For the first time, scientists analysing the DNA of Denisovans — an extinct group of hominins that was discovered around a decade ago — have offered a glimpse of what they might have looked like.
Ever since archaeologists uncovered the first fragmented Denisovan remains in a Siberian cave, researchers have scoured the globe for clues to how the mysterious hominins looked. Denisova Cave has yielded a few more small fossils, mostly teeth. A jawbone from the Tibetan Plateau added detail this year, as did information on a missing finger bone that moved between labs in Russia, California and Paris. But none of these fossils is large or complete enough to reconstruct many anatomical details.
Now, computational biologists have produced a rough sketch of Denisovan anatomy based on epigenetic changes — chemical modifications to DNA that can alter gene activity. Their approach reveals that Denisovans were similar in appearance to Neanderthals but had some subtle differences, such as a wider jaw and skull1.
“It does help to paint a clearer picture of how they might have looked. Just the idea that it’s possible to use the DNA to predict morphology so well is very impressive,” says Bence Viola, a palaeoanthropologist at the University of Toronto in Canada who has analysed Denisovan remains, but was not involved in this research.
Epigenetic modifications to DNA have a profound influence on development, disease and most biological traits throughout life. They can help to determine differences between cells with otherwise identical genomes. One of the best-studied epigenetic changes is the addition to a DNA base of a methyl chemical group — made up of one carbon atom and three hydrogens — which often quells the activity of a gene.
The methyl group degrades after death, so cannot be spotted in ancient DNA. But a team co-led by Liran Carmel, a computational biologist at the Hebrew University of Jerusalem, discovered a way to identify parts of ancient DNA that had once been methylated, by analysing patterns of chemical damage that accrues to the DNA over time. In 2014, Carmel’s team mapped methylation patterns across the genomes of Neanderthals and Denisovans, and identified a limb-development gene for which these patterns differed between the extinct groups and modern humans2.
In the latest study, Carmel and computational biologist David Gokhman, also at the Hebrew University of Jerusalem, led a team that identified thousands more regions of the genome in which the methylation patterns of Denisovans and Neanderthals were distinct from those of modern humans. They compared these with databases of epigenetic modifications in human tissue — where the impacts on gene expression are known — and produced a list of hundreds of genes for which expression levels probably differed between archaic groups and modern humans.
To connect this list to anatomical traits that would affect the Denisovans’ appearance, the researchers looked at another database, which catalogues the physical effects of genetic mutations in people with rare conditions. Carmel and Gokhman reasoned that the reduced gene expression caused by DNA methylation was roughly analogous to the effects of the disease-causing mutations.
Before applying their method to Denisovans, Carmel and Gokhman’s team first tested whether it could successfully predict the anatomy of Neanderthals, which is known from hundreds of fossils.
The predictions about physical appearance made using this approach are qualitative and relative, Carmel explains. “I can tell you that fingers are longer, but I cannot tell you they are longer by 2 millimetres,” he says.
The team found 33 Neanderthal traits that could potentially be predicted from methylation patterns. The results accurately predicted 29 of those traits, for instance that the species had broader faces and flatter heads than modern humans. But it wrongly indicated that the indentations between fused skull bones, known as sutures, were wider in humans.
The researchers then turned the technique to Denisovans. They predicted that these hominins shared many traits with Neanderthals, such as their low foreheads and wide rib cages, but identified some differences, including wider jaws and skulls. Although it is impossible to know how accurate their picture is, some of the predictions are supported by evidence from Denisovan remains.
The best-characterized Denisovan feature in the fossil record is gigantic molar teeth. Although the researchers weren’t able to predict this — because molar size was not in the database they used — they did determine that Denisovans had long dental arches, a potential adaptation for big teeth.
The 160,000-year-old lower jawbone from the Tibetan Plateau matched Gokhman and Carmel’s predictions for 3 out of 4 traits. And a piece of skull from Denisova Cave that Viola has presented at meetings (but not yet described in a paper) suggests the group had wide heads — which matches the epigenetic reconstruction. However, a reconstruction of the Denisovan fingertip, published this month3, suggested theirs was slender like humans’ — unlike the thick Neanderthal-like fingers in the prediction.
“I think the big picture is correct, but with the individual traits, there is a lot of leeway,” says Viola. Although he is impressed by the predictions, he is unsure sure how they will help determine what Denisovans actually looked like. Potential Denisovan bones are so rare that most are already tested for DNA or protein — currently the only way to link remains to the extinct group.
This is an “absolutely valid approach”, says Manolis Kellis, a computational biologist at the Massachusetts Institute of Technology in Cambridge who works with epigenetic data. The authors do a good job of accounting for uncertainties that feed into their predictions, he adds. “The resulting findings are quite robust.”
In the future, scientists might use epigenetics to reconstruct the anatomy of hominins known from fragmentary fossils or perhaps even DNA from dirt, says Pontus Skoglund, a population geneticist at the Francis Crick Institute in London. But he thinks the approach could be most useful in predicting traits, such as behaviour, that don’t leave an impression in the fossil record.
Nature 573, 475-476 (2019)