datajuicer/LLaMA-1B-dj-refine-150B
Text Generation
•
Updated
•
46.2k
text
stringlengths 3.47k
75.5k
| meta
dict | stats
dict | simhash
float64 101,270,602B
18,334,047,329B
|
|---|---|---|---|
Background
==========
Congenital fibrosis of the extraocular muscles (CFEOM) and Duane syndrome (DS) are complex strabismus disorders that present with congenital restrictive ophthalmoplegia with or without ptosis. These disorders were traditionally believed to reflect primary structural extraocular muscle (EOM) anomalies and have been referred to as \'congenital fibrosis syndromes\' \[[@B1]\]. Neuropathology studies of DS \[[@B2],[@B3]\] and one form of CFEOM (CFEOM1) \[[@B4]\], and the identification of *ARIX* as the gene mutated in a second form of CFEOM (CFEOM2) \[[@B5]\], however, support our hypothesis that CFEOM results from maldevelopment of the oculomotor (nIII) and/or trochlear (nIV) nuclei and DS results from maldevelopment of the abducens (nVI) nucleus. The continued definition of these phenotypes and identification of the underlying disease genes will assist clinical diagnostics and lead to a better understanding of the unique developmental features of the oculomotor lower motor neuron unit.
Although several distinct CFEOM phenotypes have been defined \[[@B6]-[@B8]\], each likely resulting from maldevelopment of a unique combination of alpha motor neurons in nIII and/or nIV, most reports of CFEOM families describe a stereotypical clinical phenotype. The affected members of these pedigrees are born with bilateral ptosis and restrictive ophthalmoplegia. The primary vertical position of each eye is downward and cannot be elevated above the midline. On forced duction testing there is resistance to passive movement of the globe. Although the primary position of both eyes is infraducted, there is variability in the secondary position of each eye (i.e. exotropic, esotropic, or neutral), and the degree of residual horizontal movement within the lower quadrants (full to completely restricted). This CFEOM phenotype was first described in the medical literature in 1840 \[[@B9]\] and was recognized to occur as a familial trait in 1879 \[[@B10]\]. Subsequently, families segregating this phenotype have been published under myriad names \[[@B1],[@B4],[@B6],[@B11]-[@B21]\]. We now refer to individuals with this phenotype as having \"classic CFEOM\" and to families in which all affected members have this phenotype as \" CFEOM1 pedigrees\". We previously mapped a CFEOM1 locus, referred to as *FEOM1,* to a ≤ 3 cM region spanning the centromere of chromosome 12, flanked by *D12S1584 (AFM136xf6)* on the p-arm and *D12S1668 (AFMb320wd9)* on the q-arm \[[@B6],[@B15]\].
In addition to families with CFEOM1, we have identified several less common familial CFEOM phenotypes (CFEOM2 and CFEOM3). These phenotypes are classified as CFEOM based on the presence of affected members with congenital restrictive ophthalmoplegia affecting extraocular muscles in the nIII/nIV distribution. By definition, however, one or more affected family members do not have the classic CFEOM phenotype. In families with CFEOM2, the eyes of affected family members are fixed in an exotropic, or outward, position. Thus far, this phenotype segregates as an autosomal recessive trait and maps to the *FEOM2* locus on chromosome 11q13 \[[@B8]\], and affected individuals carry homozygous mutations in *ARIX*\[[@B5]\]. *ARIX* encodes a homeodomain transcription factor required for nIII and nIV development in mice and zebrafish \[[@B22],[@B23]\]. In families with CFEOM3, the CFEOM phenotype is variably expressed. Some affected members have classic CFEOM. Others, however, are unilaterally affected, the primary position of the eye is orthotropic rather than infraducted, and/or the eye can be raised into the upper quadrants. Thus far, this phenotype segregates as an autosomal dominant trait and maps to either *FEOM3* on 16qter \[[@B7]\] or to *FEOM1*\[[@B24]\].
In our attempt to understand phenotype-genotype correlations between specific CFEOM phenotypes and *FEOM* loci, we noted that the CFEOM1 phenotype in all pedigrees reported to date maps to the *FEOM1* locus. To determine if CFEOM1 is indeed genetically homogeneous, we identified all unpublished CFEOM1 pedigrees in our database, analyzed them for linkage to the *FEOM* loci, and found that most but not all were consistent with linkage to *FEOM1.* The two small pedigrees not linked to *FEOM1* were consistent with linkage to *FEOM3.* In addition, to further define the spectrum of human *ARIX* mutations, we identified all CFEOM1 families consistent with linkage to *FEOM2* or sporadic individuals with classic CFEOM and determined that none harbored mutations in the *ARIX* gene.
Results
=======
From our database, 33 pedigrees were of sufficient size and had sufficient clinical data and 14 sporadic individuals had sufficient clinical data to qualify for the study. Of these, 20 pedigrees met CFEOM1 and 5 sporadic individuals met classic CFEOM inclusion criteria. Although not an inclusion criterion for the study, the CFEOM1 phenotype in all 20 families was inherited as an autosomal dominant trait with full penetrance. The phenotype in 9 of the 20 pedigrees was previously demonstrated to map to the *FEOM1* locus \[[@B6],[@B15],[@B25],[@B26]\] (Table [1](#T1){ref-type="table"}). Therefore, the remaining 11 families were included in this study (Figs. [1](#F1){ref-type="fig"} &[2](#F2){ref-type="fig"}, Table [1](#T1){ref-type="table"}, and see additional files 1--3 \[[Additional File 1](#S1){ref-type="supplementary-material"}, [Additional File 2](#S2){ref-type="supplementary-material"}, [Additional File 3](#S3){ref-type="supplementary-material"}\].
{ref-type="fig"} were assigned as linkage studies were performed are not equivalent when compared between families. Black bars denote the potential disease-associated region. Diagonally hatched or white bars highlight the inheritance of the non-disease-associated haplotypes. References to specific individuals within the text refer to the generation number (Roman numeral) and position within generation (Arabic numeral). In all 9 pedigrees each family\'s disease-associated haplotype is inherited by all CFEOM1 individuals and by no asymptomatic individuals.](1471-2156-3-3-1){#F1}
{ref-type="fig"}. In each family the CFEOM1 phenotype is co-inherited with *FEOM3* markers and not with *FEOM1* markers.](1471-2156-3-3-2){#F2}
::: {#T1 .table-wrap}
::: {.caption}
######
Summary of the genetic analysis of CFEOM1 pedigrees
:::
**Pedigree** **CFEOM Phenotype** **Inheritance** **Forced Ductions** **Cytogenetic analysis** ***FEOM1*** ***FEOM3*** ***FEOM2*** ***ARIX* mutations** **Publication**
-------------- --------------------- ----------------- --------------------- -------------------------- ----------------- ----------------- ------------------- ---------------------- ------------------------
A CFEOM1 AD \+ normal **LINKED** refs. \[[@B6],[@B15]\]
B CFEOM1 AD \+ normal **LINKED** refs. \[[@B6],[@B15]\]
C CFEOM1 AD \+ not done **LINKED** ref. \[[@B15]\]
H CFEOM1 AD \+ normal **LINKED** ref. \[[@B15]\]
AA CFEOM1 AD \+ normal **LINKED** ref. \[[@B15]\]
AC CFEOM1 AD \+ normal **LINKED** ref. \[[@B15]\]
AD CFEOM1 AD \+ normal **LINKED** ref. \[[@B15]\]
CD CFEOM1 AD \+ not done **LINKED** ref. \[[@B25]\]
CB CFEOM1 AD not done not done **c/w linkage** nc/w linkage NOT LINKED ref. \[[@B26]\]
BJ CFEOM1 AD \+ normal **LINKED** current
CZ CFEOM1 AD \+ normal **LINKED** current
AG CFEOM1 AD \+ normal **c/w linkage** NOT LINKED NOT LINKED current
AJ CFEOM1 AD \+ not done **c/w linkage** nc/w linkage nc/w linkage current
AH CFEOM1 AD \+ normal **c/w linkage** nc/w linkage nc/w linkage current
T CFEOM1 AD \+ not done **c/w linkage** nc/w linkage nc/w linkage current
CT CFEOM1 AD \+ normal **c/w linkage** nc/w linkage nc/w linkage current
BC CFEOM1 AD \+ not done **c/w linkage** nc/w linkage (c/w linkage) none current
E CFEOM1 AD \+ normal **c/w linkage** **c/w linkage** NOT LINKED current
K CFEOM1 AD \+ normal nc/w linkage **c/w linkage** **(c/w linkage)** none current
BT CFEOM1 AD \+ normal NOT LINKED **c/w linkage** nc/w linkage current
AD = autosomal dominant; + = positive forced duction testing for restriction; c/w linkage = consistent with linkage; nc/w li nkage = not consistent with linkage.
:::
The 11 families are ethnically diverse, not consanguineous and, to the best of our knowledge, unrelated. Eight of the families reside in the US and are of mixed European ancestry (BJ, AG, AJ, AH, CT, E, K, BT), while families CZ, T, and BC are of Italian, Irish, and Japanese ancestry, respectively. Family history of CFEOM in several previous generations was documented in two pedigrees (AG, T), and a previous family history was recounted but could not be corroborated in five others (BJ, AJ, AH, E, K). In contrast, in pedigrees CZ, BC, CT and BT, neither the parents nor more distant relatives of the eldest affected family member were reportedly affected. Examination of individuals I--1 and I--2 of pedigrees BC and CT confirmed their unaffected status. Individuals I--1 and I--2 in pedigrees CZ and BT were deceased. These data suggest that the CFEOM1 mutation rate is not negligible. Cytogenetic analyses, when performed, did not reveal abnormalities (Table [1](#T1){ref-type="table"}).
Nine of the 11 CFEOM1 pedigrees contain too few participants to establish linkage to a specific locus. Haplotype analysis of these families using multiple markers that span the critical region of a given locus can, however, eliminate linkage to the locus, determine genetic heterogeneity, and guide future mutation analyses. If the phenotype results from a mutation at a given locus, haplotype analysis at that locus will be consistent with linkage. If the phenotype does not result from a mutation at a given locus, however, haplotype data from a small family may be consistent or inconsistent with linkage. Thus, haplotype data in a small family that is consistent with linkage can result either from a disease mutation at that locus or by chance. Haplotype data in a small family that is inconsistent with linkage strongly suggests that the family\'s phenotype is not linked to the locus.
Linkage to *FEOM1*
------------------
Genetic analyses of the two largest families (BJ, CZ) established linkage of their phenotype to the *FEOM1* locus (Fig. [1](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"} & see [Additional File 1](#S1){ref-type="supplementary-material"}). Maximum lod scores of 3.01 were obtained at a theta value of zero for the fully informative markers *D12S59* and *D12S1048* in family BJ, and the fully informative markers *D12S1648, D12S345,* and *D12S59* in family CZ.
Linkage to *FEOM1* was ruled out in family BT (see [Additional file 1](#S1){ref-type="supplementary-material"}). A lod score of -2 was obtained at a theta value of 0.04 for the fully informative markers *D12S1621, D12S59* and *D12S1668.* The *FEOM1* critical region is ≤ 3 cM, and a theta of 0.04 corresponds to a genetic distance of approximately 4 cM, thus eliminating linkage of this family\'s CFEOM disease gene to the entire *FEOM1* critical region. Exclusion of linkage to the *FEOM1* locus is further supported by haplotype analysis of this family (Fig. [2c](#F2){ref-type="fig"}). The affected sister and brother inherit different *FEOM1* haplotypes from their affected mother, and the brother\'s affected daughter inherits her unaffected paternal grandfather\'s *FEOM1* haplotype, thus proving non-association between the *FEOM1* haplotype and the disease phenotype.
The eight remaining CFEOM pedigrees were too small to produce statistically significant lod scores; however, seven of the eight families displayed haplotype and linkage data consistent with linkage to the *FEOM1* locus (Fig. [1](#F1){ref-type="fig"}, see [Additional File 1](#S1){ref-type="supplementary-material"}). Genetic and haplotype analysis of all nine families consistent with linkage to the *FEOM1* locus did not reveal any recombination events within the previously defined *FEOM1* critical region.
The smallest family, K, revealed haplotype and linkage data that was inconsistent with linkage to the *FEOM1* locus (Fig. [2a](#F2){ref-type="fig"}, see [Additional File 1](#S1){ref-type="supplementary-material"}). A lod score of -2 was obtained at a theta value of 0.002 for the only two informative markers, *D12S59* and *D12S1090.* These theta values eliminate linkage to only 0.8 cM of the \< 3 cM *FEOM1* critical region and, therefore, this locus cannot be formally ruled out. Nevertheless, the minimum number of recombination events in this family occurs only if the affected son and daughter inherit different *FEOM1* haplotypes from their affected mother, thus strongly suggesting that the disease gene in this family does not map to the chromosome 12 locus.
Genetic heterogeneity was tested taking into account the eleven new families. Admixture analysis of the two-point data with the HOMOG program showed evidence for linkage to *FEOM1* with heterogeneity for both markers tested (*D12S345* and *D12S1048*). Chi-squares of 34.601 and 40.377 were obtained for *D12S345* and *D12S1048* respectively which resulted in significant likelihood ratios of 3.26 × 10^7^ and 5.86 × 10^8^. Alpha (the proportion of linked families) was 0.90.
Linkage to *FEOM3*
------------------
All families except BJ and CZ were analyzed for linkage to the 5.6 cM *FEOM3* locus flanked by *D16S486* and 16qter. Of the seven small families consistent with linkage to *FEOM1,* only the largest (AG) can be definitively excluded from linkage to *FEOM3* (see [Additional File 2](#S2){ref-type="supplementary-material"}). Five of the remaining families (AJ, AH, T, BC, CT) showed haplotype data inconsistent with linkage to *FEOM3,* but the theta values obtained at lod scores of -2 were insufficient to rule out the entire *FEOM3* critical region.
The two families whose phenotype did not map to *FEOM1* (K, BT) had haplotype and linkage data consistent with linkage to *FEOM3* (Fig. [2b](#F2){ref-type="fig"} &[2d](#F2){ref-type="fig"}, see [Additional File 2](#S2){ref-type="supplementary-material"}). In addition, one of the small families consistent with linkage to the *FEOM1* locus (E) had haplotype and linkage data that was also consistent with linkage to the *FEOM3* locus (see [Additional File 2](#S2){ref-type="supplementary-material"}).
Linkage to *FEOM2*
------------------
All families whose phenotype was not linked to *FEOM1* or *FEOM3* were tested for linkage to *FEOM2.* Assuming autosomal dominant inheritance with complete penetrance, two families (AG, E) are not linked and five families are inconsistent with linkage (AJ, AH, T, CT, BT) to the *FEOM2* locus (see [Additional File 3](#S3){ref-type="supplementary-material"}). Family BC is consistent with linkage to both the *FEOM2* and *FEOM1* loci (maximum lod score 0.3 at both loci), and family K is consistent with linkage to both the *FEOM2* and *FEOM3* loci (maximum lod score 0.3 at both loci).
*ARIX* mutation analysis
------------------------
Genomic DNA samples from affected member of pedigree BC and K and 5 sporadic individuals with classic CFEOM were used as templates to sequence the three *ARIX* exons and flanking introns. No mutations were identified.
Discussion
==========
We have established clinical criteria for classic CFEOM and CFEOM1, and have identified 5 sporadic individuals with classic CFEOM and 20 pedigrees with CFEOM1. Of these 20 pedigrees, 18 are linked, or consistent with linkage, to the *FEOM1* locus. Two small CFEOM1 pedigrees are not consistent with linkage, however, establishing that CFEOM1 is genetically heterogeneous.
Eleven of the 20 pedigrees are large enough to establish linkage to a specific locus; we previously reported that the CFEOM1 phenotype in 9 of these pedigrees maps to the *FEOM1* locus \[[@B6],[@B15],[@B25],[@B26]\] and we now demonstrate that the remaining two also map to *FEOM1.* Our analysis of the remaining 9 CFEOM1 pedigrees demonstrates that 6 most likely result from mutations in the *FEOM1* gene. Five of these 6 are consistent with linkage to *FEOM1* and are either not linked or not consistent with linkage to *FEOM2* and *FEOM3.* One is consistent with linkage to *FEOM1* and not *FEOM3* and, although consistent with linkage to *FEOM2, ARIX* mutations were not identified. Therefore, although not proved, the CFEOM1 phenotype in these 6 families seems likely to result from mutations in the *FEOM1* gene. The phenotype of a seventh family, E, is consistent with linkage to both *FEOM1* and *FEOM3* and will be screened for mutations at both these loci.
In contrast to the 18 pedigrees whose CFEOM1 phenotype is consistent with linkage to *FEOM1,* pedigree BT is not linked to *FEOM1* and pedigree K is inconsistent with linkage to this locus. It is notable that haplotype analysis of both these small CFEOM1 families demonstrates co-inheritance with the *FEOM3* locus. In the reported family whose autosomal dominant CFEOM3 phenotype maps to the *FEOM3* locus, 9 of the 17 affected members had classic CFEOM. Our current data now suggests that, at least in small pedigrees, CFEOM1 can also map to the *FEOM3* locus. It will require the identification of additional large CFEOM1 families to determine if they too can map to this locus.
*ARIX,* which encodes a transcription factor critical to nIII and nIV development in mice and zebrafish \[[@B22],[@B23]\], was recently identified as the *FEOM2* gene mutated in affected members of CFEOM2 families \[[@B5]\]. It was unknown, however, if classic CFEOM may also result from mutations in this gene. We now find that we are unable to identify *ARIX* mutations underlying classic CFEOM in either sporadic cases or in individuals from CFEOM1 families.
This finding is consistent with our prediction that, compared to *ARIX,* the genes mutated in CFEOM1 may have a more restricted function in the development of nIII and that their expression may actually be regulated by *ARIX.* This prediction is based on the CFEOM1 phenotype and on the neuropathological study of an affected member of a CFEOM1 pedigree whose disease gene maps to the *FEOM1* locus. This study revealed absence of the superior division of the oculomotor nerve and the corresponding central caudal and medial nIII subnuclei, and marked abnormalities of the levator palpebrae superioris and superior rectus muscles \[[@B4]\]. These findings suggest that while *ARIX* is necessary for both nIII and nIV development, the CFEOM1 genes may be necessary for the development of only these two specific nIII subnuclei.
Conclusions
===========
The genetic analysis of the 11 CFEOM1 pedigrees in this report demonstrates that this disorder is genetically heterogeneous. While the phenotype of all large CFEOM1 pedigrees analyzed thus far map to the *FEOM1* locus, smaller CFEOM1 pedigrees may harbor mutations in the *FEOM3* gene. The CFEOM1 phenotype does not, however, appear to result from mutations in *ARIX.* The CFEOM1 families identified in this study contribute critical alleles toward the identification of the mutated *FEOM1* and *FEOM3* genes. Once identified, we anticipate that the study of the function of these genes will contribute to our understanding of midbrain motor neuron development.
Materials and Methods
=====================
Subjects
--------
We established inclusion criteria for this study as follows. First, we established criteria for a \"classic CFEOM individual\" as an individual with congenital nonprogressive bilateral ophthalmoplegia and ptosis, an infraducted primary position of each eye with the inability to raise either eye above the midline, and forced duction testing positive for restriction, if testing was performed. We then established criteria for a \"CFEOM1 pedigree\" as a family in which every affected member met the criteria for \"classic CFEOM\". Second, we reviewed all participants enrolled in our ongoing CFEOM study and determined which individuals and which pedigrees met these two criteria. For a pedigree to be considered, we required clinical examination records and/or photographs/videos of the primary positions of gaze for all affected study participants. Third, from these CFEOM1 pedigrees we identified those with a family structure sufficient for linkage analysis. Pedigrees were required to have affected study participants in three or more generations, or affected study participants in two generations with at least two participating offspring of affected individuals, or two or more affected study participants within one generation. In this way we did not assume a mode of inheritance. Lastly, sporadic individuals with classic CFEOM were screened for *ARIX* mutations. The study was approved by the Children\'s Hospital institutional review board, and all study participants signed informed consent forms. Our methods adhered to the Declaration of Helsinki for research involving human subjects.
Molecular studies
-----------------
Blood samples were obtained from all participating family members, and lymphocyte DNA was extracted using the Puregene kit (Gentra, Research Triangle Park, NC). Chromosome analyses of GTG banded metaphase cells at a 400 band level minimum resolution were performed on one or more affected family members of each family whenever possible to rule out cytogenetic abnormalities. All families were analyzed for linkage to the *FEOM1* locus. Families not linked to the *FEOM1* locus were also analyzed for linkage to the *FEOM2* and *FEOM3* loci. Linkage studies were conducted using three or more locus specific polymorphic DNA micro satellite markers for each family. The *FEOM1* markers included *D12S1648, D12S61, D12S1584, D12S1621, D12S345, D12S59, D12S2080, D12S1048, D12S1668,* and *D12S1090*\[[@B6],[@B15]\]. The *FEOM2* markers included *D11S1337, D11S4162, D11S4196, D11S1314,* and *D11S1369*\[[@B8]\]. The *FEOM3* markers included *D16S539, D16S3077, D16S498, D16S486, D16S476, D16S3063, D16S689, D16S2621, D16S303,* and *D16S3407*\[[@B7]\]. The primer sequences for these polymorphisms are available from the Genome Database <>. Unlabeled primers were purchased from Genosys Biotechnologies, Inc. <>. Fluorescently labeled primers were purchased from Research Genetics, Inc. <>. Radioactive products were made by 30 cycles of PCR amplification of 10-μl reaction volumes containing 10--30 ng of genomic DNA, 40 ng of each primer, 200 μM each of dATP, dTTP, dGTP, and dCTP, 1 μCi α-^32^P-dCTP (3,000 Ci mmol^-1^) and 0.5 U Taq polymerase (Perkin Elmer). The radioactively labeled PCR products were separated on 6% denaturing polyacrylamide sequencing gels, and the alleles were visualized by autoradiography \[[@B6]\]. For the fluorescently labeled products, α-^32^P-dCTP was omitted, fluorescent primers were used, and the products were analyzed in an ABI PrismTM 377 DNA Sequencer (Perkin Elmer) following the manufacturer\'s specifications.
Lod score calculations
----------------------
An individual was scored as affected based on clinical examination records and/or photographs of primary position of gaze. Lod scores were calculated using the Fastlink version 3.0 package of programs \[[@B27]\], assuming autosomal dominant inheritance with complete penetrance, and a disease incidence of 1 in 1,000,000 births, as described previously \[[@B7],[@B8]\]. Because of the absence of specific allele frequencies for each of the ethnic groups represented in the study, we assumed ten marker alleles of equal frequency. To assess nonallelic heterogeneity linkage data from two *FEOM1* markers spanning the *FEOM1* critical region (*D12S345* and *D12S1048*) were analyzed using a HOMOG version 3.35 program \[[@B28]\].
ARIX mutation detection
-----------------------
*ARIX* was PCR amplified from genomic DNA using five primer sets and sequenced on an automated ABI 377 DNA seqeuncer (PE-Applied Biosystems) as described previously \[[@B5]\].
Supplementary Material
======================
::: {.caption}
###### Additional file 1
Table 2: Two-point linkage data between CFEOM1 and the *FEOM1* locus
:::
::: {.caption}
######
Click here for file
:::
::: {.caption}
###### Additional file 2
Table 3: Two-point linkage data between CFEOM1 and the *FEOM3* locus
:::
::: {.caption}
######
Click here for file
:::
::: {.caption}
###### Additional file 3
Table 4: Two-point linkage data between CFEOM1 and the *FEOM2* locus
:::
::: {.caption}
######
Click here for file
:::
Acknowledgements
================
This work was supported by National Eye Institute EY12498 and EY13583 and by the Children\'s Hospital Mental Retardation Research Center (P30 HD18655). | {
"from": "PMC100320.md"
} | {
"alnum_ratio": 0.6874214965,
"avg_line_length": 162.9532163743,
"char_rep_ratio": 0.156842332,
"flagged_words_ratio": 0,
"lang": "en",
"lang_score": 0.9039446115,
"max_line_length": 1949,
"num_words": 5212,
"perplexity": 1484.7,
"special_char_ratio": 0.3397452001,
"text_len": 27865,
"word_rep_ratio": 0.0997501441
} | 1,533,004,202,812,425,200 |
Background
==========
Identification of species with molecular probes is likely to revolutionize taxonomy, at least for taxa with morphological characters that are difficult to determine otherwise. Among these are the single cell eucaryotes, such as Ciliates and Flagellates, but also many other kinds of small organisms, such as Nematodes, Rotifers, Crustaceans, mites, Annelids or Insect larvae. These organisms constitute the meiofauna in water and soil, which is of profound importance in the ecological network. Efficient ways for monitoring species identity and abundance in the meiofauna should significantly help to understand ecological processes.
Molecular taxonomy with sequence specific oligo-nucleotide probes has been pioneered for bacteria \[[@B1],[@B2]\]. Probes that are specific to particular species or groups of related species can be used in fluorescent in situ hybridization assays to detect the species in complex mixtures or as symbionts of other organisms \[[@B3],[@B4]\]. Alternatively, the microarray technology is increasingly used for this purpose, allowing potentially the parallel screening of many different species. Most of the species-specific sequences that are used so far for this purpose are derived from ribosomal RNA sequences. However, any other sequence is also potentially suitable, as for example mitochondrial D-loop sequences in eucaryotes.
The species-specific probes are usually derived from an alignment of the respective sequences, where conserved and non-conserved regions are directly visible. A program has been developed for ribosomal sequences that helps to build the relevant database, and supports the selection of suitable specific sequences (ARB \[[@B5]\]). In this, a correct alignment is crucial for finding the optimal probes, but alignments are problematical in poorly conserved regions. These, on the other hand, have the highest potential to yield specific probes. Moreover, the current implementation of probe finding calculates only the number of mismatching position to discriminate between the probes, but does not take into account the position of the mismatches within the stretches, which could influence the hybridization behavior. We have therefore devised here a new algorithm that allows working with datasets that need not to be carefully aligned and that takes the position of mismatches along the recognition sequence into account.
The algorithm
-------------
The algorithm includes three parts. The first one aims to provide a function that calculates the relative stability of matching oligos in dependence of the number and position of mismatches. The second one provides a strategy for probe finding that scans all possible sequence combinations, but works time efficient. The third part deals with matches caused by single nucleotide outloops of a given sequence.
Stability function
------------------
Extensive studies exist for assessing the thermodynamic consequences of internal mismatches in short oligo-nucleotides (see fro example \[[@B6],[@B7]\]). These show that there are no simple rules and that the exact influence on the stability of a hybrid depends on the nature of the mismatch, as well as its flanking nucleotides. For example, mismatches including a G (i.e. G-G, G-T and G-A) tend to be less destabilizing than the other types of mismatches \[[@B7]\], although this can not directly be predicted from steric considerations. Comparable systematic studies on the relative influence of the position of the mismatch within the oligonucleotide do not exist yet, although it is clear that the influence is lower at the ends than in more central positions \[[@B7],[@B9]\]. Preliminary evidence with an oligo-dT stretch harboring A mismatches along the sequence suggests that the position dependence could be a continuous function \[[@B8]\]. We have therefore decided to use an *ad hoc* approach for the stability calculation that is mainly designed to discriminate against sequences with more central mismatch positions.
We model the relative stability of mismatched oligos as follows. The position of the mismatch can be considered to be a \"weak point\". The location of the \"weak point\" is expressed as a probability function that takes into account the differential contribution of central versus terminal positions. The probability that the \"weak point\" is at position x is defined by p~1~. Under the experimental conditions of melting, the presence of the \"weak point\" is true, meaning that \[sum(p1) for all x\] = 1.
We assume a Gauss distribution as the respective probability function, with the maximum in the middle of the duplex and the integral value along the duplex length set to 1 (Equation 1).
Equation 1. \"Weak point\" location probability. L -- duplex length, σ -- distribution parameter, x -- duplex position.
Note that the function in Equation 1 refers to discrete positions within the sequence, while the Gauss distribution is continuous and the integration from -∞ to +∞ is set to yield 1. The parameter σ is therefore chosen such that the discrete sum approaches 1 at any intended precision. In the program discussed below the accuracy of the sum value is 0.999.
Although the preliminary experimental evidence \[[@B8]\] suggests that the destabilization function can be approximated with the Gauss distribution, the program implementation allows also to use a flat distribution, i.e. where a position-independent effect on the melting is assumed as an alternative, to compare the outputs of the two different assumptions.
For assessing the relative amount of destabilization caused by a certain mismatch, we assume that the mismatch disturbs the surrounding base pairs from (y-n) to (y+n) positions, n can be called a border parameter that will need to be experimentally verified in the future. Because n can currently only be guessed, it is set as a program variable with a default value of 5. n might also depend on the nature of the mismatch, i.e. some types of mismatches might influence the surrounding bases less than the others. We therefore implemented further program variables that allow to define a different n depending on the nature of the mismatch (i.e. it is possible to set a particular n value for each possible type of mismatch).
The overall relative stability of a given duplex is then expressed as a probability function. It is expressed as the sum of products of the individual position probabilities p~1~ (determined by the stability function) and p~2~ (determined by the border parameter). The value of p~2~ it the probability of \"melting\", conditioned that the \"weak point\" is disturbed. (Equation 2).
Equation 2. L -- the length of the duplex, p~1~ -- the \"weak point\" location probability, p2 -- the \"melting\" probability due to the disturbance of the \"weak point\".
p~2~ is a conditional probability of \"melting\" with p~2~ = 1 if the \"weak point\" is disturbed (in the region y ± n) and p~2~ = 0 at non-affected positions. This allows transforming Equation 2 into Equation 3.
Equation 3. y -- the mismatch position, n -- the border parameter
p~1~ can then be substituted by the function in Equation 1, to yield Equation 4.
Equation 4: x -- the duplex position, y -- the mismatch position, n -- the border parameter, σ-distribution parameter
In the case of several mismatches, the summing is done along all the respective mismatch regions. If the mismatches occur next to each other, their disturbed regions simply overlap and the summing is performed across the respective region.
Probe finding
-------------
The probe finding strategy is devised in a way (i) to avoid the need for exact alignments, (ii) to check probe specificity along the whole available sequence and (iii) to optimize performance. The workflow is depicted in Figure [1](#F1){ref-type="fig"}. It starts with a database in which each organism is represented by a single continuous sequence, such as a defined region of the 18S or 28S ribosomal genes. From this it takes first the sequences of the In-group organism(s) for which specific probes should be found and cuts these into short pieces of the specified oligo-nucleotide length (set as a program variable), following an approach proposed by Bavykin et al \[[@B11]\]. This is accomplished by a sliding window scheme with 1-nucleotide shifts across the whole length of the sequence(s). Two separate lists are created in this way. The first list is simply a straight list of all possible fragments from all In-group organisms. The second one consists of an array of lists for each of the In-group organisms (the two lists are identical if only one In-group organism is chosen). All duplicate oligos from the first list are then removed and each of the remaining oligos is checked whether it matches with each of the In-group organisms in the second list. A match is positive, when the relative melting probability is within the range of 0--25%, employing the function of Equation 4. Thus, this first calculation simply ensures that all candidate probes match with all In-group organisms. This calculation would be largely dispensable, if only a single In-group organism is chosen.
{#F1}
The next step is to subtract all oligos that match in any of the Out-group organisms. To avoid the comparison of all candidate oligos against all Out-group sequences, we identify first a group of sequences that is closely related to the In-group. For this one requires a rough alignment of all sequences, to calculate percentage similarity between them. Note that this serves only to identify a subgroup of sequences for speeding up the calculations, i.e. mistakes in the alignment are of no concern. The similarity calculator in the program extracts this related group of sequences by a simple percentage identity calculation across the given alignment. All sequences that are at least 90% similar to the In-group are used as Related-group. This percentage can be set as a program variable and should be set such that the Related-group does not become more than 5--10% of all sequences.
The sequences of the Related-group are again converted into a fragment list as above, duplicates are removed and all candidate oligos are matched with this list. Now only those oligos are retained, which have a melting probability of at least 75% (the exact percentage values are program variables). The majority of oligos is removed in this step. The remaining candidate oligos are then matched against the remaining sequences in the Out-group with the same cut-off criterion.
This stepwise selection scheme allows to significantly speed up the calculations even for very large datasets, but still ensures that all oligo-nucleotides of the desired length were directly or indirectly matched against all possible other oligos in the database.
Single nucleotide loops
-----------------------
Structure analysis with experimental oligo-nucleotides has shown that in a pair of hybridized oligos, one nucleotide can loop out, without interfering much with the stability of the hybridized pair \[[@B12]\]. This implies that one base of one strand of a duplex can loop out from the duplex and the rest of the strand can shift one position. This is depicted in Figure [2](#F2){ref-type="fig"}. A standard linear scanning algorithm would recognize the situation at the left as one with 11 mismatches, i.e. would suggest it as a specific probe. However, if the single nucleotide loop is taken into account, the match would be perfect and the probe would have to be considered as unspecific. Our scanning algorithm takes this problem into account by re-checking all candidate probes after the completion of the filtering steps. It does this by sequentially removing one nucleotide from the candidate probe and shifting the remainder by one position. The melting probability of the new oligo is then calculated and checked. The removed nucleotide is then reinserted and the cycle is repeated for the next position. The same procedure is done for the target sequence, so that outloops are considered to be possible on both strands of the duplex. Note that outloops of two nucleotides are considered to destabilize the helix too much to warrant a separate analogous calculation.
{#F2}
Parallel computation
--------------------
A parallel program version allows probe finding to be done in parallel on several processors. Essentially the same algorithm is used in the parallel version of the program, whereby the parallelism is introduced in the matching steps. Each process takes its own part of the database and performs the matching as well as the stability calculations. The results are then gathered by the root process and superimposed.
Program implementation
----------------------
The algorithm is implemented in a program called PROBE. The program consists of three modules that can be used independently. The first module finds the probes based on the given task (specificity group, length of probes, source database).
The second one is the analytic module, which can be used if it is impossible to design a probe for a given organism group. This module depicts the situation with the given In-group and enables to find the closest group for which the task can be accomplished. The use of the analytic mode comes into play when PROBE fails to identify a set of probes for the given organism group. Such a failure can have two reasons -- either there is no probe, which identifies all organisms in the specificity group, or there is another organism outside the specificity group, which is also identified by all candidate probes suitable for the specificity group.
For the first case, the specificity group must be broken down into several subgroups and the probes must be identified for these subgroups separately. For the second case, the organism that is very similar to the specificity group should be added to the specificity group and this may then have to be broken down into smaller subgroups.
The analytic module creates a table with the organisms of the specificity group as well as the most related organisms. This table depicts then the matching or non-matching patterns for each of the possible probes, allowing a simple visual inspection of the best specificity groups. The output can be viewed and modified with spreadsheet programs such as Excel.
The third module provides a report for the identified probe, including the mismatches in the duplexes within the specificity group, the best match out of the group and some other information.
The program is written in standard C++ in a platform independent manner. Therefore, the program can be easily compiled for Linux and Windows without any modifications. The program binary files for Linux and Windows are available from the <> as freeware accompanied with all its source files, and a manual that describes further details.
Results
=======
As an example of the performance of the program we have used the full SSU database (RDP, release 8.1) \[[@B13]\] containing approximately 16.000 sequences to find a specific oligo-nucleotide probe with a length on 20 nt for *Thermotoga maritima.* The search was done on a Pentium III (800 MHz, 512 MB RAM) PC and took about 1.5 hours without outlooping and 16 hours with outlooping, indicating that the most time intensive step is the outlooping subroutine. The parallel version running on a cluster with 24 nodes (with the slowest node being a Pentium II -- 400 MHz with 256 MB RAM) took 2 hours for the same full task.
Figure [3](#F3){ref-type="fig"} depicts the output from the check module, which allows comparing the oligos and their specificity that were found in this particular comparison. It shows that ARB suggests two oligos that are rejected by PROBE either because of mismatches occurring only at the ends, or under the outloop routine. Both programs find one oligo with acceptable high specificity.
{#F3}
Discussion
==========
The algorithm presented here does not take into account the effect of relative GC content and stacking interactions of neighboring bases on the melting temperature of the oligo-nucleotides. Accordingly, the oligo-nucleotides suggested by the program can differ significantly in melting temperature. However, as this can easily be adjusted after the selection is made, we have not included a subroutine that takes GC content into account during the primary search, because this would slow down the calculations. Furthermore, we expect that GC content differences may be of less importance for the applications envisioned here, because they can be largely compensated by the choice of experimental conditions, such as buffers that compensate stability differences \[[@B13]\].
A more general problem is our way of calculating the relative stability factor. This does currently not take the nucleotide composition into account either. The reason is that there are too few experimental data as yet, that would allow to unequivocally include this in the calculations. The current experimental data sets focus on the types of mismatches in particular contexts, but not systematically on position specific effects \[[@B7],[@B15]\]. Moreover, they deal with relatively short model oligos only (up to 12 nt). However, the probes used for species identification are longer and the different effects can currently not be accurately assessed from experimental data for such longer probes. In our equation, it is mainly the border parameter n that would be affected by base composition and nearest neighbor interactions and we have therefore left this as a variable that can be set according to experimental results. In principle, it seems possible that n differs for different sequence compositions, i.e. GC-rich stretches have a smaller n than AT-rich ones. Thus, if one chooses a low n, one would risk that GC-rich oligos are suggested as specific probes that still show cross hybridization. However, it seems that these can easily be eliminated after the selection is made. Still, if experimental data indicate that this is a major problem, the program could easily accommodate such new insights.
Finally, the stability function proposed in Equation 1 could possibly also have other shapes than Gaussian. Again this is a factor that needs further experiments. If it turns out that other functions are more appropriate, one can include this as additional options into the program. At the present we offer the extreme, namely a flat function, as an alternative option.
Conclusion
==========
We have designed a versatile algorithm for finding optimal species- and group-specific probes for molecular taxonomy that is sufficiently open to implement further experimental insights into the nature of the stability of mismatched oligo-nucleotides.
Acknowledgements
================
We are grateful to Dr. Lysov from the Engelgardt Institute of Molecular Biology, Russian Academy of Sciences for the supporting A.P. in the initial phase of the project. We thank Prof. Speckenmeyer at the Institute of Informatics, University of Cologne for providing access to their LINUX cluster and Jens Rühmkorf for his help with installing the parallel version. This project was supported by a grant from the Ministerium für Schule Wissenschaft und Forschung des Landes Nordrhein-Westfalen. | {
"from": "PMC100321.md"
} | {
"alnum_ratio": 0.7977999211,
"avg_line_length": 181,
"char_rep_ratio": 0.06109658,
"flagged_words_ratio": 0,
"lang": "en",
"lang_score": 0.9257469773,
"max_line_length": 1593,
"num_words": 3702,
"perplexity": 489.1,
"special_char_ratio": 0.2119179163,
"text_len": 20272,
"word_rep_ratio": 0.008665042
} | 9,161,345,934,003,701,000 |
Background
==========
Nerve growth factor (NGF), a prototypical neurotrophic factor and a member of the neurotrophin family, promotes a wide range of responses in its target cells. These range from neuronal differentiation, maintenance of survival, and regulation of metabolic activities \[for review see \[[@B1]-[@B3]\]. Many of these actions include and require transcriptional regulation \[[@B4],[@B5]\]. However, the greater part of the changes in gene expression that underlie the NGF response remain to be elucidated. The PC12 line of rat pheochromocytoma cells \[[@B6]\] has proved to be a particularly favorable system for detecting NGF-responsive changes in gene expression \[[@B7]-[@B11]\]. These cells resemble sympathico-pheochromo-blasts and upon exposure to NGF cease proliferation and acquire, in a transcription dependent mechanism, many of the properties of post-mitotic sympathetic neurons including neurite outgrowth and electrical excitability. The robust nature of the response of PC12 cells to NGF coupled with their capacity to be examined both before and at various times after exposure to the factor has greatly facilitated study of the NGF mechanism of action, including gene regulation \[[@B12]\].
To obtain a comprehensive and quantitative over-view of NGF-promoted gene regulation, we \[[@B11]\] have used SAGE (Serial Analysis of Gene Expression). In this technique \[[@B13]-[@B15]\] cellular transcripts are converted to SAGE \"tags\" which are sequenced, quantified and, in many cases, matched with known genes. By comparing SAGE profiles for cells in different states (as for example before and after exposure to NGF), it is thus possible to obtain a comprehensive view of gene expression and regulation. Moreover, if a sufficient number of SAGE tags are analyzed, changes in expression levels of individual transcripts can be associated with a high level of statistical significance \[[@B13]\].
Initial analysis of approximately 157,000 SAGE tags from PC12 cells cultured without or with NGF for 9 days revealed nearly 800 transcripts (of a total of at least 21,000) that are regulated by ± 6-fold or greater in response to NGF \[[@B11]\]. Of these, approximately 150 were assignable to named genes of known functions that regulate cellular behaviors ranging from actin and microtubule cytoskeleton assembly/disassembly, gene transcription, RNA processing, neurotransmission, and energetics. A variety of criteria supported the reliability of the quantitative findings revealed by our SAGE analysis \[[@B11]\].
In the present study, we have extended our SAGE profiling of naïve and long-term NGF-treated PC12 cells to over 163,000 tags representing over 22,000 unique transcripts. Analysis of these reveals the presence of transcripts encoding 74 different ribosomal proteins (RPs). Surprisingly, we find that long-term NGF exposure leads to statistically significant changes in relative abundance of at least half of these transcripts by factors of up to nearly 5-fold. In addition, a time course for one of the RP transcripts (encoding RP L9) reveals that its relative abundance begins to change within 1 hr and is maximally regulated by 8 hr of NGF exposure.
Results
=======
SAGE libraries
--------------
SAGE libraries were generated and analyzed as previously described \[[@B11]\] from matched sets of PC12 cells before and after 9 days of exposure to NGF. The present analysis is based on 76,280 15-mer tags from NGF-untreated PC12 cells and 87,004 tags from NGF-treated cells (after exclusion of duplicate ditags, mitochondrial transcripts and repetitive elements). Consideration of tags observed twice or more between the two libraries indicated the presence of approximately 22,000 unique transcripts. Of these, approximately 10% were regulated by more than 3-fold in response to NGF and approximately 4% by 6-fold or more.
Detection and quantification of transcripts encoding ribosomal proteins
-----------------------------------------------------------------------
Transcripts represented by SAGE tags were identified by direct matches of tags with the appropriate sequences of known rat genes (present in GenBank) or through matches of tags with appropriate sequences of rat ESTs (present in NCBI Unigene) that were in turn found to overlap with known rat genes. Positive identification required that the transcript or EST have a poly adenylation signal and poly A tail, and that the tag followed the most 3\' CATG of the transcript. In this way 74 tags were unambiguously assignable to known rat transcripts encoding proteins described as ribosomal components. Available sequences encoding rat ribosomal proteins L15 and S8 do not contain a CATG sequence and hence SAGE tags for these could not be identified. In addition, rat sequence data for transcripts encoding RPs L2, L25, L33 and S1 are not presently available in GenBank or Unigene.
Table [1](#T1){ref-type="table"} lists the tags in our libraries corresponding to transcripts for ribosomal proteins along with their relative abundances. Considering that the eukaryotic ribosome contains approximately 82 proteins, our analysis includes transcripts encoding a major proportion of the known ribosomal proteins.
::: {#T1 .table-wrap}
::: {.caption}
######
The effect of NGF (9 days treatment) on expression levels of transcripts for various ribosomal proteins in PC12 cell cultures
:::
TAG RP ACC\# \#TAGS- \#TAGS+ FOLD CHANGE +NGF *P* value
------------- ---------- ----------- --------- --------- ------------------ -----------
GGACCGCTCAA L3 X62166 44 27 ↓ 1.6 0.06
TTGAAGCTGAA L4/L1 X82180 49 39 ↓ 1.3 0.2
CTGCTATCCGA L5 X06148 40 28 ↓ 1.4 0.15
TACCCTCACAA L6 X87107 16 24 ↑ l.5 0.05
AGATCTATACA L7 Ml 7422 8 7 ↓ 1.1 0.5
CACCACTGTTG L7A X15013 63 47 ↓ 1.3 0.2
AATCCTGTGGA L8 P25120 119 58 ↓ **2.1** 0.0002
ATCAAGGGTGT L9 X51706 7 19 ↑ **2.7** 0.01
TTCAATAATAA L10 X87106 24 17 ↓ 1.4 0.2
GGCAAGCCCCA L10A X93352 26 30 ↑ 1.2 0.2
CGCTGGTTCCA L11 X62146 18 23 ↑ .3 0.1
ACATCATAGAT L12 R7RT12 59 15 ↓ **3.9** \< 0.0001
GCCCGAGCCAA L13 X78327 48 57 ↑ .2 0.1
AGGTCGGGTGG L13A X68282 87 118 ↑ **1.3** 0.002
AGGAGGCTACA L14 X94242 31 51 ↑ **1.6** 0.003
GCACGGGAATA L17 X58389 8 26 ↑ **3.2** 0.0007
GGTGTTGACAT L18 M20156 47 12 ↓ **3.9** \< 0.0001
AAGGTGGAAGA L18A JC4231 56 33 ↓ 1.7 0.03
GATCAGTCATT L19 X82202 65 63 ↓ 1.0 0.3
GCCTAATGTAT L21 M27905 43 32 1.3 0.2
TTTTGTATTAA L22 X78444 5 4 1.3 0.7
GTGATGGCCAC L23 X58200 72 63 ↓ 1.1 0.4
AAGGTCGAGCT L24 X78443 95 28 ↓ **3.4** \< 0.0001
CCCAGTTTTCA L26 X14671 31 19 ↓ 1.6 0.1
CCCACAAGGTA L27 X07424 16 30 ↑ 1.9 0.007
ATCCGAAAAAA L28 X52619 42 10 ↓ **4.2** 0.0001
GCCAAGGGTCG L29 X68283 97 40 ↓ **2.4** \< 0.0001
CCAGAACAGAC L30 X52619 36 9 ↓ **4.0** 0.0002
AAGGAGATGGG L31 X04809 64 111 ↑ **1.7** \< 0.0001
CTGCCTAGCGG L32 X06483 26 20 ↓ 1.3 0.2
TGCGCCAAGTG L34 X14401 27 30 ↑ 1.1 0.2
AAGAGAAGCTG L35 X51705 115 70 ↓ **1.6** 0.009
GTTCGTGCCAA L35A X03475 33 16 ↓ 2.1 0.04
CGGAAGGCGGC L36 X68284 38 77 ↑ **2.0** \< 0.0001
GATTCCGTGAA L37 X66369 58 79 ↑ **1.4** 0.01
AAAACAGTGGC L37A X14069 55 90 ↑ **1.6** 0.0003
TGACTATTAAA L38 X57007 25 21 ↓ 1.2 0.5
TCTTCTCACAA L39 X82551 15 27 ↑ 1.8 0.03
CAGATCTTCGT L40 X82636 89 32 ↓ **2.8** \< 0.0001
AGAGCGAAGTG L41 X82550 185 80 ↓ **2.3** \< 0.0001
CAAGGTGACAG S2 U92700 52 47 ↓ 1.1 0.4
CCTCAGCCAGT S3 X51536 21 35 ↑ 1.7 0.03
GTGAAGGCGGT S3A X75161 46 43 ↓ 1.1 0.4
ATGAAATCAAA S4 X14210 44 20 ↓ **2.2** 0.01
CCTTTGAGATC S5 X58465 72 34 ↓ **2.1** 0.004
GCAGAGTGCGC S6 NM_017160 63 27 ↓ **2.3** 0.0006
TTCAGCTCGAG S7 X53377 41 31 ↓ 1.3 0.3
CCCGTGTGCTC S9 X66370 54 32 ↓ 1.7 0.06
CAGTCTCTCAA S10 X13549 57 36 ↓ 1.6 0.02
TCTGTGCACCT S11 K03250 23 18 ↓ 1.3 0.4
TATGTCAAGCT S12 M18547 92 52 ↓ **1.8** 0.006
GTGTGGCACAG S13 X53378 65 65 1.0 0.2
TTGGCTGCCCA S14 X15040 99 50 ↓ **2.0** 0.0008
GTGGGTGTGTA S15 NM_017151 254 60 ↓ **4.2** \< 0.0001
AAGAGGCAAGA S15A X77953 35 48 ↑ 1.4 0.04
TGGCCCAAATT S16 X17665 155 48 ↓ **3.2** \< 0.0001
GGCCGCGTTCG S17 K02933 18 39 ↑ **2.1** 0.003
CAGAACCCACG S18 X57529 101 26 ↓ **3.9** \< 0.0001
ACCAAGATCTA S19 X51707 136 28 ↓ **4.8** \< 0.0001
CCTACCAAGAC S20 X51537 8 16 ↑ 2.0 0.07
GGTCTGGCTAG S21 X79059 0 3 ↑ 3.0 na
CCGTGGGTGAT S23 X77398 52 68 ↑ 1.3 0.03
GCCTTTATGAG S24 X52445 310 87 ↓ **3.6** \< 0.0001
CCGCCCAAAGA S25 X62482 90 43 ↓ **2.1** 0.0003
GAAAAATAAAA S26 X02414 35 16 ↓ 2.2 0.03
CACAAACGGTA S27-1 AF184893 99 46 ↓ **2.2** 0.0004
GGTAGCCACTT S27A X81839 33 77 ↑ **2.3** \< 0.0001
GAATGACCTGC S28 X59277 40 37 ↓ 1.1 0.4
CTAGTCTTTGT S29 X59051 125 123 ↓ 1.0 0.3
GTTCTCTGGCT S30 X62671 29 33 ↑ 1.2 0.3
GGATTCGGTCT P0 Z29530 40 55 ↑ 1.4 0.04
TCCAATAAAGA P1 R5RT12 96 48 ↓ **2.0** 0.0002
GGATTTGGCCT P2 X15098 83 84 ↑ 1.0 0.3
GGAGGTTATGC 40 KD RP D25224 83 34 ↓ **2.4** \< 0.0001
Rat SAGE tags and the corresponding ribosomal proteins (and GenBank accession numbers) are given along with the number of times each tag was detected. For cells before and after NGF treatment, a total of 76,280 and 87,004 11 bp tags were analyzed, respectively. Tag numbers for non-treated cultures were normalized against those for NGF treated cultures. *P* values were calculated by Monte Carlo simulations using SAGE software. Fold changes in which *P* ≤ 0.01 are expressed in **bold.**
:::
Relative expression of transcripts for ribosomal proteins
---------------------------------------------------------
Among the information provided by analysis of SAGE data is the relative abundances of transcripts. The data given in Table [1](#T1){ref-type="table"} and Figure [1](#F1){ref-type="fig"} show the relative abundances of transcripts for PC12 cell ribosomal proteins with respect to one another as well as with respect to the total cell complement of transcripts. As noted above, our analysis has detected at least 22,000 unique transcripts in PC12 cells. The 74 ribosomal protein transcripts identified here thus account for no more than 0.33% of this total. By contrast, for NGF-untreated and -treated PC12 cells, tags corresponding to the 74 identified RP transcripts represent 5.2% and 3.5%, respectively, of total tags analyzed. As anticipated, this clearly places transcripts for ribosomal proteins as a whole in the high abundance category.
{#F1}
The most abundantly expressed ribosomal transcripts in NGF-untreated PC12 cells included those encoding ribosomal proteins S15, S24 and L41. These each accounted for approximately 0.2--0.35% of total cell transcripts. The high relative abundance of these ribosomal transcripts in NGF-untreated cells can be appreciated by the observation that of the 4 tags encountered more than 200 times in our analysis, 2 (S15 and S24) encoded ribosomal proteins; moreover, RP transcripts represented 6 of the 13 tags encountered from 100--200 times and 23 of the 36 tags encountered from 50--100 times.
The data in Table [1](#T1){ref-type="table"} and Figure [1](#F1){ref-type="fig"} show that NGF treatment resulted in an overall decrease in relative abundance of transcripts encoding RPs (from 5.2% to 3.5% of total). This may reflect in part our observation that NGF treatment increases the numbers of transcripts representing low-abundance genes \[[@B11]\]. Nevertheless, transcripts for ribosomal proteins remained among the most highly expressed in the NGF-treated cells. For instance, tags for RPs L13A, L31, L37A, S24 and S29 had relative abundances of 0.1% or more. Of the 56 most abundantly expressed tags in NGF-treated cells, 21 represent ribosomal transcripts.
NGF selectively regulates the expression of transcripts encoding ribosomal proteins
-----------------------------------------------------------------------------------
In addition to yielding an overall decrease in the abundance of RP transcripts relative to total cellular transcripts, long-term NGF treatment also promoted selective changes in relative expression of transcripts encoding individual RPs (Table [1](#T1){ref-type="table"}, Figs. [2](#F2){ref-type="fig"},[3](#F3){ref-type="fig"}). Monte-Carlo simulation analysis of the SAGE data indicated that nearly half (a total of 35) of the 74 RP transcripts underwent changes that were significant at the *P* ≤ 0.01 level (see Table [1](#T1){ref-type="table"}). Of these 35 transcripts, 2/3 were down-regulated in response to NGF.
{#F2}
![**Incremental changes in expression of RP transcripts evoked by long-term treatment with NGF.** NGF-promoted changes in expression of RP transcripts are given in incremental units. Up-regulated transcripts are given as \[(number of normalized tag numbers + NGF/number of normalized tag numbers - NGF) - 1\] and are thus expressed as positive values. Transcripts that are down-regulated by NGF are given as --\[(number of normalized tag numbers - NGF/number of normalized tag numbers + NGF) - 1\] and are thus expressed as negative values. For transcripts that do not change expression in response to NGF, the incremental change is zero. **A.** Transcripts for RP L3-L41 and P0-P2. **B.** Transcripts for RP S2-S30 and 40 KD RP.](1471-2202-3-3-3){#F3}
Figure [2](#F2){ref-type="fig"} shows the distribution of fold-changes in transcript levels in response to NGF. 47 of the 74 RP transcripts were either invariant or showed changes of 2-fold or less; 10 individual transcripts decreased by 3--5 fold (L12, L18, L24, L28, L30, S15, S16, S18, S19 and S24) and one was elevated by more than 3-fold (L17).
In a past study \[[@B11]\], the differences in NGF-promoted gene expression revealed by SAGE analysis of our libraries were found to be highly reliable based on 1) the absence of regulation shown by a number of anticipated \"housekeeping\" genes including β-actin, 2) the observation that the vast majority of transcripts were not responsive to NGF, 3) the detection of anticipated changes in expression of a number transcripts previously reported to be NGF responsive, and 4) the agreement in relative expression of 20 genes as revealed by SAGE and northern blot analyses. To further confirm the reliability of our SAGE findings, we used real time quantitative PCR to compare levels of 5 different RP transcripts in PC12 cells treated with or without NGF for 9--12 days. As shown in Figure [4](#F4){ref-type="fig"}, both techniques revealed similar changes in expression.
{#F4}
Rapid regulation of an RP transcript by NGF
-------------------------------------------
We next determined the time course with which NGF regulates an RP transcript. For this purpose, we studied RP L9 which undergoes a 2.7-fold elevation in response to long-term NGF treatment. Because the changes in expression are relatively small, we used real time PCR for this end. As shown in Figure [5A](#F5){ref-type="fig"}, the relative abundance of transcripts encoding RP L9 were significantly upregulated by 1 hr of NGF treatment (1.8 ± 0.2-fold, n = 14) whereas there was no significant change in transcripts encoding RP S29, which does not undergo long-term NGF regulation. A time course (Figure [5B](#F5){ref-type="fig"}) revealed that elevation of L9 transcripts reaches maximal levels within 8 hrs of NGF exposure.
![Rapid regulation of RP L9 transcripts by NGF. **A.** Elevation of RP L9 transcripts after 1 hr of NGF exposure. PC12 cell cultures were treated with or without NGF for 1 hr and used for preparation of RNA and cDNA. Relative levels of L9 and S29 transcripts were determined by real-time PCR with normalization against levels of transcripts for GAPDH. Data are given as ratios for values of \[NGF treated/untreated\] and are expressed as means ± SE (n = 14 for L9 and n = 6 for S29). **B.** Time course for NGF response of transcripts encoding RP L9. PC12 cultures were treated with NGF for the indicated times and used for preparation of RNA and cDNA. Relative levels of L9 transcripts were determined by real-time PCR with normalization against levels of transcripts for GAPDH. Values are given as means ± SE (n = 3).](1471-2202-3-3-5){#F5}
Discussion
==========
NGF regulates expression of RP transcripts
------------------------------------------
In the present work, we used SAGE profiling to identify and quantify the relative numbers of transcripts for 74 ribosomal proteins in rat PC12 cells before and after long-term exposure to NGF. This has permitted us to detect selective changes in transcripts encoding specific RPs.
A number of criteria support the reliability of our findings. The first regards the assignment of SAGE tags to specific RP transcripts. We required that all matching ESTs or transcripts have clearly definable poly A tails and poly adenylation signals. Due to the extensive data base for rat RP transcripts, it was possible to make all matches with rat sequences. In addition, we used 15 mer (CATG +11) base SAGE tags for our analysis; we have reported \[[@B11]\] that this leads to a significantly more reliable matching of SAGE tags to genes than with the often-used 14-mer tags. A second indicator of reliability stems from prior analysis of our SAGE libraries and data obtained from them \[[@B11]\]. For instance, we have shown that our SAGE libraries show little or no NGF-promoted changes in tag numbers for transcripts encoding a number of housekeeping proteins. Moreover, a number of transcripts previously found by alternative technologies to respond to NGF treatment, showed similar changes in our SAGE profiling study. Finally, Northern blot analysis verified SAGE-predicted responses of over 20 transcripts to NGF treatment. A third criterion for reliability was that Monte Carlo simulation indicated that many of the NGF-promoted responses of RP transcripts were at a probability of *P* \< 0.01. Because of the relatively high abundance of many RP transcripts, even comparably small changes in expression could be detected at this level of significance. A last criterion was that we used real time RT-PCR to verify effects of NGF on five RP transcripts.
A previous study by Lee et al. \[[@B8]\], based on comparison of a total of approximately 7,000 random ESTs from naïve and NGF-treated PC12 cells, reported an NGF-promoted increase in expression of RPL7 transcripts and a decrease in RPL19 transcripts. This contrasts with the current data which revealed no significant change in expression of either of these transcripts. The reason for this discrepancy is unclear, but could in part originate from the relatively small number of transcripts that were surveyed in the former study. To our knowledge, there are currently no other reports regarding effects of NGF on levels of ribosomal transcripts.
Relative levels of RP transcripts
---------------------------------
Our observations indicate that there is a wide range in the numbers transcripts per cell that encode individual RPs. For instance, both before and after NGF treatment, there was over a 10-fold difference in the relative numbers of transcripts for RPs L7, L22, S21 as compared to those encoding RP S29. Because of the general scarcity of antisera prepared against mammalian RPs, we do not know whether this is reflected at the protein level. However, for a number of other NGF-regulated transcripts in PC12 cells, there is a good correlation between relative levels of message and protein \[[@B11],[@B16],[@B17]\]. Thus, it may be that although levels of individual RPs are assumed to be similar to one another, some may be present in limiting numbers. Alternatively, there may be a considerable disparity between relative abundances of RP transcripts and their corresponding proteins
A recent SAGE study of targets for N-myc in a human neuroblastoma cell line \[[@B18]\], reported relative abundances for 66 RP transcripts. The existence of such data permit us to compare the relative levels of RP transcripts in two cell types (neuroblastoma and pheochromocytoma) of related origin (i.e., neural-crest-derived) that both have the potential for neuronal differentiation. The 74 RP transcripts detected here represented 5.2% of the total transcripts in NGF-untreated PC12 cells and 3.5% after NGF treatment. By contrast, the 66 RPs reported in the neuroblastoma study accounted for 4.1% of total transcripts for cells without N-myc over-expression and 12.6% for such cells transfected with N-myc. Thus, for non-N-myc transfected neuroblastoma cells and PC12 cells (± NGF) the contribution of total RP transcripts lies in a similar range and this parameter is greatly elevated in neuroblastoma cells by N-myc over-expression. With respect to transcripts for individual RP\'s, there are many similarities as well as several striking differences between the two cell types. When NGF-untreated PC12 cells are compared with non-transfected neuroblastoma cells, of the 55 RP transcripts detected in common for both systems, about half (a total of 28) have relative abundances within a factor of 2. Twenty transcripts are more than 3-fold higher in relative abundance in PC12 cells. In two of these cases (S9 and S17) no tags were detected in the non-transformed neuroblastoma cells; in several other cases (S12, S24, L24) the relative abundance in PC12 cells was over 20-fold that in the neuroblastoma cells. For most of these, expression of N-myc elevated the neuroblastoma levels to relative abundances within 2-fold of those in PC12 cells. However, for L24 and S24, the relative abundance in N-myc-expressing neuroblastoma cells was still 1/3 that in PC12 cells. Finally 7 RP transcripts (L7, L11, L21, L30, S7, S8 and P2) were of 3--5 times lower abundance in PC12 cells than in neuroblastoma cells and this difference was appreciably enhanced when the NB cells were transfected with N-myc.
We have also compared our RP results with those for SAGE carried out with pooled human adult brain (tissue supplied by Gregory J. Riggins: <>). Comparison with data for NGF-treated cells reveals similarity in relative abundance within a factor of 2 for 39 of the 57 RP transcripts in common between the libraries and 48 of 57 to within a factor of 3. However, several large differences do occur. For example the relative abundances of transcripts encoding RPs L9, L39, S24, S13, and S17 are 5--10-fold higher in NGF-treated PC12 cells than in the pooled brain library and transcripts for RP L28 are 8-fold lower. These observations reinforce the notion that expression of individual RP transcripts can be significantly variable from cell type to cell type as well as sensitive to extrinsic signals.
Potential significance of RP transcript regulation
--------------------------------------------------
Although the ribosome has been considered as a \"molecular machine\" \[[@B19]\], it is of interest that the transcripts encoding individual proteins of this organelle are subject to regulation by NGF. Two related questions emerge regarding these findings: how do these changes compare with previous reports for regulation of RP transcripts and what might be the functional consequences of these changes?
A number of studies have employed a variety of techniques to detect changes in gene expression associated with the oncogenic state and have reported selective elevation of specific RP transcripts in tumors. Examples include RPs L7a, L37 and S14 in prostate tumors and cell lines \[[@B20]\]; RPs L5, L7A, L18, S3, S6, S8, S12, S13, S28, P0 and P1 in colorectal cancers and cell lines \[[@B21]-[@B26]\]; L5 in astrocytomas \[[@B27]\]; L18a in squamous cell carcinoma \[[@B28]\]; L19 in breast tumors that over-express erbB-2 \[[@B29]\]; RPs S3A, S4, and S17 in lymphoid malignancies \[[@B30]\]; L38, S4, P0, and P1 in rhabdomyosarcoma cell lines \[[@B31]\]; and P0, P1, P2, L5, L9, L35, L39, S3A, S10, and S17 in liver tumors \[[@B32]\]. In addition, as noted above, a recent SAGE study identified targets of N-myc in a human neuroblastoma cell line \[[@B18]\]. Of 114 up-regulated genes detected, 66 encoded RPs with elevations ranging from 40% to 37-fold. Several of these genes were also up-regulated by c-myc. Taken together, these findings indicate that the transition to the transformed state is associated with elevation of various RP transcripts. A potential interpretation of such observations is that this up-regulation reflects the enhanced rate of cell division in tumor cells and the requirement for greater levels of protein synthesis. In agreement with this possibility, transcripts encoding RPs L6 and S7 are reported to be up-regulated in regenerating liver \[[@B33]\]. However, for the study involving responses of neuroblastoma cells to N-Myc, despite the massive up-regulation of RP transcripts, there was no over-all increase in the rate of protein synthesis \[[@B18]\].
Before considering potential functional consequences of the changes in RP expression described here as well as elsewhere, it must be conceded that changes in transcript abundance may not necessarily lead to changes in protein expression. The present lack of available antibodies/antiserum to most mammalian RPs hampers such a determination. However, even if cases occur in which altered RP transcript expression does not lead to changes in expression of the corresponding protein, our and others\' findings regarding selective regulation of RP transcripts would then raise the interesting issue as to why, and the mechanism by which, transcript and protein expression are uncoupled.
If at least some of the NGF-promoted alterations in RP transcript abundance lead to changes in expression of the corresponding proteins, what might be the functional consequences? In the present system, NGF converts proliferating PC12 cells to a non-proliferating neuronally differentiated state. Our findings reveal that the relative overall abundance of transcripts for RP proteins fell by 1/3 in response to NGF. In addition, nearly half of the individual RP transcripts detected showed significant changes in expression and approximately 2/3 of these were decreases. This overall decrease in NGF-promoted RP transcript expression and the preponderance of decreases in expression of specific RP transcripts compared to increases would favor the interpretation that the changes observed here reflect, at least in part, the transition to the non-dividing phenotype. On the other hand, we observed that many of the RP transcripts did not show significant responses to NGF and that a number showed increases in expression. This suggests that the situation is likely to be more complex with individual RPs perhaps playing specific roles not simply associated with the state of cell proliferative capacity. In agreement with this, we found no significant change in expression of RP S29 transcripts even though these have been reported to be present at low levels in growth phase cells and elevated in quiescent cells \[[@B34]\].
In addition to leaving the cell cycle, NGF-treated PC12 cells undergo neuronal differentiation. This raises the possibility that some of the observed changes, as for many other NGF regulated genes, is related to acquisition of the neuronal phenotype. An analysis of gene regulation in human NTERA2 cells induced to leave the cell cycle and neuronally differentiate in response to retinoic acid revealed decreases in RP transcripts including L3, L7, L8, L10, L13, L39, S2, S6, S13, S16, S20, S19, S23, S27A and P0 \[[@B35]\]. Of these, In the present study only RP transcripts L8 and S6, S16, S19 showed significant down regulation in response to NGF whereas the others found in the NTERA2 study either did not show changes that were significant at the *P* ≤ .01 level or underwent a significant increase (RP S27A). Curcic et al. \[[@B36]\] reported that differentiation of BC3H1 myocytes is accompanied by a drop in L32 gene transcription; in contrast, we found no significant change in transcripts encoding this RP. In another study, RPs L35a and S5 were down regulated during murine erythryolukemia cell differentiation \[[@B37]\]; we also observed decreases in these transcripts. Mutation of RP S19 is associated with Blackfan\'s anemia and therefore appears to play a selective role in differentiation/proliferation of erythropoetic cells \[[@B39]\]. Here, NGF resulted in a nearly 5-fold drop in S19 transcripts. Thus, although changes in expression of specific RP genes may be associated with differentiation, the pattern of such changes may reflect the particular cell type and differentiation stimulus that is involved.
Another condition associated with changes in RP expression is apoptotic death. RP L4/L1 transcripts, which were not significantly affected by NGF, are selectively upregulated in PC12 cells prior to 5 aza cytosine-induced death and over-expression of this gene in COS-7 cells induces apoptosis \[[@B39]\]. Over-expression of L7, L13A, S29 have been also reported to induce apoptosis \[[@B40],[@B41]\]. In the present study, NGF promoted little if any change in expression of these RP transcripts. Thus, although NGF is an effective anti-apoptotic factor, these actions do not appear to be mediated by down-regulation of potentially death-inducing RPs.
At present, in contrast to studies on bacterial and archael ribosomes, relatively little is known about the functions of individual mammalian ribosomal proteins (\[[@B42]\]. However, assuming that changes in message abundance lead to alterations in protein levels, it is reasonable to anticipate that the NGF-promoted effects observed here may affect the protein synthetic capacity of the cell in some manner. Such changes are unlikely to be global; NGF does affect the cellular rate of protein synthesis, but this seems to be due at least in part to post-translational modification of the translational machinery \[[@B43]\]. The alternative is that the changes reported here may result in selective effects on translation of specific messages.
A final, and important possibility to consider is that a number of RPs appear to possess extraribosomal functions \[[@B44]\] and consequently that the changes reported here may lead to responses that do not directly relate to ribosomal function. For example, RP L18, which was significantly down-regulated by NGF, has been reported to negatively regulate double-stranded RNA (dsRNA)-activated protein kinase (PKR) \[[@B45]\]. As an additional example RP S19, which is also down-regulated by NGF, has been implicated in erythropoiesis and has also been described as a monocyte chemotactic factor \[[@B46]\]. As one last example, rat RP L9 which we found to be up-regulated by NGF starting within a few hours and maximally by 8 hrs, has been shown to be the ortholog of the yeast gene *grc5* which appears to be involved in multiple cellular functions including growth control, cytoskeleton control and energy metabolism \[[@B47]\]. Identification of NGF-regulated RPs should now facilitate exploration of their potential roles in the responses of cells to neurotrophins.
Conclusions
===========
1\. SAGE analysis provides a reliable, quantitative picture of ribosomal protein expression in PC12 cells before and after long term exposure to NGF.
2\. Transcripts for ribosomal proteins are among the most abundant transcripts in the cells; however there is a wide range between numbers of transcripts for individual RP transcripts.
3\. NGF promotes an overall decrease in relative RP transcript expression (for the 74 RP transcripts detected, from approximately 5% of total transcripts to 3.5%). This drop represents a decrease in relative expression of individual RP transcripts as well as an increase in message complexity in NGF-treated cells.
4\. Long term NGF treatment of PC12 cells promotes statistically significant changes in expression of over half of the transcripts encoding ribosomal proteins. Decreases in expression outnumbered increases by a ratio of approximately 2:1. The largest observed changes in expression are in the range of 3--5 fold.
5\. For at least one RP transcript (RPL9) the response to NGF is rapid; an elevation is detected within 1 hr of NGF exposure and reaches maximum regulation by 8 hrs.
6\. Examination of the literature reveals many other reports in which expression of individual or groups of RP transcripts are regulated in response to growth factors, differentiation agents or malignant transformation. Comparison with the present data indicate that changes in ribosomal protein transcripts is regulated in a cell and state dependent fashion with a large diversity in the particular RP transcripts that are subject to regulation. Thus, although the ribosome may be regarded as a \"machine\" there is a good deal of potential for plasticity with respect to the expression of its various components. This raises the possibilities that growth factors and other cell regulators may affect ribosomal function, and thereby the capacity of cells to transcribe specific transcripts.
7\. In the absence of information about the specific functions of most individual RPs in mammalian cells, one can only speculate at present on the physiologic significance of the reported changes. However, it seems highly plausible that the observed responses may play roles in NGF-promoted neuronal differentiation. This may be mediated in part by selective effects on translation of certain messages. In addition, past findings support the possibility that at least some of the regulated RPs have extra-ribosomal actions that may affect neuronal differentiation and function.
Materials and Methods
=====================
SAGE and matching SAGE tags with RP transcripts
-----------------------------------------------
SAGE libraries were prepared from naïve and 9-day NGF-treated PC12 cells and sequenced as previously described \[[@B11]\]. To match SAGE tags with RP transcripts, tags were initially analyzed with the National Center for Biotechnology Information (NCBI) rat SAGE tag to gene mapping database <>, which matches possible 14-mer tags with known rat genes and expressed sequence tags (ESTs). With the use of sequences present in the NCBI UNIGENE rat database <> potential matches were further scrutinized to determine whether there was a match at the 15th base and to determine whether the matched sequence was at the most 3\' end of a known rat transcript or EST. We considered only cases in which a clear poly(A) tail and a polyadenylation signal were present at the 3\' end of the transcript or EST. Appropriate ESTs were further analyzed by an advanced BLAST search for matches with known rat RP genes. Monte-Carlo simulation analysis of the SAGE data was carried out using the SAGE 300 software package \[[@B13]\].
Real time quantitative PCR analysis
-----------------------------------
PC12 cells were cultured as previously described \[[@B12]\] in complete medium (85% RPMI 1640 medium, 10% horse serum, 5% fetal bovine serum). Replicate cultures were treated with 50 ng/ml rhNGF (kindly donated by Genentech, Inc.) for the indicated times. Total cellular RNA was isolated as described previously \[[@B11]\] and 1.2 -- 5 μg was used for reverse transcription with 5\'-T~30~NN-3\' primer using Superscript II Rnase H-RT according to the manufacturer\'s specifications (Life Technologies, Grand Island, NY). Quantitative real time PCR reactions were performed as described by Troy et al. \[[@B48]\]. GAPDH was used to normalize input cDNA for samples originating from short-term NGF treatments and β-actin was used for this purpose for samples with long-term NGF exposure.
Forward and reverse primers pairs used for quantitative PCR were: *S29,* 5\'GGTATCACAGGGTAGACAGT3\' and 5\'GGTATCACAGGGTAGACAGT3\'; *L9,* 5\'GAACTCTGAGGAGGGACTTC3\' and 5\'AAACTGTACTTGTTATCAGGAT3\'; *S27-1,* 5\'CGGCACGAGCGACCTCCCTA3\' and 5\'GTTCCCACTCATCTTGAATC3\'; *L24,* 5\'CAAGAAAGGACAGTCGGAAA3\' and 5\'TTCACAGGCTTCACAATCTT3\'; *S19,* 5\'TAACCAGCAGGAGTTCGTCA3\' and 5\'TTTGTTCTAATGCTTCTTGTT3\'; β-*Actin*, 5\'ATCCTGACCCTGAAGTACCC3\' and 5\'TACGACCAGAGGCATACAG3\'; and *GAPDH,* 5\'GAAACCTGCCAAGTATGATGA3\' and 5\'TCTCTCTTGCTCTCAGTATCC3\', respectively.
Acknowledgements
================
We thank Claudine Bitel for her excellent technical assistance. Supported in part by grants from the National Institutes of Health. | {
"from": "PMC100322.md"
} | {
"alnum_ratio": 0.7158799131,
"avg_line_length": 182.4504504505,
"char_rep_ratio": 0.1062847265,
"flagged_words_ratio": 0,
"lang": "en",
"lang_score": 0.9001348615,
"max_line_length": 2103,
"num_words": 7576,
"perplexity": 1556.3,
"special_char_ratio": 0.3345595497,
"text_len": 40504,
"word_rep_ratio": 0.0370027752
} | 17,676,670,310,284,350,000 |
Background
==========
Over the last 50 years, the widespread usage of fluoridated water and fluoridated dentrifices have been cited as major reasons for a decline in caries since the early 1970s \[[@B1]\], and for the appearance of a significant association between oral hygiene and caries experience \[[@B2]-[@B4]\]. An inverse relationship exists between salivary fluoride concentration and caries experience in the deciduous and permanent dentition \[[@B5]\], but fluoride concentration is excluded from most caries prediction models \[[@B6],[@B7]\]. Acids in bacterial plaques cause caries in pits, fissures and interdental regions of teeth, but they also enhance the inhibitory effect of fluoride on demineralization, confounding the ability to predict caries from the salivary fluoride concentration \[[@B8],[@B9]\].
The greater the consumption of dietary sucrose, the greater the fall in pH and fraction of acidogenic, acid tolerant bacteria in tooth adherent plaques \[[@B10],[@B11]\]. The number of these bacteria (mostly mutans streptococci and lactobacilli), and the fluoride content, discriminate between severe and mild caries in 12--15 year-old children \[[@B12],[@B13]\]. Acid-tolerant bacteria require D-alanyl glycerol lipoteichoic acid (D-alanyl LTA) in their membranes and cell surfaces \[[@B14]\]. D-alanyl LTA is made by esterifying carboxyl-activated D-alanine to glycerol in membrane LTA by means of a D-alanyl-carrier enzyme, DCP \[[@B15]\]. Strains of *Streptococcus mutans* in which DCP is inactive do not initiate growth at below pH 6.5 and make glycerol LTA without D-alanine \[[@B14]\]. In the DCP active strains, soluble D-Alanyl LTA is extruded into culture fluid *in vitro*\[[@B16],[@B17]\] or plaque *in vivo*\[[@B18]\]. The D-alanyl esters are stable at pH 6.0 at 37°C, but hydrolyze to free D-alanine and LTA with a half-life of 3.9 h at pH 8.0 \[[@B19]\]. Healthy gingival sulci have a pH of 6.5 -- 7.5 and inflamed sulci a pH of 7.5--8.5 \[[@B20]\].
About 30% of young adults have serum IgG antibodies that precipitate with D-alanyl LTA, but not with D-alanine-free LTA \[[@B17],[@B21]\]. It is likely that plaques induce these IgG antibodies from gingival sulci that contain more acid-tolerant bacteria. An elevated IgG antibody response to D-alanyl LTA may therefore indicate the subjects in whom an inhibitory effect of fluoride on caries is enhanced. The fluoride concentrations of plaque and saliva are related to whether the drinking water is fluoridated \[[@B13]\] and to oral hygiene, which nearly always involves using a fluoridated dentrifice. The aim of this study was therefore to determine whether elevated antibody responders to D-alanyl LTA show a association of DMFT with fluoride exposure and gingival health not apparent in low responders.
Methods
=======
Subjects
--------
Antibody was obtained from blood from four sources: 1) 105 dental students, 2) 147 patients seeking dental treatment, 3) 145 volunteer blood donors (volunteers), and 4) 37 siblings aged 5 through 25 from six Amish families. The dental students and patients were attending the University of Oklahoma Health Sciences Center between 1985 and 1988. The volunteers and Amish family members were attending centers elsewhere in the US at the same time. All subjects consented to provide blood for antibody analysis according to local Institutional Review Board procedures (see Acknowledgements). The student, patient and volunteer populations (397 subjects) were used to determine what IgG concentration constituted an elevated antibody response to D-alanyl LTA, to ensure that these antibodies were not unique to dental or Oklahoma populations and to examine whether the antibody concentration was sex or age-associated. The Amish family siblings were selected to determine the frequency of high antibody concentration in children and young adults. Each sibling had at least one parent high responder to increase the likelihood of exposure to an antibody-associated oral microbiota from birth.
The clinical study participants consisted of 87 dental students (88.4% male) and 64 patients (31.3% male) who were medically healthy. All had 18 or more natural teeth and were aged \>22 and \<38 years. Of these participants, 67 dental students and 35 patients provided information that permitted an estimate of exposure to fluoridated water: residence(s) from birth through age 14 in the 1980 Fluoridation Census. Subjects not using the public water supply, or resident outside of the US for more than 18 months, were excluded. Exposure to water fluoridation scored 1 for each of five 3-year age cohorts: 0--2, 3--5, 6--8, 9--11 and 12--14 to give a fluoride exposure score (F Score) of 0 (no exposure) to 5 (complete exposure) described previously \[[@B22]\].
Most dental students had mild caries and gingivitis and most patients had moderate to severe caries and gingivitis. Exceptionally healthy or exceptionally diseased subjects were therefore increased compared to a similar number of subjects obtained as a random sample. This wide distribution of clinical measurements provided more stable estimates (narrower confidence intervals) of regression coefficients (β) than would be obtained from a similar number from a random survey of the general population. Regression lines are more robust when a greater range of measurements is used \[[@B23]\].
Clinical measurements
---------------------
Dental caries experience was the number of Decayed, Missing and Filled Teeth (DMFT), excluding third molars and teeth reported missing for other reasons. Decayed teeth (DT) were also enumerated separately from missing and filled teeth (MFT). DT indicates a combination of delay in seeking therapy and faster development of new cavities. Fluoridated dentrifice use is related to oral hygiene but not toothbrushing frequency in adults aged as in the present clinical study \[[@B3]\] and young enough to have likely used fluoridated dentrifices from early childhood. Sensitive staining for plaque accumulation \[[@B24]\] was therefore used with measures of gingivitis and pocket depth at the mesio-buccal, buccal, disto-lingual and lingual surfaces of the six teeth employed for the simplified oral hygiene index \[[@B25]\], substituting adjacent teeth as necessary (24 sites sampled).
Gingivitis was determined by whether a site bled within 30 sec of gentle probing, BOP \[[@B26],[@B27]\] and pocket depth by measuring the distance (mm) from the free gingival margin to the base of the sulcus or pocket. Finally, each subject was asked to suck an erythrosin tablet for 30 sec and the sites examined for stained plaque \[[@B24]\]. For each subject, the mean prevalences of plaque (PL) and bleeding on probing (BOP), and the mean pocket depth (PD), were calculated across all sampled sites. The clinical measurements were made by two experienced clinicians who were calibrated for this study. The clinicians agreed strongly with respect to all measurements (correlation coefficients, r \> 0.85; p \< 0.001) except gingival bleeding index, for which a weaker correlation was noted (r = 0.60, p \< 0.001). The data reported are the mean measurements from the clinical examiners.
Antigen purification
--------------------
D-alanyl LTA, but not D-alanine-free LTA, is present in culture filtrates of *Streptococcus mutans* GS5 \[[@B17]\]. Bacteria were grown at 37°C in trypticase soy broth to late stationary phase (96 h), when the maximal amount of antigen is extruded into the culture fluid \[[@B18]\]. After centrifugation to remove the bacteria, culture fluid (10 1) was concentrated 20-fold over a YM10 Diaflo Membrane filter (Amicon Corp., Beverley, MA). D-Alanyl LTA in the concentrate was detected by immunoelectrophoresis, using a standard human serum identified previously \[[@B16]\]. D-Alanine-free LTA does not react with this serum IgG \[[@B17],[@B18],[@B21]\]. The D-alanyl LTA was purified by passing the concentrated culture fluid over a 90 × 2.5 cm Sephacryl column in 0.4 M NaCl buffered with 0.05 M sodium acetate pH 5. Antigen in the fractions was collected. After equilibrating with 5 mM sodium acetate buffer pH 5.0, it bound to a short Sephacryl S-200 and eluted by adding 14 mM NaCl as described previously \[[@B16]\].
Measuring antibody content and determining high and low responders
------------------------------------------------------------------
IgG antibody content was measured by enzyme-immunoassay employing a Fast Assay Screening Test System at room temperature \[[@B28]\]. Pegs protruding from a lid were placed over a 96-well plate or trough containing 14 ml of 10 μg/ml D-alanyl LTA in acetate buffer pH 5 for 2 h (Becton Dickinson, Lincoln Park, NJ). The pegs were blocked with 14 ml of phosphate buffered saline (PBS) pH 7.0 in 1.0% Tween-20 and immersed in wells containing 0.1 ml serum. After overnight incubation, excess IgG antibody from the serum was washed away by thrice transferring the pegs to troughs containing 14 ml of PBS containing 0.05 % Tween 20 (PBS-Tween) for 5 min each time. The pegs were then immersed for 2 h in a trough containing 14 ml of anti-human IgG F(ab\'2) fragment conjugated to alkaline phosphatase in PBS-Tween and developed with nitrophenyl phosphate (Sigma Chemical Co. St Louis, MO).
The concentration of antigen-specific IgG in standard serum was obtained by measuring the optimal amount of protein immunoprecipitated \[[@B16]\], and a standard curve of absorbance against concentration was obtained (Fig. [1](#F1){ref-type="fig"}). The greatest range of absorbance occurred when sera were measured at a dilution of 1:200 \[[@B28]\]. Replicate antibody assays were performed on each serum and the concentrations read off the standard curve. The antibody concentrations are shown ranked in Fig. [2](#F2){ref-type="fig"}.
{#F1}
{#F2}
Data analyses
-------------
The amount of IgG antibody to D-alanyl LTA varies with no obvious cutoff (Fig. [2](#F2){ref-type="fig"}). However, the sera containing precipitating antibody should tend to have high IgG antibody contents. The IgG antibody measurements were alternatively divided into clusters, using the unweighted pair group method with arithmetic averages \[[@B29]\] and NT-SYS, a package of multivariate statistical computer programs \[[@B30]\]. A low response supremum was obtained by taking the antilog of the mean IgG content of the non-precipitating sera plus one standard deviation, or the antilog of the highest IgG content of the cluster grouping containing the least antibody.
The effect of age was determined after splitting the subjects into decile cohorts (Table [1](#T1){ref-type="table"}) and comparing the fraction of high antibody responders in each cohort. The youngest cohort was composed of Amish family siblings who were younger than any dental students, patients or volunteers.
::: {#T1 .table-wrap}
::: {.caption}
######
Age decile cohorts for determining changes in fraction of high antibody responders
:::
Cohort No. Decile ^a^Number in Cohort High Responders
------------ -------- --------------------- -----------------
1 5--15 ^b^19 ^c^21.05%
2 15--25 ^b^97 35.05%
3 25--35 176 36.36%
4 35--45 76 40.79%
5 45--55 31 38.71%
6 55--65 13 ^c^23.08%
7 65--72 6 ^c^16.67%
^a^381 subjects after excluding all Amish family members and 16 of the 397 dental student, patient and volunteer subjects whose age was not recorded. ^b^The Amish family siblings \<15 years comprised Cohort 1 and those \>15 years were included within Cohort 2. ^c^Comparison of cohorts 1, 6 and 7 with the remainder: X^2^ = 3.21, p = 0.073 (not significant).
:::
A multiple linear regression procedure was utilized to examine the relationship of caries (DMFT) with age, F score, and measures of gingival health obtained in this study: PL, BOP, and PD. The regression on DMFT was used: 1) to estimate the partial regression coefficients (β coefficients) within the high and low antibody response group; 2) to test each β coefficient for significance after accounting for the effects of the other four variables; and 3) to examine for significant differences in β coefficients between the antibody response groups. A β coefficient is interpreted as the change in disease response (DMFT) per unit change in one of the independent variables after adjusting for all the other independent variables in the model. Within each antibody response group, the multiple regression coefficient (R^2^) provided an estimate of the proportion of variance in DMFT explained by the combination of variables tested. Stepwise regression then identified the best estimate of the variance in DMFT that was explained by multiple variables in the separate and combined high and low responder groups. These multiple regression analyses were repeated using gingivitis (BOP) as the dependent variable and OHPI, PD, DT and MFT and age as independent variables. All of the clinically examined subjects were included because F score was not an independent variable for BOP.
To ensure that obtained relationships were robust, influential points (outliers) were identified using a statistic (DFFITS) which measured the change in coefficients caused by removing the data for each subject. If this change exceeded 2√(p/n), where p was the number of independent variables and n the number of samples \[[@B31]\], the point was influential. Repeating the regressions with all the influential points removed determined the degree to which these points had affected the results.
Results
=======
Definition of high antibody response
------------------------------------
IgG antibodies were initially detected in sera irrespective of whether D-alanyl LTA was immunoprecipitated. Excluding the Amish family group, there were 288 subjects whose sera failed to immunoprecipitate antigen (detected by immunoelectrophoresis). The mean IgG antibody concentration (log ng/ml) was 3.19 (0.64 standard deviation, s.d.) compared with 4.25 (0.57 s.d.) for 109 subjects whose sera did precipitate antigen (\'t\' test p \<10^-6^). High responders therefore had a log antibody content (ng/ml) that exceeded the mean plus standard deviation of non-precipitating serum (log ng IgG /ml \>3.83). However, low IgG concentrations formed a cluster whose supremum (log ng IgG antibody/ml) was 3.861, which corresponds to 7.26 μg/ml (left side of Fig. [2](#F2){ref-type="fig"}). Subjects were therefore classified as low responders if their log IgG antibody content exceeded 3.86 rather than 3.83. The odds ratio for a serum from a high antibody responder immunoprecipitating D-alanyl LTA was 14 times greater than for a low responder.
High antibody response, age and gender
--------------------------------------
Of the 397 students, patients and volunteers, 16 did not have their age recorded. High responders had a mean age of 32.7 years ± 9.4 s.d. (136 subjects) and low responders a mean age of 33.0 years ± 10.7 s.d. (245 subjects). Table [1](#T1){ref-type="table"} shows the fraction of high responders in different age cohorts. There was a high responder frequency of 35--40% from age 15 through 54. The reduced frequencies of high response in childhood and old age were not significant. However, within the Amish family offspring high IgG responders were significantly older. Table [2](#T2){ref-type="table"} shows that the high responder offspring had a mean age of 17.1 years compared with 13.3 years for low responders. This significant difference (\'t\' statistic = 2.42, degrees of freedom, d.f. = 35; p \< 0.03) was due to few high responder children and young teenagers and a slightly greater fraction of siblings aged 15--25 years who were high responders (50%) compared with the general population.
::: {#T2 .table-wrap}
::: {.caption}
######
Age of siblings from 6 families with at least one high responder parent
:::
High responders Low responders
---------------------- ----------------- ------------------
F+M- 17 16,14.13,12,10,8
F+M+ 20,13 21,18,16,8,5
F+M+ 22,20,18 25,16,14
F?M+ None 12,11,10,9,8
F+M- 22,20,13,11 17
F+M\' 17,16,13 20,18,10,7
^b^Mean age (s.d)^c^ 17.08(3.68) 13.25(5.00)
^a^F, Father; M, Mother; +, high responder; -, low responder, ? Not known. ^b^Mean age of the high and low responders ^c^s.d., standard deviation.
:::
High antibody responders accounted for a similar fraction of subjects irrespective of whether they were in the clinical study, or dental students, or patients (Table [3](#T3){ref-type="table"}). Men were 49.1% of the 395 students patients and volunteers whose sex was recorded. Men had also a greater frequency of high response 40.21% vs 31.34% and a greater mean log ng/ml IgG antibody content, 3.52 ± 0.76 standard deviation (s.d.) vs 3.44 ± 0.80 s.d. Neither of these differences were significant (X^2^ = 3.00, p = 0.09; \'t\' statistic = 0.34, p = 0.73). In serum samples from 18 subjects aged between 22 and 38, the IgG antibody concentrations were essentially the same after 6 months as originally estimated (squared correlation coefficient, R^2^ \>0.95, p \< 0.01). The results indicate that, for subjects aged 15 -- 55, age and sex had little effect on the frequency of high D-alanyl LTA antibody response.
::: {#T3 .table-wrap}
::: {.caption}
######
Fraction of high antibody responders in or not in the clinical study.
:::
Subject group Number \% high responders
----------------------------------------- -------- --------------------
Dental students in clinical study^a^ 87 37.9
Patients in clinical study^a^ 64 32.8
Same-age subjects not in clinical study 129 31.8
Other subjects not in clinical study^c^ 117 39.7
All subjects 397 33.8
^a^See first section of Materials & Methods. ^b^Same-age subjects not in the clinical study were 12.4% dental students, 17.1% patients and 70.5% volunteers. ^c^Other subjects were a mixture of patients and volunteers: 70.9% older (ages \>38 and \<72 years), 15.4% younger (ages \>15 and \<22 years), and the remainder age unknown.
:::
Association of DMFT with gingival health and fluoride in high and low responders
--------------------------------------------------------------------------------
Table [4](#T4){ref-type="table"} lists the variables tested for association with DMFT and the observed β coefficients in high and low antibody responders. It was immediately apparent that the β coefficients from PL were similar in both groups, whereas those from BOP and F score were only significant in high responders and those from PD were only significant in low responders. Comparison of the differences in β coefficients between high and low responders, column 4 (column 2 -- column 3), indicated relationships of DMFT to pocket depth that were significantly different and relationships of DMFT to F score that were almost significantly different, p = 0.062 (Table [4](#T4){ref-type="table"}, column 4).
::: {#T4 .table-wrap}
::: {.caption}
######
Changes in the equality of the partial β coefficients for association of tested variables withcaries severity in high and/or low responders^a^.
:::
Variable^a^ High responders ^b^n = 35 Low responders n = 67 Difference (Hi -- Low)
------------- --------------------------- ----------------------- ------------------------
F score ^c^-0.847 0.006 ^d^-0.853
PL ^d^0.259 ^d^0.234 0.025
BOP ^c^0.447 0.167 0.280
PD -1.903 ^c^4.249 ^c^-6.152
Age 0.099 ^d^0.261 -0.162
^a^Variable names are defined under \"Clinical Measurements\" in the Methods ^b^n = number of subjects ^c^p \<0.05 for value of constant or variable (β) or for all values in indicated model. ^d^*p* \> 0.05 & \< 0.15. If no subscript, p \> 0.15 ^e^HLS: high response = 1, low response = 0.
:::
Stepwise regression confirmed the similar associations of DMFT with plaque prevalence in both high and low IgG antibody responders, but significant associations with only F score and BOP prevalence in high responders and with only PD and age in low responders. In high responders, DMFT increased as plaque and BOP prevalences increased and fell as fluoride exposure increased. The equation obtained was: DMFT = 4.60 + 0.28 PL + 0.39BOP - 0.88FScore (R^2^ = 0.51, F statistic = 10.75, p \< 0.0001). By contrast, in low responders, DMFT increased with plaque prevalence, pocket depth and age. The equation was: DMFT = -13.37 + 4.58 PD + 0.27 Age + 0.30 PL (R^2^ = 0.26, F statistic = 7.41, p \< 0.0003). Within each equation, the constant and the respective β coefficients were significant (p \< 0.05), except for the β coefficient of age in low responders (p = 0.062). High responders receiving fluoridated drinking water for all 14 years of childhood (F score = 5), had a significantly lower DMFT than those never receiving fluoridated water (F score = O): DMFT = 7.50 ± 4.52 (s.d.) vs 11.60 ± 4.06 (s.d.); \'t\' test p \< 0.04. This was not true of low responders in whom the difference between F score 5 and F score 0 was not significant (DMFT = 9.33 ± 5.72 vs 12.07 ± 5.85; \'t\' test p = 0.14).
When antibody was ignored in stepwise regression (control), DMFT increased with age, PL and BOP, and decreased with F score: DMFT = -2.11 + 0.17 Age + 0.25 PL + 0.27 BOP - 0.39 FScore (R^2^ = 0.29, F statistic = 6.62, p \< 0.0001). PL and BOP were individually significant. F Score and age were borderline, p = 0.09 and 0.14 respectively, and PD was not significant, being entirely replaced with age. Despite the subjects increasing to 102 (from 35 or 67) the strength of association was similar to that of low responders only.
Association of DT with gingivitis in high and low responders
------------------------------------------------------------
Because caries experience associated with fluoride and gingival health in high responders, poor gingival health should increase the number of decayed teeth (DT) more than in low responders. When BOP was regressed against the variables in Table [5](#T5){ref-type="table"}, only DT (column 4) differed significantly between the high and low responders. Although DT alone significantly correlated with BOP in both response groups (p \< 0.01), it associated with BOP much more strongly in high responders (R^2^ = 0.57) than in low responders (R^2^ = 0.12). Fig. [3](#F3){ref-type="fig"} shows the respective correlations, and also the patient data (filled circles) skewed by few healthy subjects and the student data (unfilled circles) skewed by few moderate and no severely diseased subjects. Clearly, combining students and patients strengthened the respective associations (β coefficients). Stepwise regression indicated that, excluding DT, BOP associated with PL and PD similarly (BOP = -14.02 + 0.42 PL + 4.8 PD in high responders and BOP = -11.07 + 0.34 PL + 4.10 PD in low responders; R^2^ = 0.40 and 0.41 respectively; p \< 0.001). However, DT explained more variance (BOP = -1.46 + 0.24 PL + 0.95 DT, R^2^ = 0.62, p \< 0.001) in high responders, and less variance (BOP = -3.25 + 0.41 PL + 0.33 DT, R^2^ = 0.31, p \< 0.001) in low responders.
{#F3}
::: {#T5 .table-wrap}
::: {.caption}
######
Changes in the equality of the partial β coefficients for association of tested variables withgingivitis (BOP) in high and/or low responders.
:::
Variable^a^ High responders n = 54 Low responders n = 97 Difference (Hi -- Low)
------------- ------------------------ ----------------------- ------------------------
PL ^d^0.182 ^c^0.321 -0.140
PD ^c^2.529 ^c^3.588 -1.059
DT ^c^0.906 0.202 ^c^0.703
MFT 0.154 0.054 0.099
Age 0.022 -0.053 0.074
^a^Variable names are defined under \"Clinical Measurements\" in the Methods ^b^n = number of subjects. ^c^p \< 0.05 for value of constant or variable (β). ^d^*p* \> 0.05 & \<0.15. If no subscript, p \> 0.15.
:::
Influential points (outliers), whose presence might have affected the strength and significance of these complex regression analyses, were identified in five low responders and two high responders. When these subjects were omitted, the respective regression coefficients or their significance were little changed, indicating that the different, partial, linear regression coefficients in high or low responders were not artifacts of influential or outlying points.
Discussion
==========
This study has demonstrated that IgG antibodies to D-alanyl LTA are widespread in US adults. The fraction of high responders was essentially constant from early adulthood through middle age, but reduced in children (\<15 years) and old age (\>55 years). Within the adults (ages 15--55 years) a change from low to high response or *vice versa* was found unlikely from repeated measurements over 2--6 months. A similar lack of change in this IgG antibody concentration was reported 2--3 months after an additional 26 similarly aged patients had received oral hygiene therapy in another study \[[@B32]\]. Finally, the family studies established that a high antibody response was probably induced during the mid-teenage years. In order to apply the results of this study to children and young teenagers, longitudinal studies of the antibody response in relation to age and the clinical measurements in this study may need to be undertaken.
Despite few investigations of caries risk in 22--38 year old subjects compared with a younger or older group \[[@B7]\], the association of DMFT with PL in this study agrees with that obtained from 35 year old Norwegians \[[@B3]\]. Plaque (simplified oral hygiene index measurement) accounted for 15% of the variance in number of decayed/filled teeth surfaces in that study, and for 19% of the DMFT variance within all 151 clinically examined subjects in this study (ignoring age, antibody and all other variables). Other studies have shown that the amount of fluoride applied from dentrifices is measured better from oral hygiene or plaque accumulation, as in this study, and not from the reported frequency of dentrifice use \[[@B2],[@B3]\]. Finally, because subjects aged more than 38 are unlikely to have used fluoridated toothpastes until later in life, they were omitted to avoid confounding the results.
The rationale behind this study was that acidic plaque environments increase the amount of D-alanyl LTA and promote its immunogenicity. Accordingly, caries-protection by fluoride in the water supply and dentrifices was strong in high IgG antibody responders, accounting for just over 50% of the variance in DMFT. In addition, gingivitis (BOP prevalence) associated strongly and significantly with the number of decayed (untreated) teeth, suggesting that preventing gingivitis increased fluoride exposure from dentrifices and reduced the number of decayed teeth. Increased exposure to fluoridated water, and dentrifices associated with good gingival health, may result in fluoride inhibiting enamel remineralization at the acidic plaque pH likely present in high responders.
In low responders, the fewer antibodies to D-alanyl LTA suggest less colonization by acid-tolerant bacteria and a weaker cariogenic attack. DMFT associated with age, as reported for other subjects whose sera did not precipitate D-alanyl LTA \[[@B22]\], and also with pocket depth. An increase in pocket depth is caused by periodontopathic bacteria that associate with an alkaline environment in the sulci over many years \[[@B20]\] and a microbiota that is neither acidic nor acid-tolerant \[[@B10]\]. The coefficient of DMFT association with PD in low responders therefore differed significantly from high responders within whom F score and gingival health were stronger covariates.
Conclusions
===========
An increased mutans streptococcal challenge accompanying low plaque pH (high antibody response to D-alanyl LTA) allows much of the variation in caries experience to be controlled by water fluoridation and by the use of fluoridated dentrifices associated with maintaining oral health. High IgG antibody responders are therefore better protected from caries in an optimally fluoridated environment. The concept that fluoride protects better from caries in a low pH environment \[[@B12]\] was recently used to explain why there is a poor association between caries experience and pH fall after a 10% sucrose rinse \[[@B33]\]. In low responders, increased fluoride exposure from dentriflce use to maintain oral health, or from water fluoridation, associate relatively poorly with caries experience. Although this study has indicated that the IgG antibodies to D-alanyl LTA do not become elevated until after age 17, when much caries may have already developed, they may be elevated to a lower level in children who eventually become high responders. The D-alanyl LTA antibody response is not detectable in saliva (unpublished studies), but it can be measured from only a thumb-prick of blood. Longitudinal studies of the D-alanyl LTA response in children could improve current efforts to predict caries susceptibility by relating it to fluoride or the fluoride ion product for fluoroapatite in saliva and the pH change after a sucrose rinse \[[@B5],[@B12],[@B13],[@B33]\].
Competing interests
===================
None declared
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<>
Acknowledgements
================
This investigation was largely supported by USPHS Research Grant 1R01 DE-06740 from the National Institute of Dental Research, National Institutes of Health, Bethesda. We sincerely thank Dr. W. Bias, Immunogenetics Laboratory, Johns Hopkins University, Baltimore and Dr. F. Bach, Immunobiology Research Center, University of Minnesota Medical School, Minneapolis for donations of human serum from their patients; E. Carter and D. LeFlore for technical assistance; S. Pitts, J. Chowning and Dr. A. Cucchiara, Computing Center, University of Oklahoma Health Sciences Center, for computing assistance; Drs. R. Reynolds, M. Martin and L. Coggins Dept. of Oral Diagnosis for clinical assistance. | {
"from": "PMC100323.md"
} | {
"alnum_ratio": 0.7283410941,
"avg_line_length": 147.5186915888,
"char_rep_ratio": 0.1010139417,
"flagged_words_ratio": 0,
"lang": "en",
"lang_score": 0.9154974222,
"max_line_length": 1468,
"num_words": 5944,
"perplexity": 1015.7,
"special_char_ratio": 0.3088472869,
"text_len": 31569,
"word_rep_ratio": 0.0296545914
} | 6,339,294,179,856,051,000 |
Background
==========
In 1990 the National Institutes of Health Consensus Development Conference \[[@B1]\] concluded that breast conserving surgery (BCS) followed by radiation is an appropriate method of primary treatment for the majority of women with early stage breast cancer (AJCC stages \[[@B2]\] I, IIA, and IIB). Numerous clinical studies have shown that survival after BCS followed by radiation therapy is equivalent to survival following mastectomy for women in these stages \[[@B3]-[@B5]\]. Some absolute indicators for mastectomy remain \[[@B6],[@B7]\], in particular widespread malignant-type microcalcifications, previous radiotherapy, and a relation of tumor size to breast size that would not allow a cosmetically satisfactory result. Although there has been an increase in the use of BCS since the early 1990s, an apparent under-utilization of BCS among women for whom such treatment was not contraindicated has been documented \[[@B8]\]. Geographic location, type of hospital and health plan, and personal preferences have been investigated as possible factors to explain the continued use of mastectomy \[[@B9]-[@B12]\]. Comorbidity \[[@B13]\] has been found to be an important predictor among older women.
The Surveillance, Epidemiology and End Results (SEER) cancer registries do not collect sufficient detail to examine some of these issues and medical records reviews are time consuming. Insurance claims data are a cost-effective alternative; they have already been collected and put into an electronic format by the insurance carriers. They cover large segments of the population, allow follow-up, use standardized codes \[[@B14],[@B15]\], and do not rely on subject recall. In cancer research, insurance claims data have been used to estimate the effectiveness of cervical cancer screening \[[@B16]\], to assess the role of screening practices in the incidence of prostate cancer \[[@B17]\], and to estimate mammography participation \[[@B18]\]. Medicare enrollees in Seattle and San Francisco who received care from a Health Maintenance Organization (HMO) were found to receive more BCS than women in Fee-For-Service (FFS) plans \[[@B19]\]. Whereas a Medicare/SEER registry linked data set has been used extensively to examine treatment and screening issues in individuals 65 years and older \[[@B20]\], linking of insurance claims data from private health plans in younger populations has been more difficult because of the large number of health plans in most geographic areas. However, Hawaii provides unique opportunities for insurance claims research because the majority of medical care is received within the state and more than 90% of the population \[[@B21]\] are covered by a limited number of health plans. As a pilot project, we were able to link data from the Hawaii Tumor Registry and from a health plan in Hawaii. The objectives of this analysis were to describe breast cancer treatment and comorbid conditions using the insurance claims data and to examine possible determinants of BCS *vs.* mastectomy for breast cancer patients with stage I and II disease.
Methods
=======
The study protocol was reviewed and approved by the University of Hawaii\'s Committee on Human Studies and by the Hawaii Tumor Commission that oversees the Hawaii Tumor Registry (HTR). Before linking of databases was initiated, a memorandum of agreement between all parties involved in this research project was signed. The resulting agreement safeguarded patients\' privacy at the highest possible level. All names and identifying information were deleted from the datasets and replaced with arbitrary numbers to be used for the data analyses.
Data Sources
------------
The HTR has maintained a database of all cases of cancer diagnosed in the State of Hawaii since 1960 and became part of the SEER program in 1973. The HTR record contains demographic characteristics such as age, ethnicity, marital status, island of residence, as well as information on tumor size, extent of disease, lymph node involvement, and tumor grade. From medical records in hospitals and physicians\' offices, information is collected on the initial course (six months in the Hawaii Tumor Registry) of cancer-directed treatment following diagnosis. Quality control reviews have shown that case-ascertainment through HTR has been virtually complete \[[@B22]\]. Over 99% of cancer cases reported to the registry are histologically confirmed. The HTR also maintains a link with Hawaii Department of Health, which allows for death information to be captured in the HTR database.
Linking
-------
The insurance claims for this study were obtained from a local insurer who was a party to the memorandum of agreement. The linked data set contained cancer cases diagnosed from 1995 through 1998. During the linking process, a list of health plan members who had at least two cancer diagnostic codes in their claims history were matched against the HTR using a probabilistic method. For each matched record, the health plan furnished all claims data for that period. They also indicated whether the individual was enrolled in a Fee-For-Service (FFS) or a capitated (HMO) plan, a choice provided by the insurer. However, the majority of linked cancer cases belonged to the FFS plan and the HMO plan provided by this insurer differs considerably from a typical HMO. Individuals aged 65 years and older were included only if they were still working and had primary coverage through the health plan or if they were covered under Medicare but also had secondary coverage through the health plan. The dataset furnished by the health insurer contained all claims processed for cancer patients in this study. The data elements included: date of service, International Classification of Disease, version 9 (ICD-9) diagnosis codes \[[@B14]\], Current Procedural Terminology (CPT) codes \[[@B15]\], and the provider specialty code. The dataset contained claims for services from physicians, laboratories, freestanding facilities, as well as outpatient services provided by hospital facilities.
Data set
--------
Breast cancer cases accounted for 16.6% of all cancer cases recorded in the HTR during the period from 1995 to 1998. In the linked dataset, 27.7% of the cases were breast cancer cases. Using the ICD-9-CM code range \[[@B14]\] for breast cancer (174.0 to 174.9), we identified 1,377 female breast cancer cases. We then excluded 265 cases diagnosed after June 30, 1998 because we would not have a complete history of claims data covering at least six months of treatment. We also identified 30 women who had been recorded twice in the linked dataset because breast cancer was diagnosed in both breasts, as identified by the laterality codes. For 24 women, the diagnosis date was identical, in which case we considered the two diagnoses as one case. For the six cases with different diagnoses dates, we retained both records because there was a separate completed course of treatment for the cancer in each breast. We eliminated one case that was in the linked dataset and had inpatient data but no outpatient data, resulting in a dataset containing 1,088 breast cancer cases.
Staging
-------
We used the HTR information in the Extent of Disease (EOD-10) field \[[@B23]\] to determine the American Joint Committee on Cancer (AJCC) TNM stage \[[@B2]\], where T represents the primary tumor size, N refers to lymph node involvement, and M refers to presence of metastases. In cases where the tumor size was unknown, we assigned a stage only if the size was not a determinant in the TNM grouping. If the extension was unknown, we equated it to extension 10 (confinement to breast tissue and fat). For the small number of cases (N = 8) for whom lymph node involvement code was unknown, we considered this the same as no lymph node involvement because the tumor size was so small for these cases that a lymph node dissection was not considered necessary at the time of this study. The results of the TNM staging agreed well with the SEER summary stage codes.
Cancer treatment
----------------
We used the CPT codes \[[@B15]\] in the claims dataset to identify the type of surgery performed. For mastectomies, we included codes for simple and subcutaneous mastectomies (19180 and 19182), radical mastectomies (19200 and 19220) and modified radical mastectomy (19240). BCS was identified by the CPT codes for partial mastectomies (19160, 19162). We also included codes for excision of breast cysts or lesions (19120 and 19125) as these codes meet the definition of a lumpectomy, although we recognize that some surgeons may have billed these codes for excisional biopsies as a diagnostic procedure. If BCS was initially performed, but followed by a subsequent mastectomy, the subject was classified in the mastectomy group. This situation would have occurred when the BCS code was actually used for a diagnostic biopsy and the subsequent surgical treatment was a mastectomy or in situations where a lumpectomy was first selected, but a mastectomy became necessary because pathological examination revealed cancerous cells in the margins of the excised tissue. CPT codes 77261 to 77799 were selected to determine whether radiation therapy was received. Several CPT codes and also a number of other codes used by the health plan for billing were considered evidence of chemotherapy treatment.
Comorbidities
-------------
Information on the existence of comorbidities was extracted from the claims data using the occurrence of ICD-9 codes associated with comorbid conditions included in the Charlson Index \[[@B24]\]. We also analyzed data for conditions that were not included in that index, but occurred at a high enough frequency to warrant examination as possible comorbidities, in particular hypertension and lipid disorders. We included the following twelve comorbid conditions into our index: diabetes with complications, diabetes w/o complications, hypertension, heart disease, cerebrovascular disease, chronic pulmonary disease, peripheral vascular disease, kidney disease, liver disease, rheumatological diseases, hypothyroidism, and lipid metabolism disorders. Comorbidities were based on physicians\' claims with the respective ICD-9 code \[[@B14]\] during the 12-months period preceding the month in which the cancer was diagnosed. The month of diagnosis was excluded to avoid identifying as comorbidities any complications or conditions directly resulting from cancer treatment. Laboratory, radiology, and other diagnostic services were excluded from the comorbidity identification process because tests may have been done to rule out the condition. Each subject was assigned a comorbidity score of 0 (no condition), 1 (one condition), or 2 (two or more conditions).
Other variables
---------------
In addition to stage at diagnosis and size of tumor, we examined several other variables that were possible determinants in the utilization of BCS *vs.* mastectomy in early stage breast cancer, including age at diagnosis, ethnicity, island of residence, and marital status. For ethnicity, we used the five major groups in Hawaii (Japanese, Caucasian, Hawaiian, Filipino, and Chinese). All other ethnicities were grouped into an \"Other\" category. The specific island of residence was coded in the HTR, but since close to 80% of the population resides on Oahu \[[@B25]\], we classified residence as either Oahu or non-Oahu. For marital status, we grouped all women who were identified as single, separate, divorced or widowed in the unmarried category. Women under age 50 were considered pre-menopausal and women 50 years and older were classified as postmenopausal. Tumor grade information from the HTR record was grouped into grade I (well-differentiated cells) *vs.* all other grades (II, III, IV and unknown).
Statistical analysis
--------------------
All analyses were performed with SAS version 8.00 for Windows (SAS Institute, Cary, NC). Simple Kappa statistics (κ) was calculated to validate the treatment information from the insurance claims and the HTR \[[@B26],[@B27]\]. BCS was used as reference group throughout the analysis. It was defined as the dependent (outcome) variable and coded as a dichotomous variable, with 1 indicating BCS received and 0 indicating mastectomy received. Age was entered in units of ten years and tumor size was grouped into units of ten millimeters (one centimeter). For analyses of ethnicity as a predictor of treatment selection, Caucasian was used as the control group and indicator variables for all other ethnic groups were created. Logistic regression \[[@B28]\] was used to explore the influence of each variable on the use of BCS *vs.* mastectomy. First, we considered each independent variable by itself in a model and then we entered all variables simultaneously in a logistic regression model. Odds Ratios (OR) \[[@B29]\] with corresponding 95% Confidence Intervals (CI) were calculated to measure the degree of influence of variables on the utilization of BCS *vs.* mastectomy.
Results
=======
Based on the TNM stages, we identified 722 women who had stage I, IIA, or IIB breast cancer (66.4% of the 1,088 breast cancer cases in the linked data set). These 722 cases represented 32.8% (722 of 2,203 cases) of all early stage breast cancer cases recorded in the HTR during the study period. Approximately two-thirds of the women in the study population (Table [1](#T1){ref-type="table"}) were diagnosed in stage I of the disease and only 24% and 12% in stage IIA and IIB, respectively. Approximately 30% of the study subjects were aged less than 50 years, close to half were 50 to 64 years old, and 21% were 65 years and older. Three out of four women resided on the island of Oahu. Overall, 52.8% of the women had received BCS. Of the cases diagnosed at stage I, 57% of the women had received BCS and 47% a mastectomy. This decreased to 50% for stage IIA and to 34% for stage IIB. Among women, 65 years and older, 56% had BCS, compared to 50% of those under 50 years of age and 53% of the 50--64 year old women. While 56% of women on Oahu underwent BCS, only 43% of the women residing on the outer islands received BCS. We observed no statistically significant differences between the BCS and mastectomy group in terms of ethnicity, comorbidity count, menopausal status, marital status, and insurance plan. We found very high agreement between HTR data and claims data in identifying BCS \[κ = 0.91 (95% CI 0.88, 0.94)\], only 32 women were misclassified. For the majority of women who received BCS (92.1%), the lumpectomy was followed by radiation therapy. Additional chemotherapy was given to 34.1% and 42.5% of the women who had a lumpectomy or a mastectomy, respectively.
::: {#T1 .table-wrap}
::: {.caption}
######
Determinants of Breast Conserving Surgery Among 722 Cases of Early Stage Breast Cancer
:::
**Variable** **No. of patients** **Odds ratio (95% CI)**
--------------------- ------------ --------------------- ------------------------- --------------------------- -----------------------
BCS Mastectomy Univariate analysis Combined analysis
All 381 341
Ethnicity Caucasian 71 66 1 1
Japanese 170 147 1.08 (0.72--1.61) 0.85 (0.55--1.31)
Hawaiian 48 40 1.12 (0.65--1.91) 1.24 (0.71--2.17)
Filipino 35 47 0.69 (0.39--1.20) 0.65 (0.37--1.15)
Chinese 32 25 1.19 (0.64--2.22) 0.97 (0.50--1.86)
Other 24 16 1.45 (0.71--2.96) 1.33 (0.64--2.77)
Residence Oahu 299 232 **1.71 (1.23--2.39)** **1.67 (1.17--2.38)**
Other 82 109 1 1
Age at diagnosis \<50 yrs 110 109 Continuous (per 10 years)
50--64 yrs 186 166 **1.01 (1.001--1.028)** 1.01 (0.99--1.04)
65+ yrs 85 66
Menopausal status Pre Post 110 271 109 232 1 1.16 (0.84--1.59) 1 0.75 (0.46--1.23)
Marital status Married 262 244 0.88 (0.64--1.21) 0.94 (0.68--1.32)
Unmarried 119 97 1 1
Health plan FFS 330 300 0.89 (0.57--1.37) 0.87 (0.55--1.38)
HMO 51 41 1 1
TNM stage I 266 200 1 1
IIA 86 85 0.76 (0.54--1.08) 0.99 (0.66--1.51)
IIB 29 56 **0.39 (0.24--0.63)** 0.61 (0.32--1.19)
Tumor size \<1 cm 113 83 Continuous (per 1 cm)
1--1.9 cm 177 139 **0.97 (0.96--0.99)** 0.98 (0.96--1.00)
2 + cm 91 119
Tumor grade Well diff. 77 49 **1.51 (1.02--2.24)** 1.25 (0.83--1.89)
Other 304 292 1 1
Comorbid conditions 0 230 201 Continuous (per score)
1 109 102 0.97 (0.78--1.19) 0.91 (0.72--1.14)
2+ 42 38
:::
In the univariate models (Table [1](#T1){ref-type="table"}), we found that tumor size and grade, island of residence, age, and stage at diagnosis were predictors of breast cancer treatment. Women residing on Oahu were considerably more likely to have BCS than women living on all other islands in the state. For each one-centimeter increase in tumor size, there was a 3% lesser chance of undergoing BCS. Since size of tumor correlates to breast cancer stage, this also decreased the likelihood for women in stage IIA and IIB to receive BCS. Women with well-differentiated tumor grades were 50% more likely to undergo BCS as compared to women with all other grades. We found that for each ten-year increment in age, the chances of having BCS increased by 1%. We also observed that average tumor size was inversely related to age. Mean tumor sizes (in cm) with standard deviations were 1.82 ± 1.1, 1.59 ± 1.2, 1.39 ± 0.94 for women younger than 50 years, 50 to 64 years, and 65 years and older, respectively. Therefore, the smaller tumor sizes may account for the greater likelihood of BCS among older women.
In a combined model with all independent variables in a logistic regression, island of residence remained the only significant predictor of BCS in this population. Women living on Oahu were 67% more likely to have BCS than women on the outer islands. Although all other variables lost their statistical significance, associations for age, TNM stage, tumor size, and tumor grade remained similar in magnitude as in the univariate models. Although none of the ethnicity variables was significant, it appeared that Filipino women were less likely to receive BCS than women from all other groups. Residence on outer islands did not explain this observation. Women with a TNM stage of IIB were still 40% less likely to receive BCS than women diagnosed at stage I, but the relation lost its statistical significance due to the small number of cases. Menopausal or marital status, type of health plan, or the number of comorbidities were not related to the type of surgery.
Discussion
==========
The place of residence at the time of diagnosis was the most important predictor for receiving BCS in this dataset of health plan members. Women with early stage breast cancer living on Oahu were 70% more likely to have BCS than women living on outer islands. Age at diagnosis and tumor size were also related to breast cancer treatment although they were not statistically significant in the combined model. Contrary to our expectation and a previous publication \[[@B13]\], the number of comorbid conditions did not effect treatment in this analysis. The importance of geographic location can be explained by the availability of radiation facilities on the different Hawaiian Islands. Whereas on Oahu treatment facilities can be reached within one hour, distances on the outer islands are much farther. Only the Island of Hawaii and Maui have a radiation facility, but they are hard to reach from many parts of the islands. Women residing on Kauai, Molokai, and Lanai have to fly to Honolulu daily to receive the course of treatment or they have to remain there temporarily. Because this may be not be economically feasible for some women, as well as physically and psychologically challenging, it is possible that many women choose to undergo a mastectomy instead. Alternatively, patients and physicians in rural areas may favor mastectomy because of differences in education or cultural attitudes toward medical advances. We do not have information to explain why a small proportion of women who underwent BCS (6.9%) did not receive radiation. However, compared to other reports \[[@B13],[@B30],[@B31]\], the proportion of women who received radiation is very high. An interview study with cancer patients \[[@B32]\] showed that some patients decide against their physicians\' recommendation for radiation because they fear the treatment or hold certain beliefs about cancer. For other patients, radiation may have been contraindicated due to comorbid conditions.
In 1994, a national BCS rate of 42.6% was reported \[[@B33]\] with a rate of 46.7% for the Pacific region (Hawaii, California, Alaska, Washington, and Oregon). Our results for 1995 to 1998 indicate a slightly higher rate for Hawaii (52.8%). Geographic differences in BCS use have repeatedly been described in the literature. In western Washington \[[@B34]\], women residing outside the Seattle area were less likely to have BCS than women residing inside the county, in particular if radiation therapy facilities were not available in their county. Comparing rates across the country showed that BCS was much more common in the Northeast than in the South \[[@B33]\]. Low BCS rates were also found in rural areas, such as North and South Carolina \[[@B10]\], Minnesota \[[@B35]\], and the Southwest \[[@B36]\]. An analysis of 1991--1992 SEER data \[[@B37]\] demonstrated a strong association between BCS and distance to the next radiotherapy facility. The likelihood of undergoing BCS was 50% lower among women living more than 15 miles from the next radiotherapy facility.
In contrast to our results, a number of studies have shown that the likelihood of BCS decreases with increasing age \[[@B11],[@B30],[@B33]\], possibly due to less concern about cosmetic results. Because of the relatively small proportion (21%) of women 65 years and older and the fact that the women 65 years and older in our study had a health plan other than Medicare, they were not directly comparable to the populations in the published literature. Therefore, our ability to observe an effect among older women was limited. In particular, we were not able to examine the treatment of women 80 years and older who have been described as receiving more BCS than women age 65 to 79 years \[[@B13]\]. Our findings of lower BCS in smaller and well differentiated tumors agree with previous studies \[[@B33]\], but the absence of an effect of comorbid conditions disagrees with a study in older women \[[@B13]\] that showed a higher rate of BCS in women with comorbidities. Type of health plan did not predict type of treatment in our study probably because the majority of women in this data set had a FFS plan. In previous analyses with a larger variety of health plans, HMO members were considerably more likely to receive BCS in San Francisco and in Seattle \[[@B19]\]. Medicare and Medicaid patients received more BCS in a North Carolina hospital \[[@B7]\]. However, a study using administrative data \[[@B38]\] described lower BCS use among Medicaid patients and treatment was similar by health plan \[[@B39]\] in Northern California.
Our study had several limitations, including the possibility of incomplete claims histories if patients changed health plans during the course of treatment, but the high agreement with HTR information indicates that the validity of insurance claims information was high. Continuing efforts by the health plan through feedback loops has improved diagnostic coding \[[@B40]\] in Hawaii, but as documented for Medicare claims, problems remain \[[@B41]\]. As discussed above, our study sample of breast cancer patients does not represent the population of the state, excluding in particular Medicare and Medicaid recipients, members of a typical HMO, and the relatively small number of uninsured patients. The exclusion of Medicare claims lead to a younger population, whereas the lack of Medicaid claims biased the cases toward a higher socioeconomic status. On the other hand, the study population represented the population quite well in terms of ethnicity (Hawaii Department of Business 1998 1705 /id). Because our data set included 42.4% of all breast cancer cases under 65 years diagnosed during the study period, the results represents this population much better than the population of older women. Also, we were unable to measure some factors that have been found to be important in other reports, in particular, influence of physician age \[[@B11]\], physician specialty \[[@B42]\], type of hospital (teaching hospitals and hospitals with radiation facilities, private, county and public) \[[@B36],[@B38],[@B39],[@B43]\], socio-economic status and education \[[@B42]\], tumor-breast ratio and the expected cosmetic results \[[@B7],[@B9]\], physician/patient interactions before surgery \[[@B42]\], and psychological factors, such as fear of radiation or cancer \[[@B9]\]. The strengths of using of insurance claims data for this analysis were twofold. First, we had more detailed treatment information available than collected by the tumor registry. Second, the diagnostic codes in the claims data allowed us to estimate the number of comorbid conditions before the cancer diagnosis.
The results of this pilot study combining tumor registry and insurance claims data raises an important issue for cancer practice in the State of Hawaii and elsewhere. Health care providers and insurance plans need to work with other agencies to develop viable solutions to facilitate access to radiation facilities for women residing in remote locations, such as islands with no radiation facility. Attitudes of patients and physicians living in rural areas may be important in the choice of treatment, but could not be investigated in this project. Future research that includes interviews with patients and physicians may investigate attitudes toward certain types of treatment and identify additional barriers to BCS, such as psychological problems. In order to understand how treatment decisions were made in practice, qualitative information from physicians and breast cancer patients should be collected and examined in detail. Further validation of treatment and comorbidity from insurance claims data would strengthen this type of research in the future. The inclusion of a larger proportion of breast cancer cases is needed to establish the validity of our findings for the population at-large.
Competing interests
===================
None declared.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<>
Acknowledgements
================
We are very grateful to the employees of the health plan for their time and willingness to assist with this project. Special thanks go to Andrew White, PhD for his long-term support of our research efforts. The help of Marc Goodman, PhD and the staff of the Hawaii Tumor Registry is greatly appreciated. This research was funded by a special study grant from the National Cancer Institute, Surveillance, Epidemiology, and End Results program under contract number N01-PC67001. | {
"from": "PMC100324.md"
} | {
"alnum_ratio": 0.6880668102,
"avg_line_length": 217.4042553191,
"char_rep_ratio": 0.1366944037,
"flagged_words_ratio": 0,
"lang": "en",
"lang_score": 0.956718564,
"max_line_length": 2089,
"num_words": 4795,
"perplexity": 923.6,
"special_char_ratio": 0.3423696744,
"text_len": 30654,
"word_rep_ratio": 0.0048056832
} | 14,593,834,344,152,842,000 |
Background
==========
Health information systems (HIS) provide a scientific and technological framework to gather, manage, and interpret data to inform the public, policymakers, administrators, and health-care workers about the distribution and determinants of health conditions. Further, they can (and should) guide and measure the impact of interventions \[[@B1]\]. Public health surveillance -- a subset of HIS -- has been defined as *the ongoing, systematic collection, analysis, and interpretation of outcome-specific data for use in the planning, implementation, and evaluation of public health practice*\[[@B2]\]. Public health surveillance can be used to 1) assess the overall health status of a population, 2) describe the natural history of disease, 3) monitor disease trends, 4) detect epidemics, 5) evaluate the effect of prevention and control measures, 6) generate hypotheses, and 7) facilitate epidemiologic and laboratory research \[[@B3]\].
Before 1991, the Soviet Union centrally planned the infectious diseases surveillance systems (IDSS) of its 15 republics. Approximately 300 million persons were covered under the IDSS. Central monetary and technical support for the IDSS ended in 1991. As a result, the republics have struggled to maintain their respective IDSS. The former Soviet Union (FSU)-wide diphtheria outbreak in the 1990s \[[@B4]\] and the re-emergence of malaria in Tadjikistan in 1991 \[[@B5]\] and in Armenia \[[@B6]\] and Azerbaijan \[[@B7]\] in 1994 indicated that financial constraints resulting in the disruption of public health infrastructure and services had increased the risk of the re-emergence of infectious diseases \[[@B8]\].
After 1991, this transition to nationalism, privatization, and social reorganization posed new challenges to each republic of the FSU. The loss of centralized training, public health expertise, and resources especially impacted the surveillance systems in each republic. \[[@B9]\] Further, privatization of the FSU medical systems resulted in under-budgeted public health services and inadequately paid personnel \[[@B10]\]. Combined with these infrastructure problems, the increased population migration -- both within the FSU and internationally -- contributed to morbidity and mortality through population dislocation \[[@B8]\]. This has increased the risk of exposure to re-emerging microbial and environmental pathogens and limited access to health services and good nutrition \[[@B9],[@B10]\].
The most important of these changes was the financial crises resulting from the severance of economic ties among all republics of the FSU. Public health officials were challenged to transform the bulky, state-sponsored IDSS. With high inflation and unemployment, they also suffered from shortages of vaccines, hospital supplies, and essential drugs. Provision of basic public health services were compromised, including repairing antiquated water and sewer systems, resulting in increased risk for gastroenteritis and infections with hepatitis A \[[@B11]\]; the largest documented outbreak of typhoid in this century occurred in Tajikistan in February 1997 \[[@B12]\]. Practices such as reusing syringes during vaccination and poor sterilization procedures during dentistry have contributed to nosocomial outbreaks of HIV and a high prevalence of infections with hepatitis B \[[@B13]\].
The objective of this work was to assess the current status and functioning of various IDSS, so as to guide reform efforts. At the invitation of seven Ministries of Health (MoHs), we performed assessments in the Russian Federation and in the Republics of Kazakhstan, Tadjikistan, Uzbekistan, Turkmenistan, and Armenia, and in the Kyrgyz Republic. We found striking homogeneity in comparing the IDSS from one republic to another; and, for clarity, we present here representative findings by using the Armenian component of the IDSS (AIDSS) as a prototype.
Discussion
==========
We used the CDC guidelines to assess the seven IDSS \[[@B14]\]; these guidelines have recently been republished in revised form \[[@B15]\]. This strategy includes an assessment of public health importance, objectives and usefulness, operation of the system, cost, and the seven system attributes (i.e., simplicity, flexibility, acceptability, sensitivity, predictive value positive, representativeness, and timeliness). The assessment involves gathering both qualitative (i.e., simplicity, flexibility, and acceptability) and observations of the quantitative (i.e., sensitivity, predictive value positive, representativeness, and timeliness) attributes.
In Armenia, we attempted to gather as much information as possible with respect to the construct and utility of the AIDSS from those most integral to its functioning and application. Therefore, we conducted face-to-face interviews and focus group discussions with approximately 50 epidemiologists at the Ministry of Health, the National Sanitary Epidemiologic Service, the Institute of Epidemiology, and two regional, three districts, and one city Sanitary Epidemiologic Service (SES) office. We also interviewed 23 health-care workers at village health centers, polyclinics, hospitals, and laboratories, which serve as the primary reporting units for the AIDSS.
At the time of this assessment, the AIDSS was moving toward reform, and we chose not to use the limited resources to gather quantitative data (e.g., through chart reviews) to assess the quantitative attributes of the AIDSS (i.e., sensitivity, predictive value positive, representativeness, and timeliness). Rather, we relied on qualitative observations gleaned in the course of the interviews. We present here the qualitative observations made of the AIDSS and recommendations for reform. We believe these observations and recommendations reflect the status of the IDSS of the other FSU republics.
Description of the AIDSS
------------------------
A republic of 3.3 million, Armenia gained its independence from the FSU on September 21, 1991 (Figure [1](#F1){ref-type="fig"}). The country is divided into 11 regions, which are subdivided into 37 districts (Figure [2](#F2){ref-type="fig"}). Responsibility for the AIDSS rests with the Department of Hygiene and Epidemiologic Surveillance -- a department of the Armenian MoH (AMoH) -- composed of 52 functional units known as the Sanitary Epidemiologic Service (SES). The units or stations of the SES parallel the geopolitical divisions of Armenia. There exists one SES station in each district. Each city has one SES station, except Yerevan, the capital. Yerevan has eight districts, each with one SES station. The national SES is also located in Yerevan.
{#F1}
{#F2}
The SES, *per se,* developed from a model created in the late 1800s in russia and uniformly developed over many years in the republics of the FSU. Its principle functions are to collect and analyze public health surveillance data and to implement and enforce strategies for the prevention and control of infectious diseases. Traditionally, the SES had approximately 10% of the entire medical person-power and budget of the AMoH, and was separate from the curative medical care system \[[@B16],[@B17]\]. The SES was staffed by epidemiologists (physicians), microbiologists, sanitary hygienists, and other health workers (paramedics and physician assistants). The district SES was the basic public health unit that monitored infectious diseases, investigated outbreaks, attended to child and adolescent health, inspected the food-service industry, monitored water purity, and dealt with occupational and environmental health problems throughout Armenia \[[@B6]\]. City-, regional-, and national-level SES administrations were larger, with specialized staff.
The SES collected infectious diseases data from all health-care facilities throughout the country. Before 1991, Armenia had a comprehensive and free health-care delivery system accessible to all citizens with health facilities and health-care workers employed under the auspices of the AMoH. Outpatient facilities (village health centers and polyclinics) and hospitals reported to the AIDSS. Each district had 12--45 village health centers, two polyclinics, and one hospital. The seven cities in Armenia had variable numbers of polyclinics and hospitals.
Altogether, Armenia had 830 village health centers, 228 polyclinics, and 179 hospitals (Figure [3](#F3){ref-type="fig"}). Data from these health facilities were reported to the 52 districts and city SES and then to the national SES, which forwarded aggregated data to the Department of Hygiene and Epidemiologic Surveillance in the AMoH.
{#F3}
Objectives and Data Collection, Reporting, Analyses, and Response
-----------------------------------------------------------------
The objectives of the AIDSS were to identify cases of infectious diseases, document outbreaks, and monitor trends in disease occurrence. It collected aggregated and case-based data on new cases of 64 infectious diseases. This list included some diseases with low pathogenicity (e.g., pediculosis and scabies) and some diseases with inadequate or non-existent preventive measures (e.g., infectious mononucleosis and parapertussis). There existed no tiered (e.g., confirmed, probable, and suspected) standardized case definitions. The reporting of confirmed or suspected cases of some infectious diseases required immediate reporting via telephone or in person within 12--24 hours. These included epidemic prone diseases (e.g., diphtheria, polio, plague).
Data were provided monthly and yearly from the district to national SES. Reports from the national SES were sent monthly to the AMoH and the other 14 republics of the FSU and back to the district SES twice a year. The district and city SES reported immediately to the national SES.
At all levels, epidemiologists used descriptive statistics for data analyses. These included the calculation of aggregated case numbers and incidence and prevalence rates based on estimates of the population size provided by the state agency charged with gathering census data; limited stratification by person, time, and place; and the assessment of trends. Epidemiologists did not use analytic methods to assess risk factors for diseases, even though they were collected.
Cases and contacts of every disease were investigated by a district or city SES epidemiologist within 24 hours of the receipt of the case report, using a standardized form known as the *epid carta.* This form was not disease specific, yet lengthy (46 questions, many of which required subjective responses). Little or no feedback was provided to the original reporting sources and no routine or formal sharing of data and information occurred between the district SES and health-care facilities. Table [1](#T1){ref-type="table"} summarizes the public health practice activities in Armenia, stratified by health facility.
::: {#T1 .table-wrap}
::: {.caption}
######
Public health surveillance and action core and support activities, by health level, Republic of Armenia, 1996
:::
**Organization Level** **Public Health Surveillance and Action Core and Support Activity**
--------------------------- --------------------------------------------------------------------- --- --- --- --- --- --- --- --- ----- ---
Primary Health Facilities X X X X X X X X *X* X
District SES X X X X X X X
District Lab X X X X X X X X
District Lab X X X X X X X
Regional SES X X X X X X X
Regional Lab X X X X X
National SES X X X X
National Lab X X X X
:::
Qualitative Attributes
----------------------
### Simplicity
Referring to both its structure and ease of operation, the AIDSS was complex. Epidemiologists gathered voluminous information on each case; parallel vertical public health programs reported duplicative information; much time was required to collect, register, and report case information; and many staff maintained the system.
### Flexibility and Acceptability
Addressing the extent to which the AIDSS could adapt to changing information needs or operating conditions and reflecting the willingness of participants to provide information and monitor the system, the AMoH, itself, determined both its flexibility and acceptability. Because the AMoH provided salaries for health-care personnel, it could enforce compulsory reporting, through monetary fines and professional demotions. Changes in reportable conditions or criteria were made rapidly because administrative SES employees carried out orders quickly and completely at all levels in the system.
Quantitative Attributes
-----------------------
While we did not quantitatively measure the attributes of sensitivity, predictive value positive, representativeness, and timeliness, we did make qualitative observations.
### Sensitivity
Assessing the AIDSS\'s ability to account for all incidents of a disease (i.e., the proportion of cases detected, correctly diagnosed, and reported), we learned that detection of most reportable infectious conditions did come to medical attention. This was enhanced because all citizens received free health care and because primary care physicians were responsible for care (including community outreach) of persons in their assigned territories. However, due to constrained resources, laboratory confirmation of reportable infectious diseases was limited in practice. As such, reporting was based on clinical or epidemiologic, rather than laboratory information.
### Predictive Value Positive
The likelihood that a disease report constituted a true case of that disease was diminished because of the lack of laboratory confirmation and standardized case definitions.
### Representativeness
We felt that, ideally, the AIDSS accurately described the distribution of diseases in the population by person, time, and place because disease reporting was mandatory and failure to report was a punishable offense. And, because of the penalties, all official reporting to the national level did occur on a monthly basis. However, it was common practice for epidemiologists to conceal cases of infectious disease and willfully underreport epidemic morbidity, because outbreaks meant that the epidemiologists were not performing their duties of preventing and controlling infectious diseases. This paradox resulted in epidemiologists managing two sets of information: one officially reported to higher-ups and one unofficially kept (with more accurate numerators).
### Timeliness
Information for action or for long-term planning was available because mandatory reporting to the SES within 12--24 hours of diagnosis for most conditions under surveillance allowed rapid implementation of control and prevention measures.
### Costs
Because budgets were relatively non-existent in the FSU, historical data on the costs of operating the AIDSS were not available. In the FSU centrally planned economy, resources were obtained from cost centers (e.g., utilities were not metered, office supplies were requested by quantity and not cost, and salaries were provided from the central budget). However, numerous observations led us to conclude that the AIDSS was relatively inefficient and costly; in large part, because the expenditure was paid from the public sector.
The AIDSS was labor intensive. Being paper-driven, reporting dieases for which no practical public health interventions exist misallocated scarce resources. Other common public health activities with high opportunity cost in the AIDSS were indiscriminate disinfection of homes and work sites. Environmental background monitoring practices by district and regional SES included routine collection of specimens and laboratory testing (e.g., air and water samples, food products, and items that children may come in contact with such as toys or eating utensils) in addition to evaluation of physical factors (noise, vibration, microclimate, electromagnetic fields, levels of lighting, and ionic radiation) at several sites (e.g., work places and day-care centers).
When cases of hepatitis A, acute gastroenteritis, tuberculosis, diphtheria, or pediculosis were reported, disinfection of homes, schools, day-care centers and work places was conducted by public health workers who used chloramine application and steam cleaning of all hard surfaces and laundering of all clothing and bedding materials. The effectiveness of such disinfection practices or environmental background monitoring has not been documented, and is likely of doubtful public health utility.
During disease outbreaks, it was common practice to investigate every case and culture all available materials, and decontamination efforts were instituted regardless of epidemiologic evidence. It was common practice to hospitalize children \< 1 year of age with pneumonia for 7--14 days, and children of all ages with acute gastroenteritis for 7--15 days. It was also common to hospitalize both adults and children with hepatitis A for 21 days, with syphilis for two weeks, with gonorrhea for three weeks, and with tuberculosis for one year. These isolation practices, meant to prevent disease transmission to the community, were consequences of central planning in which emphasis was placed on input indicators such as the occupancy rates of hospitals. Incentives (budget allocations) placed on input rather than output measures led to a level of medical infrastructure that has been difficult to maintain given current levels of funding available for the health sector.
Because financial issues were a major driving factor, cost analyses of surveillance practices and control measures could identify areas for cost-savings. Analyses of the surveillance system in Ukraine (using 1996 budget figures) revealed that excessive culturing represented 47% of the cost *per capita* expenditure of the L\'viv Regional SES and disinfection procedures accounted for almost 30% of the entire Pustomity District SES\'s budget (V. Carande-Kulis, CDC, personal communication).
Recommendations
---------------
Based on this assessment, we developed recommendations with respect to the three main surveillance functions of data collection, analysis and interpretation, and retrospective and prospective responses.
### Data Collection
• Eliminate punitive consequences to obtain accurate reporting;
• Restrict the number of routinely reportable diseases based on measures of mortality, morbidity, severity, communicability, and preventability;
• Categorize events under surveillance into a three-tiered surveillance system:
â—‹ disease elimination (e.g., polio);
â—‹ case-based (e.g., diphtheria); and
â—‹ indicator-based (e.g., number of children immunized by two years of age);
• Simplify reporting procedures and forms by
â—‹ limiting urgent reporting of diseases to those that require prompt institution of control measures;
â—‹ requiring only information necessary to direct control measures and perform basic analyses; and
â—‹ developing disease-specific forms with diseases chosen for case-based surveillance;
• Develop tiered (confirmed, probable, and suspected) standardized case definitions for all events under surveillance; and
• Computerize demographic and risk-factor data for systematic and detailed analysis of reported diseases and rapid dissemination of information.
### Analysis and Interpretation
• Provide ongoing capacity for training in analytic epidemiology; and
• Base interventions on epidemiologic evidence. Use analytic epidemiology (case-control and cohort studies and presentation of data using 2 × 2 tables, odds ratios, relative risks, and tests of significance) for hypothesis generation, risk factor identification, outbreak investigations and intervention design and monitoring.
### Retrospective and Prospective Responses
• Provide feedback to all reporting sources and share information across vertical program lines and with officials throughout the public health community in a timely fashion (e.g., via a monthly public health bulletin). A bulletin could include descriptions of important outbreak investigations, disease-specific analyses of surveillance data, graphic and tabular information on selected diseases, indicators of community health, and recommendations for public health concerns.
Current HIS Reform Efforts in Armenia
-------------------------------------
This assessment stimulated and guided reform efforts that were initiated in December 1992 through a cooperative project among the AMoH, United States Agency for International Development (USAID), and CDC. This project provided technical and material assistance toward reform of the Armenian HIS. The approach focused on training a cohort of public health officials and epidemiologists in the modern aspects of epidemiology, biostatistics, surveillance techniques, and scientific communications; developing Armenia-specific case definitions; facilitating HIS reform strategies through workshops and training sessions; and developing the capacity to publish an epidemiologic bulletin.
Since 1996, this HIS reform activity has been self-sustained with no additional monetary support from USAID \[[@B6]\]. In 1996, the AMoH created a national HIS program for the development and reform of the HIS and the Armenian National Health Analytic-Information Centre \[[@B18]\]. The system has been transformed into a comprehensive HIS and includes chronic diseases, maternal and child health, and injury data. Diseases are now categorized by a three-tiered approach: disease elimination (e.g., polio), case-based (e.g., diphtheria) and indicator-based (e.g., number of children immunized by two years of age).
Preparations are now being made for additional training and to assess and improve clinic case diagnoses, management, and recording, and clinic records. New regional centers equipped with computers and faxes have been organized for the collection, analyses, and reporting of health information. National and regional public health bulletins are being published monthly in three languages - Armenian, Russian and English -- and distributed to wide audiences. Tiered, standardized case definitions and essential health indicators for decision-making at the clinic and community level have been developed and disseminated. These include health status, performance, and resource indicators.
Comprehensive HIS reform is critical throughout the FSU. Timely, accurate, and relevant health information are necessary to assess the burden of disease and disability; understand changing health patterns; measure the needs for and improve services; address inequities in health; provide information for policy formulation and planning; and provide a basis for intra- and international comparisons on health status and care utilization \[[@B1]\]. Timeliness, accuracy, and relevancy are augmented by efficiency. Integrating all sources of data into one comprehensive HIS prevents duplicate recording and reporting across services and programs, averts labor-wasting inefficiencies, and saves scarce resources.
A comprehensive HIS includes the capacity to obtain data from vital registries, clinical, administrative, and other records; from provider and population-based surveys and sentinel systems for infectious and chronic diseases, and disabilities; and maternal and child health, nutrition, and program implementation indicators, including access, coverage, and service quality \[[@B1]\]. This reform activity should include the development of an indicator monitoring system based on selected essential, action-oriented indicators of health status, service performance, and resources that can be used for decision-making at the local level.
Summary
=======
We found the AIDSS to be a complex and sensitive, yet costly and inefficient surveillance system for infectious diseases. Despite the lack of standardized case definitions, feedback of information, and computer technology, it functioned fairly well before 1991. However, the functioning and continuation of the AIDSS has been affected both directly and indirectly by events of the past decade.
Overall, the former AIDSS was useful because it detected cases of infectious diseases, estimated morbidity, monitored trends in disease occurrence, and documented outbreaks. The comprehensive no-cost health-care delivery system and compulsory reporting of diseases to the AMoH enhanced its flexibility, representativeness, and timeliness. The strengths of the AIDSS stemmed from the large numbers of health facilities and trained personnel and the separation of preventive from curative medicine, that secured an independent status (including separate budget) for preventive medicine and public health practitioners.
The AIDSS also had weaknesses. Though the system meticulously tracked persons (from birth to death), very few of these data were computerized, analyzed, or used to develop, direct, or evaluate public health policy. In most cases, when used, data guided regulation and punishment rather than public health decision-making. Simply put, data were used to fix blame and punish rather than to find and implement effective interventions.
Epidemiologists were motivated to perform actions that both pleased their supervisors and avoided the punishment of monetary loss or demotion. For example, it was common practice for epidemiologists to hide select cases of infectious disease and willfully underreport epidemic morbidity because outbreaks meant that the epidemiologists were not performing their duties of preventing and controlling infectious diseases. These disincentives to thoroughness and honesty resulted in surveillance data and reports that did not reflect true incidence and prevalence of disease, circumstances, needs, responses, or impacts \[[@B9]\].
Because the AIDSS was but a component of the systems designed for the entire FSU, the AIDSS did not address Armenia-specific needs. Further, because of the lack of tiered, standardized case definitions, there existed a potential for misclassification of diseases. Lack of feedback to reporting sources hampered improvements in clinical practice. Monitoring conditions for which there were no practical public health interventions, the multi-tiered and duplicative reporting processes, and the use of expensive and indirect monitoring and control measures such as excessive culturing, disinfection, and prolonged hospitalization led to waste of resources. As a result, the system did not guide control measures optimally nor use resources efficiently.
As the republics of the FSU embrace various aspects of democratization, improvement of public health surveillance systems such as the IDSS should be a goal if decision makers are to use credible data for informed public health practice.
Competing interests
===================
We certify that we have participated sufficiently in the conception and design of this work, as well as its execution and the analyses of the data. Further, we have collaboratively written the manuscript and take public responsibility for it. We believe the manuscript represents valid work. We have reviewed the final version of the submitted manuscript and approve it for publication. Neither this manuscript nor one with substantially similar content under our authorship has been published or is being considered for publication elsewhere. If requested, we shall produce the data upon which the manuscript is based for examination by the editors.
We certify that we have no affiliations with or involvement in any organization or entity with a direct financial interest in the subject matter or materials discussed in the manuscript. Drs. Wuhib, Chorba, MacKenzie, and McNabb were employees of the U.S. federal government when this work was performed and prepared for publication; therefore, it is not protected by the Copyright Act, and there is no copyright of which the ownership can be transferred. Dr. McNabb serves as corresponding author; his address is listed.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<>
Acknowledgements
================
We acknowledge all SES epidemiologists in the respective Ministries of Health who shared their knowledge of the IDSS and contributed to the discussions on recommendations, and individuals from CDC including Drs. Daniel Bleed, Robin Ikeda, Lyle Conrad, Scott Wetterhall, Siiri Bennett, Gulbanu Altynbaeva and Scott Deitchman and Mr. Bruce Ross. | {
"from": "PMC100325.md"
} | {
"alnum_ratio": 0.7789434624,
"avg_line_length": 148.22,
"char_rep_ratio": 0.0869917327,
"flagged_words_ratio": 0,
"lang": "en",
"lang_score": 0.9450731277,
"max_line_length": 1055,
"num_words": 4855,
"perplexity": 376.9,
"special_char_ratio": 0.2283767373,
"text_len": 29644,
"word_rep_ratio": 0.0210482872
} | 5,471,717,537,466,399,000 |
"Background\n==========\n\nThe manner in which accuracy of clinical tests is mathematically summaris(...TRUNCATED) | {
"from": "PMC100326.md"
} | {"alnum_ratio":0.6352050976,"avg_line_length":116.7906976744,"char_rep_ratio":0.2102012353,"flagged_(...TRUNCATED) | 1,726,739,655,135,848,400 |
"Background\n==========\n\nMost current methods of cancer early detection, such as mammography or ce(...TRUNCATED) | {
"from": "PMC100327.md"
} | {"alnum_ratio":0.7641745939,"avg_line_length":170.2865497076,"char_rep_ratio":0.0723462728,"flagged_(...TRUNCATED) | 12,080,472,256,741,220,000 |
"Background\n==========\n\nThe introduction of small subunit ribosomal RNA as a tool in microbial ta(...TRUNCATED) | {
"from": "PMC100357.md"
} | {"alnum_ratio":0.6278115347,"avg_line_length":187.89373297,"char_rep_ratio":0.1876776701,"flagged_wo(...TRUNCATED) | 4,875,588,495,623,949,000 |
"Background\n==========\n\nBlood leukocytes are found in two sub-populations constituting the circul(...TRUNCATED) | {
"from": "PMC100780.md"
} | {"alnum_ratio":0.7616492516,"avg_line_length":138.4748603352,"char_rep_ratio":0.0874162564,"flagged_(...TRUNCATED) | 4,695,717,970,301,022,000 |
End of preview. Expand
in Dataset Viewer.
The Pile -- PubMed Central (refined by Data-Juicer)
A refined version of PubMed Central dataset in The Pile by Data-Juicer. Removing some "bad" samples from the original dataset to make it higher-quality.
This dataset is usually used to pretrain a Large Language Model.
Notice: Here is a small subset for previewing. The whole dataset is available here (About 83G).
Dataset Information
- Number of samples: 2,694,860 (Keep ~86.96% from the original dataset)
Refining Recipe
# global parameters
project_name: 'Data-Juicer-recipes-pubmed-central'
dataset_path: '/path/to/your/dataset' # path to your dataset directory or file
export_path: '/path/to/your/dataset.jsonl'
np: 50 # number of subprocess to process your dataset
open_tracer: true
# process schedule
# a list of several process operators with their arguments
process:
- clean_email_mapper:
- clean_links_mapper:
- fix_unicode_mapper:
- punctuation_normalization_mapper:
- whitespace_normalization_mapper:
- alphanumeric_filter: # 89217
tokenization: false
min_ratio: 0.2787 # 3sigma
- average_line_length_filter: # for code
max_len: 1200 # < 3sigma (1478) -- 7410
- character_repetition_filter:
rep_len: 10
max_ratio: 0.3741 # 3sigma -- 65849
- flagged_words_filter:
lang: en
tokenization: true
max_ratio: 0.00195 # 3sigma -- 8305
- language_id_score_filter: # remove language filter
min_score: 0.5 # 272359
- maximum_line_length_filter: # for code
max_len: 7328 # remove 23808 samples
- perplexity_filter:
lang: en
max_ppl: 8000 # remove 173883 samples
- special_characters_filter:
max_ratio: 0.842 # remove 87661 samples
- text_length_filter:
max_len: 136028 # 3sigma -- 15118
- words_num_filter:
lang: en
tokenization: true
min_num: 20 # remove 176537 samples
max_num: 23305 # remove 15016 samples
- word_repetition_filter:
lang: en
tokenization: true
rep_len: 10
max_ratio: 0.5981 # 3sigma -- 93843
- document_simhash_deduplicator:
tokenization: space
window_size: 6
lowercase: true
ignore_pattern: '\p{P}'
num_blocks: 6
hamming_distance: 4
- Downloads last month
- 47
Size of downloaded dataset files:
2.92 MB
Size of the auto-converted Parquet files:
1.38 MB
Number of rows:
100