Yesterday I pointed out that David Reich had a moderately dismissive attitude toward the new paper in PNAS, Effect of ancient population structure on the degree of polymorphism shared between modern human populations and ancient hominins. Here's what Reich said:
...But Reich believes that the discussion would have been different if it had happened in the open. The PNAS paper questioning the Neanderthal admixture addresses issues swirling around two years ago, but not Reich and Slatkin’s latest work. “It’s been an issue for several years. They were right to work on this,” says Reich. But now,
“it’s kind of an obsolete paper,”
Here's what Nick Patterson, Reich's colleague told me via email:
Ancient structure in Africa was considered when we wrote the Green et al. paper, and we were aware that this could explain D-statistics. But the hypothesis is no longer viable as the major explanation of Neandertal genetics in Eurasia. This was discussed in the recent paper of Yang et al. (MBE, 2012). (Not referenced by the PNAS paper). A very simple argument, that convinces me, is that the allelic frequency spectrum of Neandertal alleles in Eurasia falls off very quickly. A bottleneck flattens out the spectrum, and it turns out that the Neandertal gene flow has to be placed after the out of Africa bottleneck or the spectrum is much too flat. The paper on the arXiv from the Reich lab (Sankararaman et al.) is trying to do something much more subtle than this and date the flow. I personally am no longer interested in explaining the introgression as ancient structure. That ship has sailed. Of course the question of what was the genetic structure of Ancient Africa is quite open, and remains very interesting.
Let's back up for a moment and set the stage. What did they do in the PNAS paper which claims that one can not reject the model that the Eurasian affinity to Neandertals is due to ancient African population structure (i.e., the African ancestors of Eurasians already had a closer affinity to Neandertals, perhaps due to continuous gene flow)? Basically they created an explicit spatial model with a temporal dimension. The authors simulated parameters of gene flow (and lack thereof) as well as bottlenecks, etc., and found that ancient structure easily generated the D-statistic which the original authors of the Neandertal admixture paper relied upon. So why so dismissive from Reich & Patterson? Because the Yang et al. paper admits this problem, and formulates a way to test alternative scenarios which generate just those D-statistics, but exhibit different demographic histories. What they found in Yang et al. is that a model where a population bottleneck occurs followed by admixture is the best fit to the site frequency spectrum that you see in real populations today. In other words, they also simulated situations where ancient structure generated equivalent D-statistics to admixture, and then furthermore explored scenarios where other population genetic statistics could further prune the alternatives. One could say that the appropriate follow up paper to the PNAS contribution was actually published before it. The paper on arXiv (to be published in PLoS Genetics) goes much further. Using patterns of the linkage disequilibrium in the genome they produce a date when the admixture occurred. The statistical genetics here is somewhat opaque to the casual reader, so interpretation of these results probably should be conditional on the Yang et al. paper, whose results are more elegant and easy to digest. After all is said and done David Reich's judgment is not atypical. Several people who I know personally and are deeply immersed in human population genomics are simply not impressed by the PNAS paper. That happens, and there's no shame in it. But Reich has a point: a speedier process of publication and review would have saved a lot of people some energy. Related:Dienekes' comments.