Peter Frost over at his blog has a long post on the transition to agriculture and pastoralism in Northern Europe.

He tagged me on Twitter, so presumably, he’s soliciting my opinion/response.

The post starts off with a quick reference to the attempt to leverage massive replacement in Northern Europe eight to four thousand years ago in the interests of contemporary politics. I’m not going to address that because I’m not very interested in how these topics relate, and I won’t post comments (or will delete) that engage with that. I will focus on the science.

First, I tried to leave a comment on his weblog and blogger ate it. So I’m just going to put a post here in the interests of open exchange. I also think many readers here have some of the same opinions as Peter, or suspicions, so it might be best to clear things up.

I don’t think his Peter’s argument can really be understood without reading his 2006 paper, European hair and eye color: A case of frequency-dependent sexual selection? My opinion in regards to this hypothesis is that I think it’s probably wrong and I’m skeptical. More skeptical than I was when I first read the paper because we have more understanding of the process of the settlement of Europe during the late Pleistocene and early Holocene. But, there is still a small window for it to be correct, as one can see in Peter’s post.

The argument hinges a lot on the pigmentation profiles of proto-European groups based on predictions from algorithms which use modern Europeans as a training set. These predictions are in the papers themselves, so Peter isn’t doing anything that the authors didn’t do. But, I have come to the conclusion that they’re probably not trustworthy. These ancient populations were very different from modern Europeans, and their genetic architecture for pigmentation may have been different (modern Europeans are a compound of several groups).

Though Mesolithic Western European hunter-gatherers were probably darker in complexion than modern Europeans, I believe it is likely that they were not nearly as dark as pigmentation prediction algorithms suggest. Second, it is true that alleles correlated with blonde hair in Europeans within the KITLG locus are found in Siberia nearly 20,000 years ago. But it is not true that “Ancient DNA from Afontova Gora has shown that people had blond hair in mid-Siberia as early as 18,000 years ago.”

What has been found is that Europeans who carry the derived variant at rs12821256 are more likely to have blonde hair. Those who are heterozygote are twice as likely, while homozygotes are four times as likely. At least against the population base rate. The frequency in Scandinavia of the derived variant is ~20%. Many blonde people don’t have the derived variant. And, not all people who have the derived variant have blonde hair.

Of my three children two are heterozygotes for the derived variant (they carry one copy). Probably not coincidently these two have lighter hair than the third. But neither are really blonde, though perhaps they are blond(ish) during certain times of the year. More accurately their hair is probably sandy brown. Why? I’m their father, and as a normally complected South Asia, I give them a host of alleles at other loci which make them different from the typical European genetic architecture of pigmentation.

As I said earlier Peter can’t really be blamed for making these inferences because they are in the scientific literature themselves. But just because they’re there doesn’t make them true (though I do think Peter should be careful about extrapolating from odds ratios against a particular base rate probability to some deterministic relationship).

A final issue is the idea that the alleles that define modern Northern European pigmentation were present in Scandinavian and Eastern European hunter-gatherers. This is correct. But again, modern prediction algorithms are trained groups with modern genetic backgrounds. In mixed populations, the largest effect QTLs explain only half the variance in pigmentation. The rest of it is accounted for by “genomic ancestry”, which basically means there are loci associated with ancestral groups that haven’t been discovered yet. But a second and more important issue is that the frequency of some the alleles in modern Northern European groups is different from what you find in the ancient ones. The ancestral variant on SLC24A5 is almost impossible to find in Northern Europe in indigenous people today (in Europeans the ancestral variant is most often found in Spain, due to admixture with Africa during the Moorish period). I don’t need to review the literature, but there is evidence for a fair amount of selection on these loci within the last 4,000 years. Even SHG and EHG still segregated ancestral variants at higher frequencies that modern Europeans.

The second major theme in the blog post has to do with hunter-gatherer ancestry. There’s a section on haplogroup U where Peter suggests that its disappearance is due to selection, not a replacement. U is associated with hunter-gatherer ancestry. This may be true, but mtDNA and Y need to be interpreted cautiously in any case (both R1b and R1a are far more common than one might predict from autosomal distributions of the ancestry of populations in which they were originally found).

Then there is the argument that bottlenecks/founder effects and natural selection might have skewed our estimates. I don’t really get the former argument at all:

Founder effects may be another causal factor. When bands of hunter-gatherers are given the opportunity to adopt farming, most of them turn up their noses and only a few will make the change. Because those few bands are not perfectly representative of the hunter-gatherer gene pool, and because their numbers may increase many times over (thanks to the increase in food supply) the resulting founder effects will be substantial.

These are verbal models, and unpersuasive to anyone who has looked at the data and generated results. Mesolithic hunter-gatherers were a genetically homogeneous lot to begin with. They didn’t have all this variance to sample from. There was later increase in hunter-gatherer ancestry into European farmers from demographic reservoirs, but the argument about founder effect doesn’t work because the two groups are so different that playing around with biasing the sample from which one mixes does not change the overall result. Replace hunter-gatherer and farmer with “Ashkenazi Jew” and “Chinese.” The latter two groups have some variance, but a bottleneck on one isn’t going to change one’s estimate of admixture in a daughter population.

The issue about selection suffers from the problem that the magnitude would have to be too large and extensive across the whole genome to reshape hunter-gatherers in this manner to be plausible. One might imagine a case where gene flow and selection on parts of the genome from the donor group inflates the donor group proportion…but I don’t think that’s Peter’s point? Theoretically, a model of admixture followed by sweeps around one population’s ancestry component is possible, but I don’t think we see evidence of that in the ancient DNA.

In any case, though the verbal argument seems reasonable on first blush, the models and dynamics don’t work out.

Peter ends:

Some of the confusion in this debate may arise from the assumption that “late hunter-gatherers” formed a single group in Europe. In fact, there were at least three such groups (WHGs, SHGs, EHGs), whose genetic profiles significantly differed from each other and whose fates were likewise different. WHGs were an evolutionary dead end. They were replaced. The same cannot be said for the hunter-fisher-gatherers of Scandinavia and the Baltic, who were able to achieve high population densities by exploiting marine resources (Price 1991). With them we see more genetic continuity than rupture, and it is possible that some genetic characteristics formerly ascribed solely to “Anatolian” farmers were in fact of SHG origin.

The people who are making the assertions that Peter is rebutting are not confused as to the nature of the populations which they named and which they modeled. Peter can download the data and replicate the analyses himself. WHG, SHG and EHG seem to exist on some sort of continuum, with post-“Villabruna cluster” ancestry at one end of the spectrum and post-Ancestral North Eurasian (ANE) ancestry at the other. WHG is mostly descended from ancestors of the Villabruna cluster, who share a common ancestry derived from late Pleistocene West Eurasians with Anatolian farmers (the latter of whom admixed with Basal Eurasians). EHG is a mix of the same Villabruna people (or at least their eastern fringe), but with a preponderance of ANE-like ancestry. SHG is between these two groups.

It also seems that European hunter-gatherers sometime in the late Pleistocene and or early Holocene recieved a small but detectable pulse of East Asian ancestry. Also, commonly shared haplotypes with West Asians on SLC24A5 (SHG and EHG) and EDAR with East Asians (SHG) indicates some gene flow with other places (though I believe SHG has no detectable East Asian ancestry).

Finally, there is much discussion of a late occupation of Northeast Europe by farmers. Since I predicted this 10 years ago I don’t have much objection to this section…except I don’t think that it supports his other points at all. That is, the persistence of hunter-gatherer populations around the Baltic does not mean that hunter-gatherers were more similar to farmers than we might think, nor does it reject the likelihood of total replacement in many areas of Europe to the south.

The overall conclusion here is two-fold:

1. The assertions about pigmentation are not necessarily wrong, but they are far weaker based on the data that might be inferred from the post. Additionally, modern Europeans have lots of evidence of recent selection and allele frequency change at several of these loci.
2. The assertions about very large misestimations of inferred mixing proportions are probably wrong.