Recent genetic studies have established that the KhoeSan populations of southern Africa are distinct from all other African populations and have remained largely isolated during human prehistory until 2000 years ago. Dozens of different KhoeSan groups exist, belonging to three different language families, but very little is known about their population history. We examine new genomewide polymorphism data and whole mitochondrial genomes for .100 South Africans from the 6 ¼Khomani San and Nama populations of the Northern Cape, analyzed in conjunction with 19 additional southern African populations. Our analyses reveal fine-scale population structure in and around the Kalahari Desert. Surprisingly, this structure does not always correspond to linguistic or subsistence categories as previously suggested, but rather reflects the role of geographic barriers and the ecology of the greater Kalahari Basin. Regardless of subsistence strategy, the indigenous Khoe-speaking Nama pastoralists and the N|u-speaking 6 ¼Khomani (formerly hunter-gatherers) share ancestry with other Khoe-speaking forager populations that form a rim around the Kalahari Desert. We reconstruct earlier migration patterns and estimate that the southern Kalahari populations were among the last to experience gene flow from Bantu speakers, 14 generations ago. We conclude that local adoption of pastoralism, at least by the Nama, appears to have been primarily a cultural process with limited genetic impact from eastern Africa.

Background: Global and local ancestry inference in admixed human populations can be performed using computational tools implementing distinct algorithms. The development and resulting accuracy of these tools has been tested largely on populations with relatively straightforward admixture histories but little is known about how well they perform in more complex admixture scenarios. Results: Using simulations, we show that RFMix outperforms ADMIXTURE in determining global ancestry proportions even in a complex 5-way admixed population, in addition to assigning local ancestry with an accuracy of 89%. The ability of RFMix to determine global and local ancestry to a high degree of accuracy, particularly in admixed populations provides the opportunity for more accurate association analyses. Conclusion: This study highlights the utility of the extension of computational tools to become more compatible to genetically structured populations, as well as the need to expand the sampling of diverse worldwide populations. This is particularly noteworthy as modern-day societies are becoming increasingly genetically complex and some genetic tools and commonly used ancestral populations are less appropriate. Based on these caveats and the results presented here, we suggest that RFMix be used for both global and local ancestry estimation in worldwide complex admixture scenarios particularly when including these estimates in association studies.