Prioritizing natural-selection signals from the deep-sequencing genomic data suggests multi-variant adaptation in Tibetan highlanders

Abstract

Human genetic adaptation to high altitudes (>2500 m) has been extensively studied over the last few years, but few functional adaptive genetic variants have been identified, largely owing to the lack of deep-genome sequencing data available to previous studies. Here, we build a list of putative adaptive variants, including 63 missense, 7 loss-of-function, 1,298 evolutionarily conserved variants and 509 expression quantitative traits loci. Notably, the top signal of selection is located in TMEM247, a transmembrane protein-coding gene. The Tibetan version of TMEM247 harbors one high-frequency (76.3%) missense variant, rs116983452 (c.248C > T; p.Ala83Val), with the T allele derived from archaic ancestry and carried by >94% of Tibetans but absent or in low frequencies (<3%) in non-Tibetan populations. The rs116983452-T is strongly and positively correlated with altitude and significantly associated with reduced hemoglobin concentration (p = 5.78 × 10^-5), red blood cell count (p = 5.72 × 10^-7) and hematocrit (p = 2.57 × 10^-6). In particular, TMEM247-rs116983452 shows greater effect size and better predicts the phenotypic outcome than any EPAS1 variants in association with adaptive traits in Tibetans. Modeling the interaction between TMEM247-rs116983452 and EPAS1 variants indicates weak but statistically significant epistatic effects. Our results support that multiple variants may jointly deliver the fitness of the Tibetans on the plateau, where a complex model is needed to elucidate the adaptive evolution mechanism.

Keywords: Tibetan; adaptive genetic variant; archaic ancestry; expression quantitative traits loci (eQTL); hemoglobin concentration; high-altitude adaptation; hypoxia; next-generation sequencing (NGS); tissue-specific expression.

Figure 1.

The landscape of the candidate AGVs in TIB and the candidate adaptive genes involved in the hypoxia-induced pathways. (A) Manhattan plot of the CMS scores across the autosomes. The candidate AGVs are labeled according to their biological effect. CPS, changing protein sequence; RGE, regulating gene expression; UCE, unknown function but conserved in evolution. (B) Proportions of different types of candidate AGVs. A majority of the candidate AGVs are located in the non-coding regions, making our analyses more comprehensive than those of previous studies. (C) The functional enrichment of AGV-related genes. The full priori gene list for each pathway or category appears in Supplementary Table 7. ‘Prior literature’ indicates genes reported by previous studies on high-altitude adaptation in human and non-human species. Here, we only show the 10 categories with odds ratios >1 (the y-axis). The horizontal line in black indicates odds ratio at 1. The adjusted p values for the enrichment of each category are shown above the bars. The red bars indicate significant enrichments (adjusted p < 0.05). (D) HIF pathways and related reactions under normoxia and hypoxia. Candidate adaptive genes (in the blue boxes) are mapped to the pathways they could possibly be involved in. Genes highlighted in red are suggested to carry genomic segments introgressed from archaic hominids (see Methods).

Figure 2.

Colocalization of eQTLs and phenotype-associated candidate AGVs. (A) Genome-wide distribution and linkage disequilibrium (LD) of the candidate AGVs. The candidate AGVs, eQTLs and phenotype-associated loci are indicated by inverted triangles in black, green and orange, respectively. The LD blocks were inferred using Haploview version 4.2 [104] and are presented using the standard color scheme. The three regions of colocalization are marked using ellipses and are labeled as ‘Coloc_Region 1’, ‘Coloc_Region 2’ and ‘Coloc_Region 3’, respectively. (B) Zoom-in plots of candidate AGVs in the three colocalization regions. In each plot, gene locations are shown above the chromosome. The cis-regulated genes are indicated by green bars, while others are indicated by gray bars. The eQTLs and phenotype-associated loci are indicated by inverted triangles in green and orange, respectively. In Coloc_Region 1, the color of each inverted triangle for the eQTL matches that of the bar for the gene regulated by this eQTL. The LD of pairwise SNPs was measured by r² using Haploview version 4.2 [104].

Figure 3.

A human-anatomy plot showing tissue- and organ-specific expression of the candidate HAA-related genes. The expression profiles for these genes were obtained from the Genotype-Tissue Expression (GTEx) database. All reported tissues and organs are shown, except for the cell lines. Genes reported to be HAA-related in Tibetans or showing significant associations with phenotypes in our analysis of 2,849 Tibetans are highlighted in red, while the others are in blue. A full list of the expression patterns of all candidate HAA-related genes is given in Supplementary Table 6. The human-anatomy image was constructed at pngtree.com.

Figure 4.

Signature of local adaptation at rs116983452 and its functional associations. (A) Global distribution of rs116983452-T. Each triangle represents a sampling locality for a population. The map is adapted from http://bzdt.ch.mnr.gov.cn (GS(2016)1665, approved by the Ministry of National Resources of the People’s Republic of China). (B) Median-joining network for TMEM247. The gray area highlights a group of Tibetan-enriched haplotypes with Denisovan origin, all of which carry rs116983452-T. (C) Correlation between the altitude and the derived allele frequency at rs116983452. Each dot represents an Asian population, from both public datasets and our unpublished data. Populations analysed in this plot include various Tibetan populations (labeled), as well as Uyghur, Tajik, Kazak, Hui, Han Chinese, Japanese and Malaysian peoples (unlabeled). (D) Estimation of extended haplotype diversity (EHH) in TIB and HAN around rs116983452. (E) Significant associations between rs116983452 and various quantitative traits. Associations validated in a larger Tibetan population are indicated with fonts in red. (F) The expression of TMEM247 in three groups of Tibetan samples with different genotypes.

Figure 5.

Effects of EPAS1-rs4953354 and TMEM247-rs116983452 on the adaptive traits in Tibetans. The intronic SNP rs4953354 reported in Beall et al. in 2010 [2] was selected as a representative adaptive EPAS1 variant in comparison with the rs116983452 in TMEM247. The locations on chromosome 2 of the two genes are presented by the pink bars against the coordinates above and the positions of the two SNPs are indicated by arrows. For each SNP, the frequency of the adaptive allele and that of three genotypes are shown by blue bars. The genetic effects of rs4953354 and rs116983452 on red blood cell count (RBC), hemoglobin (HGB) and hematocrit (HCT) were tested using three linear-regression models, as illustrated in Methods. The effect size of each variant and the genetic contribution of each model are shown in the green bars below. Significant p values (p < 0.05) are denoted with asterisks. In Model 3, the effect size of the interaction of the two variants is not significant (p > 0.05) and thus is not shown in the figure. Detailed results can be found in Supplementary Table 13.