Dispersals and genetic adaptation of Bantu-speaking populations in Africa and North America

+ See all authors and affiliations

Science  05 May 2017:
Vol. 356, Issue 6337, pp. 543-546
DOI: 10.1126/science.aal1988

On the history of Bantu speakers

Africans are underrepresented in many surveys of genetic diversity, which hinders our ability to study human evolution and the health of modern populations. Patin et al. examined the genetic diversity of Bantu speakers, who account for one-third of sub-Saharan Africans. They then modeled the timing of migration and admixture during the Bantu expansion. The analysis revealed adaptive introgression of genes that likely originated in other African populations, including specific immune-related genes. Applying this information to African Americans suggests that gene flow from Africa into the Americas was more complex than previously thought.

Science, this issue p. 543


Bantu languages are spoken by about 310 million Africans, yet the genetic history of Bantu-speaking populations remains largely unexplored. We generated genomic data for 1318 individuals from 35 populations in western central Africa, where Bantu languages originated. We found that early Bantu speakers first moved southward, through the equatorial rainforest, before spreading toward eastern and southern Africa. We also found that genetic adaptation of Bantu speakers was facilitated by admixture with local populations, particularly for the HLA and LCT loci. Finally, we identified a major contribution of western central African Bantu speakers to the ancestry of African Americans, whose genomes present no strong signals of natural selection. Together, these results highlight the contribution of Bantu-speaking peoples to the complex genetic history of Africans and African Americans.

Linguistic and archaeological records indicate that Bantu languages, together with agriculture, expanded ~4000 to 5000 years ago from western central Africa to eastern and southern Africa (1). Population genetics studies have informed us about the genetic structure of African populations and demonstrated that the expansion of Bantu languages was accompanied by a diffusion of people (25). However, most genomic studies have focused on comparisons between farming and hunter-gathering populations (6, 7) rather than on patterns of diversity among Bantu-speaking populations across the continent. Thus, although Bantu speakers today account for one-third of sub-Saharan Africans, many aspects of their genetic history remain unknown.

One debated question concerns the routes followed by Bantu speakers during their dispersal across sub-Saharan Africa, owing to the poor population coverage in the Bantu heartland (i.e., the Nigeria/Cameroon frontier) or the limited genetic resolution of previous studies (25). Furthermore, documentation of how Bantu speakers adapted to the new environments they encountered—from the grasslands of Cameroon to the African rainforest, the East African plateau, and the Kalahari desert—is unknown. Their rapid adaptation may have been facilitated by the acquisition, via admixture, of adaptive alleles from local populations; the impact of this process on recent human evolution remains largely unexplored (7, 8). Finally, large-scale movements of Bantu peoples have not been limited to Africa, as historical records indicate that people from western central Africa were massively deported to North America during the transatlantic slave trade (9).

We dissected the genetic and adaptive history of Bantu-speaking populations (BSPs, which refers here to traditional farming groups and does not include Bantu-speaking rainforest hunter-gatherers) by generating genome-wide single-nucleotide polymorphism (SNP) data for 1318 individuals from 35 linguistically and anthropologically well-defined populations of western and western central Africa, including the Bantu homeland (Fig. 1A and table S1). After quality control (fig. S1) (10), we combined these data with data sets for other BSPs and non-BSPs from sub-Saharan Africa (table S2). We obtained a total of 548,055 high-quality SNPs in 2055 individuals from 57 populations.

Fig. 1 Genetic structure of African populations.

(A) Geographic locations of sampled populations. The inset shows the homeland of Bantu expansions. wRHG and eRHG correspond to western and eastern rainforest hunter-gatherers, respectively. (B) Clustering analysis was performed on 2055 individuals and 406,798 independent SNPs with ADMIXTURE (11). Results for varying numbers of postulated ancestral populations (K) are shown in fig. S2. (C) Haplotype-based PCA of wide-Bantu-speaking and narrow-Bantu-speaking populations from western central Africa, on 1015 individuals and 429,972 SNPs, with the software fineSTRUCTURE (13). The proportions of variance explained, expectedly larger than for unlinked SNP data, are shown in brackets.

Genetic cluster analyses (11) showed that BSPs from western central (wBSP), eastern (eBSP), southwestern (swBSP), and southeastern (seBSP) Africa clustered together (Fig. 1B and figs. S2 to S4), echoing their modest levels of genetic differentiation (analysis of molecular variance–based FST < 0.01) (table S3). This relative homogeneity reflects the recent separation of BSPs after their expansions throughout sub-Saharan Africa (5, 12). Furthermore, wBSPs, eBSPs, and seBSPs displayed moderate proportions of ancestry from western rainforest hunter-gatherers (~16%), Afroasiatic-speaking farmers (~17%), and San hunter-gatherers (~23%), respectively, suggesting admixture with local populations.

Two hypotheses have been proposed concerning the dispersal of Bantu-speaking populations across sub-Saharan Africa (24). According to the “early-split” hypothesis, the western and eastern branches split early, within the Bantu heartland, into separate migration routes. By contrast, the “late-split” model suggests an initial spread southward from the Bantu homeland into the equatorial rainforest (i.e., Gabon/Angola), followed by expansions toward the rest of the subcontinent. We tested these hypotheses by determining whether eBSPs and seBSPs were genetically closer to wBSPs from the southern part, relative to wBSPs from the northern part, of western central Africa. The populations from this core region can be distinguished along the first axis of the haplotype-based principal component analysis (PCA) (Fig. 1C and figs. S5 to S7) (13), mirroring genetic isolation due to both geography and linguistic barriers (fig. S8) (10). We overcame problems due to the levels of non-BSP ancestry detected in eBSPs and seBSPs (Fig. 1B) by using haplotype-based admixture inference with GLOBETROTTER (14) to account for potential admixture.

The GLOBETROTTER method estimated that eBSPs resulted from two consecutive admixture events (P < 0.05) occurring 1000 to 1500 years ago and 150 to 400 years ago between a wBSP (~75% contribution) and an Afroasiatic-speaking population from Ethiopia (~10% contribution) (table S4). For both events, the best-matching parental wBSP was located in Angola and support for a northern central African origin was weak (Fig. 2A and figs. S9 and S10). In southern Africa, seBSPs displayed signals of a unique admixture event (P < 0.01) occurring ~700 years ago between a parental BSP (~70% contribution) and the Ju/’hoansi San from Namibia (~23% contribution). The best parental BSP was located in Angola, with some contribution from eBSPs (Fig. 2B and figs. S9 and S10). Furthermore, eastern and southeastern Bantu speakers shared more identical-by-descent segments with Angolans, relative to northern wBSPs (Mann-Whitney test; P < 10−16) (table S5). Although additional sampling of African populations may further refine these patterns, our results, together with previous genetic data supporting the late-split model (2, 3), indicate that BSPs first moved southward through the rainforest before migrating toward eastern and southern Africa, where they admixed with local populations. This model is further supported by linguistics (15) and archaeoclimate data (16), suggesting that a climatic crisis ~2500 years ago fragmented the rainforest into patches and facilitated the early movements of BSPs farther southward from their original homeland.

Fig. 2 Reconstructing the dispersal of Bantu-speaking populations.

Haplotype-based inference of the genetic origins of (A) eBSPs and (B) seBSPs. The names of the tested admixed populations are shown in italics. Circle sizes are proportional to the relative genetic contribution of parental populations to admixed populations. Only the oldest admixture event in eBSPs is represented; the most recent admixture event and other examples are shown in fig. S9.

As they dispersed through the rainforest, Bantu speakers encountered local populations of rainforest hunter-gatherers (RHGs). We found that the RHG ancestry detected in wBSPs (Fig. 1B and figs. S2 and S5) resulted from an admixture event occurring ~800 years ago, using admixture linkage disequilibrium decay with ALDER (P < 10−8) (table S6) (17) and GLOBETROTTER (P < 0.01) (table S4) (14). These results, together with the low western RHG ancestry detected among BSPs from eastern and southeastern Africa (<5%), indicate that admixture between wBSPs and RHGs occurred mostly after BSPs had expanded throughout sub-Saharan Africa.

The adaptive history of farming BSPs, which were rapidly exposed and had to adapt to new ecosystems, remains largely unknown. We scanned their genomes for signatures of strong, recent positive selection—i.e., regions showing a high proportion of SNPs presenting both greater extended haplotype homozygosity and population differentiation, relative to a closely related reference population (10). We detected eight, five, and seven genomic regions presenting strong signatures of recent positive selection in wBSPs, eBSPs, and seBSPs, respectively (tables S7 to S9) (10).

The HLA locus, which mediates immune response, presented the genome-wide highest proportion of selection signals in both wBSPs and eBSPs (50.5 and 62.4%, respectively) (Fig. 3, A and B, and tables S7 and S8). The most prominent peaks for individual SNP scores were observed in the vicinity of HLA-D genes [rs3129302, empirical P (Pemp) = 2.9 × 10−5 and rs6907291, Pemp = 6.9 × 10−5, respectively] (Fig. 3D and figs. S11 to S14). In wBSPs, the second-strongest hit encompassed CD36 (Fig. 3A; figs. S11 and S12; and table S7), associated with susceptibility to Plasmodium falciparum malaria (18). The putatively selected SNP in CD36 was observed at 25% frequency in wBSPs, yet was essentially absent from non-BSPs from western Africa (rs3211881, Pemp = 5.8 × 10−6) (fig. S11F). Adaptive evolution has been demonstrated for a different, unlinked variant at CD36 in the western African Yoruba of Nigeria (rs3211938) (19), suggesting convergent adaptation.

Fig. 3 Genomic signatures of recent positive selection.

(A to C) Genomic signatures of recent positive selection in (A) wBSPs, (B) eBSPs, and (C) seBSPs. Blue points, and their sizes, indicate the proportion, in 100-SNP windows, of SNPs showing outlier neutrality statistics (10). (D to F) Local selection signatures for (D) the HLA region in wBSPs, (E) the LCT region in eBSPs, and (F) the GPR156 region in seBSPs. Blue points indicate selection scores for individual SNPs (10). The blue line indicates the proportion, in 100-SNP windows, of SNPs showing outlier neutrality statistics. Other candidate loci are shown in figs. S11, S13, and S15. [(A) to (F)] The green, pink, and gold solid lines indicate the local ancestry in BSPs from western RHG, Eastern African, and San populations, respectively.

In eBSPs, the next-strongest selection signal overlapped the LCT gene region, which encodes the lactase enzyme (28.7%) (Fig. 3, B and E; figs. S13 and S14; and table S8). The derived allele of the best candidate SNP at this locus (rs4954204, Pemp = 5.7 × 10−6) displayed high levels of both haplotype homozygosity and genetic differentiation and was linked to the lactase persistence allele C-14010 (20). In seBSPs, the proportions of selection signals were lower (<24%) (Fig. 3, C and F; fig. S15; and table S9), possibly reflecting a different demographic and adaptive history.

We scanned the genomes of BSPs for the presence of regions with unusually high levels of non-BSP ancestry (10). Again, the HLA region in wBSPs showed a strong excess of ancestry from rainforest hunter-gatherers, at 38%, 6.74 SD higher than the genome-wide average of 16% (Fig. 3A). Similar results were obtained when excluding the classical HLA region and restricting the analysis to data from a single SNP array (fig. S11, A and B), indicating that our findings are unlikely to result from the incorrect modeling of the complex HLA haplotype structure or misalignments of alleles between SNP arrays. Simulations under realistic demographic models showed that drift or continuous gene flow from RHGs could not account for the high frequency of introgressed HLA variants in wBSPs (P < 0.0001) (fig. S16 and table S10). Given that these introgressed variants are independent from those presenting the strongest selection signals (10), our results indicate that the HLA locus has been a hotspot of recent adaptation in BSPs.

We found a local excess of eastern African ancestry in the LCT region of eBSPs, and the introgressed variants were those that also showed the strongest positive selection scores of the region (Fig. 3, B and E) (10). Simulations indicated that the high frequency of these variants in eBSPs (up to 30% in the Bakiga eBSP and <1% in wBSPs) (fig. S13D and table S8) could not be explained by strong drift or continuous gene flow from eastern Africans (P < 0.0001) (fig. S17 and table S10). These observations support a model in which eBSPs acquired the lactase persistence trait from eastern Africans (20) and illustrate that the rapid adaptation of human populations migrating to new environments can be facilitated by admixture with local populations.

Last, we estimated the genetic contribution of Bantu-speaking populations to African Americans by analyzing the African ancestry of 5244 African Americans from various locations in North America (table S2). Consistent with previous analyses (5, 2123), the program ADMIXTURE estimated that African Americans had 73% and 78% African ancestry in the northern and southern United States, respectively (fig. S18 and table S11). GLOBETROTTER partitioned their African ancestry into different contributions: 13% from Senegambia, 7% from the Windward Coast, 50% from the Bight of Benin, and up to 30% from western central Africa, mostly from Angola (Fig. 4 and table S11). The estimated contribution of BSPs from western central Africa is consistent with historical records reporting that 23% of slaves transported to North America between 1619 and 1860 originated from this region (9). Furthermore, ADMIXTURE estimated that western RHG ancestry accounted for ~4.8% of the African ancestry of African Americans (Fig. 4 and fig. S19). Given that a direct RHG contribution to the slave trade is unlikely (table S12) (10), this result further supports that a large fraction of the genome of African Americans derives from wBSPs, who themselves have ~16% western RHG ancestry (Fig. 4). Our results indicate that the ultimate African origins of African Americans are more diverse than previously suggested (5, 21, 23).

Fig. 4 Dissecting the African origins of African Americans.

Estimated genetic contribution, indicated by blue circles, of diverse African populations to African Americans of North America (table S11). African populations were chosen to represent the historical ports from which slaves were embarked during the transatlantic slave trade (9). Green bars indicate the western RHG ancestry of African populations and of the African genome of African Americans (fig. S19).

Relaxed selective pressure at the malaria-associated HBB and CD36 genes has been suggested in African Americans, based on large allele frequency differences between African Americans and their assumed, unique African parental population, the non-BSP Yoruba from western Africa (24). We replicated this result for CD36 when considering western Africans only (rs3211938; χ2-test P = 2.7 × 10−10) (fig. S20A), but it was entirely lost when a more diverse, and realistic, set of African parental sources was used (χ2-test P = 0.42) (fig. S20B). Thus, the CD36 signal (24) is due to the use of the Yoruba as the sole source of African ancestry in African Americans. Furthermore, our analyses did not detect any excess of African ancestry in African American genomes (25), using either set of parental populations (fig. S21) (10), collectively suggesting that no major changes in selective pressure have occurred in the history of African Americans.

Our study reconstructs the genetic history of Bantu-speaking farming communities, from their initial expansions within Africa to the most recent forced migrations of a subset of these populations to North America. Additional large-scale resequencing studies of geographically and linguistically diverse populations from Africa are needed to provide insight into the evolutionary forces acting on genome diversity at a fine geographic and temporal scale, ultimately facilitating the unbiased identification of variants contributing to diseases in the Southern Hemisphere.

Supplementary Materials

Materials and Methods

Figs. S1 to S22

Tables S1 to S12

References (2654)

References and Notes

  1. See the supplementary materials.
  2. Acknowledgments: We thank all participants who donated samples and participated in this study. We thank C. Schlebusch and G. Hellenthal for helpful discussions. We thank E. Soumonni, a historian whose advice guided the recruitment of Beninese individuals, and J.-P. Chippaux (CERPAGE, Cotonou, Benin) for his help with local authorities. We thank the African Variation Genome Project, the Data Access Committee Chair for the National Human Genome Research Institute (particularly V. Ota Wang), the Electronic Medical Records and Genomics (eMERGE) Genome-Wide Association Study, the Multiethnic Cohort Study, the Gene, Environment Association Studies consortium (GENEVA), and the Health Aging and Body Composition (Health ABC) Study for kindly providing access to their data. Detailed acknowledgments can be found elsewhere (10). This work was funded by the Institut Pasteur, the Centre National de la Recherche Scientifique (CNRS), Agence Nationale de la Recherche (ANR) grant AGRHUM (ANR-14-CE02-0003-01), and the “Histoire du Génome des Populations Humaines Gabonaises” project (Institut Pasteur/Republic of Gabon). The newly generated SNP genotype data have been deposited in the European Genome-Phenome Archive under accession code EGAS00001002078.
View Abstract


Navigate This Article