The epigenomic landscape of African rainforest hunter-gatherers and farmers

Abstract

The genetic history of African populations is increasingly well documented, yet their patterns of epigenomic variation remain uncharacterized. Moreover, the relative impacts of DNA sequence variation and temporal changes in lifestyle and habitat on the human epigenome remain unknown. Here we generate genome-wide genotype and DNA methylation profiles for 362 rainforest hunter-gatherers and sedentary farmers. We find that the current habitat and historical lifestyle of a population have similarly critical impacts on the methylome, but the biological functions affected strongly differ. Specifically, methylation variation associated with recent changes in habitat mostly concerns immune and cellular functions, whereas that associated with historical lifestyle affects developmental processes. Furthermore, methylation variation--particularly that correlated with historical lifestyle--shows strong associations with nearby genetic variants that, moreover, are enriched in signals of natural selection. Our work provides new insight into the genetic and environmental factors affecting the epigenomic landscape of human populations over time.

Figure 1. Study design and genetic structure of rainforest hunter-gatherers and farmers.

(a) Geographic location of the sampled rainforest hunter-gatherer (RHG) and farmer (AGR) populations. (b) Principal component analysis (PCA) of the genotype data for the study populations, based on 456,507 independent genome-wide SNPs. The tree presented at the top right of the panel represents the branching model for these populations. (c) Schematic representation of the different population comparisons, indicated by arrows, used for the detection of differentially methylated sites (DMS) between groups.

Figure 2. DNA methylation profiles and functional differentially methylated regions.

(a–d) PCA of genome-wide DNA methylation profiles for the different population comparisons. (e,f) Gene ontology (GO) enrichment analysis for (e) recent DMS and (f) historical DMS. The top GO categories for biological processes and molecular functions are shown, together with the log-transformed FDR-adjusted enrichment P values.

Figure 3. Contribution of genetic variation to the DNA methylation levels.

(a) Proportion of methylation sites that are associated with a nearby genetic variant (in grey) and among different DMS sets (in colour). The numbers in the bars correspond to the total number of DMS per population comparison. P values were calculated by resampling. (b) Proportion of the variance of DNA methylation explained by nearby genetic variants (R²) for the various meQTL sets, in each population. The P values (Mann–Whitney U-test) obtained indicate a significant skew in the R² distribution of the various meQTL–DMS sets (in colour) with respect to that of all meQTLs (in grey) in the corresponding population. R² values are higher for meQTLs associated with historical DMS (11.5% (10.7–12.3%) and 10.0% (8.9–11.2%) in w-RHG and f-AGR, respectively) than for those related to recent DMS (6.5% (5.7–7.2%) and 6.8% (6.1–7.4%) in w-AGR and f-AGR, respectively). NS, not significant, *P<0.05, **P<0.01, ***P<0.001. (c–f) Examples of meQTLs detected in this study. The three boxplots on the left represent the distribution of M-values as a function of genotype. The minor allele frequency of each meQTL is presented for each population. Red lines indicate the fitted linear regression model for M-value ∼ genotype for each population. The forest plots on the right represent the estimated β, corresponding to the slope of the linear regression, for each population. (c–e) meQTLs detected in all populations but presenting different allelic frequencies between RHG and AGR groups. The mean F_ST values between w-RHG and f-AGR/w-AGR groups for the SNPs concerned were higher (0.15, 0.19 and 0.10, respectively) than that observed genome wide (F_ST<0.03). (f) Population-specific meQTL, where the SNP rs1534362 is associated with methylation differences in the enhancer region at 6p12.3 only in RHGs.

Figure 4. Selection signals at genetic variants associated to DNA methylation levels.

(a,b) Odds ratios measuring the enrichment in high (a) F_ST and (b) LSBL values among meQTLs, with respect to the remainder of genome-wide SNPs located in a 20-kb window surrounding each methylation probe, in the different population comparisons. P values were calculated using a Cochran–Mantel–Haenszel test, stratified by derived allele frequencies. The colours in the plots correspond to the (a) population comparisons and (b) genetic distances shown in the schematic trees below each plot. (c) Odds ratios measuring the enrichment in high |iHS| values for the different meQTL data sets (in colour). P values were estimated using a χ²-test. For F_ST, LSBL and |iHS|, we considered only SNPs with an LD r²<0.8. NS, not significant, *P<0.05, **P<0.01, ***P<0.001.