Which population is most genetically distant from Africans?

Razib had a post yesterday on the question of which population is more distant from Africans. In that post he concluded that:

That means that those of us of non-African ancestry are all equally distant from the African root.

I was planning to rebut this conclusion, but Razib beat me to the punch by correcting himself that Amerindians are more distant from Africans. First, let's deal with the original post:

He based this conclusion on some observations on a recent study:

The tree makes it clear: all non-Africans form their own independent branch from Africans. In the PCA you see that along the biggest component of variation in the genetic data the non-African groups are about the same distance from Africans. And in the ADMIXTURE analysis when you assume four ancestral populations, the Africans and non-Africans separate out cleanly excluding groups which a high likelihood of European or Arab admixture.

Neither PCA nor ADMIXTURE tells us anything about who is more distant from Africans. First of all, the clusters identified by ADMIXTURE are not phylogenetic units, nor is the pattern of splits at successive K evidence for the human phylogeny. For example, at K=2, East Eurasians split from Europeans/Africans, even though it is clear that East Eurasians do not represent a sister clade to a clade comprising of Europeans/Africans. In short ADMIXTURE (a well as frappe, STRUCTURE and assorted Bayesian clustering methods) tell us nothing about who is more distant from whom.

Nor does PCA tell us anything more. Sure at the first 2 dimensions, Africans split off from non-Africans but that is insufficient to conclude that Africans are equidistant to all Eurasians. This is due to the fact that distance at the first two dimensions of PCA lower bounds total distance (a consequence of the Pythagorean theorem, no less). What does this mean? Populations that are equidistant on the (PC1, PC2) plane may be actually non-equidistant overall.

Finally, the tree argument provides evidence for Out of Africa, but it does not provide evidence for equidistance, as it not only encodes phylogenetic relationships (who is closer to whom in evolutionary terms) but also distance. It is perfectly possible that two taxa A and B may form a clade relative to a third taxon C, but that does not guarantee that A will be closer to B than to C.

As Razib notes in his newer post, it is actually Amerindians who are most distant from Africans. Here is part of a table of Fst values that I was planning to use to make that point (he uses a different one, but all studies pretty much tell the same story). It's from the recent Korean study:

YRI GU 0.118332
YRI GJ 0.119501
YRI JC 0.1192
YRI YC 0.118593
YRI PC 0.118986
YRI GR 0.119158
YRI JJ 0.11892
YRI NJ 0.118754
YRI US 0.119086
YRI CA 0.11903
YRI VN 0.116789
YRI CB 0.108875
YRI JL 0.117513
YRI MH 0.109053
YRI CHB 0.117318
YRI JPT 0.118053
YRI AI 0.142997
YRI CEU 0.101197

Note that Yoruba (YRI) are most distant from Amerindians (AI), while all other populations fall within a very narrow range of 0.101197 to 0.119501 or about ~18% variation in distance to Yoruba.

Things become more interesting if we exclude Caucasoids (CEU) and limit ourselves to East Eurasians. Now, the range of distance from Yoruba becomes 0.108875-0.119501, or only ~10% variation in distance to Yoruba.

Yoruba are closer to (in order): to Caucasoids, East Asians, Amerindians, in that order. Distant from all, but not equidistant.

Razib advances an explanation for the Amerindian-African distance:

So what does this mean? And why is this so? I think I won’t revise my model of the out of Africa migration. I don’t think there was serious secondary migration out of Africa after the initial one (at least until recently). And yet somehow the indigenous populations of the New World are more genetically distinct. This is because of genetic drift. Specifically, a set of serial founder events, where the genetic variation was reduced and ancestral allele frequencies changed rapidly. When a population goes through a bottleneck, and then becomes isolated, it “goes its own way,” as there isn’t gene flow to requilibrate the frequencies. The push east, to Australasia and to the New World, was accompanied by founder events due to fissioning off of small groups from the main ancestral population. From what we can tell there was relatively little gene flow after the initial settlement of the New World and Oceania (actually, there may have been several waves into the New World to be fair, but it looks like there wasn’t enough Eurasian gene flow to dampen the reduction in heterozygosity caused by bottlenecks).

"Drift" gets invoked so often, so it is worthy to consider if it is responsible for the greater distance of Amerindians to Africans. "Drift" can indeed shift allele frequencies in a small population, and increase Fst. However, there are dozens of small and isolated populations throughout Eurasia, yet none of them are more distant from Africans than Amerindians are. If "drift" is responsible, then we would expect some isolated Eurasians to repeat the pattern of greater distance that Amerindians present. Good luck finding any study where any Eurasian population is more distant to Africans than (unadmixed) Amerindians are.

There is no need to invoke "drift" to explain Amerindian-African distance, as there is a simpler explanation: Amerindians, unlike Eurasians, are more distant from Africans, not because of "drift" shifting their allele frequencies, but because they lacked the opportunity for substantial gene flow with Africans.

It is worthy to revisit the Fst table: Amerindians are 1/5 further from Africans than East Asians are because they spent about 1/5 of the time since "Out of Africa" isolated from Africans. East Asians had about 60 thousand years worth of opportunities for gene flow with Africans (however limited), while the corresponding time for Amerindians is about 45-50 thousand years.

There is actual evidence for post-Out of Africa gene flow between Africa and Eurasia, but not between Africa and the Americas. It comes from the YAP marker of the human Y-chromosome. This defines the DE-YAP clade of the Y-chromosome phylogeny. This clade is absent in the Americas, but it is present in Africa, West and East Eurasia. Africa and West Eurasia share the E subclade, and East Eurasia possesses the D subclade.

What YAP tells us is that Out-of-Africa was not a one-time event: there has been subsequent gene flow linking Africa and Eurasia, and YAP is the smoking gun for this gene flow (*) There is no way around it: post-OOA men bearing YAP Y-chromosomes spread across a range spanning drom Nigeria and South Africa to the isles of Japan, but not into the Americas. These were not necessarily the only conduits for post-OOA gene flow, but they prove the existence of such gene flow.

Moreover, the greater proximity of Caucasoids to Africans can also be explained by the joint possession (by Africans and Caucasoids) of the E subclade of YAP (at the exclusion of East Asians and Negritos, who possess the D subclade).

In conclusion: the population that is most distant from Africans are Amerindians. They are more distant not because of drift but because of an earlier (10-15kya) cessation of any important gene flow between them and the rest of the species: about 1/5 less gene flow => about 1/5 more distance (relative to East Asians). Finally, this cessation of gene flow does not only make sense archaeologically, but is also evidenced genetically by the prevalence of the YAP marker, which records African-Eurasian contacts (and a nested subset of African-Caucasoid contacts) at the exclusion of Amerindians.

(*) The direction of this gene flow is contested as both Eurasian and African origin of YAP have been proposed. That's not important however, as YAP links Africa and Eurasia at the exclusion of the Americas either way.

UPDATE (Dec 24): It turns out that the greatest distance within our species is between Mbuti Pygmies and Papuans.