Genome-wide patterns of population structure and admixture in West Africans and African Americans

Genetic Structure of West African Populations.

Our study focused on West African populations, because previous genetic and historical studies suggest that region was the source for most of the ancestry of present-day African Americans (2, 27, 28). Among the sampled West African populations, Wright’s measure of population differentiation [autosomal F_ST (29)] was low (1.2%), suggesting quite recent common ancestry of all individuals in our sample or, alternatively, a large effective population size for the structured population from which the sample was drawn, with a large degree of gene flow among subpopulations. Nonetheless, we observed substantial variation in pairwise F_ST among sampled populations, suggesting genetic heterogeneity among the groups (Table 1). Differences in pairwise F_ST may reflect variation in effective population size or migration rates among the populations potentially attributable to isolation by distance or heterogeneity in geographical or cultural barriers to gene flow. For example, the Fulani appear to be genetically distinct from all other West African populations we sampled (average pairwise F_ST = 3.91%). Likewise, we found that the Bulala, Xhosa, and Mada populations consistently exhibited pairwise F_ST above 1% when compared with any other population, whereas the non-Bantu Niger-Kordofanian populations of the Igbo, Brong, and Yoruba exhibited little genetic differentiation from one another (average F_ST <0.4%). These results suggest that there are clear and discernible genetic differences among some of the West African populations, whereas others appear to be nearly indistinguishable even when comparing over 300,000 genetic markers.

To investigate whether we could reliably distinguish ancestry among individuals from these populations, we used two approaches tailored for high-density genotype data. One, FRAPPE, implements a maximum likelihood method to infer genetic ancestry of each individual, wherein the individuals are assumed to have originated from K ancestral clusters (26). Fig. 1A and Fig. S2 summarize FRAPPE results when the number of clusters, K, is varied from K = 2 to K = 7. The small number of clusters was consistent with the small overall level of population differentiation among these populations. We next undertook PCA of the matrix of individual genotype values (i.e., the matrix with entries “0,” “1,” or “2” generated by tallying the number of copies of a given allele across all SNPs in a panel for all individuals genotyped) (30).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Population structure within West Africa and relation to language and geography. (A) FRAPPE analysis of the West African populations. Individuals are represented as thin vertical lines partitioned into segments corresponding to the inferred membership in K = 2 through K = 5 genetic clusters as indicated by the colors (see Figs. S2–S5 for additional results). (B) Principal components 1 and 2 of the African individuals. (C) Principal components 1 and 2 of the African individuals, excluding the Fulani population, wherein the components have been rotated to emphasize further similarity with geography. (D) Approximate locations of sampled populations in Africa. (E and F) FRAPPE clustering of Europeans, African Americans, and West Africans. Individuals are represented as thin vertical lines partitioned into K segments corresponding to the inferred membership of the genetic clusters indicated by the colors. Values for K = 2 (E) and K = 4 (F) are shown for comparison between the two analyses.

Patterns of population structure were consistent between the two approaches (Figs. S2–S5). For example, in the FRAPPE analysis, the Fulani population was distinguished at K = 2, with Bulala, Mada, and Kaba populations showing some genetic similarity with the Fulani. PCA, likewise, separated the Fulani from other populations along the first principal component (PC1) (Fig. 1B). The two subsequent principal components, PC2 and PC3, reflect the geographical distribution of the populations. PC2 showed a Chadic and Nilo-Saharan dimension extending into inland Africa from the coast, distinguishing the Bulala, Mada, and Kaba populations. These populations belong to the Nilo-Saharan and Afro-Asiatic (Chadic) linguistic groups and live further inland. Analysis of the African populations, excluding the Fulani, gave a PC1 and PC2 that resemble the second and third principal components of the PCA with the Fulani (Fig. 1C). Rotating the PC1 and PC2 axes from the PCA without the Fulani reveals the similarity of the genetic and geographical maps (Fig. 1 C and D).

At K = 3, the FRAPPE algorithm clusters the Bulala into their own group and suggests genetic similarity of the Mada, Kaba, and Hausa, potentially indicating differentiation of Nilo-Saharan- and Afro-Asiatic-speaking populations from Niger-Kordofanian-speaking populations. At K = 4, all individuals from the Bantu-speaking Xhosa of South Africa cluster into a single group and individuals from the Bantu-speaking populations (Fang, Bamoun, and Kongo) exhibit considerable shared membership in this cluster. At K = 5, the Mada are distinguishable as a unique group, with modest genetic similarity with the Hausa and Kaba as well as with most of the Niger-Kordofanian populations. These results suggest that although these populations are quite closely related genetically, it is possible to detect meaningful population substructure given sufficient marker density [see also ref. (2)]. It is important to note that there is likely further substructure and diversity within these populations. Because we sample a modest number of individuals from each population (n = 13, on average, per population), we are not likely to have captured all the genetic variation within each population, region, or linguistic family. To compare patterns of haplotype structure and discern differences in demographic history among the African populations, we estimated linkage disequilibrium (LD) between all pairs of markers in the data for all populations (see SI Text and Fig. S6). All the African populations showed low levels of LD (even at closely linked sites) and a rapid decay of LD with distance genome-wide relative to populations of European ancestry.

Genome-Wide Patterns of Admixture in African Americans.

To understand the genetic structure of the African-American population better and to determine African-American ancestry, we used FRAPPE to evaluate African Americans together with European and African individuals genotyped on the same marker set. At K = 2, African populations (blue) were distinguished from European populations (red), with African Americans showing highly variable levels of European and West African ancestry (Fig. 1 E and F). For the African Americans, estimated mean West African ancestry was 77%, consistent with prior studies (2, 28, 31–34). Analysis at K = 4 revealed additional substructure in a North-South cline within Europe and clusters coinciding with the linguistic and geographical substructure within Africa (see SI Text, Tables S2–S4, and Figs. S7 and S8 for additional FRAPPE and population genetic analyses). PCA of the genotype value matrix of the European, West African, and African-American samples revealed the primary axis of variation (PC1) to correspond with “European” vs. “West African” ancestry (see Fig. 2A) and explained ∼ 9.8% of the genetic variance. Specifically, we observed two centroids in the data, with all the individuals of European ancestry exhibiting negative loadings along PC1, whereas all the West African individuals exhibited positive loadings. African Americans exhibited a wide range of loadings along PC1, presumably attributable to differences in European vs. West African ancestry. PC2 corresponds to population substructure within West Africa and largely mirrors the patterns discussed above.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Results of our PCA-based ancestry estimation method. (A) Graphical illustration of approach: Euclidean distances from a given individual's coordinates in PCA space (i.e., “loadings”) and the West African centroid (“a”) and the European centroid (“b”) along PC1 for PCA space that includes Europeans, African Americans, and West Africans. (B) Local ancestry estimation using the PCA sliding window approach and associated HMM for number of chromosomes for a given individual (i.e., “0,” “1,” or “2”) with African ancestry. (C–F) Individual ancestry estimates of 4 representative African-American individuals (denoted 1, 2, 3, and 4 in Fig. 2A) in our dataset of 365 individuals. The colors represent two chromosomes of West African ancestry (blue), two chromosomes of European ancestry (red), or one chromosome of West African and one chromosome of European ancestry (green). (G) Mean ancestry of 365 African-American individuals at each window across chromosome (chrom) 1, chrom 11, chrom 12, and chromosome X (X Chr). The black line shows the overall mean estimated ancestry. Red bands indicate +3 and −3 SDs from the mean ancestry. (All chromosomes are reported in Fig. S10).

Estimation of Admixture in Local Genomic Regions.

We reconstructed estimated European or West African ancestry for every African American in our dataset at every position in the genome using a PCA-based algorithm (Fig. 2A). Our method is a generalization of the approach of Paschou et al. (35) and estimates genome-wide proportion of West African ancestry for a given individual as P = b/(a + b), where b and a are the chord distances from the European and West African centroids, respectively, for the given individual along PC1. Our generalization involves undertaking the PC1 distance analysis on a grid of points along the genome (as opposed to genome-wide) centered on 15 SNP windows and using a Hidden Markov Model (HMM) for inference of ancestry state (i.e., having “0,” “1,” or “2” chromosomes of recent African origin; see SI Text, Fig. 2B, and Fig. S9). An ancestry plot summarizing the number of segments of European (i.e., “0”), West African (i.e., “2”), or admixed (i.e., “1”) ancestry for a representative African-American individual with 73.5% West African ancestry is illustrated in Fig. 2C. There is a great deal of variation among the ancestry plots of the 365 self-identified African Americans in the study, ranging from an estimate of over 99% West African ancestry to an estimate of less than 1% West African ancestry (Fig. 2F). Some patterns reflected a high level of West African ancestry and only one or two ancestry-switching events per chromosome, suggesting very recent direct African ancestry (Fig. 2D). Other patterns reflected only European and admixed ancestry throughout the genome, suggesting one parent of European ancestry and one parent of African-American ancestry (Fig. 2E).

An interesting question one can address with these kinds of data is whether regions of the genome show substantially high European or West African ancestry across all individuals in the sample [e.g., as may be the case if a particular allele from one of the ancestral populations was under strong selection (36–39)]. For our analysis, we considered genomic regions as potential candidates for increased European or West African ancestry if the mean ancestry for the region across the 365 African-American individuals was 3 SDs above or below the genome-wide average of West African ancestry (78.1%). Using this approach, we found that several genomic regions of autosomal chromosomes 5, 6, and 11 could be considered outliers from the genome-wide distribution of ancestry, although these differences were not significant after correction for multiple tests. In Fig. 2G, we show mean ancestry across two example chromosomes that do not show any outlier regions (chromosomes 1 and 12) and one chromosome showing a region falling outside the 3 SD criteria (chromosome 11). Mean ancestry estimates for all chromosomes can be found in Fig. S10, and a precise listing of molecular regions for the three outlier regions may be found in Table S4. In contrast to the autosomes, the X chromosome shows significantly high West African ancestry along the majority of the chromosome, consistent with a gender-biased model of admixture with excess European male and West African female ancestry (Fig. 2G).