Transcription factor binding and modified histones in human bidirectional promoters

Abstract

Bidirectional promoters have received considerable attention because of their ability to regulate two downstream genes (divergent genes). They are also highly abundant, directing the transcription of approximately 11% of genes in the human genome. We categorized the presence of DNA sequence motifs, binding of transcription factors, and modified histones as overrepresented, shared, or underrepresented in bidirectional promoters with respect to unidirectional promoters. We found that a small set of motifs, including GABPA, MYC, E2F1, E2F4, NRF-1, CCAAT, YY1, and ACTACAnnTCC are overrepresented in bidirectional promoters, while the majority (73%) of known vertebrate motifs are underrepresented. We performed chromatin-immunoprecipitation (ChIP), followed by quantitative PCR for GABPA, on 118 regions in the human genome and showed that it binds to bidirectional promoters more frequently than unidirectional promoters, and its position-specific scoring matrix is highly predictive of binding. Signatures of active transcription, such as occupancy of RNA polymerase II and the modified histones H3K4me2, H3K4me3, and H3ac, are overrepresented in regions around bidirectional promoters, suggesting that a higher fraction of divergent genes are transcribed in a given cell than the fraction of other genes. Accordingly, analysis of whole-genome microarray data indicates that 68% of divergent genes are transcribed compared with 44% of all human genes. By combining the analysis of publicly available ENCODE data and a detailed study of GABPA, we survey bidirectional promoters with breadth and depth, leading to biological insights concerning their motif composition and bidirectional regulatory mode.

Figure 1.

Three types of motif representation in bidirectional promoters. Motifs in the overrepresented category occur more often (e.g., four times) than the sum of occurrences in two unidirectional promoters (e.g., 2 + 1 = 3 times). Motifs in the shared category occur as often (e.g., three times) as the sum of occurrences in two unidirectional promoters. Motifs in the underrepresented category occur less often (e.g., once) than the sum of occurrences in two unidirectional promoters.

Figure 2.

Enrichment of motifs, TF-binding, modified histones, and other nonsequence-specific factors in bidirectional promoters. The number of binding sites in bidirectional and sampled unidirectional promoters are normalized by the maximum number of sites and plotted against POSSUM log likelihood scores. Counts in the unidirectional reverse gene set are not shown for clarity because they are very similar to the binding-site counts in the unidirectional forward gene set (solid gray). The solid-black lines are the result of summing binding sites in the unidirectional forward and unidirectional reverse gene sets. Error bars indicate the standard error of the mean binding-site counts derived from 10 randomizations. ^aA TF that has a corresponding TRANSFAC motif placed in the same category; ^ba TF whose categorization is consistent among multiple ChIP experiments in different cell types or conditions; ^ca nonsequence-specific factor.

Figure 3.

The distribution of Pearson correlation coefficients for (A) randomly paired genes of unidirectional promoters, (B) randomly paired divergent genes of different bidirectional promoters, and (C) divergent gene pairs of the same bidirectional promoters. Divergent genes from the same bidirectional promoters are more correlated in expression than randomly paired genes of unidirectional promoters (P = 2.2 × 10⁻¹⁶ by the Wilcoxon rank sum test) and randomly paired genes of different bidirectional promoters (P = 1.21 × 10⁻¹¹).

Figure 4.

The fraction of promoters that overlap at least 1 bp with a set of ranked ChIP-chip target lists is plotted on the y-axis (see Methods for details). (A) Percent overlap of bidirectional promoters; (B) twice the percent overlap of unidirectional promoters; (C) percent overlap of unidirectional promoters. Both H3K4me3 and MYC overlap more with bidirectional promoters than twice the overlap of unidirectional promoters.

Figure 5.

Histone modification ChIP intensities downstream of ENCODE genes. Average log2 ChIP intensity of several modified histones and formaldehyde-assisted isolation of regulatory elements (FAIRE) in all genes (Giresi et al. 2007), actively transcribed genes, and 46 divergent genes. Divergent genes have elevated histone modification signals downstream of their TSS, similar to actively transcribed genes.

Figure 6.

Discretized histone modification patterns are not uniformly distributed. The most prominent pattern is labeled 1, where ChIP intensities are elevated downstream of both genes, but diminished throughout the bidirectional promoter. Also prominent are patterns 2 and 3, where the ChIP intensities are just elevated in one downstream gene (the genomic forward direction or reverse direction, respectively). The pattern labeled 4 represents the case where ChIP intensities are elevated in the bidirectional promoter as well as both downstream genes, and 5 represents no elevation throughout the region. The number 6 represents three remaining patterns: elevated ChIP intensities in the bidirectional promoter but diminished in both downstream promoters, elevated ChIP intensities in the bidirectional promoter and downstream of the forward gene, but diminished in the reverse gene, and elevated ChIP intensities in the bidirectional promoter and downstream of the reverse gene, but diminished in the forward gene.

Figure 7.

GABPA ChIP in three region categories and four motif score cutoffs. GABPA binds preferentially to bidirectional promoters over unidirectional promoters and over nonpromoter regions at all motif score cutoffs. Regions containing high-scoring binding sites are more likely to bind GABPA than those containing only medium or low-scoring sites. NA indicates that no regions in this category were tested.

Figure 8.

Mutagenesis of a 30-bp bidirectional fragment. Bases essential for promoter activity in each direction are underlined. Transversion mutations were introduced three bases at a time and shifted two bases for a total of 15 mutants. Mutation of a 12-bp region annihilates promoter activity in both directions. The consensus CCGGAARYR is recognized by the transcription factor GABPA. Asterisks (*) indicate position of transcription start sites.