Recent Selection Changes in Human Genes under Long-Term Balancing Selection

Abstract

Balancing selection is an important evolutionary force that maintains genetic and phenotypic diversity in populations. Most studies in humans have focused on long-standing balancing selection, which persists over long periods of time and is generally shared across populations. But balanced polymorphisms can also promote fast adaptation, especially when the environment changes. To better understand the role of previously balanced alleles in novel adaptations, we analyzed in detail four loci as case examples of this mechanism. These loci show hallmark signatures of long-term balancing selection in African populations, but not in Eurasian populations. The disparity between populations is due to changes in allele frequencies, with intermediate frequency alleles in Africans (likely due to balancing selection) segregating instead at low- or high-derived allele frequency in Eurasia. We explicitly tested the support for different evolutionary models with an approximate Bayesian computation approach and show that the patterns in PKDREJ, SDR39U1, and ZNF473 are best explained by recent changes in selective pressure in certain populations. Specifically, we infer that alleles previously under long-term balancing selection, or alleles linked to them, were recently targeted by positive selection in Eurasian populations. Balancing selection thus likely served as a source of functional alleles that mediated subsequent adaptations to novel environments.

Keywords: environmental changes; natural selection; out-of-Africa..

Fig. 1

P values of the neutrality tests: HKA and MWU. The cells are colored according to the 5% significance threshold: Green for balancing selection with excess of diversity (HKA) or intermediate frequency alleles (MWU); blue for positive or negative selection, with excess of low-frequency alleles (MWU). For the results of all genes, see supplementary figure S2, Supplementary Material online.

Fig. 2

PtoD in 1000 Genomes populations. We performed the analysis in windows of 10,000 bp sliding by 100 bp. Windows with more than 40% of the sequence not passing our quality filters were excluded. (A) The ranges of PtoD (y-axis) across all windows in each gene are shown as vertical lines, with the gene symbol placed in the average PtoD. For each continent, we also show the expectation under neutrality as the 95% and 99% CIs (thicker and thinner vertical lines, respectively), calculated from 10,000 neutral simulations of the human demography (Gravel et al. 2011) using 1 × 10⁻⁸ per site per generation as average mutation and recombination rates. (B–E) PtoD along 400,000 bp region of the chromosome (x-axis) centered on each candidate gene. The dots are colored according to population (as in A); the dotted and dashed blue lines mark the 95% and 99% CIs of expected PtoD for Africans (they are a conservative representation in non-Africans, which have lower levels of genetic diversity). The rectangles on the x-axis represent genes in positive (above) and negative (below) orientation, with the candidate genes in black.

Fig. 3

Two-dimensional SFS. (A) SNPs from the control regions and (B) SNPs from the four candidate genes combined, where red dots are nonsynonymous SNPs. The histograms on the top and right side of the scatterplot are the SFS for the x and y population. The representation of the scatter plot is colored according to the SNP density. Because the SFS in each population includes sites that are monomorphic but segregate in the other population, the excess of intermediate frequencies in the candidate genes is not as evident as in classical SFS plots (see supplementary fig. S5, Supplementary Material online, for the one-dimensional SFS for each population and supplementary fig. S6, Supplementary Material online, for the SFS for each gene). Supplementary figure S7, Supplementary Material online, shows the other pairwise population comparisons, which are very similar.

Fig. 4

Allele frequency of the most differentiated nonsynonymous iAdO-alleles (one per gene) in the 1000 Genomes populations. The blue and orange portions of the pie charts represent the ancestral and derived alleles, respectively. The SNP names (as “chromosome:position”) are on the top right of each plot. The patterns are similar for all other nonsynonymous iAdO-alleles (supplementary table S5 and fig. S9, Supplementary Material online).

Fig. 5

Two-dimensional SFS for each of the candidate genes. See figure 3 for more details and supplementary figure S8, Supplementary Material online for other pairwise comparisons.

Fig. 6

Evolutionary models and ABC results. (A) An overdominant balanced polymorphism (green dot) that arose Tbs generation ago (green star) increases to intermediate frequency (∼50%) and is maintained at that frequency in African populations for all models. To illustrate the behavior of the balanced polymorphism in Eurasia in each model, we represent in horizontal lines a population of ten chromosomes at different times. The colored vertical lines illustrate one sample of the possible allele frequency trajectories (with derived allele frequency on the x-axis). We refer to the Results and Materials and Methods sections for a detailed description of the models. (B) Posterior probabilities of the hierarchical ABC approach when one B-P model (dark blue) was tested against the B-B (green) and B-N (gray) models. (C) Posterior probabilities of each of the three B-P models: B-Pcfe (aquamarine), B-Psv (light blue), and B-Pdn (blue). Supplementary figure S14, Supplementary Material online, shows very similar results when using the ABC approach that compares all five models together.

Fig. 7

Haplotype networks. The circles are proportional to the number of haplotypes, with colors representing populations. The length of the branch between two haplotypes is proportional to the number of differences. SNPs with a global count lower than six were removed to reduce complexity. The networks were generated using the function “haploNet” from the R-package “pegas” (Paradis 2010) and are cladistic trees (Templeton et al. 1992) which do not allow reticulations.