Being probably the most genetically diverse modern human population, they are occasionally viewed as akin to the ur-humans, with everyone else shedding diversity via founder effects as they moved away from a south African modern human urheimat.
They are also sometimes viewed as a basal branch of the human family tree, and they probably are -if modern humans are made to fit a tree model. But, modern humans didn't really evolve tree-like (some African farmers have Khoisan-like admixture, and the Khoisan themselves have relatively "shallow" common ancestry with other Africans and many Eurasians on account of their possession of a respectable frequency of Y-haplogroup E).
I have sometimes noted that in the case of South African groups were are lucky that the Khoisan exist as a discrete set of populations, making it easier to discern the legacy of South African hunter-gatherers in the genomes of immigrant farmers and pastoralists who converged southwards over the last few thousand years. This can be contrasted with the presumable situation in places like West Africa (the cradle of Sub-Saharan African farming), in which any indigenous hunter-gatherer groups have ceased to exist as distinct entities a long time ago.
A new AJHG paper sample south African genomes extensively and arrives at a startling conclusion. In the words of the authors:
Overall, the results of this analysis indicate that it is very unlikely that the highly divergent L0k1b/L0k2 lineages were incorporated into the Bantu-speaking populations via gene flow from a population that was ancestral to a Khoisan population in our sample but subsequently lost from the Khoisan population via drift. Instead, these results support the hypothesis that the ancestors of the Bantu-speaking populations carrying the divergent L0k lineages (who now live mainly in Zambia) experienced gene flow from a pre-Bantu population that is nowadays extinct. Alternatively, it is possible that descendants from this pre-Bantu population do exist but have not yet been included in population genetic studies; however, our extensive sampling of populations from Botswana, Namibia, andWest Zambia (which includes representatives of nearly all known Khoisan groups) makes it highly unlikely that this pre-Bantu Khoisan population has not yet been sampled.In other words, we must resist the tendency to think of the Khoisan as representatives of all pre-Bantu south Africans. The Khoisan are certainly descendants of old south Africans, and represent a part of the pre-Bantu genetic landscape that retained its cultural distinctiveness (and hence can be nowadays sampled as a distinct population). But, there were other, now submerged, peaks in that landscape that are no longer extant in distinct form, but only in absorbed form in the gene pool of south African farmers.
This is fairly interesting in itself, and certainly ought to change our belief about what Africa looked like pre-Bantu expansion. We ought to think of, perhaps, a cornucopia of groups: many of them may have gone extinct; some may have been completely absorbed into more successful ones, and perhaps only a handful survive as distinct entities. Such a view would agree with the conclusions of physical anthropology about the persistence of archaic-leaning groups in parts of Africa down to the Holocene boundary.
The American Journal of Human Genetics, 17 January 2013 doi:10.1016/j.ajhg.2012.12.010
Ancient Substructure in Early mtDNA Lineages of Southern Africa
Chiara Barbieri et al.
Among the deepest-rooting clades in the human mitochondrial DNA (mtDNA) phylogeny are the haplogroups defined as L0d and L0k, which are found primarily in southern Africa. These lineages are typically present at high frequency in the so-called Khoisan populations of hunter-gatherers and herders who speak non-Bantu languages, and the early divergence of these lineages led to the hypothesis of ancient genetic substructure in Africa. Here we update the phylogeny of the basal haplogroups L0d and L0k with 500 full mtDNA genome sequences from 45 southern African Khoisan and Bantu-speaking populations. We find previously unreported subhaplogroups and greatly extend the amount of variation and time-depth of most of the known subhaplogroups. Our major finding is the definition of two ancient sublineages of L0k (L0k1b and L0k2) that are present almost exclusively in Bantu-speaking populations from Zambia; the presence of such relic haplogroups in Bantu speakers is most probably due to contact with ancestral pre-Bantu populations that harbored different lineages than those found in extant Khoisan. We suggest that although these populations went extinct after the immigration of the Bantu-speaking populations, some traces of their haplogroup composition survived through incorporation into the gene pool of the immigrants. Our findings thus provide evidence for deep genetic substructure in southern Africa prior to the Bantu expansion that is not represented in extant Khoisan populations.