The Archaeological and Genetic Correlates of Amazigh Linguistics

Introduction The census size of Mediterranean North Africa exceeds 160 million people [1], but relatively little is known about the genetic makeup of these populations and the demographic history of migration between North Africa and neighboring regions. Mediterranean North Africans are often grouped with Near Eastern populations because populations in both regions speak primarily Afro-Asiatic languages, like Arabic, and phenotypic attributes, like lighter skin pigmentation, differentiate many North Africans from sub-Saharan Africans. Recently, geneticists have attempted to replicate disease associations identified in Europeans and Near Eastern groups with North African populations, reflecting a hypothesis of shared genetic ancestry, with mixed results [2]–[5]. In this paper, we present analysis of autosomal single nucleotide polymorphism (SNP) array data for seven North African populations (see Materials and Methods), distributed along an east-to-west transect across the continent. We clarify the population structure of North Africa and explicitly interrogate the history of gene flow into North Africa from the Near East, Europe and sub-Saharan Africa. Prior genetic studies, largely from uniparentally inherited markers, have not resolved the location origin of North African populations or the timing of human dispersal(s) into North Africa. Analyses based on the frequencies of a small number of autosomal genetic polymorphisms and uniparental markers have shown that the genetic landscape follow an east-west pattern with little to no difference between Berber- and Arab-speaking populations [6], [7]. Mitochondrial data, for example, indicate an early back-to-Africa migration [8]–[10], but Y-chromosome markers largely support a Neolithic expansion and historic period gene flow throughout the Mediterranean [11] (though see [12]). Do current North Africans retain genetic continuity with the first modern human occupants of northern Africa from more than 50,000 years ago (ya) or was northern Africa primarily repopulated during the Holocene by herding and farming populations from elsewhere? Evidence of Neolithic migration from the Near East is supported by the introduction of domestic animals like cows, sheep and goats to North Africa. But the indigenous development of ceramics in Saharan Africa by 9,000 ya is also suggestive of an incipient form of agriculture or pastoralism, prior to any demic diffusion from the Near East [13]. Less controversial is the observation that many North African populations now speak Arabic and that this language shift occurred primarily after the Arabic conquest 1,400 ya. This Arabic shift is well documented, but it remains unknown how deeply recent migrations ( 0.8. To reduce biases due to our lower sensitivity to short tracts, we only modeled tracts longer than 3 cM, and considered assigned tracts with posterior probability >0.8. With a 3 cM cutoff we expect to capture 50% of tracts from 55 generations ago and 10% of tracts from 130 generations ago (see Materials and Methods). Unassigned short tracts (i.e. the “undecided” regions, Figure 5B and Figure 6) within a long continuous migrant segment can be artificially shortened by spikes of low posterior probability. Unassigned tracts that were situated within a tract of one ancestry and which maintained a posterior probability >0.5 for the same neighboring ancestry were considered to as one long ancestry tract. 10.1371/journal.pgen.1002397.g006 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. We focus on the sub-Saharan African migrant tracts in South Moroccans (Figure 5B) and Egyptians (Figure 6). These tracts tend to be highly diverged from other ancestries in the population (Fst>0.10) and populations with similar divergence resulted in accurate haplotype assignment in prior testing [28]. Under a model of constant migration from the Bantu-speaking Kenyans and southern Moroccans started about 41 generations ago (ga) (95% CI: 39–44ga) assuming there was no migration occurring prior to this period. The confidence interval calculations, obtained by resampling sub-Saharan migrant tracts with replacement, do not take into account possible biases caused, for example, by the model assumption of a fixed migration rate. Constant Versus Episodic Migration We hypothesized that the distribution of sub-Saharan African tracts in the Moroccans and Egyptians might better reflect a single episode or “pulse” of migration. In order to test this hypothesis, we modify Pool and Nielsen's [30] approach to conform to a pulse model (see Materials and Methods). We compared the log likelihoods summed over all migrant tracts under constant and pulse migration models for each population maximized over the relevant parameters, and present the model with the higher log likelihood (Table 1). Estimates of the time of migration are more recent under a pulse model. The younger estimate occurs because the model fit must account for relatively long migrant tracts in the data; under a constant migration model these tracts represent recent migrants, but for a single episode of migration, long tracts can only be accounted for by recent migration of the entire sample. In order for the average migrant tract length to be equal in the two models, migration must have started more than twice as long ago in the constant migration model compared to the pulse model (Table 1). Our Egyptian sample of Nilotic segments (derived from Maasai) has a better log likelihood under a pulse migration model, estimated as time since admixture of 24ga (95% CI: 23–26ga) rather than 51ga under a constant migration model (Table 1). 10.1371/journal.pgen.1002397.t001 Table 1 Models of migration into North Africa. Admixed Population Migrant Population1 Migration Model2 Log Likelihood Time of migration (G)3 Bootstrap 95% CI4 Egyptians Maasai (Kenya) Continuous 2682 51 47.7–55.8 Pulse 2705 24 22.8–25.8 South Moroccans Luhya (Kenya) Continuous 5386 41 39.1–44.0 Pulse 5365 19 18.1–19.7 1 Segments from the migrant population were required to be greater than 3 cM in length. Given this minimum threshold, 815 migrant Egyptian and 1,275 migrant South Moroccan segments were discarded. 2 Two migration models were tested: a “pulse” model assumes a single episode of migration occurred at T0 followed by no further migration, a “continuous” model assumes constant migration from T0 to the present day. Log likelihoods given either model were compared and we present the model with the highest log likelihood. 3 The maximum likelihood estimate of time since migration initially began “T0” from the migrant population into the admixed: population. We assume prior migration between the populations was zero. Time since migration began T0 is indicated in generations. 4 The 95% confidence interval was estimated by sampling migrant Egyptian segments (n = 1,246) and migrant South Moroccan segments (n = 2,770) 1,000 times with replacement. Discussion Out of Africa and Back Again? By sampling multiple populations along an approximate transect across North Africa, we were able to identify gradients in ancestry along an east-west axis (Figure 1 and Figure 2). Notably, even northwestern populations with very high proportions of Maghrebi ancestry, such as the Tunisians and Saharawi, still cluster with Out-of-Africa populations in the population structure analyses (Figure 1 (k = 2), Figure 2). This observation of clustering formed the basis for further analyses to distinguish between two alternative demographic models. First, North Africans could be closer to OOA populations due to extensive gene flow, likely from the Near East, over the past ∼50 Kya. Second, North Africans could be closer to OOA populations if the two groups had diverged more recently than either had split with sub-Saharan Africans. We can reject a simple model of long-term continuous gene flow between the Near East and North Africa, as evidenced by clear geographic structure and non-zero Fst estimates. Fst estimates between the inferred Maghrebi cluster and sub-Saharan Africans are two to three-times greater than Fst between the Maghrebi and Europeans/Near Easterners ancestral clusters (Table S3). We then address whether this population structure was recent or ancient. Although Fst estimates from ascertained data may be biased, as rare alleles are under-represented in the site frequency spectrum, comparison of African-European Fst from resequencing data and the Affymetrix 500 K platform showed only a negligible difference [31]. Assuming reasonable effective population sizes for North African Maghrebi and neighboring populations [17], we first showed that all North African populations are estimated to have diverged from OOA groups more than 12,000 ya (Figure 3). After accounting for putative recent admixture (Figure 1), the indigenous Maghrebi component (k-based) is estimated to have diverged from Near Eastern/Europeans between 18–38 Kya (Figure 3), under a range of Ne and k values. We hence suggest that the ancestral Maghrebi population separated from Near Eastern/Europeans prior to the Holocene, and that the Maghrebi populations do not represent a large-scale demic diffusion of agropastoralists from the Near East. With model parameters for divergence approximately estimated, we then ask whether North African ancestral populations were part of the initial OOA exit and then returned to Africa [8], or if an in situ model of population persistence for the past 50 Kya is more likely (with variable episodes of migration from the Near East)? We can address this question only indirectly with contemporary samples; however, several auxiliary observations point toward the former hypothesis. Substantially elevated linkage disequilibrium in all of these North African population samples, compared to sub-Saharan populations [32], is consistent with a population bottleneck. Hellenthal et al. [30] also observed that the reduction in the number of haplotype founders required to reconstruct the Mozabite population, as compared to other African populations, could be explained by a population bottleneck. If North African ancestral populations persisted in situ, then we need to invoke two population bottlenecks, one in the ancestors of North Africans (including the Berbers) and one for OOA groups. Alternatively, the “OOA” bottleneck would need to occur in North Africa, rather than when groups moved out of the continent [33]. The second possibility appears at odds with most published models of the movement of modern humans outside of Africa. A scenario where North African Maghrebi ancestry is the result of in situ population absorbing Near Eastern migrants would likely need the following premises to explain the results here and elsewhere: a) an Out-of-Africa migration [concurrent with bottleneck] occurs 50–60 Kya, geographically dividing North African and Near Eastern populations; b) North Africans experience a separate bottleneck; c) gene flow maintains similarity between the two geographically distinct populations; d) the gene flow then ceases or slows roughly between 12–40 Kya in order to allow sufficiently distinct allele frequency distributions to form. In contrast, we find it more parsimonious to describe model where: a) an OOA migration occurs [concurrent with a bottleneck]; b) OOA populations and North Africans diverge between 12–40 Kya when a migration back-to-Africa occurs. These models should be further tested with genomic sequence data, which have better power to detect magnitude and timing of bottlenecks, and to estimate the true joint allele frequency spectrum. More recently, the substantial, east-to-west decline of Near Eastern ancestry (Figure 1A) could represent a defined migration associated with Arab conquest 1,400 ya or merely gene flow occurring gradually among neighboring populations along a North African | Arabian Peninsula transect. Although we observe a declining amount of Maghrebi ancestry from northwest-to-northeast, it is possible that other geographically North African samples (e.g. Egyptians further south than the sampled Siwa Oasis) do not conform to this geographic cline. Finally, we also observe European ancestry that is not clearly accounted for by the inclusion of a Near Eastern sample. Additional migration coming from Europe might be plausible, though the origin and the period where it took place cannot be determined with the present data. The less than 25% European ancestry in populations like Algerians and northern Moroccans could trace back to maritime migrations throughout the Mediterranean [34]. Alternatively, the Qatari could represent a poor proxy for an Arabic source population, causing additional diversity to be assigned European (e.g. European ancestry tracts were not reliably assigned as European with PCADMIX). In summary, although paleoanthropological evidence has established the ancient presence of anatomically modern humans in northern Africa prior to 60,000 ya [35], the simplest interpretation of our results is that the majority of ancestry in modern North Africans derives from populations outside of Africa, through at least two episodes of increased gene flow during the past 40,000 years (Figure 1, Figure 2, Figure 3). Reconstructing Multiple Admixed Ancestries Multiple local ancestry assignment methods, including PCADMIX, require thinning genotype datasets to remove alleles in high linkage disequilibrium between populations [29], [36]; this step discards information regarding haplotype patterns that tend to be more informative than genotypes when using data biased by SNP ascertainment [37]. HAPMIX incorporates both LD information and uncertainty in phase inference for haplotypes [18], but the software is currently limited to a two-population model. Our ancestral proportions of European and sub-Saharan ancestry for many North Africans at k = 2 (Figure 1) are similar to those obtained with HAPMIX by Price et al. [18] for the HGDP Algerian Mozabites, assuming a two-population mixture of northern Europeans and Yoruba. However, our results show that increasing the number of possible ancestral populations reveals multiple, diverse ancestries (e.g. Maghrebi, Near Eastern, Nilotic) and that the proportion of sub-Saharan African assignment decreases as these other ancestries are accounted for. This decrease in assigned sub-Saharan ancestry in North African samples, from a k = 2 model, is consistent with an interpretation that Maghrebi or Near Eastern diversity that is not present in the panel populations is more likely to be assigned to the more diverse, Sub-Saharan African ancestry. Using a two-population admixture model, Price et al. [18] estimated the time of migration from sub-Saharan Africa into the Mozabites to have begun about 100 generations ago (or more). Our results suggest that sub-Saharan African and Maghreb admixture is considerably more recent, 24–41 generations ago (and even the upper 95% CI estimate under either model is 55ga, Table 1). The discrepancy between these two estimates may result from our incorporation of multiple source populations, our use of non-linear models to estimate migration timing and the elimination, in Price et al. [18], of individuals with megabase long African segments. Time of Migration Estimation We use a two-population model of migration where we measure the number and length of migrant tracts observed in the admixed population. However, as argued earlier, North African populations have absorbed migrants from multiple episodes of migration. We use three- and four-population admixture deconvolution to identify the tracts from these separate migrations. One complication with this approach is the possibility that source populations that contribute migrants to North Africa are themselves exchanging migrants. For example, Near Eastern populations expanded into European continent during the Neolithic, and even an isolated population like the Spanish Basque may have genomic segments that trace back to the Neolithic expansion [38], [39]. In this case, estimation of the time of migration of Arabic individuals into North Africa would be biased by Basque segments of Arab ancestry that were contributed by Europeans, but are locally assigned to Arabic ancestry. We confine our migration estimates to those from sub-Saharan populations into North Africans because there has likely been relatively little recent gene flow between sub-Saharan Africans and the European/Near Eastern populations. Moorjani et al. [40] present evidence for recent gene flow ( 0.8 posterior probability) across chromosome 1 for each of 4 ancestries assigned in the Egyptians: Maghrebi (Saharawi), European (Basque), Near Eastern (Qatari), Sub-Saharan (Maasai). (TIF) Click here for additional data file. Figure S10 A) Distribution of the number and length in centimorgans of migrant Sub-Saharan (Luhya) tracts distributed by length found in the South Moroccan population. B) Distribution of the number and length in centimorgans of migrant Sub-Saharan (Maasai) tracts distributed by length found in the Egyptian population. Red bar indicates the minimum threshold cutoff employed in the migration parameter analysis. Please note the different scales along the X-axis. (TIF) Click here for additional data file. Table S1 Name, sample size and country of origin for populations newly genotyped in the present study as well as for populations published previously. References are included in the table. (DOC) Click here for additional data file. Table S2 Additional estimates of Fst after removed putative admixture events. (DOC) Click here for additional data file. Table S3 Significance of the comparisons of ancestry assignment using PCADMIX and ADMIXTURE. (DOC) Click here for additional data file. Text S1 Assigning local ancestry with PCADMIX. (DOC) Click here for additional data file. Text S2 Concordance between ADMIXTURE and PCADMIX. (DOC) Click here for additional data file. Text S3 Chromosome 1 Ancestry Deviations. (DOC) Click here for additional data file.

We have typed 275 men from five populations in Algeria, Tunisia, and Egypt with a set of 119 binary markers and 15 microsatellites from the Y chromosome, and we have analyzed the results together with published data from Moroccan populations. North African Y-chromosomal diversity is geographically structured and fits the pattern expected under an isolation-by-distance model. Autocorrelation analyses reveal an east-west cline of genetic variation that extends into the Middle East and is compatible with a hypothesis of demic expansion. This expansion must have involved relatively small numbers of Y chromosomes to account for the reduction in gene diversity towards the West that accompanied the frequency increase of Y haplogroup E3b2, but gene flow must have been maintained to explain the observed pattern of isolation-by-distance. Since the estimates of the times to the most recent common ancestor (TMRCAs) of the most common haplogroups are quite recent, we suggest that the North African pattern of Y-chromosomal variation is largely of Neolithic origin. Thus, we propose that the Neolithic transition in this part of the world was accompanied by demic diffusion of Afro-Asiatic-speaking pastoralists from the Middle East.