Evolutionary processes acting on candidate cis-regulatory regions in humans inferred from patterns of polymorphism and divergence

Abstract

Analysis of polymorphism and divergence in the non-coding portion of the human genome yields crucial information about factors driving the evolution of gene regulation. Candidate cis-regulatory regions spanning more than 15,000 genes in 15 African Americans and 20 European Americans were re-sequenced and aligned to the chimpanzee genome in order to identify potentially functional polymorphism and to characterize and quantify departures from neutral evolution. Distortions of the site frequency spectra suggest a general pattern of selective constraint on conserved non-coding sites in the flanking regions of genes (CNCs). Moreover, there is an excess of fixed differences that cannot be explained by a Gamma model of deleterious fitness effects, suggesting the presence of positive selection on CNCs. Extensions of the McDonald-Kreitman test identified candidate cis-regulatory regions with high probabilities of positive and negative selection near many known human genes, the biological characteristics of which exhibit genome-wide trends that differ from patterns observed in protein-coding regions. Notably, there is a higher probability of positive selection in candidate cis-regulatory regions near genes expressed in the fetal brain, suggesting that a larger portion of adaptive regulatory changes has occurred in genes expressed during brain development. Overall we find that natural selection has played an important role in the evolution of candidate cis-regulatory regions throughout hominid evolution.

Figure 1. Polymorphism and divergence across pooled sites in African and European Americans.

Figure 2. Allele frequency spectra for different classes of sites.

Unfolded site frequency spectrum of nonsynonymous, synonymous, and conserved non-coding sites in 15 African Americans (above) and 20 European Americans (below); the data was re-sampled for 16 chromosomes to account for missing data. We used the chimpanzee to infer the ancestral state of each polymorphism, and corrected for ancestral misidentification using the method of Hernandez et al. .

Figure 3. Gene-specific estimates of selection in mkprf.

Distribution of the probability that γ falls within 5 categories: strong negative selection (red, γ<−1), weak negative selection (brown, −1<γ<−0.5), nearly neutral (yellow, −0.5<γ<0.5), weak positive selection (purple, 0.5<γ<1), and strong positive selection (blue, γ>1). Data shown is for African Americans from a concurrent analysis including (A) candidate cis-regulatory regions, (B) nonsynonymous sites, and (C) synonymous sites, and an independent analysis including (D) simulated neutral data under the inferred demographic model.

Figure 4. Correlation in estimates of γ at different classes of sites within a gene.

There is a significant, yet weak positive correlation between estimates of γ in candidate cis-regulatory regions and nonsynonymous sites in both African Americans (top left) (Kendall's tau = 0.055, p = 1.8×10⁻¹¹), and European Americans (top right) (tau = 0.043, p = 2.9×10⁻⁷). There is a slightly stronger correlation between synonymous and nonsynonymous sites in both African Americans (bottom left) (tau = 0.096, p<10⁻¹⁶), and European Americans (bottom right) (tau = 0.087, p<10⁻¹⁶). Candidate cis-regulatory, synonymous, and nonsynonymous sites were run in a single, concurrent run of mkprf.

Figure 5. A comparison of signatures of selection between different classes of sites.

Cumulative distributions of (A) the probability of negative selection [Pr(γ)<−0.5] and (B) the probability of positive selection [Pr(γ)>0.5] across different classes of sites, and the distribution of γ in (C) candidate cis-regulatory and nonsynonymous sites, and (D) candidate cis-regulatory and synonymous sites. Data shown is for African Americans from a single, concurrent analysis in mkprf including all classes of sites.