In addition to genomic regions with regulators that affect single phenotypes in cis or in trans, we found genomic regions containing transcriptional regulators that influence multiple expression phenotypes. We divided the autosomal genome into 491 windows of 5 Mb and determined the number of regulators mapping to each window. We began by examining the regulators for the 142 expression phenotypes with P < 4.3 × 10−7. We found windows that contained many more ‘hits’ than expected by chance. If regulators for these phenotypes were distributed at random across the genome, the probability of six or more hits per window would be less than 6 × 10−5 and we would not expect to see any windows with more than four hits. Instead, we found two hotspots with six or more hits (P < 0.03 after Bonferroni correction): seven phenotypes mapped to one window on chromosome 14, and six phenotypes mapped to one window on chromosome 20.
When we relaxed our linkage criterion to include the 984 regions with P < 3.7 × 10−5, we found many more expression phenotypes whose regulation mapped to shared hotspots. The two regions indicated above contain trans-acting regulators for the most expression phenotypes (). Regulation for 31 of the 984 expression phenotypes mapped to the 5-Mb window on chromosome 14 (14q32), and regulation for 25 phenotypes mapped to the window on chromosome 20 (20q13).
Figure 2 Master transcriptional regulators. a, Distribution of significant linkage peaks (P < 3.7 × 10−5) in 5-Mb windows across the autosomal genome. Arrowheads indicate the two windows (located on chromosomes 14 and 20) that contain regulators (more ...)
We consider the existence of hotspots to be evidence for master regulators of the baseline level of gene expression. The mapping was done without considering possible relationships among phenotypes, but the shared expression control regions suggest co-regulation. We therefore examined the correlation in expression levels of the 31 and 25 target genes corresponding to the two master regulatory regions. The expression levels in 94 CEPH grandparents were used. In permutation tests with 1,000 replications, we found that the pair-wise correlation between any two expression phenotypes did not exceed 0.52.We therefore set 0.52 as the threshold for correlation by chance (nominal P = 0.001). Hierarchical clustering was used to summarize these results graphically and group genes by the correlations of their expression levels. We looked for clusters of expression phenotypes whose members have pairwise correlations that all exceed 0.52. Among the 31 target genes whose regulators mapped to the chromosome 14 hotspot, we found one such regulated cluster with 14 genes, and three additional clusters each with two genes (). Similarly, among the 25 phenotypes whose regulators mapped to the chromosome 20 region, we found one cluster of four and two clusters of two genes whose members have pairwise correlations that all exceed 0.52. The correlation in expression level of these genes supports the observation that they share common transcriptional regulators. However, the regulatory regions defined by mapping are still large, and there might be subgroups of co-regulated phenotypes that are influenced by distinct, but very closely linked, regulators.
Some sets of closely linked genes are influenced by the same cis
regulators, and have correlated expression profiles16–18
. In our data, some target genes whose expression levels map to a trans
-acting master regulatory region are also located very close to each other. For example, among the target genes whose expression maps to the regulatory region on chromosome 14 are four genes (MMP24
) found in a 6-Mb region of chromosome 20 (UCSC Genome Browser, version hg15). In addition, among the target genes of the regulatory region on chromosome 20 are two genes (ITM2B
) separated by less than 50 kilobases (kb) on chromosome 13. These observations reflect a complex regulatory network where master transcriptional regulators affect baseline expression levels of many genes that have similar expression profiles, and in some cases, reside close together on human chromosomes.