Last week, Cell published a new paper by David Gokhman and coworkers that tries to infer the skeletal form of Denisovans from signatures of methylation in the Denisovan genome data. The paper is here: “Reconstructing Denisovan Anatomy Using DNA Methylation Maps”.
I’m a skeptic about the paper. The authors follow an approach that has not been shown to predict the morphology of any living humans or any other species.
The study tries to justify its method by looking at Neanderthal methylation and saying that the methylation pattern can accurately predict Neanderthal bone shapes. This may look convincing on the surface. But it’s actually not. This is a case where “researcher degrees of freedom” are very high.
I am working on a comment to submit to Cell about the study, focusing on some technical issues. I hope that the editors will be interested in publishing a critical exchange on the approach.
In the meantime, I want to point readers to a related study by another group of researchers that has been mostly ignored in the press coverage on this new paper.
Genevieve Housman and coworkers released a preprint earlier this year on biorXiv that looked at methylation patterns in several primate species (chimpanzees, macaques, vervet monkeys, baboons, and marmosets) to see if methylation could predict skeletal morphology within species or between species.
Housman and coworkers carried out a careful study that includes several analyses that are not in the new Cell paper by Gokhman and colleagues. Housman and coworkers measured DNA methylation in bone cells taken from the femur of each primate individual, and they actually took caliper measurements on the femur of each individual. This means that in every case they compared like with like. That’s different from what Gokhman and colleagues could do with ancient DNA, where the samples are coming from different parts of the skeleton from a mixture of different cell types.
Housman and coworkers found some methylation differences that seemed to associate with morphology in their samples. But these were small effects that they could not separate from other possible influences (genetic and environmental differences). I would add that the small sample sizes in the study (only 4 chimpanzees, for example) make it possible that the observed effects might be spurious effects of small samples.
When they looked between species, Housman and colleagues found differences in methylation, similar to Gokhman and coworkers in this new study. Some of those differences in methylation are near genes that matter to skeletal form. Gokhman and coworkers found the same thing for Denisovans and Neanderthals. But in the living primates, the methylation differences actually did not correlate well to the phylogenetic relationships of the primate species. While Housman and coworkers suggest that the methylation differences between species might make some difference to the evolution of their traits, they did not try to predict what those differences were. That’s very reasonable considering the lack of clear signatures within species and the small proportion of the genome included in these regions with different methylation.
What I think: Studying methylation differences is a promising avenue of research, but we are a long way from understanding how differences in methylation may relate to differences in the skeleton. It is important to match like with like – in the Housman study, the methylation of femur bone cells was used to study femur form.
It’s not outlandish to think that the pattern of methylation across promoters in the genome might give a clue about morphology. But there are huge gaps in our knowledge. And there are things we know about methylation in bone that this study doesn’t seem to represent. Those make me skeptical that this study is doing much more than presenting an interesting hypothesis.
I’ve seen a number of geneticists quoted in news stories, saying that now paleoanthropologists will have to test what methylation has told us about Denisovan morphology. That’s only partly correct. A test with paleontology cannot be valid until comparative evidence supports the proposed mechanism connecting differential methylation to morphology—remember, based here upon indirect functional inference for only 2.2 genes per trait.
But if we’re looking for a paleontological test, the Xiahe mandible may already provide one. Gokhman and coworkers predict that the Denisovan dental arch should be longer than modern humans and also longer than Neandertals. The Xiahe mandible, with its complete M3 agenesis, does not have a long dental arch.