Cross-continental whole-genome sequencing dataset
We generated high-coverage whole-genome sequence data for 95 individuals from Uganda (70 Batwa and 25 Bakiga, East Central Africa), 43 individuals from the Philippines (20 Mamanwa and 23 Manobo, Southeast Asia) and 28 individuals from Sri Lanka (15 Adivasi and 13 Sinhalese, South Asia), for a total of 166 individuals. These newly sequenced genomes were integrated with data from 926 individuals contributed by collaborator-sequenced cohorts (see Supplementary Fig. 1 for the data processing workflow), bringing our dataset to a total of 1,092 individuals across Africa, Asia and Oceania (Fig. 1a and Table 1). This dataset captures a broad spectrum of subsistence strategies: among the 1,092 individuals, 670 practise agriculture (AGRs), 341 practise hunting and gathering (HGs) and 81 practise pastoralism (pastoralists (PSTs)) (Fig. 1b).
Fig. 1: Genetic diversity and structure of study populations.
a, Geographic locations of sampled individuals. b, Global PCA, with the subsistence strategy indicated in the inset along with sample sizes. See Supplementary Fig. 1 for a detailed global PCA coloured by population. c, Regional PCAs: Asia, Oceania and Africa. d, Regional ADMIXTURE analysis results at the K value with the lowest cross-validation error: K = 3 for Asia (left), K = 2 for Oceania (centre) and K = 4 for Africa (right). For all plots: triangles, HGs; circles, AGRs; and squares, PSTs. Map generated using the rnaturalearth R package.
Table 1 Summary of samples included in the study
An important goal of our study is to enable comparisons of populations with different subsistence modes while leveraging shared genetic ancestries and geographic proximities wherever possible, enabling us to approximate the effects of subsistence as a localized ecological pressure. For example, in South Asia, the Sinhalese (AGRs) and Adivasi (HGs) were chosen as a comparison pair owing to their shared genetic affinity, as previously described15 and confirmed by our principal component analysis (PCA) and ADMIXTURE analysis (Fig. 1c,d). In other regions, HGs and AGRs do not share recent genetic ancestry. For example, the Batwa (HG) and Bakiga (AGR) populations of East Central Africa diverged ~130–160 ka (refs. 16,17)—in such cases, we prioritized geographic proximity. Of the eight regions, six include at least one HG and one AGR or PST population. In East Asia and Oceania, only AGR populations are represented. This broad sampling strategy ensures regional representation for populations with historically different origins of agriculture18, while contextualizing population structure and demography at different geographic scales.
In Southern Africa, we selected a PST rather than an AGR population owing to the ≠Khomani San’s (a Khoe-San HG population) unique demographic history. The Khoe-San represent one branch of the oldest population divergence in Homo sapiens, whereas all non-Africans and Africans descend from the other branch19,20. As such, the Nama PSTs represent the most appropriate comparative group given that the Nama and ≠Khomani share recent genetic ancestry, and the Nama’s adoption of pastoralism around 2.5 ka (refs. 21,22) appears to be a local cultural process with limited genetic admixture with East African pastoralists23.
We processed all 1,092 samples through a standardized genotype calling and arrived at a baseline genotype call set with 59,774,327 biallelic single nucleotide polymorphisms (SNPs) across the 22 human autosomes. Among these SNPs, 18,722,175 were singletons and 7,121,909 were novel variants not found in dbSNP v156, adding to the pool of known variants across global populations. We define this as the ‘baseline variant call format file (VCF)’ and applied appropriate variant- and sample-level filters for downstream analyses (Supplementary Fig. 1). Briefly, the level 1 VCF excludes individuals deemed heterozygosity outliers (n = 1,078; 59,024,349 SNPs), the level 2 VCF further excludes individuals that are second-degree relatives or closer (n = 918; 57,349,118 SNPs), the level 3 VCF includes phase information (n = 918; 57,349,118 SNPs) and the level 4 VCF includes only variants with ancestral state annotations (n = 918; 54,852,761 SNPs).
Population genetic characterization of the dataset
To contextualize selection signals in the following sections, we first characterized population structure and demographic histories across the study populations. Using the level 2 VCF and excluding variants with a minor allele frequency (MAF) < 0.01 (13,924,142 SNPs remaining), we performed PCA (Fig. 1c and Supplementary Fig. 2) and ADMIXTURE analysis (K = 2–13, where K is the clustering parameter for ADMIXTURE; for all results, please refer to Supplementary Fig. 3) to broadly characterize population structures and genetic differentiation across the study samples. This was conducted at the global level (Fig. 1b and Supplementary Figs. 2 and 3) and at the supraregional level in Africa, Asia and Oceania (Fig. 1c,d). We recapitulated expected global patterns of genetic diversity, in which African populations separated from non-African populations along PC1 (Fig. 1b). PC2 and PC1 in the global (Fig. 1b) and Africa supraregional (Fig. 1c) PCAs, respectively, separate the Southern African ≠Khomani, an Indigenous Khoe-San HG population, from all other populations. The ≠Khomani harbour the greatest amount of genetic diversity across surveyed human populations and represent one of the deepest population divergences within our species24. We observed that the supraregional clustering analyses for Africa generally differentiate based on geography, with most HG, and PST and AGR, populations from Southern and Eastern Africa sharing closer genetic affinities within their respective regions despite subsistence differences (Fig. 1c,d). Notable exceptions include several admixed Nama individuals with genetic ancestries related to both the ≠Khomani and Central and East Africans25, and the Himba, Southern African PSTs who share genetic affinity with Bantu-speaking AGRs in Central Africa, as noted in other published studies26. Within Africa, contrasting genetic ancestries between subsistence groups are primarily observed in West and East Central Africa, where the HGs Batwa and Baka share greater genetic affinity than their respective neighbouring AGRs, supporting the deep divergence between rainforest HGs and farmers16,17.
The Asian and Oceanian populations cluster closely together in the global PCA (Fig. 1b), yet a finer-scale population structure emerges in the supraregional PCAs and ADMIXTURE analyses. In Asia, PCA (Fig. 1c) and ADMIXTURE analysis (Fig. 1d) highlight both shared and distinct genetic ancestries across regions and subsistence methods at a broad scale. The Korowai HGs from western New Guinea (geopolitically, Indonesia) represent a Papuan-like genetic ancestry that is differentiated from the rest of the sampled Asian populations. ADMIXTURE analysis shows a Korowai-associated component that is shared with both South Asian populations, the HG Adivasi and AGR Sinhalese, which probably represents a basal East Eurasian ancestry shared before the split of South Asians, East Asians and Australians27. Similarly, the Aeta and the Agta, HGs from the Philippines, are differentiated from other study populations, although the Agta share genetic ancestry with AGR-related sources28,29. By contrast, East Asian (Taiwanese) and Island Southeast Asian (Philippines and Indonesia) AGR populations tend to be genetically more similar to one another. While this pattern is concordant with the ‘Out of Taiwan’ expansion ~4–5 ka, which involved migrations of Austronesian-speaking, farming populations across Island Southeast Asia and the Pacific30, these affinities may also reflect deeper pre-expansion relationships among ancestral populations in Southeast Asia. We also observe examples of diverse admixture histories across Asia that are not strictly driven by geography or subsistence alone: the Mamanwa and Ati HGs from the Philippines, who are genetically more similar to AGRs from the Philippines29, and the Flores Rampasasa and Sumba AGRs from Indonesia, who have genetic ancestries from both AGR and HG sources31,32. Finally, in Oceania, where only AGR populations were sampled, we observed an ancestry gradient defined by Papuan- and Austronesian-like genetic ancestries, consistent with previous studies on the peopling of Near and Remote Oceania33,34.
Demographic history
We inferred historical effective population sizes (Ne) using MSMC2 (ref. 35) on eight haplotypes per population. Marked differences emerged in the recent past between populations practising different subsistence strategies. Across most regions, AGR populations showed stable or increasing Ne in the Holocene (Fig. 2a and Supplementary Fig. 4), consistent with demographic expansions associated with the adoption of farming36. By contrast, HG populations showed pronounced declines in recent Ne, with the opposing demographic trajectories being most evident in South Asia, East Africa, East Central Africa and, to a lesser extent, Southeast Asia.
Fig. 2: Demographic patterns observed in study populations.
a, Historical effective population sizes inferred using MSMC2 (n = 8 haplotypes per population). b, Total ROH length (Mb) in one representative HG and non-HG population per geographical region. The bars indicate median ± interquartile range (IQR), and whiskers, three times the IQR. A single ADV outlier beyond 3 s.d. has been removed for visualization purposes. Differences between HG and non-HG ROH lengths were assessed using two-sided Wilcoxon rank-sum test, with n individuals per population detailed in the source data. See Supplementary Fig. 4 for results from all populations.
We further investigated historical genetic isolation by quantifying total runs of homozygosity (ROHs) per individual across populations (Fig. 2b; see Supplementary Fig. 5 for results from all populations). Across all regions, apart from Southern Africa, HG populations showed greater total ROH lengths than AGR and PST populations of the same region, suggesting that HG populations had smaller Ne and/or experienced more genetic isolation or endogamy in the recent past. This difference in total ROH by subsistence is more pronounced in Asia than in Africa and particularly striking in Southeast Asia; for example, the Aetas (AET) had three times the mean total ROH length as the Cebuanos (PHP) (228.7 Mb versus 73.1 Mb). Interestingly, the Khomani San (SAN) showed relatively shorter total ROH than their two neighbouring PST populations despite their HG lifestyle (Supplementary Fig. 5). Some possible explanations are that the SAN have maintained large Ne over long evolutionary timescales20, and that flexible social networks and mobility among different Khoe-San populations may have facilitated intergroup gene flow. In fact, recent genetic evidence based on mitochondrial DNA (mtDNA) and autosomal data from modern and ancient individuals from across Africa suggests continuous gene flow between HG groups, including Khoe-San-speaking populations, over long periods of time37. Present-day Khoe-San populations may reflect the aggregation of lineages from genetically structured HG groups that occupied much of Southern Africa before the influx of Bantu speakers, retaining substantial genetic diversity from the region22.
Collectively, our MSMC2 and ROH results highlight the impact of subsistence transitions on recent human demography, with agriculture generally associated with larger effective population sizes and lower ROH burden. However, HG groups show remarkable diversity in social organization, mobility patterns and interactions with neighbouring populations38, resulting in heterogeneous genetic patterns.
Selection on the immune system, metabolism and endocrine function
The advent of agriculture catalysed several key changes in human diets, relationships with pathogens and reproductive patterns. We asked whether the transition to farming drove parallel adaptive shifts in these biological pathways across diverse populations facing comparable ecological challenges. Specifically, we tested for enrichment of recent signals of positive selection in the form of per-gene integrated haplotype scores (iHS) across increasing thresholds in (1) virus-interacting proteins (VIPs), reflecting pathogenic burdens in foraging versus agricultural societies; (2) metabolic genes, given documented dietary shifts to cereal-based sources; and (3) endocrine genes, which underpin growth and reproductive timing. We targeted VIPs because they represent a major substrate of recent adaptation among agricultural populations from the 1000 Genomes project39. However, as those analyses primarily focused on farming groups, it remains unclear whether similar patterns extend to populations practising alternative subsistence modes. The list of VIPs has been published40, while metabolic and endocrine-related genes were curated using Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (Supplementary Table 2).
Across all regions, VIPs showed consistent enrichment across increasing iHS thresholds regardless of subsistence strategy (χ2 P < 0.01) (Fig. 3a), challenging the prevailing assumption that agriculturalists alone experienced heightened pathogen-driven selection during the Neolithic owing to sedentism and livestock domestication41. To further investigate whether the nature of this immune selection differs between subsistence groups, we performed an over-representation analysis on VIP genes that were above the 95th percentile of iHS scores and exclusive to each subsistence mode. Both groups showed enrichment for antiviral processes, consistent with pervasive viral selective pressure across subsistence modes. However, the specific genes and pathways enriched differed qualitatively. HG populations showed enrichment for intracellular viral restriction factors, including genes involved in positive regulation of viral genome replication (false discovery rate (FDR) = 6.0 × 10−4), driven by TRIM38, STAU1, LARP1, SRPK1 and DDB1, which encode constitutively expressed factors that broadly restrict RNA and DNA virus replication at the cellular level. By contrast, AGR populations showed specific enrichment for interferon-alpha production and its regulation (FDR = 0.018), driven by IFIH1 (MDA5), DHX36, DHX9 and NMI. IFIH1/MDA5 is a cytosolic pattern recognition receptor for double-stranded RNA and a well-documented target of pathogen-driven selection42. Its exclusive enrichment in AGRs is consistent with selective pressure imposed by novel RNA viruses associated with dense settlement and livestock proximity. Together, these findings suggest that while both HGs and AGRs have undergone strong selection on antiviral VIPs, the molecular targets may reflect their distinct pathogenic environments: broad intracellular restriction in HGs and innate RNA sensing and interferon induction in AGRs.
Fig. 3: Enrichment of selection signals in virus-interacting, metabolic and endocrine genes.
a, Fold enrichment of genes categorized into high-level biological functions across increasingly extreme gene iHS percentiles. Each panel shows one representative population for each region and subsistence group. Colour-filled circles indicate two-sided χ2 test of independence P < 0.01. b, Logistic regression testing for enrichment of top 5% iHS genes in subcategories of metabolic and endocrine gene sets. Fully solid data indicate two-sided χ2 test of independence P < 0.05. ***P < 0.0005; **P < 0.005; *P < 0.05. See Supplementary Table 3 for full pathway-level results. Exact n and P values for each panel can be found in source data.
In contrast to the broad enrichment of VIPs across geographic regions, metabolic genes showed significant depletion at high iHS percentiles in a more region-specific manner, including in Southeast Asian and West Central African HGs and AGRs, and East Central African AGRs (χ2 P < 0.01) (Fig. 3a). Meanwhile, endocrine genes showed no consistent evidence for recent selection across populations (Fig. 3a). When we examined functional subcategories of metabolic and endocrine genes, we found that genes involved in specific pathways operated in opposite directions (Fig. 3b). Most populations demonstrated significant enrichment for genes involved in carbohydrate metabolism (odds ratio (OR) = 1.6–3.9), indicating robust signals for positive selection. We also observed a striking enrichment of oestrogen signalling genes in most populations (OR = 2.2–8.7), suggesting selective pressure related to hormonal regulation, reproductive timing or fertility. Enrichment results for all other subcategories are provided in Supplementary Table 3.
Collectively, these results highlight pervasive immune-related selection across populations, while selection on general metabolic and endocrine pathways appears more context dependent. We reason that variation in local dietary ecologies, food-processing practices, and even recent market integration and access to imported foods may modulate selective pressures on metabolic pathways, such that broad-scale subsistence labels alone may lack the resolution to capture fine-scale lifestyle-driven pressures on selection.
Selection on immune and metabolic regulatory elements
Given that the most pervasive signals of recent selection across populations were observed in VIPs, we next investigated whether selection can be detected in other components of the immune system. Following inspection of iHS outlier SNPs in VIP-encoding regions, we noticed that there was no significant enrichment for nonsynonymous variants bearing signals of positive selection (Supplementary Fig. 6), suggesting that recent selection is acting predominantly on non-coding regions.
An increasing body of work has shown that regulatory regions are often targets of natural selection12,13,14,43, as variation in these regions can alter gene expression, chromatin accessibility, interactions with transcription factors and ultimately organismal fitness. We therefore tested whether recent positive selection preferentially targeted immune candidate cis-regulatory elements (cCREs) from immune and metabolically active tissues and whether enrichment patterns vary with subsistence strategy. We considered three classes of cCREs: promoter-like signatures (PLS), distal enhancer-like signatures (dELS) and proximal enhancer-like signatures (pELS). Immune cCRE annotations were generated in B cells, CD4 T cells, CD8 T cells, erythroblasts, myeloid cells, NK cells and progenitor cells, and metabolic cCRE annotations were derived from adipose, liver and pancreas tissue44,45.
We modelled the presence or absence of each cCRE class against the top 1% of windows by absolute iHS, allowing us to quantify how immune and metabolic cCREs respond to selection across varying ecological contexts and demographic backgrounds. Across populations, single-predictor models showed consistent enrichment of recent selection in windows containing immune cCREs (Fig. 4a). All three classes—PLS, pELS and dELS—showed clear evidence of selection, with the strongest enrichments in enhancer-like signatures, particularly among AGR groups, followed closely by HG and PST populations. We observed comparable patterns for metabolic tissue cCREs, with significant enrichment across all three regulatory classes (Fig. 4a). While PLS and pELS elements from adipose, liver and pancreas tissue showed consistent enrichment across populations and subsistence strategies, dELS elements from metabolic tissues showed a stronger and more AGR-specific signal, mirroring the pattern observed for immune cCREs and suggesting that the regulatory architecture of metabolic adaptation may be particularly sensitive to the lifestyle changes associated with the adoption of agriculture.
Fig. 4: Recent positive selection on regulatory variation.
a, OR ± 95% CI from logistic regression enrichment of the top 1% iHS 100-kb windows in tissue cCREs. Enrichments were modelled separately for each tissue–cCRE type. Populations with fewer than 15 individuals were excluded. The grey-shaded regions indicate agriculturalist populations. b, OR ± 95% CI from fixed-effect meta-analysis (n = 26 populations) across logistic regression enrichments of the top 1% windows in tissue cCREs. All tissues and cCRE types were modelled jointly. c, OR ± 95% CI from fixed-effect meta-analysis (n = 26 populations) across logistic regression enrichments of the top 1% windows for immune-cell-type-specific cCREs. Each cell type–cCRE combination was modelled separately. d, Fold enrichment of cis-eQTLs across windows of increasing iHS percentiles for five populations with available expression data (Batwa, Bakiga, Korowai, Mentawai and Sumba); all data points are significant as determined by two-sided χ2 test of independence at P < 0.05. ***P < 0.0005; **P < 0.005; *P < 0.05. Exact n and P values for each panel can be found in source data and Supplementary Table 5 (d).
To assess whether these signals are tissue specific, we conducted fixed-effect meta-analyses incorporating immune, metabolic and brain cCREs simultaneously as predictors46 (Fig. 4b). This joint model revealed a clear contrast by tissue: brain enhancer-like elements were significantly depleted for signatures of recent selection (PLS, P = 3.5 × 10−80; pELS, P = 7.1 × 10−6; dELS, P = 1.1 × 10−4), probably reflecting strong evolutionary constraint on neural developmental programmes. By contrast, both immune and metabolic enhancer-like elements were significantly enriched for selection signals (immune PLS, P = 8.0 × 10−23; immune dELS, P = 9.4 × 10−60; adipose pELS, P = 3.7 × 10−58; liver pELS, P = 1.4 × 10−3; liver dELS, P = 1.8 × 10−5). This analysis also suggests that the broader enrichments across all cCRE classes observed in the single-tissue analyses may be mediated by pleiotropic regulatory elements active across immune and metabolic tissues, which become less distinguishable in the joint model.
Together, these results suggest that both immune and metabolic regulatory elements have been recurrent targets of recent positive selection across populations practising different subsistence strategies. Notably, fixed-effect estimates for immune and metabolic cCRE classes were significantly higher among AGRs than among HGs (Supplementary Table 4). This suggests that the adoption of agriculture enhanced selection on both pathogen-facing and diet-related regulatory elements, potentially driven by novel immunological exposures, dietary shifts towards cultivated carbohydrate-rich foods and greater efficacy of selection in populations with larger Ne. Importantly, the presence of selection signals in HGs and PSTs underscores that these groups also faced substantial immune challenges, including exposure to zoonotic and vector-borne pathogens, such as malaria in rainforest HGs and livestock-associated enteric parasites in PSTs47,48, which probably also contributed to adaptive changes in immune cCREs.
Finally, to examine whether recent selection on immune regulatory elements shows cell-type-specific heterogeneity, we implemented similar fixed-effect meta-analysis models incorporating seven immune-cell-type-specific annotations across the three cCRE classes (Fig. 4c). This allowed us to estimate the contribution of regulatory elements associated with different arms of the immune system. Signals of recent selection were strongest in regulatory elements specific to the lymphoid lineages, primarily CD4+ T cells and B cells involved in adaptive immune responses. By contrast, myeloid lineage cCREs, key mediators of innate immunity sensing, showed no enrichment in PLS and dELS signatures, and only modest enrichment in pELS (OR = 1.27, P < 5 × 10−4). Regulatory elements specific to progenitors were consistently depleted for selection signals (PLS, OR = 0.68; pELS, OR = 0.70; dELS, OR = 0.75; P < 5 × 10−4 for all three classes), probably reflecting strong evolutionary constraint on essential developmental functions.
Together, these findings suggest that recent positive selection has preferentially targeted regulatory elements in adaptive immune cell types, particularly at enhancer regions, while progenitor and innate immune cell programmes have remained largely unaffected by similar selective pressures.
Selection on genetic variants regulating gene expression
If recent selection has acted on candidate regulatory elements, as suggested by our results, genetic variants impacting gene expression—expression quantitative trait loci (eQTLs)—should also show enrichment for selection signals. To test this hypothesis, we intersected SNP-level iHS scores with cis-eQTLs mapped in five populations that overlap with our dataset. These included the Batwa (TWA) and Bakiga (KIG) populations from Uganda, for which eQTLs were mapped in peripheral blood mononuclear cells49, and the Korowai (KWI), Mentawai (MTW) and Sumba (SMB) populations from Indonesia, for which eQTLs were mapped in whole blood50.
We calculated the fold enrichment of eQTLs across increasing iHS thresholds by comparing the proportion of eQTL SNPs with non-eQTL SNPs exceeding each threshold within the corresponding population (Fig. 4d). Across all five populations, we observed a consistent pattern of increasing enrichment at more extreme iHS percentiles, with fold-enrichment ratios approaching or exceeding 2-fold at the 99th percentile (χ2 P values are provided in Supplementary Table 5). These patterns were not driven by differences in allele frequency, as eQTLs and non-eQTLs showed nearly identical MAF distributions (Supplementary Fig. 7).
These results provide evidence that recent selection has acted on variants with functional effects on gene expression, reinforcing our broader conclusion that regulatory variation, particularly that modulating immune function, has been a recurrent target of selection across human populations.
Patterns of convergent evolution associated with subsistence strategies
Most classical selection scans, such as iHS, are agnostic to ecological information and therefore provide limited insight into the environmental drivers of adaptation. To explicitly test whether genomic loci involved in local adaptation are associated with subsistence strategies, we used Bayenv to identify SNPs whose allele frequencies correlate with HG or non-HG (AGR and PST) lifestyles across populations51. As this framework leverages population-level covariance in allele frequencies, it is primarily sensitive to longer-term selective processes and less influenced by very recent changes in subsistence practices. We tested 7,124,658 common SNPs (MAF > 0.05) using 5 independent Markov Chain Monte-Carlo (MCMC) runs, treating subsistence mode as the environmental variable. This approach identified 4,145 SNPs significantly correlated with the subsistence mode (Supplementary Data 1), based on a median log10(Bayes factor (BF)) > 1.5 (corresponding to a likelihood ratio > 30) and median absolute Spearman correlations ≥ 0.05. After linkage disequilibrium (LD) pruning, these outliers condensed to 1,883 approximately independent signals.
Among the top 25 genic variants by BF, we identified signals belonging to immune, metabolic, nervous system and spermatogenesis processes (Fig. 5a). The most significant signal was rs9296077 in the 3′ untranslated region (UTR) of HLA-DPB1 (log10(BF) = 121.67), which encodes one of the major histocompatibility complex (MHC) class II molecules expressed in antigen-presenting cells. Overall, we found that outlier SNPs were significantly enriched in proximal enhancer-like regions and UTRs (Supplementary Fig. 8c), consistent with a broader trend of selection acting on gene regulatory elements. In addition, Bayenv outlier SNPs were strongly enriched for variants with high posterior probabilities of positive selection over the past 10,000 years (4-fold enrichment, P = 2.02 × 10−59; Supplementary Fig. 8d) as inferred from temporal allele frequency shifts in ancient Eurasian genomes52. This time window spans the global transition to agriculture and the later Bronze Age, marked in Europe by the expansion of pastoralist Yamnaya-related populations, major demographic growth and a period increasingly proposed to involve intensified pathogen-driven selection53,54. The strong overlap therefore supports the view that the convergent signals we detect reflect genuine adaptations to shifts in subsistence, population structure and infectious pressures.
Fig. 5: Patterns of convergent evolution associated with subsistence strategies.
a, Manhattan plot of log10(BF) for all tested SNPs, with significant SNPs coloured red. For visual clarity, SNPs not meeting the Spearman correlation threshold were brought to log10(BF) = 1.5, and SNPs above log10(BF) = 50 were brought to 50. The top 25 variants labelled by unique genes are annotated with the gene’s biological function. b, Population alternate allele frequency (AAF) subtracted from the regional mean AAF for four representative SNPs. The horizontal bars represent the mean deviation by subsistence within each region. See Supplementary Data 1 for the full list of significant SNPs and source data for exact n values. c, We performed gene set enrichment using the one-sided hypergeometric test of the genes overlapped by significant SNPs, showing significant gene sets after FDR correction pertaining to biological functions of interest. See Supplementary Data 2 for the full list of enriched gene sets. Exact n and P values can be found in source data.
To characterize the biological processes associated with subsistence-related adaptation, we performed an over-representation analysis on the list of genes overlapping all significant SNPs. This revealed 408 significantly enriched pathways (FDR < 0.05), which are broadly clustered into five functional categories: immune function, metabolism, nervous system processes, development and muscle-related pathways (Fig. 5c and Supplementary Data 2). The most significantly enriched terms were linked to developmental and growth processes, including anatomical structure development (P = 3.7 × 10−7) and cellular developmental process (P = 2.8 × 10−6), suggesting that selection has acted on genes shaping life-history traits. Numerous metabolic and immune-related pathways were also enriched, including lipid transport (P = 1.7 × 10−2), branched-chain amino acid metabolic process (P = 8.3 × 10−3), regulation of T cell activation (P = 0.029) and MHC protein complex (P = 2.8 × 10−3), providing independent support for adaptation within the same biological systems identified through our haplotype- and regulatory-based analyses, while also pinpointing functions more tightly linked to lifestyle transitions. Finally, enrichments related to nervous system function and muscle development, such as glutamatergic synapse (P = 3.3 × 10−2) and muscle structure development (P = 1.4 × 10−2), may reflect changes in neuromuscular and physiological traits associated with sedentism, energy allocation and altered activity patterns following the adoption of agriculture. Interestingly, while 4 of the top 25 genic variants by BF were associated with genes implicated in spermatogenesis (CDYL, MYCBPAP, SPATA20 and LZTF1) (Fig. 5a), we did not observe broader enrichment of reproduction-related signals broadly within all Bayenv outliers (Fig. 5c).
Several individual SNPs illustrate the biological and evolutionary relevance of these findings. For example, we identified a cluster of ten tightly linked outlier SNPs in WNK4, including three exonic variants associated with pseudohypoaldosteronism type II (for example, rs9916754 shown in Fig. 5b). Given the pleiotropic roles of WNK4 in modulating nutrient absorption and lung inflammatory responses55,56,57, it emerges as a potential candidate of selection driven by dietary and immunological shifts. Signals in the MHC region, including rs9461833 and rs1401331217 (Fig. 5b) near HLA-DPB1 and HLA-C, respectively, showed consistent allele frequency differences between HG and non-HG groups across regions. Given the long-standing association between MHC diversity and pathogen-mediated selection58,59,60, this probably reflects convergent immune adaptation to ecological factors tied to lifestyle.
Taken together, our Bayenv results provide compelling evidence that subsistence modes have driven convergent selective pressures on biological functions or traits related to immunity, diet and energy expenditure.