We tested 121 and 291 randomly selected promoters from divergent and general genes, respectively, by ChIP in three cell lines to determine the percentage of genes from each class that were bound by GABP. Using an enrichment threshold of 3-fold for classifying a promoter as bound, we found that 86.8% of divergent promoters and 30.6% of general promoters were bound in at least one cell type. The significantly larger fraction of divergent promoters bound by GABP was also observed when comparing a subset of the two promoter types with matched CpG content. Furthermore, most divergent promoters were bound in all three cell types, suggesting that GABP regulates these promoters in a wide variety of tissues. This is not surprising given the broad expression of GABP and its tendency to regulate ubiquitously and broadly expressed genes, a class to which many divergent gene pairs belong [3
]. However, a small fraction of divergent genes, as well as a large number of other genes, showed cell line specific binding, so the fraction of promoters regulated by GABP could increase with additional cell type sampling. Collectively, these studies suggest that GABP regulates the vast majority of divergent gene pairs as well as a large number of general genes.
The finding that GABP binds to such a large percentage of divergent gene promoters is surprising, but not unexpected. Our previous study estimated that the transcription factor binds to 57% of divergent gene promoters based upon observed binding frequencies, using a 5-fold enrichment cutoff, for promoters containing a high, medium, or low scoring motif [9
]. In this study, we examined a larger number of promoters without first scanning for GABP motifs, tested an additional cell type and used a less conservative but still stringent cutoff. Applying a binding cutoff of 5-fold to our current data decreases the fraction of bound promoters from 87% to 77% for divergent promoters and from 31% to 20% for general promoters; neither change would alter our conclusions. That GABP binds to such a large fraction of both divergent and nondivergent promoters is also not surprising given that its binding site is one of the top ten most common sequence motifs in the human genome [20
The broad expression of many ets
-family transcription factors and the similarity in their binding sites led us to test whether ETS1, another ubiquitously expressed family member, might also be involved in the regulation of divergent genes [19
]. Interestingly, we found considerable overlap between GABP and ETS1 binding at both divergent and general genes in Jurkat cells. This is perhaps not surprising, as ETS1 is the most highly expressed ets
-family member in this cell line [19
]. However, for all tested cell lines, GABP bound a larger fraction of divergent promoters than did ETS1, and only in Jurkat cells did the latter factor bind more than 10% of targets. In addition to any functional differences conveyed by the binding of these ets
transcription factors, it would appear that GABP regulates divergent promoters on a widespread basis at a basal level with ETS1 and/or other family members, perhaps serving to modify this activity in a cell-specific manner.
In light of the observation that many general promoters are capable of balanced bidirectional transcriptional activity, we explored the relationship between GABP binding and bidirectional transcriptional activity at all promoters [3
]. We found that GABP binding was significantly associated with promoter activity above background in both directions (p
< 0.0001; χ2
) and with balanced bidirectional transcription (p
= 0.0103; χ2
), supporting a model whereby GABP directs bidirectional transcription. Furthermore, GABP enrichment was significantly anticorrelated with the log2
ratio of promoter activity for each direction of the corresponding promoter fragment (Spearman's ρ
0.0067). The low correlation coefficient can be partly explained by recognizing that we would not expect a perfect correlation between the degree of enrichment and a lower ratio of activities in each direction. Although the sum of motif occurrence scores correlates with the degree of enrichment [9
], there is no reason to expect that the strength or amount of GABP binding would lead to more balanced transcriptional activity. Nonetheless, these data strongly suggest that GABP binding directs bidirectional transcription at divergent and general promoters alike.
Despite strong statistical evidence that GABP binding and bidirectional activity are both correlated and associated, we observed a number of unbound promoters that demonstrated balanced bidirectional activity as well as examples of bound promoters with activity primarily or only in one direction. Many of the former cases likely represent promoters that are bound in vivo but were not scored as bound because they fell below 3-fold enrichment threshold. HeLa cells, in addition to having the lowest percentage of bound divergent promoters that are bound by GABP, had enrichment values on average half that of those observed in K562 and Jurkat cells for those promoters that were bound in all three cell types (unpublished data). In addition, there are almost certainly mechanism(s) for bidirectional activity that are not dependent on GABP binding. For example, five adjacent Sp1 sites were capable of directing bidirectional transcription in vitro [21
]. We compared Sp1 and GABP binding at the set of promoters from our previous study [9
] and were not able to observe a significant correlation between factor binding in either Jurkat or K562 cells (unpublished data). Nonetheless, it will be interesting to see the degree of overlap between GABP binding and that of Sp1 and other factors with binding sites over-represented among divergent promoters.
There are several possible interpretations for GABP bound promoters that do not conform to our definition of balanced bidirectional transcription. First, not all divergent promoters have log2
H/L expression ratios of less than 2. Although this definition of balanced bidirectional transcriptional activity encompasses the majority of divergent promoters, it may exclude many biologically relevant bidirectionally active promoters. In addition, unknown regulatory element(s) may modify the bidirectional activity conveyed by GABP binding. For example, TATA elements are known to convey tissue and start-site specificity and may also influence the directionality of transcription [22
]. We tested for a correlation between predicted TATA binding site log-likelihood scores and log2
H/L ratio among all promoters in this study and observed a significant, positive relationship (Spearman's ρ = 0.2601, p
< 0.0001), although, in another study, the introduction of a TATA and/or Inr element into the mouse thymidylate synthase promoter had no effect on its bidirectional activity [23
]. In general, it is clear that GABP binding confers bidirectional activity at a large number of promoters, but this activity may be modified in a number of ways on a gene-by-gene basis.
In this study, we observed balanced bidirectional transcriptional activity at 34% of general promoters. One possible explanation for this observation is that a strong promoter contains elements sufficient for leaky expression in the other direction when reversed in our reporter plasmids. To investigate this possibility, we looked for a correlation between activities in each direction for general promoters and found a significant relationship (Spearman's ρ
< 0.0001). However when we compared the luciferase activity values for general promoters that were categorized as having balanced bidirectional activity and those that did not, there was no significant difference (p
= 0.3419; Mann-Whitney U-test). This suggests that although there may be some expression in the reverse direction due to a strong promoter, this was not sufficient to result in balanced bidirectional activity. Another likely explanation is that the approximately 500-bp promoter fragment we cloned may not have contained all the regulatory sequences necessary for its proper function in the genome. Doubtless other transcription factor(s) and/or boundary elements such as insulators play a role in controlling the level and direction(s) of promoter activity in their endogenous genomic context. Finally, many of the general promoters for which we observed bidirectional activity may actually belong to an unannotated divergent gene pair. A subset of these may prove to be functional bidirectional promoters with a protein-coding gene in one orientation and some other type of transcript in the other. Several recent studies have reported that a much larger percentage of the genome is transcribed than was previously thought, much of this occurring outside of annotated genes [17
], and it will be interesting to see whether these transcripts are in any way specifically initiated or are merely an unintended consequence of GABP regulation of promoter function. While GABP may be responsible for the majority of balanced bidirectional activity, the regulation of directionality at promoters needs more research.
Finally, having shown that GABP is correlated with bidirectional transcriptional activity, we tested whether the introduction of a single GABP site into a promoter with no annotated reverse transcript, little to no promoter activity in the reverse direction and no evidence for GABP binding might be sufficient to induce such activity. Four of six such promoters produced significant increases in reverse luciferase activity after the introduction of a GABP site. Furthermore, for at least one of the promoters that did not show a significant increase in either direction, NM_001697, it is likely that we inadvertently disrupted other sequences necessary for promoter function, as there were significant decreases in activity in both directions. We cannot explain why the introduced site was not capable of directing balanced transcriptional activity in all cases, although it is worth noting that not all divergent promoters exhibit balanced activity. Even in the case of the small change observed in the reverse direction for AF161466, the fold increase was 1.9, which is likely to be biologically relevant for a dosage sensitive gene product. These results are particularly striking given that GABP sites were introduced without regard for site orientation or proximity to the start of transcription. Previous work demonstrated the ability of two tandem GABP sites to drive transcription in either orientation [16
], but our study is the first report to our knowledge to show that a single site can drive transcription in both directions in a functional promoter. Given that previous studies have shown that GABP sites might be required for bidirectional transcriptional activity [9
], these results argue very strongly that a GABP site can be both necessary and sufficient for bidirectional transcriptional activity.
While our data do not suggest how the paired gene arrangement came to exist, they are consistent with a role for GABP regulation in maintaining this relationship. In this model, unknown forces would bring two critical genes together, each with its own promoter, and GABP would reinforce this relationship through its ability to direct balanced, bidirectional transcriptional activity. Given its ability to be sufficient for such activity, the presence of a GABP site could make other factor binding sites redundant over evolutionary time. Thus, the short intergenic region between the gene pair would be free to accumulate mutations, each gene losing its independent regulators and becoming part of an inseparable, truly bidirectional promoter. This supports the conclusion that the promoters of divergent gene pairs do not consist of two overlapping promoters, and that in addition to the previously noted characteristics of these promoters, their bidirectional character is in large part conveyed by GABP binding. Analyses of transcription start site distributions in species from fish to mammals have revealed that although a bimodal distribution, indicative of an abundance of divergent gene pairs, seems to exist only in mammals, 83 divergent pairs have been maintained in proximity and orientation as far separated from humans as Fugu
]. A search of HomoloGene (http://www.ncbi.nlm.nih.gov/sites/entrez?db=homologene
) indicated that although GABPα has an ortholog in Drosophila
, the most distant GABPβ ortholog has been observed in Danio rerio
is not currently represented in HomoloGene). The coincident emergence of conserved divergent gene pairs and a GABP heterodimer capable of bidirectional transcriptional activity reinforces the finding that GABP is the major regulator of divergent genes.
In summary, we showed that GABP binds the majority of divergent promoters and is correlated with and sufficient for bidirectional activity. We have established its role as a major regulator of divergent genes, which carry out a variety of activities critical for the function and survival of many different cell types. In addition, GABP binds a large number of nondivergent genes, and further study will be needed to ascertain whether its ability to promote bidirectional transcription genes has widespread biological consequences through the generation of novel and/or noncoding transcripts.