As a first step in the identification of TFs in Caenorhabditid species, we generated an updated list of putative C. elegans TF genes by searching its annotated genome sequence (Wormbase WS173 release) 17 for gene ontology (GO) terms associated with transcription factors. This led to the identification of 1271 putative TF genes (Table 1). Since our criteria for selecting a TF was the presence of a well-defined DNA binding domain that selectively modulates gene transcription (for example, bHLH or homeobox), we manually inspected the above list of putative TFs. This allowed us to reject 564 genes as false positives since these encode factors that are associated with the basal transcriptional apparatus (for example, DNA polymerases), chromatin alterations, DNA packaging (histones), as well as entries that were incorrectly curated in Wormbase (Additional files 1, 2, 3). To the remaining genes (707), we added 281 TF encoding genes found in published literature and other public database entries that were not identified in our initial search (See Materials and Methods). The final C. elegans TF set included a total of 988 genes (Table 2 and additional file 4), of which 917 are shared with the previously annotated C. elegans TF set (wTF2.0, 934 genes) 18. The 17 genes in the wTF2.0 set that are not shared in our updated set either lack a known DNA binding domain or are annotated as pseudogenes (Additional file 5). The increased number of genes in the present TF set likely results from the availability of annotations published since the compilation of wTF2.0.

We used the newly defined C. elegans TF set to search for homologs in the fully sequenced genomes of C. briggsae (CB3 release) and C. remanei (11/29/2005 release) 171920. We used InParanoid 21 to identify 713 and 703 best reciprocal hit orthologs in C. briggsae and C. remanei, respectively (Table 2, see Material and Methods). To these lists, we added 282 C. briggsae and 390 C. remanei putative TF genes that were identified through Hidden Markov Model (HMM)-based searches 22. Altogether, a total of 995 and 1093 potential TF encoding genes were identified in C. briggsae and C. remanei, respectively (Table 2 and additional files 6 and 7). Among the TF orthologs in the three nematode species, we identified 652 genes that exhibit a 3-way best reciprocal BLAST orthologous relationship (hereafter referred to as the nematode TF set) (Figure 1). The proportion of C. elegans TF genes with detectable orthologs in C. briggsae (713/995, 71.7%) is significantly higher compared to the proportion of all conserved genes between the two species (12858/20621, 62.4%) 13 (χ² = 7.56, df = 1, p = 6.0 × 10^-3), which may indicate strong selective pressure to maintain these genes.

To examine the evolutionary conservation of the nematode TF set of genes in other phyla, we searched for their orthologs in the genomes of fruit fly(D. melanogaster), mouse (M. musculus), and human (H. sapiens). Using the InParanoid database 23 we identified a total of 150 TFs that exhibit reciprocal orthologous relationships between three nematode species and are conserved in fly, mouse, and human (Additional files 4 and 8).

Best-hit reciprocal orthologs could not be identified for 215 TF genes in C. elegans, 211 in C. briggsae, and 310 in C. remanei (Figure 1 and additional files 4, 6, and 7). It should be pointed out that C. briggsae and C. remanei TF genes are based on computational predictions and that the C. remanei genome has yet to be assembled; hence while many of the TF genes without detectable orthologs may have arisen by lineage-specific gene duplication, others could result from incomplete annotation of the C. briggsae and C. remanei genomes. Therefore, the actual number of divergent TF genes in these species is likely to be smaller than the numbers we have estimated. To further study this set of genes in C. briggsae (211), we searched for their closest homologs in C. elegans. This revealed 30 genes with weak sequence similarity (BLASTP E-value > 10^-10) suggesting that these most likely represent candidate C. briggsae-specific TF genes (Additional file 9). The remaining 181 appear to be species-specific paralogs, of which 69 are zinc finger-C4/nuclear hormone receptor (ZF-C4/NHR) family members (see below).

Previous studies in humans and other organisms have shown that TF genes tend to evolve more rapidly than non-transcription factor (non-TF) genes 242526, therefore we performed a similar analysis in nematodes by analyzing their coding sequence divergence and comparing it to non-TF genes. Due to the large divergence times between the three species 1315, the rate of synonymous substitution per synonymous site (d_S) for many genes is likely to be saturated (d_S > 3, Figure 2A). Therefore we restricted our analysis to the rate of non-synonymous substitution per non-synonymous site (d_N), which does not show such saturation 27. We found a significantly higher d_N for TF genes conserved among nematodes (652) when compared to 3-way conserved non-TF gene orthologs (10,827 genes; Kruskal-Wallis rank sum test, p < 2.2 × 10^-16; Figure 2B), whereas no difference was detected between TF genes with orthologs in nematodes, fly, mouse, and human (150) and non-TF genes (Kruskal-Wallis test, p = 0.3498).

We studied the distribution of protein families among C. elegans TFs based on known DNA binding domains. This analysis revealed more than 50 distinct families of which 30 were found to contain 5 or more members (Figure 3). No significant difference was observed in the representation of various families between the C. elegans set and the set conserved among three nematodes (χ² = 35.05, df = 30, p = 0.2408), indicating that the distributions of TF families in these species may be similar. The majority of genes in C. elegans and nematode TF sets (28.6% and 20.5%, respectively) were found to encode the nuclear hormone receptors (NHRs), a C4-type sub-family of zinc finger proteins that play key roles in development and homeostasis 28. The NHR genes was previously shown to have undergone extensive lineage-specific expansion in C. elegans 29. Besides NHR, HOX genes that regulate cell fate specification and embryogenesis 30 are also among the largest TF families in nematodes (11% of C. elegans TF genes, and 11.6% of nematode TF genes) (Figure 3A, C). In contrast, the distribution of TF families among the divergent C. elegans gene set (215 genes) differs significantly from that observed among the entire C. elegans TF set (χ² = 83.91, df = 30, p = 5.33 × 10^-7) due to its high proportion of NHR genes (52.6%, Figure 3B). Likewise, the representation of different families among TF genes with orthologs in nematodes, fly, mouse, and human also differs from that of the C. elegans TF set (χ² = 152.27, df = 30, p = 0, Figure 3D). Interestingly, the single largest conserved family represented among the orthologs in three different phyla is HOX (17.3%), supporting multiple previous studies indicating the importance of this family among all metazoans 183031.

Studies in C. elegans as well as other organisms have shown that genes that are co-expressed and/or functionally related are frequently clustered together on chromosomes 3233343536. To investigate whether TF genes in nematodes exhibit a similar pattern, we plotted the physical locations of C. elegans and C. briggsae TF genes using non-overlapping windows of 200 kb (the genome of C. remanei has not yet been assembled and therefore was not used in this analysis). Figure 4 shows that TF genes in C. elegans and C. briggsae as well as those that are conserved among nematodes are non-randomly distributed on chromosomes. A total of 183 C. elegans TF genes were found to be located in 25 distinct clusters (marked with stars in Figure 4A, Table 3). A similar pattern was observed in C. briggsae (184 genes in 27 clusters) (Figure 4B and Table 3). Chromosome V carries highest number of clusters (and genes) in both species (C. elegans: 97 genes in 10 clusters; C. briggsae: 64 genes in 7 clusters) (Table 3) that are primarily composed of NHR family members (92% in C. elegans and 84% in C. briggsae, red bars in Figure 4).

The analysis of the chromosomal distribution of TF genes also revealed that many members of the large TF families, such as ZF-C4/NHR, ZF-C2H2, T-box and HOX are arranged in perfect tandem arrays (defined as having a contiguous repetition of TF genes) (267 C. elegans genes in 103 arrays, 235 C. briggsae genes in 107 arrays), the largest of which consists of 8 NHR genes in C. elegans and 6 NHR genes in C. briggsae (Figure 5 and additional files 10 and 11). Although the majority of such arrays consists of genes of the same TF family (76.7% in C. elegans and 67.3% in C. briggsae), less than half of all genes found in such arrays have best-reciprocal hit orthologs between the two species (31.6% in C. elegans, 47.2% in C. briggsae) (Additional files 10 and 11) suggesting significant lineage-specific duplication and expansion of the tandem arrays.

Our findings extend Robinson-Rechavi et al.'s analysis of the extensive lineage-specific expansion of NHR genes in C. elegans 29 to the other two Caenorhabditid species (283, 232, and 256 NHR genes in C. elegans, C. briggsae, and C. remanei, respectively). The sequence analyses revealed a total of 134 NHR genes having 3-way best-reciprocal orthologs among the nematode species (Additional files 4, 6, and 7). The remaining NHRs are composed of what appear to be lineage-specific paralogs and those that have diverged sufficiently in sequence such that orthologous relationships could no longer be assigned.

We constructed a phylogenetic tree of the nematode NHR family members (437 genes, see Materials and Methods) to study their inter– as well as intra-specific relationships. The most striking feature of the phylogeny is the frequent presence of several closely related NHRs located tandemly on the same chromosome (Additional file 12). Such groupings suggest the presence of extensive tandem duplications, which could explain the mechanism behind the expansion of the NHR gene family, and perhaps the independent occurrence of some NHR genes in the lineages of each of these species. In the case of C. elegans NHRs, we found at least 15 distinct groups on chromosome V including 7 that are located in one large cluster of the phylogeny (Additional file 12).

The presence of NHRs in chromosomal clusters prompted us to study their distribution in further detail. We identified a total of 47 tandem arrays composed of contiguous repetitions of NHR genes in C. elegans, which are found on all chromosomes with the exception of chromosome III (Additional file 10). These include 10 arrays that are comprised of 5 or more genes, all of which are located on chromosome V. A similar analysis in C. briggsae identified 30 NHR arrays having 6 or fewer genes (Additional file 11). In total, 9 NHR arrays were partially or completely conserved between C. elegans and C. briggsae. One of these arrays, for instance, consists of 7 genes in C. elegans (nhr-136, nhr-153, nhr-154, nhr-206, nhr-207, nhr-208, and nhr-209) and the corresponding 4 in C. briggsae (CBG23383/Cbr-nhr-136, CBG23380/Cbr-nhr-153, CBG23380/Cbr-nhr-154 and CBG23379/Cbr-nhr-209). This suggests that either the array has expanded in C. elegans or perhaps lost 3 of the genes in C. briggsae. Examination of the C. remanei TFs revealed the presence of best reciprocal hit orthologs for all array members found in C. elegans with the exception of nhr-206 leading us to propose that nhr-207 and nhr-208 were most likely lost in the C. briggsae lineage. This analysis, however, carries a caveat in that the annotations of the C. briggsae and C. remanei genomes are based on computational predictions and lack experimental validation.

Finally, we found that 7 tandem arrays in C. briggsae are composed of NHR genes that lack best reciprocal hit orthologs in C. elegans and C. remanei (Additional file 11). The largest of these is comprised of 6 NHR genes (CBG01243, CBG01244, CBG01245, CBG01246, CBG01247, CBG01248) (Figure 5). These C. briggsae-specific arrays may be caused by lineage-specific expansion although the possibility of a selective loss of their orthologs in other species cannot be ruled out.

We investigated the relationship between sequence conservation and function of TF genes in C. elegans. From a comprehensive list of 13,647 RNAi phenotypes associated with 4,351 genes 14, we identified 281 TFs that exhibit one or more mutant phenotypes (Additional file 13). These consist of more than half of all TF genes conserved among nematodes, fly, mouse, and human (52.7%, 79 of 150), over one-third of genes conserved among the three nematode species (36.5%, 238 of 652), and one-fifth of the TF genes in C. elegans that did not have identifiable orthologs in the other nematode species (20%, 43 of 215). We also determined the number of distinct mutant phenotypes associated with TF genes in each of the above three groups as well as with non-TF genes. This analysis revealed that TF genes conserved among nematodes, fly, mouse, and human are linked to a significantly greater number of mutant phenotypes in C. elegans when compared to the other sets (4.38 ± 2.31, 3.36 ± 2.09, 2.91 ± 1.82 and 3.19 ± 1.84 phenotypes per gene for TF genes conserved across phyla, conserved in nematodes, C. elegans TF genes without detectable orthologs in the other nematode species and non-TF genes, respectively; Kruskal-Wallis rank sum test, p = 8.58 × 10^-3, 1.8 × 10^-3, 1.32 × 10^-15, respectively, after Bonferroni correction). No difference was found in pairwise comparisons between the other gene sets (Kruskal-Wallis rank sum test p = 1, in all comparisons after Bonferroni correction).

To further analyze the roles of C. elegans TF genes in specific tissues and developmental processes, RNAi phenotypes were sorted into six broad categories: viability (embryonic and post-embryonic growth and survival), fertility (germline and germ cells), sex (sex determination and reproductive system), vulva (vulval cell proliferation and morphogenesis), body (cuticle, size, and morphology), and behavior (movement and feeding) (Additional files 13 and 14). Among the six categories, "viability" ranks highest in terms of the proportion of TF and non-TF genes (Figure 6). However, it is important to keep in mind that this may be linked to a greater interest in identifying transcription factors that are involved in growth and survival of C. elegans. A further sub-classification of this category into "embryonic viability" and "post-embryonic viability" (based on the phenotype when lethality occurs in RNAi-treated animals) revealed that among the "embryonic viability" class TF genes conserved among nematode species are significantly under-represented when compared to the non-TFs (χ² = 7.39, df = 1, p = 6.56 × 10^-3) (Figure 6), while no difference was observed among the datasets for genes affecting post-embryonic viability (χ² = 0.77, df = 1, p = 0.38). By contrast, the TF genes conserved among nematodes, fly, mouse, and human showed no enrichment for any of these two categories (χ² = 3.59 and 0.39, df = 1, p = 0.059 and 0.53, respectively). Among other categories, we observed an over-representation of mutant phenotypes for nematode-conserved as well as nematode-fly-mouse-human-conserved TF gene sets associated with vulval development (χ² = 8.24 and 11.75, df = 1, p = 4.1 × 10^-3 and 6.08 × 10^-4, respectively) and sex determination and reproductive system-related processes (χ² = 9.51 and 8.78, df = 1, p = 2.04 × 10^-3 and 3.59 × 10^-3, respectively) when compared to the non-TF gene set.

The above findings that more than half of C. elegans TF genes conserved across phyla are associated with RNAi phenotypes prompted us to examine their mutant phenotypes in other organisms. We found that 69 (46%) of TF genes conserved among nematodes, fly, mouse, and human are associated with lethal phenotype in fly (Additional file 15). In the case of mouse, out of a total of 81 orthologs for which knock out and mutant phenotypes are described (see Materials and Methods), 75 (92.6%) exhibit defects ranging from mild to gross abnormalities, including lethality (Additional file 15). A similar analysis in human revealed 35 TF genes linked to various diseases and genetic disorders (Table 4). In total, 44 (29.3%) TF genes regulating C. elegans development and behavior are also essential either in mouse or human or both. These include 30 genes that control viability in the fruit fly. Overall, 121 (80.7%) TF genes conserved among nematodes, fly, mouse, and human play important roles in at least one of these organisms. This is likely an underestimate due to technical limitations of RNAi experiments (e.g., strains, redundancy of factors or pathways) and that comparisons between organisms involve different experimental approaches (e.g., RNAi in C elegans and chromosomal mutations in D. melanogaster). Thus, functional studies of conserved TFs in C. elegans promise to elucidate mechanisms involved in biological processes conserved across phyla.

To further explore the mechanism of transcription factor function in metazoans, we generated an interaction map of C. elegans TF genes based on known physical and genetic interactions 3738. The map consists of 1594 interactions involving 277 TF genes and their direct non-TF interactors (Figure 7A and additional file 16). The network appears to be scale free as seen by the presence of several nodes with high degree of connectivity (such as lin-35, which shows the highest number of interactions and is connected to more than one-third of all existing nodes; 521 of 1340) (Figure 7B). lin-35 is an ortholog of the human Retinoblastoma (Rb) gene which plays an important role in cell proliferation 3940. Among the lin-35 interacting genes, 43 (8%) encode TFs, of which 18 have best reciprocal hit orthologs in mouse and human. Other prominent hubs include pal-1 (conserved among nematode species), as well as other TF genes with orthologs in nematodes, fly, mouse, and human: tag-331, eya-1, and sma-4 (Figure 7B). Each of these genes plays important role in C. elegans development, and RNAi-mediated knock-downs cause defects such as slow growth (pal-1), lethality (pal-1, tag-331), larval arrest (eya-1, tag-331), uncoordinated movement (eya-1), and small size (sma-4) 41424344. Interestingly, the subnetwork comprising of the hub gene tag-331 (human ortholog RNF113A) and its 32 direct interactors appears to be largely isolated. A closer examination revealed that two-thirds of these 22 genes are conserved in nematodes yet lack best reciprocal hit orthologs in fly, mouse, or human genomes. The remaining third includes four genes conserved in nematodes, fly, mouse, and human (zfp-1, R11F4.1, apl-1 and fcd-2) and whose human homologs are linked to disorders (AF10/MLLT10: leukemia, Glycerol kinase: hyperglycerolemia, APP: Alzheimer's, and FANCD2: Fanconi anemia). It remains to be determined if the human genes interact with RNF113A as well as whether RNF113A is involved in any of these diseases. Among the remaining hub genes, the eya-1 mammalian orthologs promote development of tissues and organs 414546 whereas the sma-4 ortholog SMAD4/DCP4 acts as a tumor suppressor 474849.

In addition to analyzing the prominent hubs in the interaction network, we also examined the relationship between connectivity of TFs, sequence conservation, and known function. The results revealed a significantly greater number of interactions among C. elegans TFs that are conserved in nematodes, fly, mouse, and human, as compared to those that are not (Kruskal-Wallis rank sum test, p = 0.0207, after Bonferroni correction). We also found that TF genes associated with mutant phenotypes in RNAi assays exhibit significantly more interactions when compared to those that lack a detectable phenotype (Kruskal-Wallis rank sum test, p = 0.0069). These results are consistent with previous studies showing that highly connected hubs tend to be enriched in essential genes 5051.

The C. elegans TF-encoding genes were searched using 8 GO terms (Table 1) within WS173 release of Wormbase. The C. briggsae and C. remanei TFs were identified using the HMMER 2265 and InParanoid programs 21. The complete genome sequences of each of the three Caenorhabditid species were downloaded from WormBase (C. elegans release WS173, C. briggsae release WS173 and C. remanei release 11/29/2005) 17. As the C. remanei predicted peptide dataset is known to contain redundant copies of genes due to heterozygosity in the sequenced genome, (E. Schwartz, personal communication) we used the CD-HIT program (version 2007-0131) 66 in order to cluster and remove all additional transcripts that had greater than or equal to 98% sequence similarity to other transcripts at the protein level. The original dataset of 25,948 transcripts was truncated down to 24,267 non-redundant transcripts that were used in further analysis 27.

InParanoid was run with default values, using blastall version 2.2.14 with –VT emulation, on all three complete genome predicted peptide datasets in pairwise comparisons. The results were collected and placed into species-specific paralogs, 2- and 3-way best-hit reciprocal ortholog categories using custom PERL scripts. Each category was searched for genes from the C. elegans TF set and the number of TFs in each category was identified (Additional files 6 and 7). HMM alignment-based searches were carried out on the C. briggsae and C. remanei predicted peptides using previously established techniques 2267. The HMMER signature files (profiles) of known DNA binding domains were retrieved from Pfam 68. In most cases, a cut-off score of 0.1 was used. If a HMMER predicted TF gene in non-elegans species lacked a homolog in C. elegans, it was considered false positive and therefore removed creating the final, conservative datasets that were used in the study.

The C. elegans orthologs of D. melanogaster, M. musculus and H. sapiens TFs were retrieved using the data available on the InParanoid database 2369.

Genes were grouped into different families based on the presence of known DNA-binding domains according to the WormBase 14, Pfam 68, and InterPro 70 databases. Only well defined and unambiguous domains that are known to be involved in transcriptional regulation were considered. Families with fewer than 5 members were placed together in a miscellaneous category. The TF families shown in Figure 3 are as follows. AP2: Activator protein-2 family; AT hook: AT hook DNA binding motif (preference to A/T rich region) family; bHLH: basic helix-loop-helix family; bZIP: basic leucine zipper family; CBFB/NF-YA: CCAAT binding factor family; CSD: Cold shock DNA binding domain family; HMG box: High mobility group box family; HOX: Homeobox family; MADF: Myb DNA binding domain family; SAND: DNA binding domain family named after Sp100, AIRE-1, NucP41/75, DEAF-1; SANT: Myb-like DNA binding domain; SMAD: SMAD (Mothers against decapentaplegic (MAD) homolog) domain family; T-box: T-box family; WH: Winged-helix family; WH-FH: Winged-helix and Forkhead domain family; WH-ETS: Winged-helix and ETS domain family; ZF-C2H2: C2H2-type zinc finger protein family; ZF-C2H2-BED: C2H2 and BED-type zinc finger protein family; ZF-BED: BED-type zinc finger family; ZF-C2H2-RING: C2H2 and RING-type zinc finger protein family; ZF-C4/NHR: C4-type zinc finger/Nuclear hormone receptor family; ZF-CCCH: C-x8-C-x5-C-x3-H class of zinc finger family; ZF-DHHC: DHHC-type zinc finger family; ZF-FLYWCH: FLYWCH-type of zinc finger family; ZF-GATA: GATA class of zinc finger family; ZF-PHD: C4HC3 zinc-finger-like motif family; ZF-others: zinc finger family members not listed above; ZF-DM: DM (dsx and mab-3) zinc finger family; ZF, AT hook: AT hook and zinc finger domain family; ZF, SANT: SANT and zinc finger domain family; Misc: Miscellaneous TF family not listed above.

The physical locations of C. elegans and C. briggsae TF and non-TF genes were retrieved from Wormbase (WS173 release) and grouped into non-overlapping windows of 200 kb (similar to the 250 kb used by 33). A 400 kb window analysis was also performed and the conclusions remain the same (data not shown). Since many genes are alternatively spliced, we eliminated transcript-specific bias by focusing on single open reading frame for each transcription factor. In the case of C. briggsae, a total of 1329 genes were not assigned to any of the chromosomes and hence were excluded from the analysis. For simplicity, we only used the average between the start and end positions as a proxy for the gene position. The significance of TF clustering on chromosomes was determined by comparing their frequency with the overall frequency of genes in a given window using a χ² test 33. Clusters with p value less than 0.05 were considered significant.

The predicted C. elegans NHR gene dataset (283 genes) was used to identify orthologs and paralogs in C. briggsae and C. remanei using the complete genome INPARANOID datasets (see above). 204 and 152 potential homologs were identified in C. briggsae and C. remanei, respectively. The peptide dataset was aligned using Dialign 2.2 71 and then manually inspected. We identified two large conserved blocks within most predicted peptides and removed all sequences that did not align within these blocks. The remaining sequences were then realigned with Dialign 2.2 and truncated only to retain the two conserved domains. As per Robinson-Rechavi et al. 29 we chose to use only ungapped sites and removed first sequences missing significant portions of the conserved domains and finally excluded all gapped sites. In the end, we retained 437 sequences (213 C. elegans, 106 C. briggsae and 118 C. remanei) for phylogenetic analysis.

The phylogeny was constructed using a maximum likelihood based method as implemented in PhyML 72 using the JTT substitution model 73 with the default proportion of invariable sites (0.0) and rate heterogeneity between sites corrected by a gamma law (using the default gamma parameter of 1.0 and eight rate categories). The phylogeny was then bootstrapped by generating 1000 randomized datasets using SEQBOOT and assessing the percentage of consensus trees using CONSENSE, both in the PHYLIP package 74.

DNA sequences from C. elegans, C. briggsae and C. remanei were aligned according to their protein alignment using Dialign 2.2 75 and RevTrans 1.4 76. Rates of synonymous substitutions per synonymous site (d_S) and non-synonymous substitutions per non-synonymous site (d_N) were estimated using codeml from PAML 77. Evolutionary rates between TF and non-TF data sets were compared using a permuted Kruskal-Wallis rank sum test using 10,000 permutations.

The RNAi phenotypes of all known C. elegans genes were retrieved from Wormbase (WS170 release). A total of 13,648 phenotypes associated with 4,351 genes were analyzed and sorted into 82 different categories (Unc, Dpy, Vul etc.) (Additional files 13 and 14).

For phenotypes associated with C. elegans TF orthologs in fly, mouse, and human, we searched Flybase 78, NCBI OMIM 79, PubMed 80, and other public databases (http://www.informatics.jax.org, http://www.bioscience.org/knockout/alphabet.htm, http://www.dsi.univ-paris5.fr/genatlas, http://www.genetests.org). Only those phenotypes that were unambiguous and did not show discrepancy between different published sources were included. In order to reduced any effect linked to a differential amount of genes annotated as involved in particular mutant phenotypes, all the analyses were performed within each phenotypic class by comparing the distribution of genes with mutant phenotypes among the different sets (non-TF genes, TF genes, C. elegans TF genes, TF genes conserved among the three nematode species, and TF genes with orthologs in nematodes, fly, mouse, and human).

The C. elegans gene network was built using the genetic and protein-protein interaction data for transcription factors curated by BioGRID (version 2.0.27 release) 3738. The network was visualized by using Cytoscape 81.