Genomic ancestry of North Africans supports back-to-Africa migrations

Abstract

North African populations are distinct from sub-Saharan Africans based on cultural, linguistic, and phenotypic attributes; however, the time and the extent of genetic divergence between populations north and south of the Sahara remain poorly understood. Here, we interrogate the multilayered history of North Africa by characterizing the effect of hypothesized migrations from the Near East, Europe, and sub-Saharan Africa on current genetic diversity. We present dense, genome-wide SNP genotyping array data (730,000 sites) from seven North African populations, spanning from Egypt to Morocco, and one Spanish population. We identify a gradient of likely autochthonous Maghrebi ancestry that increases from east to west across northern Africa; this ancestry is likely derived from "back-to-Africa" gene flow more than 12,000 years ago (ya), prior to the Holocene. The indigenous North African ancestry is more frequent in populations with historical Berber ethnicity. In most North African populations we also see substantial shared ancestry with the Near East, and to a lesser extent sub-Saharan Africa and Europe. To estimate the time of migration from sub-Saharan populations into North Africa, we implement a maximum likelihood dating method based on the distribution of migrant tracts. In order to first identify migrant tracts, we assign local ancestry to haplotypes using a novel, principal component-based analysis of three ancestral populations. We estimate that a migration of western African origin into Morocco began about 40 generations ago (approximately 1,200 ya); a migration of individuals with Nilotic ancestry into Egypt occurred about 25 generations ago (approximately 750 ya). Our genomic data reveal an extraordinarily complex history of migrations, involving at least five ancestral populations, into North Africa.

Figure 1. Map of samples and population structure of North Africa and neighboring populations.

A) On the map, star symbols indicate the location of new population samples and circles indicate previously published samples. The decreasing proportions of Maghrebi ancestry is indicated in a west-to-east gradient of blue across North Africa. An unsupervised clustering algorithm, ADMIXTURE was used to analyze population structure among 13 African, 2 European and 1 Near Eastern populations based on approximately 300 K autosomal SNP loci in common. The two main gradients of ancestry in North Africa (at k = 8), Maghrebi and Near Eastern, are emphasized with arrows. Other population colors in open circles match the colors displayed in the population structure analysis. B) We plot the full dataset assuming k = 2,4,6,8 ancestral populations. k = 10 and log likelihoods are presented in Figure S1.

Figure 2. Multidimensional scaling components discriminating genetic clusters in Africa.

We used multidimensional scaling (MDS) to discriminate clusters of genetic variation within Africa and neighboring regions. MDS was applied to the pairwise, individual identity-by-state (IBS) matrix of 279,500 SNPs using PLINK 1.07 software . The top three MDS components were plotted together using R 2.11.1. A) In axis one is plotted dimension 1 and in axis two is plotted dimension 2. B) In axis one is plotted dimension 1 and in axis two is plotted dimension 3. Population colors match Figure 1B (k = 8), with the exception that the Fulani group was highlighted as a distinct population, indicated in light green. North African populations are all indicated in turquoise except for Tunisians that are shown in dark blue.

Figure 3. Divergence time estimates among European, Maghrebi, and Near Eastern ancestral populations.

We use population-based F_st to estimate divergence time between each of the North African populations and the Qatari (green dots) and Tuscans (purple dots), respectively. We assume point estimates of effective population sizes based on autosomal haplotype heterozygosity estimates from Li et al. (Material and Methods). We further estimate ancestral population clusters assigned at k = 5–8 with ADMIXTURE. Assuming that the ancestral clustering procedure has removed, or at least mitigated, the effect of recent migrations into Mediterranean populations, we then use F_st to estimate divergence times between these ancestral clusters. The range of k-based estimates for Maghrebi versus Near Eastern ancestry is indicated with light green polygon. The range of k-based estimates for Magrebi versus European ancestry is indicated with light purple polygon.

Figure 4. Distribution of long segments that are identical by descent (IBD) for pairs of individuals.

The cumulative amount of DNA shared IBD between each pair of individuals within a population was summed. We display counts of the number of pairs sharing cumulative IBD in A) Tunisian Berbers, B) Saharawi and B) North Moroccans.

Figure 5. Correlation between ancestry proportions inferred from ADMIXTURE and PCADMIX.

We compare the proportions of ancestry inferred from assuming 3 ancestral populations in individuals from South Morocco A) using a clustering algorithm set to k = 8 (Figure 1) summing the sub-Saharan ancestry, both Qatar and European ancestry, and Maghrebi ancestry. We compared these estimates (left bar) to our PCA-based local ancestry assignment estimates (right bar). The three ancestral populations were Saharawi [SAH], Bantu-speaking Luhya [LWK], and Spanish Basque [BAS]. B) Genome admixture deconvolution on chromosome 1 of sixteen South Moroccans. Using a principal component-based method of admixture deconvolution, we assign local ancestry to South Moroccan individuals. We implement our PCA-based method for k = 3, and choose the ancestral populations based on the three ancestral populations were Saharawi [SAH], Bantu-speaking Luhya [LWK], and Spanish Basque [BAS]. Chromosome 1 for all sixteen South Moroccans is presented for both the maternal and paternal haplotypes.

Figure 6. Genome admixture deconvolution karyogram of an Egyptian.

A single Egyptian individual is presented for ancestry assuming k = 4 source populations: Saharawi [SAH], Nilotic-speaking Maasai [MKK], Spanish Basque [BAS] and Arabic Qatari [QAT]. Maasai segments (which were inferred from k = 3 and were highly diverged from the SAH, QAT, BAS segments) are layered on top of the inferred Maghrebi/Qatari/Basque ancestral karyogram, for k = 4 putative source populations.