Each of the 1,317 gene families included members with similar functional roles. The family sizes varied in a wide range between 2 to 223 members (Figure 1). The number of gene families was found to bear an inverse exponential relation to family size. About two-fifths of the gene families were duplex. Only three gene families had more than 100 members per family: Immunoglobulin heavy chain (162 genes), Zinc finger proteins (200 genes) and Solute carrier (223 genes).

The functional classification of 1,317 gene families comprising 7,928 genes in the six functional classes unveiled that the Signaling and communication is largest with 529 families and 3,072 genes (Figure 2). The Cell cycle is the smallest with 82 families and 470 genes.

Of the 1,317 gene families, 131 were entirely intrachromosomal. Chromosome 1 had the largest number with 17 families followed by chromosomes 19 and 11 with 13 and 12 families respectively. The remaining chromosomes had less than 10 intrachromosomal gene families per chromosome. The functional classification of these 131 intrachromosomal gene families revealed that the highest number (45) belonged to the class of 'Immune and related functions' closely followed by the class of Signaling and communication with 40 families. The remaining classes had the following distribution of gene families: Metabolism (24), Information (15), Structure and motility (5) and Cell cycle (2). These observations indicate that the creation of intrachromosomal human gene families was driven by large number of duplications followed by divergence in selected functional classes.

Of the 1,317 gene families, 732 families had (TG/CA)_n repeats in at least one of their members and 326 families had repeats in all their members. Of the 7,928 genes in 1,317 families, 3,986 genes had intragenic (TG/CA)_n repeats of length greater than or equal to 6 units. All 3,986 genes had repeats in their introns. Only 277 genes had (TG/CA)_n repeats in exons indicating that these repeats are mainly present in introns.

The distribution of genes with (TG/CA)_n repeats in the six functional classes is displayed in Figure 2. It is apparent that the class of Signaling and communication had the highest number of genes with (TG/CA)_n repeats. Comparison of the proportion of genes with repeats in each class with the global proportion showed that the class of Signaling and communication had significantly higher than the expected proportion (p < 0.0001, Binomial test). In contrast, the classes of Immune and related functions and Information had significantly lower than the expected proportion of genes with repeats (p < 0.0001 and p < 0.0002 respectively). The proportion of genes with repeats was not significantly different from the global proportion in the Cell cycle, Metabolism and Structure and motility classes. These observations show that the (TG/CA)_n repeats exhibit positive association with the genes belonging to Signaling and communication whereas, they are negatively associated with the genes belonging to Immune and related functions and Information.

It has been shown that the human genome has an isochore structure that varies in GC content 5. This variation raises the possibility that the observed selective distribution of (TG/CA)_n repeats might have arisen due to fluctuations in the local %(G+C) content of the genomic region as opposed to function. We examined this by comparing the average %(G+C) content of the genes in the six functional classes with the corresponding proportions of genes with repeats. The average %(G+C) content was observed to be in the narrow range (47–49%) in the six functional classes whereas, the proportion of genes with repeats varies widely in the range 29.6–61%. These observations indicate that the proportion of genes with repeats is significantly determined by function instead of small fluctuations in %(G+C) content.

Comparison of the proportion of genes containing (TG/CA)_n repeats with the average lengths of genes in each of the six functional classes revealed a linear relationship (Figure 3, correlation coefficient R = 0.93, p < 0.007). The signaling and communication class had the longest average gene length (74.07 kb) along with the highest proportion of genes with (TG/CA)_n repeats (61.23%). The class of Immune and related functions had the shortest average gene length (21.26 kb) with the lowest proportion of genes with (TG/CA)_n repeats (29.65%). These observations show that the proportion of genes with (TG/CA)_n repeats bears a linear relationship to the length of genes.

In order to examine the characteristics of distribution of (TG/CA)_n repeats with respect to multiple functional roles principally governed by their length, we analysed the repeats stratified into three categories: type I (6 ≤ n < 12), type II (12 ≤ n < 23) and type III (n ≥ 23). The results are displayed in Figure 4. The number of genes containing (TG/CA)_n repeats decreases with increasing repeat length. It is also apparent that several genes (53% of the total) have multiple types of repeats in various combinations.

Classification of the distribution of genes with (TG/CA)_n repeats stratified into three categories into six functional classes is shown in Figure 5. It is evident that the proportion of genes containing repeats decreases in the order I > II > III in all classes. The proportion of genes containing (TG/CA)_n repeats of Signaling and communication were significantly higher than the expected proportion in all three categories of repeats (p < 0.0001, type I, II and III). On the other hand, the proportion of genes with (TG/CA)_n repeats of Immune and related functions and Information were significantly lower than expected proportion in all three categories: Immune and related functions (p < 0.0001, type I, II and III), Information (p < 0.0001, type I and II, p < 0.004, type III). The proportion of genes with type III repeats was marginally lower than the expected proportion in Metabolism class (p < 0.01) and marginally higher than the expected proportion in Structure and motility class (p < 0.02). The proportion of genes with repeats in the three categories was not significantly different from the expected value in the class of Cell cycle. These observations show that repeats of all types are positively associated with the genes of Signaling and communication whereas they are negatively associated with the genes of Immune and related functions and Information.

The distribution of average number of (TG/CA)_n repeats per gene in the three categories in the six functional classes is displayed in Figure 6. Comparison of the average number of repeats per gene in the three categories with the global distribution pattern revealed that in most cases the observed number was significantly lower than the expected value, except for the genes belonging to Signaling and communication and Structure and motility, which had significantly higher average number of repeats per gene than the expected value (p < 0.0004 in all three categories, both classes). The average number of type III repeats per gene in the class of Cell cycle was not significantly different from the expected value. These observations show that the repeat densities were higher in the genes belonging to Signaling and communication and Structure and motility classes whereas, the genes belonging to other classes had lower repeat densities.

As a special case of this study, we examined the distribution of (TG/CA)_n repeats in the top 2% large families (27). The family sizes of this category varied widely from 32 to 223 members. Functional classification of these large families revealed the following distribution: Immune and related functions (9), Signaling and communication (8), Information (6), Metabolism (2), Structure and motility (1) and Cell cycle (1).

The proportion of genes with (TG/CA)_n repeats in large families is displayed in Table 1. Comparison with the global distribution showed that the proportion of genes with repeats was significantly higher than expected value in the Signaling and communication and Structure and motility classes (p < 0.0001, Binomial test). There was no significant difference between the observed and the expected proportion of genes with repeats in the class of Metabolism. In the remaining classes, the proportion of genes with repeats was significantly lower than the expected value (p < 0.0001, Binomial test). As observed with all gene families, a linear relationship was observed between gene lengths and proportion of genes with (TG/CA)_n repeats (correlation coefficient R = 0.79, p < 0.0001).

The large Collagen gene family belonging to the class of Structure and motility had the highest proportion of genes containing repeats (86.5%). In order to analyze this further, we examined the sequence conservation of the region flanking 200 bases upstream and downstream in addition to the repeats by comparing the human sequence with the available genome sequence of Chimpanzee (Pan troglodytes), a nearest ancestor to human 46. We observed, that of the 268 sequence segments including repeats from human Collagen genes, 244 were conserved with greater than 92% identity in the chimpanzee. Of these 244 repeats in human Collagen genes, 73 repeats were identical in length, 142 repeats displayed length polymorphism in the chimpanzee, 27 repeats had point mutations and in 2 cases there were no repeats in the corresponding segments in the chimpanzee. These observations show that both human and chimpanzee Collagen genes have high repeat content, high conservation of position and flanking sequences of the repeats. However, majority of repeats exhibited length polymorphisms, which is consistent with their characteristic property 6.

Sequences of 35,114 human genes (build number 33) were retrieved from LocusLink http://www.ncbi.nlm.nih.gov/LocusLink/43 using a JavaScript program. A sum of 192 genes could not be retrieved because of either inaccessibility to the LocusLink page or absence of the link for retrieving the gene sequence. A gene in this analysis is considered as the nucleotide sequence from the start of first exon to the end of last exon. If alternate splicing was reported, the gene length considered was the start of first exon to the last known exon including all alternatively spliced products for that gene.

Perl scripts, 'SimRep' and 'RepGene' were written for the identification and mapping of perfect intragenic (TG/CA)_n repeats of length n ≥ 6 units in genes 17. Throughout this work we have used n ≥ 6 units as the minimum cut-off to identify (TG/CA)_n repeats. All repeats were scored in the intragenic region (exons and introns only).

We grouped (TG/CA)_n repeats into three categories (types I, II and III), according to their length and biological properties. Type I (TG/CA)_n repeats, in the range 6 ≤ n <12 units, are short repeats based on the observation that a repeat length of 8 units (n = 8) is minimum to be likely polymorphic 3435. Type II (TG/CA)_n repeats comprise of 12 ≤ n < 23 units and is based on the observation that more than 93% of the (CA)_n repeats of n ≥ 12 units are polymorphic 6. Further, repeats of this length have also been shown to have preferential binding to nuclear factors compared to short repeats 36 and can also stimulate mRNA splicing 2122. Type III repeats consist of relatively long reiterations of (TG/CA)_n (n ≥ 23 units) and have propensity to adopt structures such as Z DNA 8937. Other studies have shown that (TG/CA)_n repeats of length greater than 22.5 units can stimulate recombination 181920.

Functional roles of a large number of human genes are not well known. Presently, these genes are assigned hypothetical annotations. Genes labeled as 'LOC', 'DFKZP', 'FLJ', 'HSPC', 'HSU', 'HT', 'KIAA', 'ORF', 'hypothetical', 'PRO' and 'pseudogenes' without clear functional details were filtered out. A total of 22,688 genes were removed in this filtering exercise. Out of the remaining 12,426 genes, a total of 8,778 genes (25% of total) were clustered into gene families based on their gene root symbols as defined in the guidelines of Human Gene Nomenclature Committee (2002) 4. The remaining 3,648 genes could not be clustered into gene families and are solitary.

The HGNC guidelines consider sequence and functional similarity of proteins encoded by genes while grouping them into gene families 43839. A root symbol signifies a gene family. The family members are designated by Arabic numerals placed immediately after the gene root symbol, for example GPR1, GPR2, GPR3 for genes of the G protein-coupled receptor family. A Perl script namely Clustergene was written to cluster 8,778 human genes into 1,556 gene families. The Perl script called ChromoCluster was written to report gene families located on the same chromosome. Subsequently these gene families were classified into the six functional classes as described below.

The gene families were classified into six functional classes namely, 'Information', 'Cell cycle', 'Metabolism', 'Signaling and communication', 'Immune and related functions' and 'Structure and motility' based on the scheme defined by Adams et al. 40. We combined the functional classes of replication, transcription, RNA processing and translation into 'Information' class based on Andrade et al. 41.

'Cell cycle' includes cell cycle, apoptosis, chromosomal structure and DNA repair; 'Immune and related functions' includes immunology, homeostasis, carrier proteins/membrane transport and stress response; 'Information' includes protein synthesis, translation factors, ribosomal proteins, post-translational modification/targeting, protein degradation, tRNA synthesis/metabolism, RNA synthesis, transcription factors, RNA polymerase, RNA processing, RNA degradation, DNA synthesis/replication and DNA repair; 'Metabolism' includes amino acids, nucleotides, sugars, lipids, cofactors, protein modification, energy and carrier proteins/membrane transport; 'Signaling and communication' includes receptors, hormone/growth factors, intracellular transducers, effectors/modulators, metabolism, cell adhesion and channels/transport proteins; 'Structure and motility' includes cytoskeletal, microtubule-associated proteins/motors and extracellular matrix.

Assignment of gene families to each of the functional classes was carried out according to their annotations in Gene Ontology 42 and LocusLink 43 databases. Out of the total 1,556 gene families, 1,317 could be classified into any of the six functional classes. The remaining 239 families could not be classified unambiguously due to limited information on gene function. Subsequent analysis, with respect to functional classification and distribution of (TG/CA)_n repeats, presented here is from 1,317 gene families comprising of 7,928 genes.

The repeats present in human Collagen genes were aligned with Chimpanzee (Pan troglodytes) genome by using 'BLAT' software available at UCSC Genome Bioinformatics Site http://www.genome.ucsc.edu/cgi-bin/hgBlat49. Nucleotide segments including the repeats and containing 200 nucleotides upstream of the start and 200 nucleotides downstream from the end of each of the (TG/CA)_n repeat were extracted for human Collagen genes 48. These segments were aligned with the Chimpanzee genome (Build 1, version 1, Nov 2003) using BLAT. Only those segments that showed more than 92% identity were noted as conserved.

Significance of the differences between the proportions of genes containing repeats and repeats densities in the six functional classes compared with global distribution was tested using Binomial proportions test. The observed proportion in each class was tested against the expected proportion, which was computed assuming no preference with respect to function. Correlation coefficient (R) was computed to examine the relationship between average gene length of gene families belonging to a functional class and the proportion of genes with (TG/CA)_n repeats in the corresponding functional classes. The 'Interactive Statistical Calculation Pages' website http://members.aol.com/johnp71/javastat.html was used to perform the statistical tests.