Introduction

Northeast Africa has undeniably been a key region in human evolutionary history. The out-of-Africa migrations need to have passed through, if not originated in the region. East Africa is also home to some of the most important fossil evidence for human evolution from early bipedal species such as Australopithecus afarensis [1], to the emergence of early anatomically modern humans such as the Omo fossils [2, 3].

Numerous overlapping migrations of farmers and herders over the last several thousand years have also played a critical role in reshaping the current socioeconomic and linguistic diversity of the region. It is clear that back migrations into Northeast Africa have had a major impact on the genetic ancestries of the peoples in the region today [4–6]. Ethiopian populations, for instance, harbor a large proportion of “non-African” ancestry, as high as, ∼ 40% in some groups—see for instance the Amhara in Fig 1c in [7]. What is clear is that some current-day Northeast Africans can trace much of their ancestry from other sources than the original hunter-gatherers in the region, such as the Mota individual, an Ethiopian male who lived around 4500 years ago [8, 9]. It is also clear that these back migrations into Africa have been ongoing for a long time period. For North Africa, seven individuals from Morocco that had a high affinity to Middle Eastern populations, dated to 15 000 years ago, suggesting the possibility that similar deep-in-time admixtures might have occurred in other parts of Africa [10].

In 2017, several ancient genomes were sequenced in an attempt to uncover the demographic patterns in African prehistory [11]. The study contained data from 16 ancient African individuals from 8 100—400 BP. They found that ancient East African hunter-gatherers form a cline of ancestry with modern-day southern African hunter-gatherer (San) groups. This indicates that in the past, hunter-gatherers with a gradient of shared ancestry ranged from eastern to southern Africa. The fact that these hunter-gatherer groups existed until the relatively recent past allows for the possibility of interactions between them and later pastoralist and agricultural groups in East Africa. This data was later re-analyzed with several new individuals, particularly from East Africa [5]. They proposed a four-stage model where initially Sudanese Nilotic speakers admixed with groups with Eurasian ancestry (either from Northern Africa or the Levant) within Northeast Africa. In step two, the descendants of these groups migrated to East Africa reaching Lake Turkana by around 5 000—4 000 BP and central Tanzania by around 3000 BP and mixed with local hunter-gatherer groups throughout this process [5]. The first signs of pastoralism in East Africa coincide with these events. Thirdly the second wave of Sudanese-related groups migrated into the area and contributed to the pastoral Iron Age populations. Lastly, West African ancestry (genetically similar to Bantu speakers) appeared alongside the advent of crop farming in the region. These findings were then yet again revised in 2020 [12]. By analyzing 20 additional ancient individuals, additional resolution was given to the picture and several new patterns emerged. Mainly, [12] propose that the pastoralists probably arrived in East Africa in multiple waves from several different locations, or that severe population structure was present (distinguishing between the two was not possible). Both [12] and [5] conclude that there was no single event of hunter-gatherer and herder introgression, neither in space nor in time. Instead a complex “moving frontier” is proposed with diverse patterns of interactions along the contact zones between hunter-gatherer and herder groups.

In the last decade, several genetic studies on modern-day populations have focused on the genetic demographic history of Ethiopia and found patterns of linguistic stratification within Ethiopian populations, i.e. populations within the same language family are more similar to each other than populations belonging to other language families [4, 13, 14]. It is less clear if this pattern holds true in Northeast Africa as a whole, as [15] found a stronger association between geography and genetics than between genetics and linguistic family. By studying modern-day genetic variation, [6] investigated the non-African part of Ethiopian populations and were able to conclude that there has been Eurasian admixture, likely coming from Levantine, rather than Arabian populations. This event was estimated to have occurred around 3 000 years ago.

By leveraging one of the largest datasets of Northeast African populations to date, we aim to add resolution to Eurasian admixture in Northeast African populations. Specifically, we aim to improve the estimation of the best proxies for the origin of Eurasian admixture in modern-day Northeast African populations by using more Northeast African and Middle Eastern, and Eurasian reference populations. In this study, we follow the approach of [4, 6, 16] in that we employ local ancestry methods to identify the Eurasian fragments of East African genomes and extract those segments from the surrounding genomes, a process referred to here as ancestry-deconvolution. We then identify the current-day populations that best match those segments. We also date the events to get a better understanding of historic and prehistoric movements in the region. Using the information of possible sources for admixture and dating of these, we construct a model representative of the population history in the region. Overall we find a complex history of Eurasian admixture in Northeastern Africa, related to the spread of languages, the Muslim conquest, and trade routes along the Red Sea.

Materials and methods

Results

After quality control, removal of related individuals, and down-sampling to a maximum of 30 individuals per population the dataset consisted of 2066 individuals from 101 population groups and 199 422 SNPs. Note that some populations are represented multiple times from different original publications, resulting in a total of 97 unique populations, see S6 Table.

Discussion

In this study, we investigate the patterns of different genetic ancestries in Northeast African populations, focusing on Eurasian back migrations. We inferred population structure using both global and local ancestry methods. Using the local ancestry method MOSAIC we identify regions of Northeast African populations’ genomes with Eurasian ancestry. We also attempt to date this admixture. In our approach, we start from the modern-day groups and try to infer patterns of past interactions by analyzing their genomes. We however acknowledge that Northeast Africa is a region with a complex history spanning thousands of years. The expansions and contractions, rise and fall of states, kingdoms, and empires across the region have had a major impact on the formation, dissolution, and current distributions of the sampled communities in this study. We, therefore, recognize that the groups included in this study are modern-day populations that were created by introgression/interaction/assimilation events in the past and should not be seen as unchanged entities that represent exact past distributions of groups. For example, the interpretation of the dating for Nilotic speakers from East Africa, the Maasai, Turkana, and Samburu is complex since they only relatively recently reached their current-day distributions through expansions from Sudan and Uganda within the last few centuries [43].

Dating the admixture of different groups with each other is of great interest to population geneticists. Having a date for when the mixing of two groups occurred allows us to incorporate other types of independent evidence into our analyses, such as written or oral history or linguistic information, thus a big part of the effort in this study and discussion is spent on our inferred dates of admixture. The “best by f₃” analysis is our attempt to propose a scenario that best fits the previously known genetic history of the region, whilst the “best by R²” analysis, based on the genomic fit (R²), is intended as a less constrained scenario for picking out Eurasian ancestry in Northeast African groups. As shown in Fig 5 the two approaches result in similar dates for most of the populations.

Population structure inferences illustrate the complex genetic history of Northeast African populations. Larger patterns of genetic associations between many of the world’s distinct human lineages are reflected in Northeast African genomes. The hunter-gatherer’s ancestry highlights the deep history of the region and its people and that this ancestry remains within the East African populations. The southern part of the region has a closer genetic affinity to West African groups, a result of the Bantu expansion and several of these populations also speak Bantu languages today. That the Bantu expansion did not continue further into the region could be a result of geographical barriers such as the Ethiopian Highlands and the dry regions of the Horn of Africa, indicated by our FEEMS analysis in Fig 1D or as suggested by [15] that the Northeast African Nilotic speaking herders (such as the Dinka and Nuer), who have remained relatively isolated from other groups, could have formed a buffer against the Bantu expansion continuing further into Northeast Africa.

Eurasian admixture has had a large influence on the genomes of Northeast African groups. The Egyptian and Sudanese Copt populations for instance are genetically very similar to Middle Eastern groups rather than to other African populations. The pattern is true also for the rest of North Africa and present as early as at least 15 000 years ago [10] though not investigated here. The Copts look genetically similar to the Egyptians from Cairo, see Fig 1A and 1C, this is not unsurprising given that the Copts arrived in Sudan around 200 years ago from Egypt and seem to have lived relatively isolated since then [15]. Our admixture date for the Copts (with Eurasians) was inferred to be 27.5 for the f₃ analysis and 25.7 for the R² and around 22 generations for the Egyptians. Thus this admixture took place around the 14th century.

Further south in the region, we continue to see the impact of past Eurasian admixture. Northeast African populations from Sudan and Ethiopia positions’ in the PCA plots are being drawn towards Eurasian populations, S6 Fig. ADMIXTURE analyses recapitulate this pattern where Northeast African groups share the component maximized in Middle Eastern groups (pink component at K = 6, Fig S2 Fig). The Sudanese data in our study is mainly from [15] who also investigated the time and sources of admixture in Sudanese populations. [15] investigated a simpler admixture scenario with only two putative sources, namely the Sudanese Nuer and the TSI (Tuscan) to represent the admixture of a Sudanese basal population with a Eurasian source. This is most similar to our R² approach in which we picked the scenario with the best genomic fit (R²) and for two Eurasian sources in each run and then picked the two Eurasian sources that produced the best genomic fit (R² value). Our findings are generally in agreement, particularly for the Eurasian admixture dates that is the primary focus of our study.

In the area of current-day Sudan and South Sudan, there is a clear divide between the Eastern Sudanic- and Semitic-speaking groups from Sudan, and the South Sudanese groups, as well as the Saharan-speaking Sudanese groups. This divide can be seen both with regards to global ancestry as well as their inferred admixture dates for their Eurasian ancestries. Dongola had been the capital of the Nubian Kingdom and the fall of Dongola in 1317 to Mameluke forces meant the start of Arab and Islamic dominance south of the borders of Egypt. Many of the Semitic speakers in our dataset have their Eurasian admixture dated to this time—around 20 generations ago. The exception is mainly the Southern Semitic speakers such as the Beni-Amer and Tygray whose dates are slightly older at around 30 generations ago. Around 30 generations ago is also the inferred date for the Ethiopian Cuschitic-speaking Afar and Oromo (though Oromo had a generation time of 21 for the best by f₃). South Sudanese groups however stayed largely isolated, this pattern is evident in the ADMIXTURE analysis, as the populations around South Sudan are represented by a specific component (the black component at K = 5 and onward) with very little of the non-African (pink) component that we find in most other North-East African groups, indicating their isolation and genetic homogeneity compared to other populations.

Previous studies that investigated Ethiopian population structure, observed clustering based on linguistic families [4, 14, 15]. This pattern is recapitulated in our analysis, both from the population inference methods as well as the admixture dating. The Omotic-speaking Ari populations form their own small cluster (PC 1 vs PC 3 in S6 Fig), a reflection of their segregated status within Ethiopia [4]. The Gumuz (Language isolate) and Anuak (Nilotic) display very little Eurasian admixture, and given that the date that we infer is only a few generations ago, it could be that they received this Eurasian admixture through secondary admixture with another neighboring group a few generations ago, or that it’s an effect of recent or ongoing admixture.

Within the northeast African geographic space, the analysis using FEEMS recapitulates expected natural barriers to migration such as the Red Sea, the Gulf of Aden, the Persian Gulf, and the Sahara Desert. In addition to clear geographical barriers, we also see evidence of linguistic and cultural barriers. One obvious example is the low migration rate between the Ethiopian Somali and the other Ethiopian populations, and as expected high migration rate is inferred between the different Somali groups. The Great Rift Valley forms a natural barrier across Ethiopia with highlands on both sides of the rift. A previous study looking at Ethiopian genetics found a significant association of genetic similarity to elevation, ethnicity, and first language, and interestingly not a second language nor religion [13].

Along the Red Sea coast of Eritrea and Sudan, we find a region of high gene flow extending into northern Ethiopia and into the Great Rift Valley, Fig 1D. This region corresponds well to the pink component in Fig 1A and 1B which seems to represent Yemeni ancestry. f3 visualizations also indicates higher geneflow from Arabian groups in this area relative to more northern and southern latitudes, Fig 2. It is also a region in which we infer some of the oldest inferred admixture dates. These observations, as well as the shared linguistics of South Semitic (as South Semitic languages that are found in Yemen, Oman, Eritrea, and Ethiopia [44]), indicate a close connection between Eritrea, and Ethiopia to the south of the Arabian peninsula and present-day Yemen. The Kingdom of Aksum (or the Aksumite Empire) encompassing Eastern Sudan, Northern Ethiopia, Eritrea, Djibouti and across the Red Sea into Yemen, thrived between the 1:th and 7:th century AD, as trade along the Red Sea increased and the trade along the Nile decreased. Both Rome and Byzantium traded with the Indian Subcontinent and artifacts from these Kingdoms can be found at Aksumite sites, [45, 46]. The Semitic-speaking Ethiopian populations also group together with the Middle Eastern populations in the UMAP analysis, S20 Fig. These admixture events could come as the result of the Red Sea trade. Aksum collapsed in the 8th century as Islam started to expand and control over the Red Sea trade shifted to the Near East [47].

Previous ancestry deconvolution studies pointed at Levantine sources for the Eurasian admixture in Northeast Africans rather than Arabic groups [4, 6]. We find that the pattern is more complex with different source populations in different regions, see Fig 1B and 1C as well as Fig 2. Levantine contributions are seen more towards the north and contributions from Arabian peninsula groups are seen more at lower latitudes, Figs 1B, 1C and 2.

The fact that both approaches for admixture dating produced populations from the same country that had the most extreme difference in Eurasian admixture dating, highlights the heterogeneous nature of North-East African genetics and how little explanatory power country borders have on population structure. It is, not unexpectedly so, rather geographic, linguistic, and cultural borders that explain the degree of genetic interconnections between groups.

The major linguistic family was the only factor that was significant (and only for the best by f₃) in our ANOVA test of the available categories, S4 Table. The linear regression analysis of distance from the Levant, S21A and S21B Fig, also produced a significant fit with a negative coefficient indicating more recent admixture dates further from the Levant—this is likely driven by the younger dates for the populations in and around South Sudan. The same pattern was observed when comparing the distance to Sanaa, albeit with a smaller slope of the line and larger p-values (S21C and S21D Fig).

One possible explanation for this phenomenon could be that populations with little or no previous Eurasian admixture would have their inferred admixture date affected more by recent Eurasian admixture than populations that experienced larger admixture in the past. In other words, most, if not all, of the populations in this study have or have had admixture with populations from the Middle East during the Arab expansion, and this newer admixture is obscuring older admixture patterns. The groups with younger inferred dates in our analysis thus likely have less older admixture contributions.

Our study thus points to that the distribution of Eurasian-like ancestry in Eastern and North-Eastern African populations is mostly an effect of more recent migrations (many of them recorded in historical texts) rather than ancient events related to the advent of pastoralism in the region at large, as indicated by ancient DNA studies [5]). Identifying the impact of ancient events on populations was not feasible when the original pattern has been distorted or masked by subsequent admixture events. To fully explore the question of Eurasian admixture into Africa over larger timescales likely requires population-level aDNA, especially of the early East African hunter-gatherers such as Mota, and the various in-moving groups, including those containing Eurasian admixture.

Conclusions

North-Eastern Africa is a vast region with complex histories of migrations and admixture. It was not possible to identify one source or origin of Eurasian admixture in the region, rather different populations have experienced admixture at different times, at varying degrees, and from different external sources. Although slight trends were observed linked to language grouping and geography, the overall pattern proved to be complex and specific to certain population groups. Previous studies have highlighted these events in distinct regions or countries in Northern and Eastern Africa, whilst we in this study have tried to combine them with a specific emphasis on the Eurasian admixture in modern-day populations.

Supporting information

Acknowledgments

The computational analyses were done through the Swedish National Infrastructure for Computing (SNIC) at Uppmax. Authorized NIH Data Access Committee (DAC) granted data access to Carina Schlebusch for the controlled-access genetic data analyzed in this study that were previously deposited by Scheinfeldt et al. 2019 in the NIH dbGAP repository (accession code phs001780.v1.p1; date of approval: 2019-05-17). For the genome-wide genotype data from the Patin et al. 2017 study (EGA accessory number EGAD00010001209), data access was granted via European GenomePhenome Archive (EGA) by the GEH Data Access Committee EGAC00001000139. A special thanks to the author of MOSAIC Michael Salter-Townshend for the discussion on how the best perform the ancestry deconvolution, to Carolina Bernhardsson for help with plotting, and Cesar Fortes-Lima for help with the study design.