How widespread is Denisovan ancestry today?

Last month, David Reich and colleagues Reich:Denisova:2011 reported on estimates of Denisovan ancestry for island and mainland Asian populations. Their most memorable conclusion was that they could find no substantial sign of Denisovan ancestry anywhere on the Asian mainland, or indeed on any island that had ever been connected by land to Asia.

The distribution was stark, as illustrated by the map from the paper:

I wrote about the paper when it was released (“Denisovan DNA in the islands, and an Australian genome”), noting:

Notice the apparent lack of Denisovan ancestry in anyone who lives anywhere that was once connected by land with mainland Asia. I say "apparent" deliberately: Abi-Rached and colleagues reported last month on the widespread distribution of Denisovan HLA types among today's Asian populations, and those may well be products of Denisovan genes that were later selected. I've already identified a handful of other loci that seem to reflect Denisovan ancestry in mainland Asian people. According to the comparisons by Reich and colleagues, such loci must be exceptions.

Abi-Rached and colleagues Abi-Rached:2011 had argued that HLA alleles found in the Denisovan genome are presently common in some parts of Asia, and likely reflect local adaptive introgression. Substantial introgression of a small number of genes would not be enough to create a strong genome-wide appearance of Denisovan ancestry. Still, it was a little odd that the first genes anybody looked closely at would provide strong evidence of introgression.

Now, Pontus Skoglund and Mattias Jakobsson Skoglund:Jakobsson:2011 say that Denisovan ancestry is widespread across China and Southeast Asia.

That conclusion contradicts Reich and colleagues, so why do the studies come to such different results?

Skoglund and Jakobsson suggest that they have succeeded in finding introgression where others failed because their model accounts for ascertainment bias in the available datasets. SNP data come from genotyping chips, which have been designed using known polymorphisms. Five years ago, we knew much more about polymorphisms in Europe than other parts of the world, and so the HGDP, and HapMap to a lesser extent, do a good job of sampling rare alleles in Europe but miss many rare alleles in Africa and other populations. This is the ascertainment bias.

Some of the most obvious signs of introgression today are cases where rare alleles are shared with an archaic genome. If ascertainment bias causes you to miss the rare alleles, you’ll miss the introgression.

But that explanation isn’t really sufficient to explain the differences between these papers. For one thing, Reich and colleagues Reich:Denisova:2011 also worked hard to account for ascertainment biases in their SNP samples. For another, whole genome comparisons between East Asian samples and the Denisova genome have not yielded evidence of Denisovan ancestry, even though whole genomes have no ascertainment bias. The number of whole genomes so far compared is very small, and so the statistical ability to detect introgression is lower, but Skoglund and Jakobsson actually replicate that null result in their current paper.

Probably most important, it’s not clear that Skoglund and Jakobsson’s result can actually be explained by rare alleles. Here is Figure 1e from their paper: