mathematical model suggested widely varying estimates of <italic>R</italic>₀, ranging from 3.5 to 6.0 [##REF##11742399##66##], stochastic models assuming a homogeneous pattern of spread, but ignoring the pre-existing immunity level in the afflicted population, grossly underestimated <italic>R</italic>₀as slightly above unity [##UREF##21##67##]. A revised estimate by a model that accounts for the detailed intrinsic dynamics of smallpox in an initially partially immune population suggested that <italic>R</italic>₀is in the order of 6.9 (95% CI: 4.5, 10.1) [##REF##12851223##18##]. This roughly corresponds to an <italic>R</italic>₀for which 80–90% of vaccination coverage would allow sufficient herd immunity to be achieved [##REF##8174658##68##,##UREF##22##69##] (i.e., population-based protection of unvaccinated individuals due to the presence of vaccinated individuals), similarly to Bernoulli's early model, which yielded an estimate of the force of infection that can be translated to <italic>R</italic>₀= 6.7 [##REF##12387913##53##].</p>", "<p><italic>R</italic>₀plays a key role in determining the critical vaccination coverage in a randomly mixing population [##REF##7063839##70##]. If <italic>v </italic>= 80% of individuals are protected by vaccination, the average number of secondary cases is reduced to 20%. Following the model of Lotz, the number of cases in each generation is</p>", "<p></p>", "<p>Figure ##FIG##3##4b## shows the growth of cases when <italic>v </italic>= 50% are protected by vaccination: only a single case is expected in each generation (for <italic>R</italic>₀= 2). The number of cases decreases from one generation to the next if (1-<italic>v</italic>)<italic>R</italic>₀is less than 1 (cf. equation (9)). In line with this, we can formulate the most fundamental condition of immunization to achieve a sufficiently high herd immunity level. In order to eradicate an infectious disease by vaccination, the fraction protected by vaccination must satisfy [##REF##5321089##71##]</p>", "<p></p>", "<p>If <italic>R</italic>₀is 6 for smallpox, more than 1-1/6 = 83.3% of susceptible individuals need to be successfully immunized to prevent an epidemic by vaccination alone. Although the pattern of smallpox spread is most probably non-random, equation (10) can be used as an approximation to guide policymaking. If all individuals are vaccinated, <italic>v </italic>can be interpreted as the direct effectiveness of vaccination [##REF##2066239##72##,##REF##1899778##73##]. The effectiveness of smallpox vaccination remained controversial during the early 20th century, partly because of a lack of reliable estimation methods [##UREF##1##2##], but the methodologies have greatly improved since then [##UREF##23##74##, ####REF##3066628##75##, ##REF##3264824##76##, ##REF##12900143##77##, ##REF##1415152##78####1415152##78##]. During the late 19th century, when vaccine quality was not always ensured, the crude overall effectiveness of vaccination seems to have been higher than 85% [##UREF##0##1##,##UREF##2##21##].</p>", "<title>Heterogeneity and behavioural change in relation to historical data</title>", "<p>To predict the course and size of an epidemic appropriately, it is critically important to clarify the heterogeneity of transmission. The above-mentioned critical coverage for eradication assumes a randomly mixing population, but it has been established that smallpox spreads for example more easily within households than in the community [##REF##5732451##79##, ####REF##762399##80##, ##REF##5110548##81##, ##REF##4376976##82####4376976##82##]. A theoretical approach to modelling household and community transmission separately has been described [##REF##16894173##14##,##REF##7795319##83##], and a tool that allows the two different levels of transmission to be estimated has been developed [##UREF##24##84##], but it is very difficult to obtain the necessary estimates from the limited information given in historical records (e.g. detailed household transmission data are always distorted by vaccination). Age-related heterogeneity is yet another important determinant of smallpox epidemiology [##UREF##25##85##], and spatial patterns of transmission can also influence the success of interventions [##UREF##26##86##]. Unfortunately, historical records, especially those recorded during the Intensified Smallpox Eradication Programme, are considerably biased (e.g. by individual vaccination histories), and thus it is difficult to address age-related and spatial heterogeneities.</p>", "<p>Behavioural changes during an outbreak also have to be clarified to model a bioterrorist attack realistically. It has been suggested that the frequency of contact decreases after the information on an ongoing epidemic is widely disseminated [##REF##12990130##87##, ####REF##15893110##88##, ##REF##17547753##89####17547753##89##]. A mathematical model that attempted to incorporate such a declining contact frequency during an epidemic suggested that even gradual and moderate behavioural changes could drastically slow the epidemic [##REF##15913667##90##]. Methods incorporating such changes remain yet to be developed to help public health policy making. A generalized method could perhaps incorporate results of a psychological response survey [##UREF##27##91##].</p>", "<title>Public health interventions</title>", "<p>Given the basic parameters that describe the intrinsic transmission process, we are now able to examine the effectiveness of interventions. In addition to the critical level of mass vaccination that was discussed in the previous section, here we discuss further issues on vaccination strategies and other public health interventions in bioterrorism preparedness.</p>", "<title>Duration of vaccine-induced immunity and partial protection</title>", "<p>The degree of protection of vaccinated individuals in the present population is yet another important public health issue. Immunological studies showed that a fraction of previously vaccinated individuals still reacts to exposure with variola virus [##REF##12925846##92##,##REF##17560685##93##], but it is difficult to attribute each kind of immunological response to actual protection against the disease and its severity. Thus, the degree of protection of individuals who were vaccinated 30 to 50 years ago has remained an open question.</p>", "<p>As we have previously shown, an epidemiological model that partly addressed the effect of booster events estimated that primary vaccination protected for a median duration of 11.7 to 28.4 years against the disease [##REF##16804475##94##], indicating that most vaccinated individuals in the present community may no longer be protected from contracting smallpox. However, similar estimates also indicate that vaccinated individuals are still protected against severe manifestations and death from smallpox [##REF##14561660##95##]. An analysis of a statistical record of an outbreak in Liverpool from 1902–3 revealed a median duration of protection against smallpox death of 49.2 years (95% CI: 42.0–57.3) [##REF##14561660##95##,##REF##11691969##96##]. This finding (of long-lasting partial protection from severe manifestations) was further supported by statistical analyses of similar historical datasets [##REF##16804475##94##] and of individual case records extracted from historical outbreaks in Australia where booster events were extremely rare [##REF##17033746##97##]. In the event of a bioterrorist attack in the early 21st century, residual immunity could significantly decrease the individual burden of disease. However, the persistence of partial protection does not necessarily imply a positive impact on the population level. Masked symptoms may cause difficulties in case recognition and clinical diagnosis. Although it might be virologically plausible that previously vaccinated cases are less infectious (e.g. due to low levels of virus in their nasopharynx), reduced severity may also permit movements of infectious individuals, worsening the prospects of public health control. The ripple benefit of residual immunity has yet to be clarified.</p>", "<p>To understand the complex interplay of all partial effects of vaccination, various biological and social effects must be considered. In theory, vaccination does not only diminish the susceptibility of vaccinated individuals, but also reduces the degree and duration of infectiousness upon infection. The vaccine-induced reduction of infectiousness can be estimated using the household secondary attack rate (SAR), expressed as the ratio of the number of infected household contacts to the number of exposed household contacts [##UREF##28##98##]. Suppose that SAR_ijdenotes the household secondary attack rate where <italic>i </italic>and <italic>j</italic>, respectively, give the previous vaccination histories of the secondary and primary case (i.e. <italic>i </italic>or <italic>j </italic>= 1 represents previously vaccinated, whereas <italic>i </italic>or <italic>j </italic>= 0 represents unvaccinated individuals). Let us consider the following household transmission data, which were observed in India [##REF##5732451##79##]:</p>", "<p>The household SAR caused by unvaccinated cases among unvaccinated and vaccinated contacts were estimated to be SAR₀₀= 40/650 = 0.0615 and SAR₁₀= 11/583 = 0.0189, respectively. Those caused by vaccinated cases among unvaccinated and vaccinated household contacts were SAR₀₁= 10/499 = 0.0200 and SAR₁₁= 2/421 = 0.0048, respectively.</p>", "<p>The crude efficacy of vaccine in reducing susceptibility VE_S, infectiousness VE_I, and a combined effect of both VE_Tis then estimated by</p>", "<p></p>", "<p>If we make the simplifying assumption that the biological effect of vaccination was identical for all vaccinated individuals, vaccination reduced susceptibility by 69.3%, infectiousness by 67.4%, and the combined effect was 92.3%. Although an effect of vaccination on the duration of disease has rarely been observed and reported, historical epidemiological studies in Dalian, China, suggested that the mean symptomatic period was reduced by 13.7–48.5% if the case was previously vaccinated [##UREF##2##21##,##UREF##29##99##].</p>", "<title>Vaccination strategies</title>", "<p>Given that the intrinsic dynamics as well as the effects of vaccination are sufficiently quantified, vaccination strategies against smallpox can be optimized. Three issues, of which the epidemiology has been discussed though the quantitative effect has not yet been fully clarified, are discussed in the following: revaccination, ring vaccination, and post-exposure vaccination.</p>", "<p>After it became clear in the late 19th century that vaccine efficacy was not perfect and that vaccine-induced immunity waned over time, revaccination was put into practice. Revaccinated individuals were said to have contracted smallpox less often and had much milder manifestations, so that scheduled revaccinations became accepted in the early to mid 20th century [##UREF##5##26##], but the intervals from primary vaccination to revaccinations and the number of revaccinations varied widely within and between countries, making analytic evaluations very difficult. Accordingly, it is extremely difficult to quantify the effectiveness of revaccination in reducing the chance of smallpox even with statistical techniques in the present day. Crude estimates of the increased protection against smallpox death were obtained for several outbreaks; e.g. for Madras during the 1960s [##UREF##6##27##], where 87.1% fewer cases died in the revaccinated group than in the group who had only received the primary vaccination (770/3266 vs. 4/132 deaths, respectively), but this revaccination effect only measures what happened to people who were infected in spite of vaccination. (What makes an explicit interpretation of these findings even more difficult was the fact that vaccination in India was made using the rotary lancet, which left a scar even in the absence of \"take\".)</p>", "<p>Vaccination can be combined with the practice of case finding: Ring vaccination is a surveillance containment measure that consists of vaccinating and monitoring all susceptible individuals in a prescribed area around one or several index cases [##REF##5110547##100##]. This combined strategy is deemed more effective than mass vaccination [##UREF##30##101##], but combinations of vaccination and public health measures have not yet been explicitly evaluated. Ring vaccination was introduced and evaluated mainly in West and Central Africa and in Asia where it was always combined with case isolation [##REF##1083309##102##]. Although it is difficult to exclude the impact of other interventions and to estimate the net effectiveness of ring vaccination explicitly (e.g. the impact of previous vaccinations can usually not be separated [##REF##12500053##103##]), accumulated experience during the Intensified Eradication Programme strongly suggests that ring vaccination (accompanied by vigorous isolation) worked well [##UREF##30##101##]. The strategy is deemed logically effective in containing localized outbreaks, but it is important to ensure effective contact tracing if we are to rely on ring vaccination alone [##REF##15200816##9##,##REF##15298713##104##].</p>", "<p>Vaccination may still be protective if a person has already been exposed to the virus, a procedure referred to as post-exposure vaccination [##REF##14718077##105##,##REF##17121292##106##]. Despite numerous discussions [##REF##12594644##107##], the protective effect of post-exposure vaccination has remained unclear. A historical study from the early 20th century suggests that vaccination within 7 days after exposure is effective [##UREF##7##28##]. Smallpox textbooks in the 1960s and '70s claimed that 'vaccination within 72 hours almost promises protection' [##UREF##5##26##,##UREF##6##27##], a statement roughly consistent with a more recent statistical estimate based on historical data and on several assumptions concerning the hypothetical frequencies of vaccinated and protected individuals [##REF##15097002##108##], and with a laboratory study demonstrating a cell-mediated response within 4 days after exposure [##REF##15346340##109##]. A similar estimate was obtained in a Delphi analysis, which concluded that post-exposure vaccination can be assumed to be 80–93% effective during the first 3 days after exposure and 2–25% thereafter [##REF##14513416##110##]. However, as we have shown, a statistical exercise demonstrates that historical data, which only record cases who developed smallpox after post-exposure vaccination, hardly provide sufficient insight into the effectiveness of post-exposure vaccination [##REF##17321212##111##]. Information regarding the denominator is insufficient for the majority of records (i.e. we do not know how many exposed people who were vaccinated were protected from the disease). Only the effectiveness of vaccination against severe disease upon infection can be estimated from such data: the shorter the interval between exposure and vaccination, the lower the probability of developing severe smallpox. To the best of our knowledge, only one outbreak in Leicester, UK, from 1903–04 provided insight into the protection against disease by post-exposure vaccination [##UREF##31##112##]: counting from the eruption of the index case, it was reported that none of 210 individuals (0%) who were vaccinated on the first day after exposure, 2 among 359 (0.5%) who were vaccinated on the second day, 5 among 102 (4.9%) who were vaccinated on the third day, and 10 among 116 (8.6%) who were vaccinated on the fourth day or later developed the disease. Although this seems to indicate some degree of protection, the actual efficacy of post-exposure vaccination can only be determined by comparing these findings to observations in a group of individuals who were exposed for exactly the same periods of time, but refused or were denied post-exposure vaccination.</p>", "<p>Despite effective vaccination, pros and cons of vaccination practice always need to account for adverse events of vaccination [##REF##12856218##113##]. Vaccine-strain dependent differences in the frequency of adverse events have been reported, and the risk of death due to vaccination has been analyzed in detail only recently [##REF##16933957##114##,##REF##13678138##115##]. Theoretical frameworks reported to date agree with each other that we should not implement pre-attack mass vaccination in order to minimize the number of adverse events. Policy suggestions of mathematical models for post-attack vaccination strategies depend on the specific attack scenarios and need to be investigated further.</p>", "<title>Case isolation and contact tracing</title>", "<p>Rather than relying completely on vaccination, recent modelling studies have suggested that an outbreak could be contained by a combination of case isolation and contact tracing [##REF##12851224##6##,##REF##16894173##14##], owing mainly to the characteristics of the intrinsic dynamics of smallpox (e.g. the relatively long generation time and the clear symptoms of smallpox). The importance of monitoring and controlling \"contacts\" has been highlighted in a historical observation [##UREF##31##112##] and was also stressed during the Eradication Programme [##REF##12496354##51##,##REF##5489073##116##,##REF##5060374##117##]. A public health system's capability in conducting contact tracing may determine whether or not a smallpox outbreak can be controlled without vaccination. This should also take into account response logistics and the limited number of public health practitioners [##REF##15298713##104##]. A mathematical exercise suggested that the optimal intervention also depends on the initial attack size: whereas an outbreak caused by few cases could easily be controlled by isolation and contact tracing alone, regional (targeted) mass vaccination is recommended if the initial attack size is big and <italic>R</italic>₀is large [##REF##17222358##118##].</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>HN reviewed the literature, analyzed the data and drafted an early version of the manuscript. ME reviewed the early version of the manuscript and assisted in editing the manuscript. SOB participated in the writing and revision of the manuscript. All authors have read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This review would not have been possible without technical support and input from Klaus Dietz and Isao Arita. This study was in part supported by European Union project INFTRANS (FP6 STREP; contract no. 513715). The study of HN was supported by The Netherlands Organisation for Scientific Research (NWO, ALW-IPY-NL/06-15D).</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Incubation period distribution of smallpox fitted to a lognormal distribution (n = 131)</bold>. The vertical arrow indicates the maximum likelihood estimate of the 99th percentile of the incubation period [##UREF##11##42##]. The median and the coefficient of variation are 12.5 days and 18.0%, respectively.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Relative frequency of secondary transmissions of smallpox by disease-age</bold>. Expected daily frequency of secondary transmissions with corresponding 95% confidence intervals [##REF##17156499##46##]. The percentages indicate the fraction of transmissions among all transmissions that occurred in the given intervals. The disease-age <italic>t</italic> = 0 denotes the onset of fever.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Age-specific case fatality of smallpox</bold>. Observed (grey bars) and fitted (continuous line) age-specific case fatalities of unvaccinated cases in (A) Verona, Italy, 1810–1838 [##REF##12387913##53##,##UREF##16##55##] and (B) Sheffield, UK, 1887–8 [##UREF##15##54##].</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p><bold>Theoretical initial courses of smallpox outbreaks following a geometric growth</bold>. A. The infection tree (i.e. transmission network) of smallpox is shown by generation, following equation (8). For simplicity, <italic>R</italic>₀is assumed to be 2. B. Infection tree under vaccination. Vaccination is assumed to reduce the number of secondary transmission by 50%, and thus only 1 case in each generation is expected.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Reported estimates of <italic>R</italic>₀for smallpox</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\">Location</td><td align=\"center\"><italic>R</italic>₀</td><td align=\"center\">Range (min, max)</td><td align=\"center\">Assumptions</td></tr></thead><tbody><tr><td align=\"center\">Unspecified [##REF##8174658##68##,##UREF##22##69##]</td><td align=\"center\">5.0</td><td align=\"center\">-</td><td align=\"left\">Calculated from proposed goal of vaccination coverage</td></tr><tr><td align=\"center\">Abakaliki, Nigeria, 1967 [##UREF##21##67##]</td><td align=\"center\">1.1</td><td align=\"center\">(1.0, 1.2)^‡</td><td align=\"left\">Population mixes randomly, initially fully susceptible</td></tr><tr><td align=\"center\">Various outbreaks in Europe and the US, 18–20th centuries [##REF##11742399##66##]</td><td align=\"center\">3.5–6.0</td><td align=\"center\">(3.4, 10.8)</td><td align=\"left\">Population mixes randomly, initially fully susceptible</td></tr><tr><td align=\"center\">Paris, 17th century [##REF##12387913##53##]</td><td align=\"center\">6.7</td><td align=\"center\">-</td><td align=\"left\">Population is fully susceptible at birth</td></tr><tr><td align=\"center\">Abakaliki, Nigeria, 1967 [##REF##12851223##18##]</td><td align=\"center\">6.9</td><td align=\"center\">(4.5, 10.1)^‡</td><td align=\"left\">Initially partially immune, heterogeneous mixing</td></tr></tbody></table></table-wrap>" ]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1742-4682-5-20-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>f</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>;</mml:mo><mml:mi>μ</mml:mi><mml:mo>,</mml:mo><mml:msup><mml:mi>σ</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mi>σ</mml:mi><mml:msqrt><mml:mrow><mml:mn>2</mml:mn><mml:mi>π</mml:mi></mml:mrow></mml:msqrt></mml:mrow></mml:mfrac><mml:mi>exp</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mo>−</mml:mo><mml:mfrac><mml:mrow><mml:mo>−</mml:mo><mml:msup><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>ln</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>−</mml:mo><mml:mi>μ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:msup><mml:mi>σ</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1742-4682-5-20-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>L</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>μ</mml:mi><mml:mo>,</mml:mo><mml:msup><mml:mi>σ</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∏</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>n</mml:mi></mml:munderover><mml:mrow><mml:mi>f</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>;</mml:mo><mml:mi>μ</mml:mi><mml:mo>,</mml:mo><mml:msup><mml:mi>σ</mml:mi><mml:mn>2</mml:mn></mml:msup><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM3\"><label>(3)</label><italic>s </italic>= <italic>l </italic>+ <italic>f</italic></disp-formula>", "<disp-formula id=\"bmcM4\"><label>(4)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1742-4682-5-20-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>s</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>t</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:mrow><mml:msubsup><mml:mo>∫</mml:mo><mml:mn>0</mml:mn><mml:mi>t</mml:mi></mml:msubsup><mml:mrow><mml:mi>l</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>t</mml:mi><mml:mo>−</mml:mo><mml:mi>τ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mi>f</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>τ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mi>d</mml:mi><mml:mi>τ</mml:mi></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM5\"><label>(5)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1742-4682-5-20-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>L</mml:mi><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∏</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>n</mml:mi></mml:munderover><mml:mrow><mml:mi>s</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mstyle><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∏</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>n</mml:mi></mml:munderover><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:mrow><mml:msubsup><mml:mo>∫</mml:mo><mml:mn>0</mml:mn><mml:mrow><mml:mi/><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:msubsup><mml:mrow><mml:mi>l</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>τ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mi>f</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>τ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mi>d</mml:mi><mml:mi>τ</mml:mi></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM6\"><label>(6)</label><italic>c</italic>(<italic>a</italic>) = <italic>α </italic>exp(-<italic>βa</italic>) + <italic>γ</italic>(1 - exp(-<italic>δa</italic>))²</disp-formula>", "<disp-formula id=\"bmcM7\"><label>(7)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1742-4682-5-20-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>L</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>α</mml:mi><mml:mo>,</mml:mo><mml:mi>β</mml:mi><mml:mo>,</mml:mo><mml:mi>γ</mml:mi><mml:mo>,</mml:mo><mml:mi>δ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mstyle displaystyle=\"true\"><mml:munder><mml:mo>∏</mml:mo><mml:mi>i</mml:mi></mml:munder><mml:mrow><mml:mi>c</mml:mi><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mrow><mml:msub><mml:mi>M</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:msup><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>c</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>a</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mrow><mml:msub><mml:mi>N</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:msup></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM8\"><label>(8)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1742-4682-5-20-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>a</mml:mi><mml:msub><mml:mi>R</mml:mi><mml:mn>0</mml:mn></mml:msub><mml:mo>,</mml:mo><mml:mi/><mml:mi>a</mml:mi><mml:msubsup><mml:mi>R</mml:mi><mml:mn>0</mml:mn><mml:mn>2</mml:mn></mml:msubsup><mml:mo>,</mml:mo><mml:mi/><mml:mn>...</mml:mn><mml:mo>,</mml:mo><mml:mi/><mml:mi>a</mml:mi><mml:msubsup><mml:mi>R</mml:mi><mml:mn>0</mml:mn><mml:mi>n</mml:mi></mml:msubsup><mml:mo>,</mml:mo><mml:mi/></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM9\"><label>(9)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1742-4682-5-20-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>a</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>v</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:msub><mml:mi>R</mml:mi><mml:mn>0</mml:mn></mml:msub><mml:mo>,</mml:mo><mml:mi>a</mml:mi><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>v</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:msubsup><mml:mi>R</mml:mi><mml:mn>0</mml:mn><mml:mn>2</mml:mn></mml:msubsup><mml:mo>,</mml:mo><mml:mn>...</mml:mn><mml:mo>,</mml:mo><mml:mi>a</mml:mi><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mi>v</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mi>n</mml:mi></mml:msup><mml:msubsup><mml:mi>R</mml:mi><mml:mn>0</mml:mn><mml:mi>n</mml:mi></mml:msubsup></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM10\"><label>(10)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1742-4682-5-20-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>v</mml:mi><mml:mo>&gt;</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>R</mml:mi><mml:mn>0</mml:mn></mml:msub></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM11\"><label>(11)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1742-4682-5-20-i9\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:msub><mml:mrow><mml:mtext>VE</mml:mtext></mml:mrow><mml:mi>S</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mrow><mml:mtext>SAR</mml:mtext></mml:mrow><mml:mrow><mml:mn>10</mml:mn></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mrow><mml:mtext>SAR</mml:mtext></mml:mrow><mml:mrow><mml:mn>00</mml:mn></mml:mrow></mml:msub></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mn>0.693</mml:mn></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow><mml:msub><mml:mrow><mml:mtext>VE</mml:mtext></mml:mrow><mml:mi>I</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mrow><mml:mtext>SAR</mml:mtext></mml:mrow><mml:mrow><mml:mn>01</mml:mn></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mrow><mml:mtext>SAR</mml:mtext></mml:mrow><mml:mrow><mml:mn>00</mml:mn></mml:mrow></mml:msub></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mn>0.674</mml:mn></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow><mml:msub><mml:mrow><mml:mtext>VE</mml:mtext></mml:mrow><mml:mi>T</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>−</mml:mo><mml:mfrac><mml:mrow><mml:msub><mml:mrow><mml:mtext>SAR</mml:mtext></mml:mrow><mml:mrow><mml:mn>11</mml:mn></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mrow><mml:mtext>SAR</mml:mtext></mml:mrow><mml:mrow><mml:mn>00</mml:mn></mml:mrow></mml:msub></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mn>0.923</mml:mn></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:semantics></mml:math></disp-formula>" ]

[]

[]

[]

[]

[]

[ "<table-wrap-foot><p>^‡The ranges for the outbreak in Abakaliki are 95% confidence intervals.</p></table-wrap-foot>" ]

[ "<graphic xlink:href=\"1742-4682-5-20-1\"/>", "<graphic xlink:href=\"1742-4682-5-20-2\"/>", "<graphic xlink:href=\"1742-4682-5-20-3\"/>", "<graphic xlink:href=\"1742-4682-5-20-4\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>Insulin, secreted by pancreatic islet <italic>β</italic>-cells, is the principal regulating hormone of glucose metabolism. In humans, plasma insulin exhibits oscillatory characteristics across several time scales independent of changes in plasma glucose [##REF##11815487##1##, ####REF##1644923##2##, ##REF##10545148##3##, ##REF##8770017##4####8770017##4##]. These oscillations are caused by pulsatile insulin secretion [##REF##6345247##5##,##REF##3311858##6##]. Loss of insulin pulsatility is observed in patients of both type 1 diabetes (T1D) and type 2 diabetes (T2D) [##REF##6345247##5##,##REF##10867718##7##,##REF##7014311##8##], and in relatives with mild glucose intolerance or in individuals at risk for diabetes [##REF##10910002##9##, ####REF##9124326##10##, ##REF##3283553##11##, ##REF##1541379##12####1541379##12##]. However, the role of insulin pulsatility in glucose metabolic control and diabetes is still not well understood.</p>", "<p>The pulsatile insulin release is driven by the electrical burst of <italic>β</italic>-cell membrane. Theoretically single isolated <italic>β</italic>-cells can burst, and can be induced <italic>in vitro </italic>to release insulin under tightly controlled conditions. But due to the extensive heterogeneity among individual <italic>β</italic>-cells, not all cells will respond to glucose, and for those that do respond, the amplitude, duration and frequency of oscillations are variable [##REF##10545148##3##,##REF##3510882##13##]. In contrast, in <italic>β</italic>-cell clusters or islets where the cell-cell communication is intact, all cells respond to glucose with regular and synchronized oscillations [##REF##10545148##3##,##REF##3510882##13##,##REF##9051586##14##].</p>", "<p>Inter-<italic>β </italic>cell coupling is mediated through the gap junction channels formed between adjacent <italic>β</italic>-cells. Gap junctions are specific membrane structures consisting of aggregates of intercellular channels that enable the direct exchange of ions. Such channels result from the association of two hemichannels, named connexons, each contributed separately by the two adjacent cells. Each connexon is an assembly of six transmembrane connexins, encoded by a family of genes with more than 20 members. Using rodent models it was found that connexin36 (Cx36) is the only connexin isoform expressed in <italic>β</italic>-cells [##REF##12766175##15##, ####REF##10905480##16##, ##REF##17395748##17##, ##REF##15625088##18####15625088##18##]. Recent study found that Cx36 is also expressed in human islets [##REF##12688633##19##]. Cx36 gap junctions have weak voltage sensitivity and small unitary conductance [##REF##10559394##20##]. This unique combination of properties makes them well suited as electrical coupler, which is important for the regulation of insulin release from <italic>β</italic>-cells [##REF##17395748##17##].</p>", "<p>The critical functional role of the gap junctional coupling between <italic>β</italic>-cells has been demonstrated in many experiments. Studies on pancreatic islets and acinar cells revealed that cell-to-cell communication is required for proper biosynthesis, storage and release of insulin, and were nicely reviewed in [##REF##17919186##21##,##REF##16931449##22##]. Single uncoupled <italic>β</italic>-cells show a poor expression of the insulin gene, release low amounts of the hormone, and barely increase function after stimulation [##REF##8844334##23##, ####REF##1334972##24##, ##REF##1697604##25####1697604##25##]. Alterations in Cx36 level are associated with impaired secretory response to glucose [##REF##12766175##15##,##REF##17395748##17##,##REF##12697840##26##,##REF##12540616##27##]. Lack of Cx36 results in loss of <italic>β</italic>-cell synchronization, loss of pulsatile insulin release, and significantly higher basal insulin release in the presence of sub-stimulatory glucose concentration from isolated islets [##REF##15919802##28##]. Blockage of gap junctions between <italic>β</italic>-cells also similarly abolish their normal secretory response to glucose [##REF##10545148##3##,##REF##1697604##25##,##REF##2440035##29##]. Restoration of <italic>β</italic>-cell contacts is paralleled by a rapid improvement of both insulin biosynthesis and release [##REF##8844334##23##, ####REF##1334972##24##, ##REF##1697604##25####1697604##25##]. Further support for this concept comes from the finding that a number of tumoral and transformed cell lines that do not express connexins show abnormal secretory characteristics [##REF##8522612##30##]. Transfection of the cells with a connexin gene corrected the coupling and some of the secretory defects [##REF##8522612##30##]. In addition to the functional role in insulin secretion, study with transgenic mice overexpressing Cx36 showed that it protects <italic>β</italic>-cells against streptozotocin (STZ) and cytokine (IL-1<italic>β</italic>) damage, and loss of the protein sensitizes <italic>β</italic>-cells to such damages [##REF##16931449##22##]. On the other hand, impaired glucose tolerance can compromise the gap junctional channels. In vitro study of freshly isolated rat islets has found that short exposure (30 min) to glucose can modify gap junction configuration [##REF##3526916##31##] whilst a chronic increase in glucose decreases Cx36 expression [##REF##16263767##32##], suggesting that compromise of <italic>β</italic>-cell coupling may be implicated in the early glucotoxicity and desensitization phenomena, and may therefore be relevant to diabetes pathophysiology.</p>", "<p>Theoretical models were developed to describe the <italic>β</italic>-cell oscillation [##REF##6305437##33##, ####REF##3300186##34##, ##REF##11106596##35##, ##REF##2850029##36##, ##REF##1646657##37##, ##REF##8369400##38####8369400##38##], which also revealed how an increased regularity of glucose-dependent oscillatory events was achieved in clusters as compared to isolated islet <italic>β</italic>-cells [##REF##11106596##35##, ####REF##2850029##36##, ##REF##1646657##37##, ##REF##8369400##38####8369400##38##]. Together, these experimental and modeling results strongly indicate the essential role of cell-cell communication in normal <italic>β</italic>-cell function, which may account for the hierarchical organization of <italic>β</italic>-cell mass. The insulin secreting <italic>β</italic>-cells, together with the other endocrine cells, comprise only about 1–2% of the total pancreatic mass. Rather than being distributed evenly throughout the pancreas, they reside in a highly organized micro-organ, the pancreatic islet, with specific 3D morphostructure, copious intercellular coupling and interactions, and are governed by sensitive autocrine and paracrine regulations. This organization, not individual <italic>β</italic>-cells, is the basis for generating the insulin oscillation and a proper glucose dose response. Therefore one would expect that the morphostructural integrity of islets, namely, the interactions and the three-dimensional architecture among various cell populations in islets, is critical for islet function. Indeed, in islet transplantation studies it has been found that these characteristics are predictive of <italic>in vivo </italic>function and survival of islets, as well as the clinical outcome after transplantation [##REF##17259521##39##]. Despite the many published models of pulsatile insulin release, a quantitative investigation of the functional role of islet <italic>β</italic>-cell's cytoarchitectural organization was not available until recently [##REF##17912360##40##].</p>", "<p>In our previous work we have proposed that a <italic>β</italic>-cell cluster can be described by three key architectural parameters: number of <italic>β</italic>-cells in the cluster <italic>n</italic>_{<italic>β</italic>}, number of neighboring <italic>β</italic>-cells that each <italic>β</italic>-cell is coupled with <italic>n</italic>_c, and intercellular coupling strength <italic>g</italic>_c[##REF##17912360##40##]. Traditional islet simulation has assumed a simple cubic packing (SCP) arrangement of <italic>β</italic>-cells, with 6 nearest neighbors for each cell, i.e. <italic>n</italic>_{<italic>c</italic>,max}= 6. We found that this model significantly underestimates the neighboring cells each <italic>β</italic>-cell has, with which potential intercellular coupling could be formed [##REF##17912360##40##]. It is therefore limiting to investigate the effect of varying proportions of non-<italic>β </italic>cells (which do not couple with <italic>β</italic>-cells), or the functional consequence of architectural perturbations such as compromised degree of intercellular coupling resulting from <italic>β</italic>-cell death. We therefore introduced a new hexagonal closest packing (HCP) model with 12 nearest neighbors for each cell, and <italic>n</italic>_{<italic>c</italic>,max}= 12. It provides a much more accurate approximation to the cytoarchitectural organization of cells in islet tissue. Experimental studies of islet <italic>β</italic>-cell clusters also implicated a hexagonal organization of cells [##REF##7821725##41##,##REF##11095104##42##] (see figure 7 on page S15 of [##REF##7821725##41##], figure ##FIG##4##5## on page 40 of [##REF##11095104##42##], for example). Further, it was estimated that in rodent islets about 70% of the cells are <italic>β</italic>-cells; this corresponds to an effective <italic>n</italic>_c~ 8.4 (as 30% of the 12 nearest neighbors are non-<italic>β </italic>cells) in our HCP model, which is consistent with laboratory measurements of the degree of inter-<italic>β </italic>cell coupling [##REF##10903339##43##]. Human islets are believed to contain proportionally much less <italic>β</italic>-cells, at ~50% [##REF##15923354##44##,##REF##16461897##45##], which corresponds to <italic>n</italic>_c~ 6.</p>", "<p>Using this new <italic>β</italic>-cell packing model, we examined, for the first time, the functional dependence of islet oscillation on its architecture. Optimal values of <italic>n</italic>_{<italic>β</italic>}, <italic>n</italic>_cand <italic>g</italic>_cat which functional gain is maximized are obtained [##REF##17912360##40##]. In this study, we further investigate islet-bursting phenomenon as reflected in three functional measures: fraction of <italic>β</italic>-cells that could burst <italic>f</italic>_b, synchronization index <italic>λ</italic>, and bursting period <italic>T</italic>_b. We will specifically examine the influence of structural perturbation to <italic>n</italic>_cand <italic>g</italic>_c, and if a composite measure of islet morphostructural integrity can be defined from them. As in previous study, we focus the investigation from the perspective of high frequency oscillation resulting from the feedback loops of intracellular calcium currents, which is in the time scale of ~10–60 sec. We reserve the more comprehensive investigation of <italic>β</italic>-cell oscillation at different time scales in future work.</p>" ]

[ "<title>Methods</title>", "<title>Mathematical model of the electrical excitability of <italic>β</italic>-cells</title>", "<p>As we have previously described in [##REF##17912360##40##], we adopt the formulation developed by Sherman et al [##REF##8770032##67##,##REF##11276531##68##] of the Hodgkin-Huxley model for <italic>β</italic>-cell electrical excitability, for its simplicity:</p>", "<p></p>", "<p>The ionic current terms include the fast voltage-dependent L-type Ca²⁺-channel current <italic>I</italic>_Ca, the glucose sensitive K_ATPchannel current <italic>I</italic>_{<italic>KATP</italic>}, the voltage-dependent delayed rectifier K⁺current <italic>I</italic>_{<italic>K</italic>}, and a slow inhibitory K⁺current <italic>I</italic>_{<italic>S</italic>}, given by:</p>", "<p></p>", "<p>where <italic>g</italic>_{<italic>KATP</italic>}, <italic>g</italic>_{<italic>Ca</italic>}, <italic>g</italic>_{<italic>K</italic>}, <italic>g</italic>_{<italic>S </italic>}are channel conductance. The activation parameters <italic>n</italic>, <italic>s </italic>are given by</p>", "<p></p>", "<p>with , , being the fraction of open channels for the corresponding currents respectively at steady state. The parameters <italic>V</italic>_{<italic>m</italic>}, <italic>V</italic>_{<italic>n</italic>}, <italic>V</italic>_{<italic>s</italic>}, and <italic>θ</italic>_{<italic>m</italic>}, <italic>θ</italic>_{<italic>n</italic>}, <italic>θ</italic>_{<italic>s </italic>}are constants that describe the dependence of channel activation on membrane voltage V. The change in intercellular calcium concentration is given by</p>", "<p></p>", "<p>where <italic>f </italic>is the fraction of free Ca²⁺and <italic>k</italic>_Cais the removal rate of Ca²⁺in the intracellular space. <italic>α </italic>is a conversion factor from chemical gradient to electrical gradient. For a more detailed explanation of the model equations, parameters and their values, and the implementation, please refer to [##REF##17912360##40##]. The numerical simulation was performed for the 4 ODEs given in equations 3, 5, and 6.</p>", "<title>The HCP model of <italic>β</italic>-cell cluster</title>", "<p>We have previously introduced the HCP model of islet cytoarchitecture to simulate the functional consequence of varying structure [##REF##17912360##40##]. In this model each cell has 6 nearest neighbors in 2D (<italic>n</italic>_{<italic>c</italic>,max}= 6), and 12 in 3D (<italic>n</italic>_{<italic>c</italic>,max}= 12). Setting up the simulation for HCP <italic>β</italic>-cell clusters is more intricate than the SCP model, and we have developed a cell labeling algorithm [##REF##17912360##40##]. Briefly, given a <italic>β</italic>-cell cluster with edge size <italic>n</italic>, labeling of cells starts with the center or the primary layer. It is a 2D regular hexagon of edge size <italic>n</italic>, with a total of 3<italic>n</italic>²-3<italic>n</italic>+1 cells. The remaining <italic>n</italic>-1 layers on each side (top and bottom) of the primary layer, starting from immediate layer adjacent to it, alternate between being an irregular hexagonal (IH, the six sides and internal angles are not all equal) layer, and a regular hexagonal (RH) layer. The edge size decreases each time when traversing up or down. The number of cells in IH and RH layers is given by 3(<italic>r</italic>-1)²and 3<italic>r</italic>²-3<italic>r</italic>+1 respectively where <italic>r </italic>is the edge size of that layer. When <italic>n </italic>is even, a 3D HCP cluster ends with an IH-layer on its surface and when <italic>n </italic>is odd, it ends with an RH-layer on its surface [##REF##17912360##40##]. This definition ensures that our HCP clusters are symmetric along all directions, which simulates the natural growth of pancreatic islets. Lastly, the program generates nearest neighbor list for each <italic>β</italic>-cell based on the Euclidean distance between cells.</p>", "<p>All cell j located at (<italic>x</italic>_{<italic>j</italic>}, <italic>y</italic>_{<italic>j</italic>}, <italic>z</italic>_{<italic>j</italic>}) belongs to the neighborhood of cell i at (<italic>x</italic>_{<italic>i</italic>}, <italic>y</italic>_{<italic>i</italic>}. <italic>z</italic>_{<italic>i</italic>}) if the Euclidean distance between the two cells is 1, namely:</p>", "<p></p>", "<p>This neighbor list is then utilized to set up the term in equation 3.</p>", "<p>Figure ##FIG##3##4## presents the top view of a 3D HCP-323 and a SCP-343 cell cluster. Evident from the figure is the complexity of HCP but the added advantage of a higher degree of intercellular coupling, as well as the simplicity of SCP with its limited intercellular coupling.</p>", "<title>Numerical Simulations</title>", "<p>HCP and SCP <italic>β</italic>-cell clusters of different sizes with number of <italic>β</italic>-cells <italic>n</italic>_{<italic>β </italic>}ranging from 1–343, number of inter <italic>β</italic>-cell couplings of each <italic>β</italic>-cell <italic>n</italic>_cvarying between 0–12, and coupling strength <italic>g</italic>_cspanning from 0–1000 pS, were simulated, as described in figure ##FIG##4##5##. Totally we simulated for over 800 different clusters. For each point in the structure space <bold><italic>S</italic></bold>: (<italic>n</italic>_{<italic>β</italic>}, <italic>n</italic>_c, <italic>g</italic>_c), 10 replicate clusters were simulated with the biophysical properties of individual <italic>β</italic>-cells following the heterogeneity model as previously described, in table 2 of [##REF##17912360##40##]. 500 uncoupled single <italic>β</italic>-cells were also simulated, which corresponds to point (1, 0, 0) in <bold><italic>S </italic></bold>(figure ##FIG##4##5##). This provides the baseline information for analyzing the functional characteristics of coupled cell clusters.</p>", "<p>Simulation for <italic>n</italic>_cis modulated by randomly decoupling varying percentages of <italic>β</italic>-cells from the rest. This is designed to simulate the loss of <italic>β</italic>-cell mass under pathological conditions, or the presence of non <italic>β</italic>-cells (mainly <italic>α</italic>- and <italic>δ</italic>-cells) in natural islets. It is known the non-<italic>β </italic>islet cells do not synchronize with <italic>β</italic>-cells or among themselves [##REF##10226151##69##], presumably because they do not couple to <italic>β</italic>-cells, and the coupling among themselves are too sparse to coordinate their dynamic activities. Gap conductance <italic>g</italic>_cis varied from a no coupling state (where each cell is in a quarantine-like state and functioning without any communication, <italic>g</italic>_c= 0 pS) to a strongly coupled state of 1000 pS.</p>", "<title>The Sorting Hat for <italic>β</italic>-cells</title>", "<p>We introduce the Lomb-Scargle periodogram [##UREF##0##46##,##UREF##1##47##], which describes power concentrated in a particular frequency, namely, the power spectral density (PSD), to sort the bursting status of <italic>β</italic>-cells. We adopt this method over the more commonly used Fourier method for two reasons: (1) it does not require evenly spaced time series while the Fourier method does. It may not be a major concern if we restrict to only the analysis of the intracellular calcium (figure ##FIG##0##1##, upleft), and only the steady state solution. But other parameters, particularly the membrane potential, exhibit more complex temporal patterns, with high frequency oscillation overlaying the plateau phase of the slower oscillations (figure ##FIG##0##1##, upright). (2) the Lomb-Scargle Periodogram comes with a statistical method to evaluate the significance of the observed periodicity [##UREF##1##47##] while Fourier transform method does not.</p>", "<p>Briefly, let <italic>y</italic>_ibe the time-dependent intra-cellular calcium [<italic>Ca</italic>(<italic>t</italic>)] obtained by simulation at each time <italic>t</italic>_{<italic>i</italic>}, where <italic>i </italic>= 1,2,..., <italic>N</italic>, with mean and variance <italic>σ</italic>². The Lomb-Scargle periodogram <italic>P</italic>(<italic>ω</italic>) at an angular frequency of <italic>ω </italic>= 2<italic>πf </italic>is computed according to the following equation:</p>", "<p></p>", "<p>where the constant <italic>τ </italic>is obtained from:</p>", "<p></p>", "<p>The low-limit of <italic>f </italic>is taken to be 1/<italic>T</italic>, where <italic>T </italic>is the time span and is equal to <italic>t</italic>_N- <italic>t</italic>₁. Since our simulations are carried out for a period of 120 sec, <italic>f </italic>is 0.0083 Hz. The uplimit of <italic>f </italic>is taken as the Nyquist frequency, <italic>N</italic>/(2<italic>T</italic>), where <italic>N </italic>is the length of the dataset. This gives a value of 1.0 Hz. Scargle showed that the null distribution of the Lomb-Scargle periodogram at a given frequency is exponentially distributed, namely the cumulative distribution function of <italic>P</italic>(<italic>ω</italic>) is given by Pr [<italic>P</italic>(<italic>ω</italic>) &lt;<italic>z</italic>] = 1 - <italic>e</italic>^{-<italic>z </italic>}[##UREF##1##47##]. Therefore, once <italic>P</italic>(<italic>ω</italic>) is calculated for different frequencies, the significance of the principal peak, <italic>max</italic>(<italic>P</italic>(<italic>ω</italic>)) can be evaluated by [##UREF##1##47##]:</p>", "<p></p>", "<p>where <italic>M </italic>equals number of independent test frequencies. In our case it equals the number of data points <italic>N</italic>. Expression (10) tests the null hypothesis that the peak is due to random chance. When <italic>p</italic>-value of the principal peak is small, the time series is considered to contain significant periodic signal, and in our case, the cell can be considered a burster with regular oscillatory pattern. In this study the threshold <italic>p</italic>-value for burster cell is set to be 0.005. Among non-bursters, cells whose maximum and minimum membrane voltages differ by less than 30 mV, Δ<italic>V </italic>= |<italic>V</italic>_max- <italic>V</italic>_min| &lt; 30 mV, are sorted as silent cells and the rest as spikers. The flowchart of the complete sorting process is presented in figure ##FIG##5##6##. For the burster <italic>β</italic>-cells, their bursting periods <italic>T</italic>_band degree of synchronization in bursting were then determined.</p>", "<title>Synchronization Analysis</title>", "<p>Briefly, the instantaneous phase of each <italic>β</italic>-cell was first determined using the Matlab command: <italic>φ</italic>_{<italic>j</italic>}(<italic>t</italic>) = unwrap(angle(Hilbert(detrend(<italic>V</italic>_{<italic>j</italic>}(<italic>t</italic>)))). A mean field value of phase Φ is determined by taking the circular mean of the individual phase angles of all bursting <italic>β</italic>-cells</p>", "<p></p>", "<p>The synchronization strength to mean field by each <italic>β</italic>-cell can be calculated by</p>", "<p></p>", "<p>A cluster synchronization index (CSI) is then defined by</p>", "<p></p>", "<p>It measures how cells in the whole cluster are coupled in their oscillation. When synchronization is evaluated among bursting <italic>β</italic>-cells only, a simpler approach that measures the mean pair-wise phase difference can be taken. The synchronization of each pair of cells <italic>j </italic>and <italic>k </italic>is calculated by</p>", "<p></p>", "<p>The mean of all pair-wise synchronization are then determined by:</p>", "<p></p>", "<p>For each cluster both the mean value and the distribution of CSI and <italic>λ </italic>are evaluated. The results are compared to reveal if there is modular pattern within the cluster, namely, if there are sub-regions within the whole cluster where the <italic>β</italic>-cells within each region is well synchronized, but not with <italic>β</italic>-cells in the other sub-regions. In the <italic>β</italic>-clusters we have simulated, the results of CSI and <italic>λ </italic>are not significantly different, and therefore for simplicity we only report the results of <italic>λ</italic>.</p>" ]

[ "<title>Results</title>", "<title>Sorting cells using Lomb-Scargle periodogram</title>", "<p>The first step post simulation of a <italic>β</italic>-cell cluster is to determine the bursting status of each <italic>β</italic>-cell in the cluster. In general it can be a burster, a spiker, or a silent cell [##REF##17912360##40##]. A burster is defined as a cell capable of producing a sequence of well-defined regular bursts which correlate with the period between consecutive peaks and nadirs in the calcium signal or membrane action potential. In contrast, a spiker usually produces uncontrolled continuous voltage spikes and does not spend any significant time in the plateau phase of sustained oscillation, thereby being unable to generate a glucose dose response. A silent cell is one which remains in the hyperpolarized state throughout, and thus remains inactive in the insulin secretion process. In our previous work, we used an empirical rule based on the peak and nadir information of the <italic>s</italic>(<italic>t</italic>) signal (the slow variable of the potassium channel, see equations 4–5 in methods) to distinguish between spikers and bursters. In this study we introduce a more analytical method. The sorting hat (Rowling J.K.) we utilized is the Lomb-Scargle periodogram [##UREF##0##46##,##UREF##1##47##], which describes power concentrated at particular frequencies. We applied it to intracellular calcium concentration [<italic>Ca</italic>(<italic>t</italic>)].</p>", "<p>Figure ##FIG##0##1## presents the calcium and membrane voltage profiles of three sample cells – a burster, a spiker and a silent cell, along with their computed Lomb-Scargle periodograms. As we can see, the spiker and the silent <italic>β</italic>-cells have a broad frequency spectrum and power is spread out over a wide-range of frequencies, whereas for the burster <italic>β</italic>-cell, the distribution is much narrower and the major peak frequency was observed at 33 mHz. The <italic>p</italic>-value of the principal frequency component of the burster cell assumes a significantly low value with <italic>p </italic>&lt; 10^-12, while it is &gt;0.4 for the spiker and silent cells. In this study the threshold <italic>p</italic>-value for burster cell is set to be 0.005. We find that this algorithm distinguishes well the burster cells from the rest. Figure ##FIG##0##1d## presents the distribution of <italic>p</italic>-values for 819 <italic>β</italic>-cells from three <italic>β</italic>-cell clusters: a HCP-323, a SCP-343, and a HCP-153 cluster. Cells with regular bursting clearly segregate from others into a distinct group. Spikers with a very regular spiking frequency can also have marginally significant principal peaks, but normally with <italic>p </italic>&gt; 0.05. The algorithm was tested extensively and zero misclassification was found for all the clusters we have simulated. Hence we believe that the <italic>f</italic>_bestimation using the Lomb-Scargle periodogram is accurate.</p>", "<title>The hyperbolic relationship between <italic>g</italic>_cand <italic>n</italic>_c, and the cluster coupling index CCI</title>", "<p>To investigate the functional role of islet structure characterized by (<italic>n</italic>_{<italic>β</italic>}, <italic>n</italic>_c, <italic>g</italic>_c), we simulated for over 800 different structural states of islet (see figure ##FIG##4##5## in methods). Our previous study has revealed a quantitative dependence of islet function on the 3D morphostructural organization of its <italic>β</italic>-cells. This raises the question if a composite measure of islet architectural integrity can be defined to capture the dependence and to develop predictive models of islet function. Given a <italic>β</italic>-cell cluster, the architecture intactness of the whole cluster depends critically on both the individual pair-wise cell coupling strength (<italic>g</italic>_c) and the number of couplings each <italic>β</italic>-cell has (<italic>n</italic>_c).</p>", "<p>Specifically, the coupling term in equation 3 (see methods) can be written as:</p>", "<p></p>", "<p>where is the mean field value of all the nearest neighbors of cell <italic>i</italic>. This suggests that mean (<italic>n</italic>_c·<italic>g</italic>_c) can be a measure that describes the coupling integrity of the islet.</p>", "<p>For a normal islet, the distribution of (<italic>n</italic>_c·<italic>g</italic>_c) is around a constant.</p>", "<p>We have evaluated the three functional measures <italic>f</italic>_b, <italic>λ</italic>, and <italic>T</italic>_bfor all <italic>β</italic>-cell clusters that we have simulated. Figure ##FIG##1##2## presents the results for the HCP-323 and SCP-343 clusters on a <italic>g</italic>_c-<italic>n</italic>_cplane. It is of interest to note that they indeed follow a hyperbolic response to <italic>g</italic>_cand <italic>n</italic>_cat lower values of <italic>g</italic>_cor <italic>n</italic>_c, and plateau at higher values. Other clusters with different <italic>n</italic>_{<italic>β </italic>}emulate these responses.</p>", "<p>The islet cell coupling and cytoarchitecture are likely compromised during the onset and progression of diabetes. During prediabetic development of disease, as well as after diabetes onset, significant loss of <italic>β</italic>-cell mass occurs [##REF##15662003##48##,##REF##18317728##49##]. This will reduce the number of available <italic>β</italic>-cells for coupling, thus reducing the value of <italic>n</italic>_c. During T1D specifically, the infiltrating immune cells will further reduce <italic>n</italic>_c, as many neighboring cells would be replaced by the immune cells. Though the role of gap junction conductance in human diabetes has not been investigated in depth, animal model studies have indicated its potential involvement in both T1D and T2D [##REF##16931449##22##]. The gap junction conductance <italic>g</italic>_cbetween each pair of cells is the product of number of gap junctional channels formed between them and the specific conductance of each channel, with the latter depending on the channel configuration among other factors. Using transgenic rodent models, it has been shown that the amount of gap junctions directly affects the cell-cell communication and the synchronization of <italic>β</italic>-cell oscillation [##REF##15919802##28##,##REF##17395748##50##]. Reduced amount of gap junctions leads to loss of regular oscillation and the pulsatile insulin release at stimulatory levels of glucose, and increased insulin output at basal glucose. These characteristics of pancreatic dysfunctions mimic those observed in diabetes, and are suggestive of a role of gap junction in the pathophysiology of diabetes [##REF##16931449##22##]. Conversely, gap junctions are dynamic structures, their number, size, and configurations are readily affected (regulated) by environmental conditions, including the glucose level [##REF##3526916##31##,##REF##16263767##32##]. Therefore diabetes progression likely can also affect the value of <italic>g</italic>_c.</p>", "<p>Bearing in mind the significance of the combined effect of <italic>g</italic>_cand <italic>n</italic>_cin determining cluster coupling, and their potential importance in the pathological development of disease, we propose a dimensionless cluster coupling index:</p>", "<p></p>", "<p>as an islet cytoarchitectural integrity descriptor, where <italic>C</italic>₀= (<italic>n</italic>_c,0·<italic>g</italic>_c,0) is a normalization constant, and <italic>n</italic>_c,0and <italic>g</italic>_c,0are their corresponding normal physiological values. In normal rodent islets, ~70% of the islet cells are <italic>β</italic>-cells, which gives <italic>n</italic>_c,0~ 8.4 assuming hexagonal arrangement. The gap junctional conductance has been measured, and found to distribute around <italic>g</italic>_c,0~200 pS [##REF##7473686##51##,##REF##2015391##52##]. Therefore <italic>C</italic>₀~ 1680 pS•cell. Less is known about human islets except that the proportion of <italic>β</italic>-cells is smaller, at ~50% [##REF##15923354##44##,##REF##16461897##45##], which gives <italic>n</italic>_c,0~ 6.0. The <italic>g</italic>_c,0value of human islets is still to be measured. It would of interest to examine if human islets have higher <italic>g</italic>_c,0(most likely by forming more gap junction channels between pairs of neighboring <italic>β</italic>-cells) compared to rodent islets, to compensate for the smaller <italic>n</italic>_c,0value.</p>", "<p>Figure ##FIG##2##3## presents the dependence of the three functional measures on CCI for all HCP <italic>β</italic>-cell clusters we simulated, assuming <italic>C</italic>₀= 1680 pS•cell. Clearly when CCI&lt;1.0, all three measures increase monotonically with increasing CCI value. Little additional functional gain is obtained in the region of CCI&gt;1.0. Values of CCI greater than 1.0 represent higher states of coupling in the islet network system. Islet is robust in its function with strong inter-communication and synchronization. The functional gain of increasing either <italic>g</italic>_cor <italic>n</italic>_cwhen the other is intact, is not of much therapeutic value. This region is of interest to investigate the uplimit of islet connectivity and how this might have evolved. It would also be of interest to study the CCI values of real islets, their distribution, and the upper limit of islet evolution in terms of developing gap junctions and neighborhood coupling.</p>", "<p>During diabetes <italic>n</italic>_cand <italic>g</italic>_cvalues are likely compromised, either contributing to or resulting from problems in glucose tolerance. Reduction either in <italic>n</italic>_cor <italic>g</italic>_cwill lower the value of CCI. When CCI&lt;1.0, extensive variation in all three measures is evident, indicating functional impairment and instability. For consideration of potential therapeutic treatment, this is the critical region for investigation of mechanisms to restore islet structural integrity and functionality by improving <italic>g</italic>_cand/or <italic>n</italic>_c, and bringing CCI back to its desired value. For this reason we denote CCI&lt;1.0 as the region of interest (ROI) for potential therapy (shaded areas in figure ##FIG##2##3##).</p>" ]

[ "<title>Discussion</title>", "<p>Previously we have, for the first time, studied the functional dependence of islet pulsatile insulin release on its cytoarchitectural organization of <italic>β</italic>-cells [##REF##17912360##40##]. In the current study, we further investigated two key islet structural parameters <italic>g</italic>_cand <italic>n</italic>_con islet bursting properties, which are likely involved in the pathophysiology of diabetes. Although numerous experiments have demonstrated the importance in islet function of cell-cell communication between <italic>β</italic>-cells mediated through the gap junction channels, few studies have examined quantitatively the functional role of density and strength of the gap junctions. As synchronization of <italic>β</italic>-cells in their electrical burst and insulin release is the hallmark of normal islet function, we focused on three related functional measures: fraction of <italic>β</italic>-cells that can burst <italic>f</italic>_b, synchronization index <italic>λ</italic>, and bursting period <italic>T</italic>_b. We specifically examined the hyperbolic response of <italic>β</italic>-cell cluster function to the combined input of <italic>g</italic>_cand <italic>n</italic>_c. This means islet functionality can be preserved by manipulating any one or both of them. For example under weak <italic>g</italic>_ccaused by low expression of gap junction proteins (Cx36), increasing the value of <italic>n</italic>_cwill result in improved number of burster cells, bursting pattern and synchronization, and improved islet function. Similarly, when infiltration of immune cells and <italic>β</italic>-cell loss leave few well-connected neighboring <italic>β</italic>-cells (reduced <italic>n</italic>_c), targeting the gap junction strength (improving <italic>g</italic>_c) of existing couplings can improve the bursting and synchronization.</p>", "<p>We characterized the hyperbolic effect of <italic>g</italic>_cand <italic>n</italic>_con islet function in a dimensionless composite measure CCI. We showed that this measure correlates well with islet functional performance. We believe that CCI has the potential to be an index of islet's well-being that is predictive of islet function, and thus a key factor linking structure and function. It can provide insight to the intrinsic compensation mechanism of islet cells when damage occurs. The complexity of islet function can be better understood when associating it with CCI.</p>", "<p>Human islet biology is difficult due to tissue inaccessibility. Most of our current knowledge is obtained and extrapolated from animal studies. However, recent studies revealed cytoarchitectural differences between human and animal islets [##REF##15923354##44##,##REF##16461897##45##]. Specifically, in the frequently used rodent models, an islet contains significantly lower proportions of non-<italic>β </italic>cells compared to in humans, ~30% versus ~50% (this gives, on average, <italic>n</italic>_c~ 8.4 versus <italic>n</italic>_c~ 6, in our HCP cell cluster model). It was further estimated that about 70% of <italic>β</italic>-cells exclusively associate with <italic>β</italic>-cells in rodent islets (namely 70% <italic>β</italic>-cells have <italic>n</italic>_c~ 12), whilst in human islets, this number can be as low as 30% (only 30% <italic>β</italic>-cells have <italic>n</italic>_c~ 12) [##REF##15923354##44##,##REF##16461897##45##]. These reports suggest that rodent islets may have much higher <italic>n</italic>_cthan human ones. The functional implication of such architectural difference is still not known, but clearly cannot be extrapolated linearly. We believe that our work, aimed at achieving a quantitative understanding of islet function and cytoarchitecture, will help us to study human islet biology utilizing animal models. For example, it will also be of interest to examine if CCI is conserved across species, and if it can serve as a scale-invariant index that unveils a common reigning principle across species of islet functional dependence on structure.</p>", "<p>Investigation of islet function and structure is no doubt of interest to the study of glycemic control, diabetes pathogenesis, and the related metabolic syndromes. Such a study is <italic>sine qua non </italic>for understanding pathological progression of <italic>β</italic>-cell mass and function loss, and islet tissue engineering and transplantation, to name a few [##REF##17259521##39##,##REF##17912360##40##]. Under many physiological/pathological conditions, such as pregnancy, puberty, and diabetes, <italic>β</italic>-cell mass is modified. Often the modification is more profound than a mere change of islet size or islet number. For example in T1D the infiltrating immune cells spread from peripheral islet vessels to the centre of a given islet, causing <italic>β</italic>-cell apoptosis across the islet [##REF##3309229##53##] and modification of islet architecture in addition to its total <italic>β</italic>-cell mass. To many with T1D, islet transplantation represents a viable hope to control hyperglycemia; however, significant loss of islet mass and function are observed both short term and long term after transplantation [##REF##15983207##54##]. It is still not clear what exactly the transplanted islets go through. Predictive models of islet function and survival post transplantation are much needed. Several commonly used parameters in islet preparation quality control: islet size (<italic>n</italic>_{<italic>β</italic>}), percent of cells that are <italic>β</italic>-cells (affects <italic>n</italic>_c), non-apoptotic <italic>β</italic>-cells (affects both <italic>n</italic>_cand <italic>g</italic>_c), etc [##REF##17259521##39##], actually constitute the structural framework of the islet. Very recently, it has been explicitly pointed out that the morphostructural integrity of the islets is critical and predictive of <italic>in vivo </italic>function and clinical outcome in islet allotransplantation, and should be studied more [##REF##17259521##39##]. We believe our study provides a starting point for better understanding these issues.</p>", "<p>In this study, we focused the investigation on islet architectural measures, and how they affect islet oscillation. For simplicity, as in previous study, we adopted an oscillation model that describes only the high frequency (at the time scale of ~10–60 sec) component resulting from the feedback loops of the intracellular calcium currents. To have a more comprehensive physical description and better understanding of the pulsatile insulin secretion from islets, and how it depends on islet cytoarchitecture, the other components, especially the intracellular metabolism and the signal transduction pathway of glucose induced insulin release need to be included: the oscillation of glycolysis, ATP/ADP ratio, cAMP, and the other metabolic factors such as NADPH, glutamate, glutamine; the cytosolic calcium, and the exchange of calcium with ER and the effect of ER stress; etc [##REF##15347584##55##, ####REF##4263005##56##, ##REF##15178199##57##, ##REF##15834002##58##, ##REF##12583611##59##, ##REF##15294427##60##, ##REF##17928534##61##, ##REF##15985450##62##, ##REF##12644446##63##, ##REF##16832733##64##, ##REF##11874447##65##, ##REF##12719219##66####12719219##66##]. These coupled with the electrical current oscillation, would generate an additional slow rhythm at the time scale of 2–10 min. The latter is important as it is at a more readily measurable time scale with available laboratory techniques. It would be of interest to investigate how the intracellular pathways and intercellular connections are coupled in determining the islet function, how the properties of individual <italic>β</italic>-cells affect the islet function through the network of coupled <italic>β</italic>-cells, and whether in a coupled network, the islet is more robust to defects in individual <italic>β</italic>-cells such as problems in the intracellular pathways. In this sense, our work only represents the first step towards developing practical models and quantitative measures of islet architecture and investigating its role in islet function. More sophisticated models and laboratory studies are needed. The electrophysiology of islet and <italic>β</italic>-cell oscillation, and evaluation of islet architectural organization, are all experimentally challenging. We believe that such theoretical analysis, though may only represent an initial minimal model approach, are meaningful to gain some insight, and to help design the most relevant and feasible experiment to examine the key factors in these issues.</p>" ]

[]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Insulin, the principal regulating hormone of blood glucose, is released through the bursting of the pancreatic islets. Increasing evidence indicates the importance of islet morphostructure in its function, and the need of a quantitative investigation. Recently we have studied this problem from the perspective of islet bursting of insulin, utilizing a new 3D hexagonal closest packing (HCP) model of islet structure that we have developed. Quantitative non-linear dependence of islet function on its structure was found. In this study, we further investigate two key structural measures: the number of neighboring cells that each <italic>β</italic>-cell is coupled to, <italic>n</italic>_c, and the coupling strength, <italic>g</italic>_c.</p>", "<title>Results</title>", "<p><italic>β</italic>-cell clusters of different sizes with number of <italic>β</italic>-cells <italic>n</italic>_{<italic>β </italic>}ranging from 1–343, <italic>n</italic>_cfrom 0–12, and <italic>g</italic>_cfrom 0–1000 pS, were simulated. Three functional measures of islet bursting characteristics – fraction of bursting <italic>β</italic>-cells <italic>f</italic>_b, synchronization index <italic>λ</italic>, and bursting period <italic>T</italic>_b, were quantified. The results revealed a hyperbolic dependence on the combined effect of <italic>n</italic>_cand <italic>g</italic>_c. From this we propose to define a dimensionless cluster coupling index or CCI, as a composite measure for islet morphostructural integrity. We show that the robustness of islet oscillatory bursting depends on CCI, with all three functional measures <italic>f</italic>_b, <italic>λ </italic>and <italic>T</italic>_bincreasing monotonically with CCI when it is small, and plateau around CCI = 1.</p>", "<title>Conclusion</title>", "<p>CCI is a good islet function predictor. It has the potential of linking islet structure and function, and providing insight to identify therapeutic targets for the preservation and restoration of islet <italic>β</italic>-cell mass and function.</p>" ]

[ "<title>Abbreviations</title>", "<p>CCI: cluster coupling index; CSI: cluster synchronization index; HCP: hexagonal closest packing; SCP: simple cubic packing; T1D: type 1 diabetes; T2D: type 2 diabetes; ROI: region of interest; PSD: power spectral density.</p>", "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>AN and XW both contributed to the development of the modeling method. AN wrote the Matlab code and ran the simulation. Both contributed to the writing of the manuscript, read and approved the final manuscript.</p>" ]

[ "<title>Acknowledgements</title>", "<p>This work is supported in part by a special fund from Children's Hospital Foundation, Children's Research Institute of Wisconsin and Children's Hospital of Wisconsin. Most simulations were run on the cluster Zeke of the Computational Bioengineering group at MCW, courtesy of Dr. Dan Beard. We thank Gregg McQuestion for administration assistance with the cluster.</p>" ]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p>Cell sorting using Lomb-Scargle periodogram. (a) Calcium profiles, (b) membrane action potential profiles, and (c) Lomb-Scargle periodogram, of a burster cell, a spiker cell and a silent cell. The burster has a clear peak frequency at <italic>f </italic>= 33 mHz (0.033 sec^-1), whereas the spiker and silent cells have broad spectra. (d) Distribution of the principal peak <italic>p </italic>values. All cells with <italic>p </italic>&lt; 10^-12were plotted at <italic>p </italic>= 10^-12. The burster cells form a distinct group from others, with <italic>p </italic>&lt; 0.005 (dashed black line).</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Fraction of burster cells <italic>f</italic>_b, synchronization index <italic>λ</italic>, and bursting period <italic>T</italic>_bplotted for a HCP-323 <italic>β</italic>-cell cluster (a, c and e) and a SCP-343 cluster (b, d, and f) on the <italic>g</italic>_c-<italic>n</italic>_cplane. A clear hyperbolic relation is visible.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p>Islet functional measures versus CCI exhibiting potential ROI for therapy (shaded areas). (a) Fraction of burster cells <italic>f</italic>_b. (b) Synchronization Index <italic>λ</italic>. (c) Bursting period <italic>T</italic>_b.</p></caption></fig>", "<fig position=\"float\" id=\"F4\"><label>Figure 4</label><caption><p>3D HCP and SCP cell clusters projected on a 2-dimensional <italic>x-y </italic>plane. (a) A HCP-323 cluster with edge size 5. Each cell is connected with <italic>n</italic>_c= 12 neighbors, 6 from the same layer and 6 from the layers above and below. (b) A conventional SCP 7 × 7 × 7 cluster with <italic>n</italic>_c= 6 for each cell.</p></caption></fig>", "<fig position=\"float\" id=\"F5\"><label>Figure 5</label><caption><p>Simulation schema.</p></caption></fig>", "<fig position=\"float\" id=\"F6\"><label>Figure 6</label><caption><p>The flowchart of our cell sorting algorithm. Intra-cellular calcium signals, <italic>Ca</italic>(<italic>t</italic>) are passed to a program which calculates the periodogram PSD and the probability of the peak PSD values. If <italic>p </italic>&lt; 0.005, cell is sorted as a burster cell. Among non-bursters, cells which satisfy the condition Δ<italic>V </italic>= |<italic>V</italic>_max- <italic>V</italic>_min| &lt; 30 mV are considered silent cells, and the rest as spiker cells.</p></caption></fig>" ]

[]

[ "<disp-formula id=\"bmcM1\"><label>(1)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1742-4682-5-17-i1\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munder><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mtext>all cells coupled to </mml:mtext><mml:mi>i</mml:mi></mml:mrow></mml:munder><mml:mrow><mml:msub><mml:mi>g</mml:mi><mml:mtext>c</mml:mtext></mml:msub><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>V</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>=</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mtext>c</mml:mtext></mml:msub><mml:mo>⋅</mml:mo><mml:msub><mml:mi>g</mml:mi><mml:mtext>c</mml:mtext></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>×</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mover accent=\"true\"><mml:mi>V</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1742-4682-5-17-i2\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mover accent=\"true\"><mml:mi>V</mml:mi><mml:mo>¯</mml:mo></mml:mover><mml:mi>i</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mi>c</mml:mi></mml:msub></mml:mrow></mml:mfrac><mml:mstyle displaystyle=\"true\"><mml:munder><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mtext>all cells coupled to </mml:mtext><mml:mi>i</mml:mi></mml:mrow></mml:munder><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM2\"><label>(2)</label><italic>CCI </italic>= (<italic>n</italic>_c·<italic>g</italic>_c)/<italic>C</italic>₀</disp-formula>", "<disp-formula id=\"bmcM3\"><label>(3)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1742-4682-5-17-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable><mml:mtr><mml:mtd><mml:mrow><mml:msub><mml:mi>C</mml:mi><mml:mrow><mml:mi>m</mml:mi><mml:mo>,</mml:mo><mml:mi>i</mml:mi></mml:mrow></mml:msub><mml:mfrac><mml:mrow><mml:mi>d</mml:mi><mml:msub><mml:mi>V</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:mi>d</mml:mi><mml:mi>t</mml:mi></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mo>−</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:mi>C</mml:mi><mml:mi>a</mml:mi><mml:mo>,</mml:mo><mml:mi>i</mml:mi></mml:mrow></mml:msub><mml:mo>+</mml:mo><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mrow><mml:mtext>ATP</mml:mtext></mml:mrow></mml:msub><mml:mo>,</mml:mo><mml:mi>i</mml:mi></mml:mrow></mml:msub><mml:mo>+</mml:mo><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:mi>K</mml:mi><mml:mo>,</mml:mo><mml:mi>i</mml:mi></mml:mrow></mml:msub><mml:mo>+</mml:mo><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:mi>S</mml:mi><mml:mo>,</mml:mo><mml:mi>i</mml:mi></mml:mrow></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>−</mml:mo><mml:mn>.....</mml:mn></mml:mrow></mml:mtd></mml:mtr><mml:mtr><mml:mtd><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munder><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mtext>all cells coupled to </mml:mtext><mml:mi>i</mml:mi></mml:mrow></mml:munder><mml:mrow><mml:msub><mml:mi>g</mml:mi><mml:mtext>c</mml:mtext></mml:msub><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>V</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM4\"><label>(4)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1742-4682-5-17-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable columnalign=\"left\"><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mrow><mml:mtext>ATP</mml:mtext></mml:mrow></mml:msub></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:msub><mml:mi>g</mml:mi><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mrow><mml:mtext>ATP</mml:mtext></mml:mrow></mml:msub></mml:mrow></mml:msub><mml:msub><mml:mi>O</mml:mi><mml:mrow><mml:msub><mml:mi>K</mml:mi><mml:mrow><mml:mtext>ATP</mml:mtext></mml:mrow></mml:msub></mml:mrow></mml:msub><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>V</mml:mi><mml:mo>−</mml:mo><mml:msub><mml:mi>V</mml:mi><mml:mi>K</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:mi>C</mml:mi><mml:mi>a</mml:mi></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:msub><mml:mi>g</mml:mi><mml:mrow><mml:mi>C</mml:mi><mml:mi>a</mml:mi></mml:mrow></mml:msub><mml:mo>⋅</mml:mo><mml:msub><mml:mi>m</mml:mi><mml:mi>∞</mml:mi></mml:msub><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>V</mml:mi><mml:mo>−</mml:mo><mml:msub><mml:mi>V</mml:mi><mml:mrow><mml:mi>C</mml:mi><mml:mi>a</mml:mi></mml:mrow></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:msub><mml:mi>I</mml:mi><mml:mi>K</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:msub><mml:mi>g</mml:mi><mml:mi>K</mml:mi></mml:msub><mml:mo>⋅</mml:mo><mml:mi>n</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>V</mml:mi><mml:mo>−</mml:mo><mml:msub><mml:mi>V</mml:mi><mml:mi>K</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:msub><mml:mi>I</mml:mi><mml:mi>S</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:msub><mml:mi>g</mml:mi><mml:mi>S</mml:mi></mml:msub><mml:mo>⋅</mml:mo><mml:mi>s</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>V</mml:mi><mml:mo>−</mml:mo><mml:msub><mml:mi>V</mml:mi><mml:mi>K</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM5\"><label>(5)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1742-4682-5-17-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtable columnalign=\"left\"><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:mfrac><mml:mrow><mml:mi>d</mml:mi><mml:mi>n</mml:mi></mml:mrow><mml:mrow><mml:mi>d</mml:mi><mml:mi>t</mml:mi></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>τ</mml:mi><mml:mi>n</mml:mi></mml:msub></mml:mrow></mml:mfrac><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mi>∞</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>n</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mtd></mml:mtr><mml:mtr columnalign=\"left\"><mml:mtd columnalign=\"left\"><mml:mrow><mml:mfrac><mml:mrow><mml:mi>d</mml:mi><mml:mi>s</mml:mi></mml:mrow><mml:mrow><mml:mi>d</mml:mi><mml:mi>t</mml:mi></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>τ</mml:mi><mml:mi>s</mml:mi></mml:msub></mml:mrow></mml:mfrac><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>s</mml:mi><mml:mi>∞</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>s</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mtd></mml:mtr></mml:mtable></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1742-4682-5-17-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>m</mml:mi><mml:mi>∞</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mi>exp</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mrow><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>m</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>V</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mo>/</mml:mo><mml:mrow><mml:msub><mml:mi>θ</mml:mi><mml:mi>m</mml:mi></mml:msub></mml:mrow></mml:mrow></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1742-4682-5-17-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mi>∞</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mi>exp</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mrow><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>n</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>V</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mo>/</mml:mo><mml:mrow><mml:msub><mml:mi>θ</mml:mi><mml:mi>n</mml:mi></mml:msub></mml:mrow></mml:mrow></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1742-4682-5-17-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>s</mml:mi><mml:mi>∞</mml:mi></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mi>exp</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mrow><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>s</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>V</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mo>/</mml:mo><mml:mrow><mml:msub><mml:mi>θ</mml:mi><mml:mi>s</mml:mi></mml:msub></mml:mrow></mml:mrow></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM6\"><label>(6)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1742-4682-5-17-i9\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mfrac><mml:mrow><mml:mi>d</mml:mi><mml:msub><mml:mrow><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>C</mml:mi><mml:msup><mml:mi>a</mml:mi><mml:mrow><mml:mn>2</mml:mn><mml:mo>+</mml:mo></mml:mrow></mml:msup><mml:mo stretchy=\"false\">]</mml:mo></mml:mrow><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:mi>d</mml:mi><mml:mi>t</mml:mi></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mi>f</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mo>−</mml:mo><mml:msub><mml:mi>α</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:msub><mml:mi>I</mml:mi><mml:mrow><mml:mi>C</mml:mi><mml:mi>a</mml:mi><mml:mo>,</mml:mo><mml:mi>i</mml:mi></mml:mrow></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>k</mml:mi><mml:mrow><mml:mi>C</mml:mi><mml:mi>a</mml:mi><mml:mo>,</mml:mo><mml:mi>i</mml:mi></mml:mrow></mml:msub><mml:msub><mml:mrow><mml:mo stretchy=\"false\">[</mml:mo><mml:mi>C</mml:mi><mml:msup><mml:mi>a</mml:mi><mml:mrow><mml:mn>2</mml:mn><mml:mo>+</mml:mo></mml:mrow></mml:msup><mml:mo stretchy=\"false\">]</mml:mo></mml:mrow><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM7\"><label>(7)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M10\" name=\"1742-4682-5-17-i10\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtext>Nbr</mml:mtext><mml:mo stretchy=\"false\">(</mml:mo><mml:mtext>Cell </mml:mtext><mml:mi>i</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mrow><mml:mo>{</mml:mo><mml:mrow><mml:mtext>Cell </mml:mtext><mml:mi>j</mml:mi><mml:mrow><mml:mo>|</mml:mo><mml:mrow><mml:msqrt><mml:mrow><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>x</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>x</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>y</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>y</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:msup><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:msub><mml:mi>z</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>z</mml:mi><mml:mi>j</mml:mi></mml:msub><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:msqrt><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow></mml:mrow></mml:mrow><mml:mo>}</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></disp-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M11\" name=\"1742-4682-5-17-i11\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munder><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>=</mml:mo><mml:mtext>all cells coupled to </mml:mtext><mml:mi>i</mml:mi></mml:mrow></mml:munder><mml:mrow><mml:msub><mml:mi>g</mml:mi><mml:mtext>c</mml:mtext></mml:msub><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>V</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:msub><mml:mi>V</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M12\" name=\"1742-4682-5-17-i12\" overflow=\"scroll\"><mml:semantics><mml:mover accent=\"true\"><mml:mi>y</mml:mi><mml:mo>˜</mml:mo></mml:mover></mml:semantics></mml:math></inline-formula>", "<disp-formula id=\"bmcM8\"><label>(8)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M13\" name=\"1742-4682-5-17-i13\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>P</mml:mi><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>ω</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mi>σ</mml:mi><mml:mn>2</mml:mn></mml:msup></mml:mrow></mml:mfrac><mml:mrow><mml:mo>{</mml:mo><mml:mrow><mml:mfrac><mml:mrow><mml:msup><mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>y</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mover accent=\"true\"><mml:mi>y</mml:mi><mml:mo>˜</mml:mo></mml:mover></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>cos</mml:mi><mml:mo>⁡</mml:mo><mml:mi>ω</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:msup><mml:mrow><mml:mi>cos</mml:mi><mml:mo>⁡</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mi>ω</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac><mml:mo>+</mml:mo><mml:mfrac><mml:mrow><mml:msup><mml:mrow><mml:mrow><mml:mo>[</mml:mo><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>y</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mover accent=\"true\"><mml:mi>y</mml:mi><mml:mo>˜</mml:mo></mml:mover></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mi>sin</mml:mi><mml:mo>⁡</mml:mo><mml:mi>ω</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow><mml:mo>]</mml:mo></mml:mrow></mml:mrow><mml:mn>2</mml:mn></mml:msup></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:msup><mml:mrow><mml:mi>sin</mml:mi><mml:mo>⁡</mml:mo></mml:mrow><mml:mn>2</mml:mn></mml:msup><mml:mi>ω</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mi>τ</mml:mi></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow><mml:mo>}</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM9\"><label>(9)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M14\" name=\"1742-4682-5-17-i14\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>tan</mml:mi><mml:mo>⁡</mml:mo><mml:mo stretchy=\"false\">(</mml:mo><mml:mn>2</mml:mn><mml:mi>ω</mml:mi><mml:mi>τ</mml:mi><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:mi>sin</mml:mi><mml:mo>⁡</mml:mo><mml:mn>2</mml:mn><mml:mi>ω</mml:mi><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow><mml:mrow><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>=</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mi>N</mml:mi></mml:munderover><mml:mrow><mml:mi>cos</mml:mi><mml:mo>⁡</mml:mo><mml:mn>2</mml:mn><mml:mi>ω</mml:mi><mml:msub><mml:mi>t</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM10\"><label>(10)</label><italic>p </italic>= 1 - (1 - <italic>e</italic>^{-max(<italic>P</italic>(<italic>ω</italic>))})^{<italic>M</italic>}</disp-formula>", "<disp-formula id=\"bmcM11\"><label>(11)</label>Φ(<italic>t</italic>_{<italic>k</italic>}) = arg ∑ exp (<italic>iφ</italic>_{<italic>j</italic>}(<italic>t</italic>_{<italic>k</italic>}))</disp-formula>", "<disp-formula id=\"bmcM12\"><label>(12)</label><italic>ρ</italic>_{<italic>j </italic>}= |⟨ exp(<italic>iφ</italic>_{<italic>j</italic>}(<italic>t</italic>_k) - Φ(<italic>t</italic>_{<italic>k</italic>}))⟩|</disp-formula>", "<disp-formula id=\"bmcM13\"><label>(13)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M15\" name=\"1742-4682-5-17-i15\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mtext>CSI</mml:mtext><mml:mo>=</mml:mo><mml:mrow><mml:mo>〈</mml:mo><mml:mrow><mml:msub><mml:mi>ρ</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow><mml:mo>〉</mml:mo></mml:mrow><mml:mo>=</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mi>β</mml:mi></mml:msub></mml:mrow></mml:mfrac><mml:mstyle displaystyle=\"true\"><mml:munder><mml:mo>∑</mml:mo><mml:mi>j</mml:mi></mml:munder><mml:mrow><mml:msub><mml:mi>ρ</mml:mi><mml:mi>j</mml:mi></mml:msub></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></disp-formula>", "<disp-formula id=\"bmcM14\"><label>(14)</label><italic>λ</italic>_{<italic>j,k </italic>}= |⟨ exp(<italic>i</italic>(<italic>φ</italic>_{<italic>j</italic>}(<italic>t</italic>) - <italic>φ</italic>_{<italic>k</italic>}(<italic>t</italic>))⟩|</disp-formula>", "<disp-formula id=\"bmcM15\"><label>(15)</label><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M16\" name=\"1742-4682-5-17-i16\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mi>λ</mml:mi><mml:mo>=</mml:mo><mml:mrow><mml:mo>〈</mml:mo><mml:mrow><mml:msub><mml:mi>λ</mml:mi><mml:mrow><mml:mi>j</mml:mi><mml:mo>,</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub></mml:mrow><mml:mo>〉</mml:mo></mml:mrow><mml:mo>=</mml:mo><mml:mfrac><mml:mn>2</mml:mn><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mi>β</mml:mi></mml:msub><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mi>β</mml:mi></mml:msub><mml:mo>−</mml:mo><mml:mn>1</mml:mn></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:mfrac><mml:mstyle displaystyle=\"true\"><mml:munderover><mml:mo>∑</mml:mo><mml:mrow><mml:mi>j</mml:mi><mml:mo>,</mml:mo><mml:mi>k</mml:mi><mml:mo>&gt;</mml:mo><mml:mi>j</mml:mi></mml:mrow><mml:mrow><mml:msub><mml:mi>n</mml:mi><mml:mi>β</mml:mi></mml:msub></mml:mrow></mml:munderover><mml:mrow><mml:msub><mml:mi>λ</mml:mi><mml:mrow><mml:mi>j</mml:mi><mml:mo>,</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub></mml:mrow></mml:mstyle></mml:mrow></mml:semantics></mml:math></disp-formula>" ]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1742-4682-5-17-1\"/>", "<graphic xlink:href=\"1742-4682-5-17-2\"/>", "<graphic xlink:href=\"1742-4682-5-17-3\"/>", "<graphic xlink:href=\"1742-4682-5-17-4\"/>", "<graphic xlink:href=\"1742-4682-5-17-5\"/>", "<graphic xlink:href=\"1742-4682-5-17-6\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The patellofemoral (PF) joint consists of the distal femur and patella. PF pain syndrome – also known as anterior knee pain or chondromalacia – is very common and is widely believed to be caused by abnormal motion of the patella relative to the femur (often referred to as patellar tracking). Abnormal patellar tracking is thought to alter the mechanical interaction between the patella and femur, and may progress to cartilage degeneration and osteoarthritis.</p>", "<p>Accurately measuring <italic>in-vivo</italic> PF joint motion remains a significant challenge. PF joint motion has been measured in cadaver specimens using electromagnetic sensors [##REF##9021506##1##, ####REF##1513139##2##, ##REF##9418625##3##, ##REF##8914726##4####8914726##4##], three-dimensional (3D) video analysis of markers [##REF##7706335##5##,##REF##10843125##6##], x-ray stereophotogrammetry [##REF##8119024##7##,##REF##2324856##8##], goniometers [##REF##10211457##9##], and coordinate measuring machines [##REF##10716285##10##]. These studies have provided valuable insight into factors that may influence patellar tracking, but cadaveric experiments are unable to duplicate the <italic>in-vivo</italic> motions, forces, or muscle firing patterns common to live human subjects. <italic>In-vivo</italic> studies of PF joint motion have traditionally relied upon static two-dimensional (2D) radiographs [##REF##8231783##11##, ####REF##7573659##12##, ##REF##4433362##13##, ##REF##500753##14##, ##REF##12016081##15####12016081##15##], 2D video digital fluoroscopy [##REF##8352119##16##], intracortical bone pins [##REF##1583014##17##,##REF##1517258##18##], x-ray photogrammetry [##REF##489632##19##], electromagnetic sensors [##REF##15664880##20##], static CT [##REF##3956015##21##,##REF##7584173##22##], and static MRI [##REF##15958222##23##, ####REF##15288456##24##, ##REF##2929285##25##, ##REF##15111080##26####15111080##26##]. While these studies have also provided helpful information about patellar tracking, static analyses can not quantify PF joint function during dynamic activities, 2D analyses are incapable of capturing the complex 3D relationship of the patella relative to the femur, and bone pins [##REF##1583014##17##,##REF##1517258##18##] limit the number of willing volunteers and make serial studies over time impractical since bone pins can not be reliably reattached in the exact location.</p>", "<p>More recently, dynamic MRI-based techniques have grown in popularity as a tool for measuring PF joint motion under <italic>in-vivo</italic> conditions. These techniques – which have been described by various names, including kinematic MRI [##REF##14669963##27##, ####REF##9626893##28##, ##REF##8327718##29####8327718##29##], cine phase contrast MRI [##REF##9596534##30##,##REF##10633267##31##], motion-triggered cine MRI [##REF##8451415##32##] or fast phase contrast MRI [##REF##12541228##33##]-acquire a series of MR images as the subject performs a periodic knee motion activity (typically flexion and extension), with each MR image acquired at a unique phase of the knee motion cycle. Thus, multiple motion cycles are required to assemble the MR images necessary to represent a single motion trial. Dynamic MRI techniques that rely upon conventional closed bore scanners are limited by the physical dimensions of the scanner. Specifically, these scanners do not allow for activities that replicate the forces and ranges of motion that produce symptoms for patients with PF pain syndrome. Furthermore, this approach implicitly assumes that there is relatively little variability in knee motion patterns between successive motion cycles.</p>", "<p>Additional techniques for assessing <italic>in-vivo</italic> PF joint motion have included dynamic CT imaging [##REF##9207527##34##] and single-plane fluoroscopic imaging combined with shape matching [##REF##16121540##35##]. Dynamic CT imaging has limitations similar to those associated with dynamic MRI. The single-plane fluoroscopic technique is a promising approach that has achieved reasonable levels of theoretical accuracy, but has yet to be validated [##REF##16121540##35##].</p>", "<p>To overcome the limitations associated with existing methods for measuring PF joint motion, our laboratory has developed a new model-based tracking technique for measuring <italic>in-vivo</italic> 3D joint motion. The purpose of the study was to assess the accuracy of this model-based tracking technique for <italic>in-vivo</italic> PF joint motion by comparing the model-based technique to an accurate radiostereometric analysis (RSA) technique that measures joint motion by tracking the position of implanted tantalum beads [##REF##12751286##36##].</p>" ]

[ "<title>Methods</title>", "<title>Overview</title>", "<p>We have developed a CT model-based technique for accurately measuring <italic>in-vivo</italic> joint motion from biplane x-ray images. Specific details of this technique, which tracks the position of bones by maximizing the correlation between biplane x-ray images and digitally reconstructed radiographs (DRRs), have been published previously [##REF##16813452##37##]. To validate this new technique, we implanted small beads into the patella and femur of three cadaver knee specimens, recorded biplane radiographic images while manually flexing and extending the leg, measured the position of the patella and femur using model-based tracking, measured the position of the patella and femur with dynamic RSA [##REF##12751286##36##] – our \"gold standard\" – and then compared the results of the two techniques.</p>", "<title>Specimen preparation</title>", "<p>Three 1.6 mm diameter tantalum beads were implanted into both the patella and femur of three intact lower limbs from two cadaver specimens (72/male, 89/female). The quadriceps tendon was exposed through a 50 mm skin incision and sutured with nylon cord. The nylon cord was then placed between the skin and quadriceps muscle so that simulated muscle forces could be directed in a physiologic direction parallel to the femur's long axis. The tibia was secured to a custom testing apparatus with the leg inverted, i.e., with the femur hanging passively below the tibia (Figure ##FIG##0##1##). Although knee flexion is most often accomplished with the tibia rotating relative to a fixed femur, this experimental setup resulted in both the femur and patella moving relative to a fixed tibia and thus resulting in a more challenging assessment of PF joint motion. The specimen was then positioned with the knee centered in a biplane x-ray system [##REF##12751286##36##].</p>", "<title>Testing procedures</title>", "<p>Biplane x-ray images were acquired while manually flexing the knee from full extension (i.e., approximately 10° flexion) to approximately 90° of flexion with respect to the femoral and tibial long axes. Knee motion was achieved by manually pulling the nylon cord attached to the quadriceps tendon to cyclically flex and extend the knee. Given that accuracy was assessed by applying both the model-based tracking and dynamic RSA techniques to each trial, it was not necessary to accurately replicate <italic>in-vivo</italic> conditions (i.e., joint motion, muscle forces, or joint contact forces) or insure the repeatability of testing conditions between trials. The biplane x-ray images were acquired at 60 frames per second for 1.5 seconds with the x-ray generators in pulsed mode (70 kV, 320 mA) and video cameras shuttered at 1/500 s to eliminate motion blur. For each specimen, we acquired biplane x-ray images of four flexion-extension trials and two static trials.</p>", "<p>Following testing, we obtained axial CT images of each knee using a LightSpeed VCT (GE Medical Systems). The CT data set had 0.625 mm slice spacing and an in-plane resolution of approximately 0.4 mm/pixel. The femur and patella were segmented from surrounding bones and soft tissues (ImageJ 1.32 j, <ext-link ext-link-type=\"uri\" xlink:href=\"http://rsb.info.nih.gov/ij\"/>) and then rescaled with a feature-based interpolation technique that resulted in a 3D bone model with voxel dimensions similar to the biplane x-ray image pixel size.</p>", "<title>Model-based tracking</title>", "<p>The 3D positions and orientations of the patella and femur were measured from the biplane x-ray images using a technique referred to as model-based tracking. Briefly, this technique applies a ray-tracing algorithm to project a pair of digitally reconstructed radiographs (DRRs) from the CT-based bone model. The <italic>in-vivo</italic> position and orientation of a bone is estimated by maximizing the correlation between the DRRs and the biplane x-ray images. Using this technique, the 3D position and orientation of the patella and femur were determined independently for all frames of each trial. The final step involved determining the position of the tantalum beads within the CT bone model and then expressing their 3D position relative to a laboratory coordinate system.</p>", "<title>Dynamic RSA</title>", "<p>For comparison, the 3D position of each implanted tantalum bead was also determined from the biplane images using a previously validated and well-established dynamic RSA technique [##REF##12751286##36##]. This process determined the 3D location of each implanted tantalum bead relative to the laboratory coordinate system to an accuracy of within ± 0.1 mm. These data enabled a direct comparison with the model-based tracking results.</p>", "<title>Kinematics</title>", "<p>PF joint kinematics were determined using transformations between each bone's 3D position and orientation (determined from the model-based tracking and dynamic RSA results) and anatomical axes determined from the CT bone model. Specifically, patellar motion was quantified in terms of shift (i.e., medial-lateral translation relative to the femur), flexion (i.e., rotation about a medial-lateral axis relative to the femur), tilt (i.e., rotation about the patella's long axis), and rotation (i.e., angular position relative to the patella's anterior-posterior axis) [##REF##12012037##38##]. These four parameters are believed to represent the most clinically relevant motion variables. For completeness, anterior-posterior translation and superior-inferior translation of the patella relative to the femur were also measured, even though these two translations are less meaningful from a clinical perspective.</p>", "<title>Comparison of techniques</title>", "<p>Accuracy of the model-based tracking technique was quantified in terms of bias and precision [##UREF##0##39##]. Measurement bias was defined as the average difference between the two techniques. Precision was defined as the standard deviation of the model-based tracking results when applied to only the static trials. Thus, any frame-to-frame variability in measurement error when no motion occurred provided an estimate of the precision of the model-based tracking technique. In addition, to provide a single measurement of accuracy, we assessed the overall dynamic accuracy by calculating the RMS error between the two measurement techniques. These measures of accuracy (i.e., bias, precision, overall dynamic accuracy) were first computed using the 3D position of the implanted tantalum beads as reported by both the model-based and dynamic RSA measurement techniques. This allowed us to assess the amount of error associated with the tracking of each bone. These three measures of accuracy were also calculated for each of the six kinematic measurements (i.e., three translations, three rotations).</p>" ]

[ "<title>Results</title>", "<p>There was very high agreement between the results from the model-based tracking and RSA techniques (Figure ##FIG##1##2##). In comparing the position of the implanted tantalum beads, bias ranged from -0.174 to 0.248 mm (depending on coordinate direction), precision ranged from 0.023 to 0.062 mm, and overall dynamic accuracy was better than 0.335 mm (Table ##TAB##0##1##). When the results were compared using kinematic parameters, bias ranged from -0.293 to 0.320 mm for the three translational parameters (patellar shift, anterior-posterior translation, proximal-distal translation, Table ##TAB##1##2##) and ranged from -0.090° to 0.475° for the three rotational parameters (flexion, tilt, rotation, Table ##TAB##1##2##). Precision ranged from 0.042 to 0.114 mm for the three translational parameters and ranged from 0.216° to 0.382° for the three rotational parameters. Overall dynamic accuracy was better than 0.395 mm for the three translational measurements, and better than 0.877° for the rotational measurements (Table ##TAB##1##2##).</p>" ]

[ "<title>Discussion</title>", "<p>Accurately measuring PF joint motion is important for understanding, among other things, the effect of conservative and surgical treatment of PF pain syndrome. A previous study that compared patellar tracking patterns between subjects with PF pain and subjects without PF pain reported average differences of approximately 5° in patellar tilt and approximately 4% in patellar offset, i.e., the percentage of the patella lateral to the midline [##REF##11002432##40##]. Assuming an average patellar width of 46 mm [##REF##15902866##41##], this 4% patellar offset corresponds to an estimated difference in patellar translation of approximately 2 mm. Thus, it is reasonable to presume that a system for measuring patellar tracking should be able to detect differences between subject populations in patellar tracking of less than 2 mm and 5°. Using the general rule that a measurement system should ideally have an accuracy that is an order of magnitude better than the smallest change you expect to measure, these data suggest that the patellar tracking technique should have an accuracy of approximately ± 0.2 mm for translations and ± 0.5° for rotations. Although the technique reported here falls short of this ideal accuracy goal, it is still four to five times more accurate than the smallest differences we would hope to detect (i.e., 2 mm of translation and 5° of rotation). From a statistical standpoint, if we assumed that all the variability within a group of subjects was due solely to measurement technique inaccuracy, then the sample size required to detect differences of 2 mm of patellar translation with a measurement system of \"ideal\" accuracy (i.e., ± 0.2 mm of error) would be 2 subjects (based on a t-test and assuming α = 0.05 and β = 0.2). In contrast, only one additional subject would be required to detect differences of 2 mm with the accuracy of the model-based tracking system reported here (i.e., ± 0.395 mm). However, since previously reported data indicates that inter-subject variability in measured knee kinematics is approximately 10 to 30 times greater than the inaccuracies associated with the measurement system reported here [##REF##11002432##40##], the authors are comfortable that the technique reported here is still within an acceptable accuracy range for detecting clinically significant differences in PF joint motion.</p>", "<p>Although a number of techniques for measuring <italic>in-vivo</italic> PF joint motion have been previously reported, the accuracy of these techniques is reported far less frequently. For example, Rebmann and Sheehan compared three cine phase contrast MR imaging protocols for measuring <italic>in-vivo</italic> knee kinematics in terms of precision and subject inter-exam variability, but did not report any explicit measures of accuracy [##REF##12541228##33##]. Similarly, Powers and colleagues have published extensively on PF joint motion and have presented measures of repeatability [##REF##9626893##28##,##REF##12112505##42##], but the authors are not aware of any report that explicitly describes the 3D accuracy of their MRI-based measurement technique. Although these measures of repeatability provide some insight into the suitability of a measurement technique – especially in contrast to studies that fail to report any measures of accuracy or reliability [##REF##8451415##32##,##REF##9207527##34##,##REF##8708048##43##] – it is important to remember that repeatability should not be confused with accuracy. Systematic errors can cause poor data accuracy, but would not necessarily affect repeatability.</p>", "<p>In contrast, several authors have carefully determined the accuracy of their techniques for measuring <italic>in-vivo</italic> PF joint motion. For example, Sheehan and colleagues used a gear-driven phantom object to assess the 3D accuracy of cine phase contrast MRI for measuring joint motion [##REF##9596534##30##,##REF##10633267##31##]. These data indicated average absolute tracking errors of less than ± 0.7 mm for in-plane motions, and slightly higher (up to 1.8 mm) of error for out-of-plane motions. Fregly and colleagues provided a rigorous theoretical accuracy assessment model-based tracking technique applied to single-plane fluoroscopic images [##REF##16121540##35##]. The authors reported good measures of accuracy (e.g., bias less than 0.75 mm and 0.4°, though precision as high as ± 4 mm and ± 1.8°) with their flat-shading technique. However, these values are from a theoretical study where all other sources of error were eliminated and it is not yet known if this level of accuracy can be achieved under experimental conditions.</p>", "<p>We believe that it is necessary to conduct a validation study for each anatomical joint to which we intend to apply the model-based tracking technique. Stated another way, we believe that it would be highly inappropriate to validate this technique for, say, the glenohumeral joint and then assume that the accuracy levels obtained in that particular validation study could be assumed to be the same for every other anatomical joint. This belief is based on the fact that the factors influencing the accuracy of the model-based technique are not the same for all anatomical joints, and that the conditions for conducting validation studies should as much as possible resemble actual <italic>in-vivo</italic> testing conditions. The specific factors influencing the accuracy of this technique include the 3D shape of a particular bone, the amount of \"internal\" bone information (i.e., variability in bone density and/or the presence of bone edges that appear in an x-ray image but do not necessarily contribute to the outline of a particular bone in all joint positions, Figure ##FIG##2##3##), the presence of surrounding soft tissues, overlap from surrounding bones, the magnitude of joint motion, and the velocity of joint motion. Although we have not yet assessed the relative influence of each of these factors to model-based tracking accuracy, this list of factors comes from first-hand experience with the technique.</p>", "<p>The advantages of this technique of combining model-based tracking with biplane x-ray imaging is that it provides accurate, 3D, non-invasive measures of PF joint motion during functional activities that are known to produce symptoms for patients diagnosed with PF pain syndrome (e.g., normal gait, stair climbing/descending). There are two primary disadvantages to this technique. The first is the that the amount of x-ray exposure associated with the CT scan and biplane x-ray imaging limits the number of trials that can be performed. However, all testing procedures have been approved by both the Institutional Review Board and the Radiation Safety Committee at Henry Ford Hospital. The second disadvantage is that the field of view is limited to the biplane x-ray system's 3D imaging volume, i.e., the region defined by the intersecting x-ray beams. Although this limitation prevents us from collecting biplane x-ray images during an entire gait cycle, we still can collect information for the vast majority of the stance phase when the muscle forces, joint forces, and pain are the highest. Another limitation of this study is that accuracy of this measurement technique was not explicitly assessed at knee flexion angles greater than approximately 90°.</p>", "<p>In summary, this model-based tracking approach is a non-invasive technique for accurately measuring <italic>in-vivo</italic> PF joint motion during dynamic activities. The results indicate that model-based tracking can measure <italic>in-vivo</italic> motion of the patella to within 0.455 mm and 0.987°. The technique achieves a level of accuracy that is necessary and sufficient for addressing clinically relevant questions regarding PF joint function. Future research will use this technique to analyze the effects of conservative and surgical treatment of PF pain syndrome.</p>" ]

[]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Accurately measuring <italic>in-vivo</italic> motion of the knee's patellofemoral (PF) joint is challenging. Conventional measurement techniques have largely been unable to accurately measure three-dimensional, <italic>in-vivo</italic> motion of the patella during dynamic activities. The purpose of this study was to assess the accuracy of a new model-based technique for measuring PF joint motion.</p>", "<title>Methods</title>", "<p>To assess the accuracy of this technique, we implanted tantalum beads into the femur and patella of three cadaveric knee specimens and then recorded dynamic biplane radiographic images while manually flexing and extending the specimen. The position of the femur and patella were measured from the biplane images using both the model-based tracking system and a validated dynamic radiostereometric analysis (RSA) technique. Model-based tracking was compared to dynamic RSA by computing measures of bias, precision, and overall dynamic accuracy of four clinically-relevant kinematic parameters (patellar shift, flexion, tilt, and rotation).</p>", "<title>Results</title>", "<p>The model-based tracking technique results were in excellent agreement with the RSA technique. Overall dynamic accuracy indicated errors of less than 0.395 mm for patellar shift, 0.875° for flexion, 0.863° for tilt, and 0.877° for rotation.</p>", "<title>Conclusion</title>", "<p>This model-based tracking technique is a non-invasive method for accurately measuring dynamic PF joint motion under <italic>in-vivo</italic> conditions. The technique is sufficiently accurate in measuring clinically relevant changes in PF joint motion following conservative or surgical treatment.</p>" ]

[ "<title>Competing interests</title>", "<p>The authors declare that they have no competing interests.</p>", "<title>Authors' contributions</title>", "<p>MJB designed this study, participated in the data collection and analysis, and drafted the manuscript. SKK participated in the data collection and analysis. ST participated in study design and data analysis. RZ developed the data analysis software. All authors read and approved the final manuscript.</p>" ]

[]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Experimental testing configuration.</bold> The tibia of each cadaveric leg specimen was rigidly attached to a custom testing fixture, with the leg suspended within the biplane x-ray system in an inverted position. The quadriceps tendon was sutured with nylon cord so that simulated muscle forces could be applied. These manually applied forces flexed the knee from full extension to approximately 80° of flexion at a rate of approximately 60° per second.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p>Single-frame model-based tracking solution for the femur (top) and patella (bottom). In each image, the two digitally reconstructed radiographs (DRRs) – i.e., the highlighted bones in each image – are superimposed over the original biplane x-ray images in the position and orientation that maximized the correlation between the DRRs and biplane images.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>The model-based tracking technique relies upon:</bold> A) internal information such as subtle differences in bone density and/or B) the presence of bone edges in an x-ray image that do not necessarily contribute to the outline of a particular bone in all joint positions.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Accuracy of the model-based technique for tracking the patella and femur was expressed in terms of bias and precision as mean ± standard deviation.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"2\">Bias</td><td align=\"center\" colspan=\"2\">Precision</td><td align=\"center\" colspan=\"2\">Overall Dynamic Accuracy</td></tr><tr><td align=\"center\">Axis</td><td align=\"center\">Patella</td><td align=\"center\">Femur</td><td align=\"center\">Patella</td><td align=\"center\">Femur</td><td align=\"center\">Patella</td><td align=\"center\">Femur</td></tr></thead><tbody><tr><td align=\"center\">X</td><td align=\"center\">-0.014 ± 0.133</td><td align=\"center\">0.207 ± 0.099</td><td align=\"center\">0.061 ± 0.027</td><td align=\"center\">0.049 ± 0.011</td><td align=\"center\">0.220 ± 0.044</td><td align=\"center\">0.234 ± 0.064</td></tr><tr><td align=\"center\">Y</td><td align=\"center\">-0.174 ± 0.114</td><td align=\"center\">-0.022 ± 0.125</td><td align=\"center\">0.062 ± 0.028</td><td align=\"center\">0.038 ± 0.005</td><td align=\"center\">0.211 ± 0.035</td><td align=\"center\">0.149 ± 0.048</td></tr><tr><td align=\"center\">Z</td><td align=\"center\">0.248 ± 0.158</td><td align=\"center\">0.218 ± 0.099</td><td align=\"center\">0.042 ± 0.007</td><td align=\"center\">0.023 ± 0.004</td><td align=\"center\">0.335 ± 0.127</td><td align=\"center\">0.276 ± 0.062</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Accuracy of the model-based technique (RMS errors, mean ± standard deviation) expressed in kinematic parameters that describe motion of the patella relative to the femur.</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td align=\"center\">Kinematic Parameter</td><td align=\"center\">Bias</td><td align=\"center\">Precision</td><td align=\"center\">Overall Dynamic Accuracy</td></tr></thead><tbody><tr><td align=\"center\">Shift (med/lat translation)</td><td align=\"center\">0.320 ± 0.105 mm</td><td align=\"center\">0.114 ± 0.039 mm</td><td align=\"center\">0.395 ± 0.079 mm</td></tr><tr><td align=\"center\">Anterior/posterior translation</td><td align=\"center\">-0.293 ± 0.201 mm</td><td align=\"center\">0.042 ± 0.011 mm</td><td align=\"center\">0.340 ± 0.162 mm</td></tr><tr><td align=\"center\">Superior/inferior translation</td><td align=\"center\">-0.107 ± 0.312 mm</td><td align=\"center\">0.058 ± 0.026 mm</td><td align=\"center\">0.315 ± 0.126 mm</td></tr><tr><td align=\"center\">Flexion</td><td align=\"center\">0.475 ± 0.420°</td><td align=\"center\">0.216 ± 0.139°</td><td align=\"center\">0.875 ± 0.237°</td></tr><tr><td align=\"center\">Tilt</td><td align=\"center\">-0.052 ± 0.651°</td><td align=\"center\">0.322 ± 0.214°</td><td align=\"center\">0.863 ± 0.156°</td></tr><tr><td align=\"center\">Rotation</td><td align=\"center\">-0.090 ± 0.290°</td><td align=\"center\">0.382 ± 0.239°</td><td align=\"center\">0.877 ± 0.090°</td></tr></tbody></table></table-wrap>" ]

[]

[]

[]

[]

[]

[]

[]

[ "<graphic xlink:href=\"1749-799X-3-38-1\"/>", "<graphic xlink:href=\"1749-799X-3-38-2\"/>", "<graphic xlink:href=\"1749-799X-3-38-3\"/>" ]

[]

{ "acronym": [], "definition": [] }

[ "<title>Background</title>", "<p>The network or directed graph description has become the preferred representation of the integrated activity of components of biological processes. The exponential growth of biological network data in the last five years has its source in recent advances in technologies such as mass spectrometry, genome-scale ChiP-chip experiments, yeast two-hybrid assays, combinatorial reverse genetic screens, and rapid literature mining techniques [##REF##16601728##1##].</p>", "<p>The science of systems biology has the aim of understanding the functional constraints and design principles of biological networks. Alon and colleagues were the first to introduce the notion of \"motifs\" in biological networks [##REF##11967538##2##,##REF##12399590##3##]. Motifs are small patterns observed to recur throughout a network, with frequencies statistically higher than expected in random networks of similar connectivity parameters. Since the introduction of this concept, motifs have been reported in many biological networks: metabolic, signaling pathway, protein-protein interaction, and ecological networks amongst others [##REF##11967538##2##, ####REF##12399590##3##, ##REF##12399584##4##, ##REF##15372033##5##, ##REF##16099987##6####16099987##6##]. Moreover, the prevalence of motifs is often considered as evidence for evolutionary selection, for implementing a <italic>specific </italic>function [##REF##11967538##2##,##REF##12399590##3##,##REF##14530388##7##]. Motifs are believed to be building blocks of the functional architecture of a biological network [##REF##12399590##3##].</p>", "<p>Consider for example the canonical set of motifs in transcription regulatory networks: Single input module (SIM), Multiple input module (MIM), and Feedforward loop (FFL) [##REF##12399590##3##]. (See Figure ##FIG##0##1##. Originally, Alon and colleagues [##REF##11967538##2##] proposed a <italic>dense overlapping regulon </italic>(DOR) as a motif; MIMs are special DORs that arose as a generalization of Bifan motif). Specific functions have been ascribed to each type of motif [##REF##11967538##2##,##REF##14530388##7##, ####REF##16406067##8##, ##REF##14607112##9##, ##REF##16729041##10##, ##REF##15107854##11####15107854##11##]: SIMs are commonly associated with temporal ordering of gene expression, MIMs with combinatorial gene regulation, and FFLs with filters that do not pass on transient signals [##REF##11967538##2##]. These functions depend not only on the topology of the subgraph, but on the logic at nodes receiving multiple inputs. The common occurrence of these motifs, relative to corresponding randomized graphs, has been taken as evidence for their selection for function.</p>", "<p>In this paper we investigate the role of small network subgraphs as building blocks of biological networks. We analysed several biological networks: transcription regulation networks of <italic>Saccharomyces cerevisiae </italic>under different physiological conditions, the transcription regulation network of <italic>Escherichia coli</italic>, and a neuronal signalling pathway network of the hippocampal CA1 neuron.</p>", "<p>Contrary to previous reports, we find that commonly accepted motifs are neither over- nor under-represented in these real networks in comparison to their random formulations. We discuss how the topology of biological networks automatically predisposes them to contain a certain distribution of motifs. This suggests that the evidence for the functional significance of motifs should be reevaluated.</p>" ]

[ "<title>Methods</title>", "<p>We use the transcription regulatory networks of <italic>Saccharomyces cerevisiae </italic>under various physiological conditions – composite, cell cycle, sporulation, diauxic shift, DNA damage, and stress response – published by Luscombe and coworkers [##REF##15372033##5##]. Their largest (composite) network contains 3459 nodes and 7014 interactions (<ext-link ext-link-type=\"uri\" xlink:href=\"http://networks.gersteinlab.org/regulation/dynamics/index2.html\"/>).</p>", "<p>To aid comparison of our work with that of Shen-Orr et al. [##REF##11967538##2##], we also use their <italic>Escherichia coli </italic>transcription network containing 424 nodes and 577 interactions (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.weizmann.ac.il/mcb/UriAlon/Network_motifs_in_coli/ColiNet-1.0/\"/>).</p>", "<p>Additionally, we use the neuronal signalling pathway network of the hippocampal CA1 neuron published by Máayan and colleagues, containing 594 nodes and 1422 interactions [##REF##16099987##6##] (<ext-link ext-link-type=\"uri\" xlink:href=\"http://www.mssm.edu/labs/iyengar/\"/>).</p>", "<p>We implemented Ullmann's algorithm for subgraph isomorphism [##UREF##0##12##] to enumerate fixed sized subgraph patterns (<italic>e.g</italic>. FFL, 3-cycle).</p>", "<p>In enumerating variable sized (maximal) subgraph patterns such as SIMs and MIMs, we used our algorithms described in [##REF##18433061##13##]. We note that Bifans are counted as MIMs with <italic>exactly </italic>two elements each in both parent and child sets. (See Definitions.)</p>", "<p>To generate random networks conserving the degree sequence of the real network, we use the method described by Shen-Orr et al. [##REF##11967538##2##]: Starting with the same number of nodes as in an original network, nodes in the random graph are assigned a specific number of in- and out-\"edge-stubs.\" Randomly chosen pairs of in- and out-edge-stubs are joined, giving rise to a random (directed) graph.</p>", "<title>Definitions</title>", "<p>A FFL is a set of three nodes (source, intermediate, and target) with one direct path, and another indirect path through an intermediate node, from source to target (See Figure ##FIG##0##1(a)##).</p>", "<p>A 3-cycle (3-CYC) is a three-node directed cyclic graph (Figure ##FIG##0##1(b)##).</p>", "<p>Single and multiple input modules (SIM and MIM) in a directed graph are <italic>maximal </italic>subgraphs comprising two non-empty disjoint sets (layers): and (standing for Parent and Child). By maximal we mean, for example, that each MIM is not contained in a larger MIM.</p>", "<p>A SIM requires that contain only one node and contain at least two nodes, such that the full graph contains an edge from the parent node to every <italic>c</italic>_{<italic>i </italic>}∈ . We also require that the <italic>indegree </italic>– number of incoming edges – of every <italic>c</italic>_{<italic>i </italic>}to be strictly equal to one: within the full network, not just within the subgraph. By this definition of a SIM, no edges can exist between any <italic>c</italic>_{<italic>i</italic>}, <italic>c</italic>_{<italic>j </italic>}∈ . It follows that is the only parent of all nodes in set .</p>", "<p>A MIM requires that both and must contain ≥ 2 nodes, that there is an edge from every <italic>p</italic>_{<italic>i </italic>}∈ to every <italic>c</italic>_{<italic>i </italic>}∈ , no edge between any <italic>p</italic>_{<italic>i</italic>}, <italic>p</italic>_{<italic>j </italic>}∈ , and no edge between any <italic>c</italic>_{<italic>i</italic>}, <italic>c</italic>_{<italic>j </italic>}∈ . A Bifan is a <italic>maximal </italic>MIM with and containing exactly 2 elements [##UREF##1##14##]. (Figure ##FIG##0##1(e)##)</p>", "<p>We note that in counting both SIMs and MIMs, we ignore self-edges.</p>", "<p>We emphasize that we impose the criterion of <italic>maximality </italic>when enumerating SIMs and MIMs. In case of SIM, the set is maximal, whereas with MIMs both and sets are maximal.</p>", "<p>These statements define the fundamental network motif set – FFL, SIM, and MIM – as, in a sense, \"orthogonal\": No subgraph can be more than one of the FFL, SIM, and MIM [##REF##18433061##13##].</p>" ]

[ "<title>Results</title>", "<p>We enumerated the occurrences of FFL, 3-CYC, SIM, MIM, and Bifan subgraph patterns (see Figure ##FIG##0##1##) in:</p>", "<p>1. the transcription networks of <italic>Saccharomyces cerevisiae </italic>(Yeast) under various physiological states [##REF##15372033##5##] (see Table ##TAB##0##1(a–f)##).</p>", "<p>2. the transcription network of <italic>Escherichia coli </italic>[##REF##11967538##2##] (see Table ##TAB##0##1(g)##), and</p>", "<p>3. the signalling pathway of hippocampal CA1 neuron [##REF##16099987##6##] (see Table ##TAB##0##1(h)##).</p>", "<p>For each network, 1000 random networks were generated conserving the degree sequence of the original network. Comparisons were made between the frequencies of appearances of various patterns in the real network, and the means and dispersions of their appearances in corresponding random networks.</p>", "<p>Table ##TAB##0##1## presents the significance profiles of various patterns. The results show that the frequencies of various subgraph patterns are <italic>not </italic>significantly over- or under-represented in real networks when compared to their random formulations. A few outliers (where |<italic>z</italic>-score| &gt; 2) appear in Table ##TAB##0##1##: FFLs in Yeast Sporulation (<italic>z</italic>-score = 2.31), 3-CYCs in Yeast Stress Response (<italic>z</italic>-score = 2.47) and neuronal signalling pathway (<italic>z</italic>-score = 2.4), and Bifans in Yeast Composite (<italic>z</italic>-score = -2.05) and Cell Cycle (<italic>z</italic>-score = -2.33). Some outliers are slightly overrepresented (<italic>z </italic>&gt; 0), and others are slightly underrepresented (<italic>z </italic>&lt; 0). We observe no outliers with |<italic>z</italic>-score| ≥ 2.47.</p>", "<p>We employ the same random model as used in earlier related works [##REF##11967538##2##,##REF##12399590##3##,##REF##15372033##5##,##REF##14530388##7##]. While conserving the degree sequence of the original network, the edges in a random network are chosen randomly so that the resultant network is free from the pressure of \"evolutionary selection\" which is incident on real biological networks. However, in addition to the conservation of the degree sequence, more sophisticated random models can be generated by embedding other connectivity constraints observed in real networks, such as rules of clustering together of nodes in a neighbourhood, and path-lengths between pairs of nodes. <italic>These additional constraints will only make the random null hypothesis more stringent to refute</italic>. Nevertheless, even using the basic random model employed in our work, we fail to gather any statistical evidence that the canonical patterns appear in real networks at non-random frequencies.</p>", "<p>We note that there are differences in the counts of various motifs reported by Luscombe et al. [##REF##15372033##5##] and this work, even though we use the same datasets (Table ##TAB##0##1(a–f)##). Our figures supersede those reported by Luscombe et al. (see [##REF##18433061##13##] for a detailed explanation).</p>", "<p>Our reanalysis of <italic>Escherichia coli </italic>transcription network provides the most direct comparison of our results with those of Alon and coworkers (see Table ##TAB##0##1(g)##). We fail to see any statistical evidence to suggest that the canonical subgraphs appear more frequently than random. On comparing our results with those published by Shen-Orr et al. [##REF##11967538##2##], we find that:</p>", "<p>1. Our definitions of fixed size subgraphs such as FFL and 3-CYC are consistent with those originally defined by Alon and colleagues [##REF##11967538##2##,##REF##12399590##3##]. Consequently, we agree on the absolute count of these subgraph patterns in the real network. Surprisingly however, our results of appearances of FFLs in random networks greatly differ. To reconfirm our results reported in Table ##TAB##0##1(g)##, we generated another set of 1000 random networks using an alternative method of random network generation – starting with the original network, over a large number of repetitions, two randomly chosen interactions are swapped. (i.e., interactions: (P1,C1), and (P2,C2) become (P1,C2), and (P2,C1)). Indeed we get similar statistical significance results using this alternative method, compared to those reported in Table ##TAB##0##1(g)##.</p>", "<p>2. Our definition of Bifan ensures that we count only those patterns where a pair of target genes are <italic>strictly </italic>regulated by a pair of transcription factors – Bifans are maximal MIMs where = = 2. We believe Shen-Orr et al. [##REF##11967538##2##] fail to maintain this strictness, thereby overcounting Bifans by including in their count two parent, two child subMIMs of larger maximal MIMs. (See Discussion.)</p>", "<p>3. Similarly, our definitions and enumeration methods of SIMs and MIMs are mathematically more rigorous than those used by Shen-Orr and colleagues [##REF##11967538##2##]. Our counts of maximal MIMs and SIMs could be converted directly to counts of non-maximal MIMs and SIMs (see below). We note therefore that the non-observance of statistically significant differences between natural and randomized networks in counts of maximal MIMs and SIMs <italic>implies </italic>that there are no statistically significant differences between natural and randomized networks in counts of non-maximal MIMs and SIMs. This comment, together with the reminder that our definitions (and counts) of FFLs and 3-CYCs are identical with those of Alon <italic>e</italic>t al., shows clearly that the discrepancies are not a simple effect of alternative definitions of SIMs, MIMs and Bifans.</p>" ]

[ "<title>Discussion</title>", "<title>The observed discrepancy in occurrence frequency of FFLs and 3-CYCs is a natural consequence of topological properties of networks</title>", "<p>Occurrences of FFLs and 3-CYCs in various biological networks (see Table ##TAB##0##1##) show patterns: there are a relatively large number of FFLs and relatively small number of 3-CYCs. In this section we explain the topological basis for these differences in their frequencies.</p>", "<p>First we note that random connectivity within three-node subgraphs itself favours FFLs. Consider a directed, complete – there is an edge between every pair of nodes – three node graph (3-graph). Excluding bidirectional edges, for any set of 3 nodes there are 2³= 8 possible directed 3-graphs. Each of these configurations is isomorphic to either a FFL or a 3-CYC – any directed complete 3-graph is either a FFL or 3-CYC. Out of 8 possibilities, 6 form FFLs, and 2 form 3-CYCs. Allowing bidirectional edges, there are an extra 19 possible configurations containing at least one bidirectional edge. Each of these possibilities gives multiple FFLs or 3-CYCs or both. With or without bidirectional edges, there is a natural 3:1 bias towards forming an FFL over a 3-CYC in a 3-graph.</p>", "<p>Global properties of biological networks also favour FFLs over 3-CYCs. Most biological networks, such as those used in our study, are <italic>scale-free </italic>[##REF##10521342##15##]. In scale-free networks, the connectivity of nodes follows the power law: the probability of a node having <italic>k </italic>neighbours is <italic>P</italic>(<italic>k</italic>) ~ <italic>k</italic>^{-<italic>γ</italic>}. Only a few nodes in such a network are highly-connected (and form <italic>hubs</italic>), while most nodes are sparsely connected [##REF##10521342##15##].</p>", "<p>We asked how many of the FFLs in various networks contain hubs among their nodes. (We consider as hubs the top 10% of nodes in the network that are highly-connected, having more than 10 neighbours.) Table ##TAB##1##2## contains the percentages of FFLs enumerated in various networks, having <italic>n </italic>= {0, 1, 2, 3} nodes as hubs. A large majority of the FFLs contain at least one hub; most common being the FFLs with hubs at two of their nodes. In the Yeast composite network, 961 of 997 FFLs have at least one common <italic>source-intermediate </italic>edge between them. These 961 FFLs can be grouped into 114 clusters (containing distinct source-intermediate edges) revealing that connected hubs often share many common children, automatically giving rise to FFLs. We believe that the principle of <italic>preferential attachment </italic>predisposes a biological network to have connected hubs that have shared children. This gives a network its robustness to random node failure [##REF##10521342##15##].</p>", "<p>We also observe that there is an imbalance between indegree and outdegree around hubs – there are significantly more outgoing edges than incoming edges. We have seen above that FFLs are naturally favoured over 3-CYCs in 3-graphs. The imbalances between in- and out-degree around the hubs further enhances the formation of FFLs. Consider a hub with <italic>m </italic>incoming edges and <italic>n </italic>outgoing edges. With a random addition of an edge between any pair of (<italic>m </italic>+ <italic>n</italic>) nodes adjacent to this hub, the probability of forming an FFL in this system is: while that of forming a cycle is: . Then, , which is symmetric in <italic>m </italic>and <italic>n</italic>. If there is a large disparity between <italic>m </italic>and <italic>n </italic>(i.e., <italic>m </italic>≪ <italic>n</italic>, or <italic>m </italic>≫ <italic>n</italic>), then one of the terms or dominates, resulting in . For example, when <italic>m </italic>= 2 and <italic>n </italic>= 20, <italic>P</italic>_FFL= 0.91, and <italic>P</italic>_3-CYC= 0.09. This shows the odds against the formation of a 3-CYC in networks with structures typical of biological networks.</p>", "<p>There have been suggestions that 3-CYC is an \"anti-motif\" – a motif that is selected <italic>against </italic>in many biological networks [##UREF##1##14##]. But, as described above, the suppression of 3-CYCs is an expected consequence of topological properties of biological networks.</p>", "<p><italic>These properties are sufficient to account for the observed profiles of FFLs and 3-CYCs</italic>.</p>", "<title>Assemblies of motifs</title>", "<p>Kashtan and colleagues [##REF##15524551##16##] observed that regulatory networks contain multi-output FFL generalizations (see Figure ##FIG##1##2(a)##) in frequencies much higher than multi-input (Figure ##FIG##1##2(d)##) and multi-intermediate (Figure ##FIG##1##2(f)##) generalisations. (These authors also suggested that multi-output FFLs were selected to achieve some information processing role [##REF##15524551##16##].)</p>", "<p>We, in contrast, observe that the varied frequencies of assemblies of multiple FFLs are a consequence of the occurrence of FFLs around hubs. Figure ##FIG##1##2## shows all possible assemblies involving two FFLs sharing a common edge. In Table ##TAB##2##3## we enumerate the occurrences of each such assembly in various networks. Clearly, the multi-output assembly of two FFLs abounds over other possibilities, simply because a large number of FFLs share a common source-intermediate edge.</p>", "<p>Thus the numbers of multi-output FFLs grow combinatorially with the number of FFLs sharing a common source-intermediate edge. The count of (<italic>k</italic>&lt;<italic>n</italic>)-output assembly of FFLs, where <italic>n </italic>is the number of FFLs sharing two common (source and intermediate) nodes, is expected to increase as ^{<italic>n</italic>}<italic>C</italic>_{<italic>k</italic>}. For example, 5 FFLs having a common source-intermediate edge (see Figure ##FIG##2##3##) will give rise to 10 non-redundant bi-output FFLs. Table ##TAB##3##4## shows the statistical significance of finding bi-ouput FFLs in various real networks used in this work, by comparing the occurrences with those observed in their corresponding random networks. Statistically, their frequencies are not significantly greater than in random networks.</p>", "<title>On SIMs, MIMs and Bifans</title>", "<p>SIMs and MIMs are variable sized subgraphs. Alon and colleagues [##REF##11967538##2##] defined the dense overlapping regulon (DOR) as a two-layered subgraph with <italic>not necessarily complete </italic>connections between them. MIMs are special DORs, a concept that arose as a generalization of the Bifan (Figure ##FIG##0##1(e)##) subgraph. These Bifans were observed to be present in large numbers in biological networks. However, some investigators fail to impose the criterion of maximality while counting MIMs. This can lead to significant inflation of counts [##REF##11967538##2##,##REF##15372033##5##]. <italic>Note that this applies equally to natural graphs and random ones</italic> (Hence we emphasize that the differences between our results and those of Alon et al. are not explicable solely on the basis of alternative definitions of some of the motifs).</p>", "<p>A maximal MIM with <italic>m </italic>parents and <italic>n </italic>children contains [2^{<italic>m </italic>}- (<italic>m </italic>+ 1)] × [2^{<italic>n </italic>}- (<italic>n </italic>+ 1)] - 1 easily enumerable non-maximal \"subMIMs\". Our definition of a Bifan ensures that we are only counting (maximal) MIMs that contain 2 parents and 2 children. Counting subMIMs as Bifans will combinatorially increase their counts, as each maximal MIM will contribute to ^{<italic>m</italic>}<italic>C</italic>₂× ^{<italic>n</italic>}<italic>C</italic>₂Bifans. For example, the Yeast composite network contains a large MIM containing 2 parents and 119 children. This alone contributes to 7021 non-maximal Bifans. The same consistency is maintained when counting SIMs. The list of subgraphs occurrences in various networks used in this paper can be downloaded from <ext-link ext-link-type=\"uri\" xlink:href=\"http://hollywood.bx.psu.edu/networks/analysis/\"/>.</p>", "<p>The natural appearance of bipartite graphs in dense general graphs has received some attention in graph theory [##UREF##2##17##]. It has also been demonstrated, using Ramsey theory [##UREF##3##18##], that bipartite cliques appear in sufficiently dense bipartite graphs [##UREF##4##19##,##UREF##5##20##]. MIMs are bipartite cliques. Biological networks contain regions in which dense bipartite graphs naturally appear, and hence giving rise to bipartite cliques. This in itself speaks against the notion of evolutionary selection of MIMs [##REF##11967538##2##].</p>", "<title>Evidence for selection of motifs?</title>", "<p>Analysis of natural networks shows that several commonly observed subgraphs identified as motifs do not appear at frequencies significantly greater than in corresponding random graphs. Instead, their frequency of occurrence is the result of the small-world character of many biological networks, and of the associated degree distribution.</p>", "<p>What does this imply about the idea that motifs have been selected, by evolution, for function? The statement that motifs are selected for function has two possible interpretations, not necessarily incompatible:</p>", "<p>1. It might be asserted that the <italic>general type </italic>of motif – for instance FFL rather than 3-cycle – is selected because of a general propriety to serve a particular function (For example, Alon et al. [##REF##16601728##1##] pointed out that a FFL with AND logic at the output node can function as a filter rejecting transient stimuli).</p>", "<p>2. Or it might be asserted that <italic>individual </italic>FFLs (or 3-cycles) within a network play specific functional roles at specific points.</p>", "<p>Statistics of frequency of occurrences of specific motifs, and the comparison of observed frequencies in natural networks relative to random networks, do not – no matter what numerical results emerge – provide evidence for or against assertions of type 2. If any individual subgraph at some node plays an essential functional role in a network, it could be selected – whether it is a commonly-occurring subgraph or not. Conversely, an observation of significantly non-random occurrence frequencies of motifs would suggest the action of positive or negative selection, acting at the level of assertions of type 1 or type 2. Indeed it seems inescapable that if assertions of type 1 are true, then at least some assertions of type 2 must also be true, but not vice versa.</p>", "<p>Our results suggest that there is no evidence for type 1 assertions.</p>" ]

[ "<title>Conclusion</title>", "<p>We have analysed several biological networks. Our results suggest that there is no evidence suggesting selection for or against subgraph patterns such as FFL, 3-CYC, SIM, MIM, Bifan. We have shown that, in contrast to the need to invoke selection to explain the structure of observed networks, it is the topological properties of networks that automatically favour the observed frequency profiles of various subgraph patterns.</p>" ]

[ "<p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License (<ext-link ext-link-type=\"uri\" xlink:href=\"http://creativecommons.org/licenses/by/2.0\"/>), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</p>", "<title>Background</title>", "<p>Inventories of small subgraphs in biological networks have identified commonly-recurring patterns, called motifs. The inference that these motifs have been selected for function rests on the idea that their occurrences are significantly more frequent than random.</p>", "<title>Results</title>", "<p>Our analysis of several large biological networks suggests, in contrast, that the frequencies of appearance of common subgraphs are similar in natural and corresponding random networks.</p>", "<title>Conclusion</title>", "<p>Indeed, certain topological features of biological networks give rise naturally to the common appearance of the motifs. We therefore question whether frequencies of occurrences are reasonable evidence that the structures of motifs have been selected for their functional contribution to the operation of networks.</p>" ]

[ "<title>Authors' contributions</title>", "<p>Both the authors contributed equally to the planning and execution of this study; both authors contributed to the draft, and have read and approved the final manuscript.</p>" ]

[]

[ "<fig position=\"float\" id=\"F1\"><label>Figure 1</label><caption><p><bold>Canonical subgraph patterns in biological networks</bold>. Canonical subgraph patterns in biological networks. (a) Feed-forward loop (FFL): contains a \"source\" (at the top), \"intermediate\" (bottom-left), and \"target\" (bottom-right) nodes. (b) 3-cycle: a three node directed cyclic graph, (c) Single-input module (SIM). (d) Multiple-input module (MIM). (e) Bifan motif. SIM, MIM, and Bifan are two-layered graphs with edges from nodes in top- to bottom-layer. A Bifan is a MIM with exactly 2 parent and 2 child nodes.</p></caption></fig>", "<fig position=\"float\" id=\"F2\"><label>Figure 2</label><caption><p><bold>Self-Assemblies of two FFLs</bold>. Various possible self-assemblies of two FFLs sharing a common edge.</p></caption></fig>", "<fig position=\"float\" id=\"F3\"><label>Figure 3</label><caption><p><bold>Example of FFLs sharing two hub nodes</bold>. Example of FFLs sharing two hub nodes that are connected.</p></caption></fig>" ]

[ "<table-wrap position=\"float\" id=\"T1\"><label>Table 1</label><caption><p>Frequencies of canonical subgraph patterns in biological networks</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"right\">FFL</td><td align=\"right\">3-CYC</td><td align=\"right\">SIM</td><td align=\"right\">MIM</td><td align=\"right\">Bifan</td></tr></thead><tbody><tr><td align=\"center\" colspan=\"6\">(a) <bold>Yeast transcription – composite</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">997</td><td align=\"right\">4</td><td align=\"right\">107</td><td align=\"right\">1551</td><td align=\"right\">186</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">993.5</td><td align=\"right\">4.2</td><td align=\"right\">76.8</td><td align=\"right\">1919.2</td><td align=\"right\">413.6</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">281.4</td><td align=\"right\">2.4</td><td align=\"right\">27.0</td><td align=\"right\">233.1</td><td align=\"right\">111.1</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">0.0123</td><td align=\"right\">-0.0977</td><td align=\"right\">0.6734</td><td align=\"right\">-1.5792</td><td align=\"right\">-2.0479</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">(b) <bold>Yeast transcription – Cell Cycle</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">103</td><td align=\"right\">3</td><td align=\"right\">27</td><td align=\"right\">56</td><td align=\"right\">15</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">79.3</td><td align=\"right\">1.9</td><td align=\"right\">28.0</td><td align=\"right\">76.6</td><td align=\"right\">31.7</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">22.6</td><td align=\"right\">1.3</td><td align=\"right\">6.9</td><td align=\"right\">11.3</td><td align=\"right\">7.2</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">1.0491</td><td align=\"right\">0.9133</td><td align=\"right\">-0.1397</td><td align=\"right\">-1.8144</td><td align=\"right\">-2.3325</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">(c) <bold>Yeast transcription – Sporulation</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">67</td><td align=\"right\">2</td><td align=\"right\">27</td><td align=\"right\">41</td><td align=\"right\">26</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">38.0</td><td align=\"right\">0.6</td><td align=\"right\">30.7</td><td align=\"right\">53.0</td><td align=\"right\">28.8</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">12.5</td><td align=\"right\">0.8</td><td align=\"right\">5.1</td><td align=\"right\">7.8</td><td align=\"right\">7.8</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">2.3148</td><td align=\"right\">1.7739</td><td align=\"right\">-0.7303</td><td align=\"right\">-1.5336</td><td align=\"right\">-0.3544</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">(d) <bold>Yeast transcription – Diauxic Shift</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">64</td><td align=\"right\">1</td><td align=\"right\">48</td><td align=\"right\">137</td><td align=\"right\">54</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">63.2</td><td align=\"right\">0.3</td><td align=\"right\">47.8</td><td align=\"right\">141.1</td><td align=\"right\">64.4</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">27.2</td><td align=\"right\">0.6</td><td align=\"right\">13.7</td><td align=\"right\">18.2</td><td align=\"right\">16.6</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">0.0301</td><td align=\"right\">1.0626</td><td align=\"right\">0.0167</td><td align=\"right\">-0.2230</td><td align=\"right\">-0.6260</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">(e) <bold>Yeast transcription – DNA Damage</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">70</td><td align=\"right\">1</td><td align=\"right\">45</td><td align=\"right\">117</td><td align=\"right\">51</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">49.0</td><td align=\"right\">0.2</td><td align=\"right\">44.9</td><td align=\"right\">117.1</td><td align=\"right\">53.4</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">25.8</td><td align=\"right\">0.5</td><td align=\"right\">12.1</td><td align=\"right\">17.0</td><td align=\"right\">14.4</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">0.8149</td><td align=\"right\">1.6548</td><td align=\"right\">0.0076</td><td align=\"right\">-0.0073</td><td align=\"right\">-0.1679</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">(f) <bold>Yeast transcription – Stress Response</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">42</td><td align=\"right\">2</td><td align=\"right\">32</td><td align=\"right\">46</td><td align=\"right\">21</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">36.1</td><td align=\"right\">0.3</td><td align=\"right\">40.5</td><td align=\"right\">52.7</td><td align=\"right\">24.0</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">14.2</td><td align=\"right\">0.7</td><td align=\"right\">9.3</td><td align=\"right\">11.7</td><td align=\"right\">6.3</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">0.4123</td><td align=\"right\">2.4005</td><td align=\"right\">-0.9102</td><td align=\"right\">-0.5698</td><td align=\"right\">-0.4761</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">(g) <bold><italic>Escherichia coli </italic>transcription</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">40</td><td align=\"right\">0</td><td align=\"right\">2</td><td align=\"right\">45</td><td align=\"right\">17</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">24.1</td><td align=\"right\">0.4</td><td align=\"right\">4.7</td><td align=\"right\">29.0</td><td align=\"right\">17.5</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">12.3</td><td align=\"right\">0.7</td><td align=\"right\">2.8</td><td align=\"right\">9.7</td><td align=\"right\">5.5</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">1.2928</td><td align=\"right\">-0.6379</td><td align=\"right\">-0.9663</td><td align=\"right\">1.6463</td><td align=\"right\">-0.1001</td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"center\" colspan=\"6\">(h) <bold>Hippocampal CA1 neuronal signalling pathway</bold></td></tr><tr><td colspan=\"6\"><hr/></td></tr><tr><td align=\"left\"><italic>n</italic></td><td align=\"right\">266</td><td align=\"right\">37</td><td align=\"right\">5</td><td align=\"right\">240</td><td align=\"right\">92</td></tr><tr><td align=\"left\"><italic>μ</italic></td><td align=\"right\">219.3</td><td align=\"right\">21.7</td><td align=\"right\">4.6</td><td align=\"right\">181.1</td><td align=\"right\">103.7</td></tr><tr><td align=\"left\"><italic>σ</italic></td><td align=\"right\">54.9</td><td align=\"right\">6.2</td><td align=\"right\">2.1</td><td align=\"right\">35.5</td><td align=\"right\">14.7</td></tr><tr><td align=\"left\"><italic>z</italic></td><td align=\"right\">0.8499</td><td align=\"right\">2.4664</td><td align=\"right\">0.1994</td><td align=\"right\">1.6590</td><td align=\"right\">0.7992</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T2\"><label>Table 2</label><caption><p>Percentage of FFLs in various networks having exactly <italic>n </italic>of its nodes as hubs</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"right\"><italic>n </italic>= 1</td><td align=\"right\"><italic>n </italic>= 2</td><td align=\"right\"><italic>n </italic>= 3</td><td align=\"right\"><italic>n </italic>= 0</td></tr></thead><tbody><tr><td align=\"left\">Yeast Composite</td><td align=\"right\">15.7</td><td align=\"right\">80.1</td><td align=\"right\">2.2</td><td align=\"right\">1.9</td></tr><tr><td align=\"left\">Yeast Sporulation</td><td align=\"right\">22.4</td><td align=\"right\">67.2</td><td align=\"right\">4.5</td><td align=\"right\">6.0</td></tr><tr><td align=\"left\">Yeast Cell Cycle</td><td align=\"right\">9.7</td><td align=\"right\">68.0</td><td align=\"right\">15.5</td><td align=\"right\">6.8</td></tr><tr><td align=\"left\">Yeast Diauxic</td><td align=\"right\">12.5</td><td align=\"right\">81.2</td><td align=\"right\">6.2</td><td align=\"right\">0.0</td></tr><tr><td align=\"left\">Yeast DNA damage</td><td align=\"right\">24.3</td><td align=\"right\">68.6</td><td align=\"right\">5.7</td><td align=\"right\">1.4</td></tr><tr><td align=\"left\">Yeast Stress response</td><td align=\"right\">21.4</td><td align=\"right\">59.5</td><td align=\"right\">19.0</td><td align=\"right\">0.0</td></tr><tr><td align=\"left\">Hippocampal pathway</td><td align=\"right\">20.9</td><td align=\"right\">58.7</td><td align=\"right\">15.5</td><td align=\"right\">4.9</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T3\"><label>Table 3</label><caption><p>Number of occurrences of various assemblies shown in Figure 2</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"center\" colspan=\"6\">Frequencies of patterns in Figure 2</td></tr><tr><td/><td align=\"right\">(a)</td><td align=\"right\">(b)</td><td align=\"right\">(c)</td><td align=\"right\">(d)</td><td align=\"right\">(e)</td><td align=\"right\">(f)</td></tr></thead><tbody><tr><td align=\"center\">Yeast Composite</td><td align=\"right\">9232</td><td align=\"right\">259</td><td align=\"right\">184</td><td align=\"right\">288</td><td align=\"right\">280</td><td align=\"right\">152</td></tr><tr><td align=\"center\">Yeast Sporulation</td><td align=\"right\">113</td><td align=\"right\">3</td><td align=\"right\">8</td><td align=\"right\">21</td><td align=\"right\">8</td><td align=\"right\">4</td></tr><tr><td align=\"center\">Yeast Cell Cycle</td><td align=\"right\">419</td><td align=\"right\">22</td><td align=\"right\">17</td><td align=\"right\">38</td><td align=\"right\">12</td><td align=\"right\">15</td></tr><tr><td align=\"center\">Yeast Diauxic Shift</td><td align=\"right\">214</td><td align=\"right\">2</td><td align=\"right\">2</td><td align=\"right\">3</td><td align=\"right\">4</td><td align=\"right\">5</td></tr><tr><td align=\"center\">Yeast DNA damage</td><td align=\"right\">140</td><td align=\"right\">6</td><td align=\"right\">6</td><td align=\"right\">11</td><td align=\"right\">4</td><td align=\"right\">8</td></tr><tr><td align=\"center\">Yeast Stress Response</td><td align=\"right\">41</td><td align=\"right\">9</td><td align=\"right\">6</td><td align=\"right\">5</td><td align=\"right\">4</td><td align=\"right\">1</td></tr></tbody></table></table-wrap>", "<table-wrap position=\"float\" id=\"T4\"><label>Table 4</label><caption><p>Frequencies of Bi-FFL assembly in various networks</p></caption><table frame=\"hsides\" rules=\"groups\"><thead><tr><td/><td align=\"right\"><italic>n</italic></td><td align=\"right\"><italic>μ</italic></td><td align=\"right\"><italic>σ</italic></td><td align=\"right\"><italic>z</italic></td></tr></thead><tbody><tr><td align=\"left\">Yeast Composite</td><td align=\"right\">9232</td><td align=\"right\">17278.2</td><td align=\"right\">13537.5</td><td align=\"right\">-0.6</td></tr><tr><td align=\"left\">Yeast Sporulation</td><td align=\"right\">113</td><td align=\"right\">52.4</td><td align=\"right\">48.1</td><td align=\"right\">1.3</td></tr><tr><td align=\"left\">Yeast Cell Cycle</td><td align=\"right\">419</td><td align=\"right\">173.8</td><td align=\"right\">132.2</td><td align=\"right\">1.9</td></tr><tr><td align=\"left\">Yeast Diauxic Shift</td><td align=\"right\">214</td><td align=\"right\">238.4</td><td align=\"right\">334.3</td><td align=\"right\">-0.1</td></tr><tr><td align=\"left\">Yeast DNA Damage</td><td align=\"right\">140</td><td align=\"right\">189.6</td><td align=\"right\">295.8</td><td align=\"right\">-0.2</td></tr><tr><td align=\"left\">Yeast Stress Response</td><td align=\"right\">41</td><td align=\"right\">67.2</td><td align=\"right\">69.3</td><td align=\"right\">-0.4</td></tr><tr><td align=\"left\">Ecoli transcription</td><td align=\"right\">0</td><td align=\"right\">0.6</td><td align=\"right\">1.1</td><td align=\"right\">-0.6</td></tr><tr><td align=\"left\">Hippocampal pathway</td><td align=\"right\">85</td><td align=\"right\">327.0</td><td align=\"right\">223.6</td><td align=\"right\">-1.1</td></tr></tbody></table></table-wrap>" ]

[ "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M1\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M2\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M3\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M4\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M5\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M6\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M7\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M8\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M9\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M10\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M11\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M12\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M13\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M14\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M15\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M16\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M17\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M18\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M19\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M20\" name=\"1752-0509-2-73-i3\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mi>z</mml:mi><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mi>n</mml:mi><mml:mo>−</mml:mo><mml:mi>μ</mml:mi></mml:mrow><mml:mi>σ</mml:mi></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M21\" name=\"1752-0509-2-73-i1\" overflow=\"scroll\"><mml:semantics><mml:mtext mathvariant=\"script\">P</mml:mtext></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M22\" name=\"1752-0509-2-73-i2\" overflow=\"scroll\"><mml:semantics><mml:mi mathvariant=\"script\">C</mml:mi></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M23\" name=\"1752-0509-2-73-i4\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mtext>FFL</mml:mtext></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>m</mml:mi></mml:msup><mml:msub><mml:mi>C</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:msup><mml:mo>+</mml:mo><mml:mi>n</mml:mi></mml:msup><mml:msub><mml:mi>C</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:mo stretchy=\"false\">)</mml:mo><mml:mo>+</mml:mo><mml:mi>m</mml:mi><mml:mi>n</mml:mi></mml:mrow><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>m</mml:mi></mml:msup><mml:msub><mml:mi>C</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:msup><mml:mo>+</mml:mo><mml:mi>n</mml:mi></mml:msup><mml:msub><mml:mi>C</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:mo>+</mml:mo><mml:mi>m</mml:mi><mml:mi>n</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M24\" name=\"1752-0509-2-73-i5\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mn>3</mml:mn><mml:mo>−</mml:mo><mml:mtext>CYC</mml:mtext></mml:mrow></mml:msub><mml:mo>=</mml:mo><mml:mfrac><mml:mrow><mml:mi>m</mml:mi><mml:mi>n</mml:mi></mml:mrow><mml:mrow><mml:mn>2</mml:mn><mml:msup><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>m</mml:mi></mml:msup><mml:msub><mml:mi>C</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:msup><mml:mo>+</mml:mo><mml:mi>n</mml:mi></mml:msup><mml:msub><mml:mi>C</mml:mi><mml:mn>2</mml:mn></mml:msub><mml:mo>+</mml:mo><mml:mi>m</mml:mi><mml:mi>n</mml:mi><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M25\" name=\"1752-0509-2-73-i6\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mtext>FFL</mml:mtext></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mn>3</mml:mn><mml:mo>−</mml:mo><mml:mtext>CYC</mml:mtext></mml:mrow></mml:msub></mml:mrow></mml:mfrac><mml:mo>=</mml:mo><mml:mn>1</mml:mn><mml:mo>+</mml:mo><mml:mfrac><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>m</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mi>n</mml:mi></mml:mfrac><mml:mo>+</mml:mo><mml:mfrac><mml:mrow><mml:mo stretchy=\"false\">(</mml:mo><mml:mi>n</mml:mi><mml:mo>−</mml:mo><mml:mn>1</mml:mn><mml:mo stretchy=\"false\">)</mml:mo></mml:mrow><mml:mi>m</mml:mi></mml:mfrac></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M26\" name=\"1752-0509-2-73-i7\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>m</mml:mi><mml:mi>n</mml:mi></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M27\" name=\"1752-0509-2-73-i8\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>n</mml:mi><mml:mi>m</mml:mi></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>", "<inline-formula><mml:math xmlns:mml=\"http://www.w3.org/1998/Math/MathML\" id=\"M28\" name=\"1752-0509-2-73-i9\" overflow=\"scroll\"><mml:semantics><mml:mrow><mml:mfrac><mml:mrow><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mtext>FFL</mml:mtext></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mi>P</mml:mi><mml:mrow><mml:mn>3</mml:mn><mml:mo>−</mml:mo><mml:mtext>CYC</mml:mtext></mml:mrow></mml:msub></mml:mrow></mml:mfrac><mml:mo>~</mml:mo><mml:mi>max</mml:mi><mml:mo>⁡</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>m</mml:mi><mml:mi>n</mml:mi></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow><mml:mo>,</mml:mo><mml:mrow><mml:mo>(</mml:mo><mml:mrow><mml:mfrac><mml:mi>n</mml:mi><mml:mi>m</mml:mi></mml:mfrac></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:semantics></mml:math></inline-formula>" ]

[]

[]