An abundance of bidirectional promoters in the human genome

Abstract

The alignment of full-length human cDNA sequences to the finished sequence of the human genome provides a unique opportunity to study the distribution of genes throughout the genome. By analyzing the distances between 23,752 genes, we identified a class of divergently transcribed gene pairs, representing more than 10% of the genes in the genome, whose transcription start sites are separated by less than 1000 base pairs. Although this bidirectional arrangement has been previously described in humans and other species, the prevalence of bidirectional gene pairs in the human genome is striking, and the mechanisms of regulation of all but a few bidirectional genes are unknown. Our work shows that the transcripts of many bidirectional pairs are coexpressed, but some are antiregulated. Further, we show that many of the promoter segments between two bidirectional genes initiate transcription in both directions and contain shared elements that regulate both genes. We also show that the bidirectional arrangement is often conserved among mouse orthologs. These findings demonstrate that a bidirectional arrangement provides a unique mechanism of regulation for a significant number of mammalian genes.

Figure 1

Distribution of distances between human genes and their nearest neighbors. (A) The distribution of distances between 5′ ends of genes on opposite strands shows a bimodal distribution, indicating that 11.8% of genes have TSSs separated by less than 1000 bp. (B) Analysis of the distribution of distances between the 5′ ends on the same strand shows that only 0.4% of gene pairs are closer than 1000 bp. (C) Analysis of the distribution of distances between the 3′ ends of genes on opposite strands shows that 7.4% of gene pairs are closer than 1000 bp.

Figure 2

Bidirectional gene pairs show correlated expression patterns in microarray experiments. (A) The frequency distribution of Pearson correlation coefficients of 500,000 random gene pairs averaged over 18 published microarray data sets. (B) The frequency distribution of correlation coefficients of nonbidirectional nearest neighbor gene pairs averaged across the 18 microarray experiments. (C) The frequency distribution of correlation coefficients of bidirectional gene pairs averaged across the 18 microarray experiments.

Figure 3

Functional activity of bidirectional and random promoters. (A) The distribution of the ratios of promoter strength in the forward and reverse directions for 56 random nonbidirectional promoters. (B) The distribution of the ratios of promoter strength in the forward and reverse directions for 258 bidirectional promoters. Based on a Kolmogorov-Smirnov test, these two distributions are significantly different (P < 0.001).

Figure 4

Bidirectional promoters function as inseparable units. Normalized luciferase activity is plotted next to each construct, with arrows indicating the direction of transcription. (A) Loss of upstream promoter sequence leads to loss of transcriptional activity, indicating that there are shared elements necessary for full promoter activity in each direction. (B) A similar series of deletion constructs illustrates that a shared 30-bp fragment is able to confer full activity in both directions.