What Did Denisovans Look Like? A New Epigenetics Approach May Give Us Some Clues.


One of the most surprising details of human evolution revealed by the sequencing of DNA from ancient fossils is that some bones thought to be Neanderthals were actually those of an entirely new kind of human. Denisovans were first identified when researchers examined the 74,000-82,000 year old genome sequenced from a pinky bone found in a Siberian cave. The genome was obviously human, but so divergent from Neanderthals and anatomically modern Homo sapiens that it pointed to a third group of humans. Later Denisovan proteins were recovered from a jaw and genomes from teeth.

Although we have intimate knowledge of some of these humans’ genomes, we know virtually nothing about their physical morphology. Without that knowledge, we have been unable to deploy the full range of tools in the paleoanthropology toolkit. Studies of diet, of behavior, of their unique evolutionary trajectory, and of many diseases present in ancient populations all depend upon having bones to study.

The original Denisovan genomes were from remains once thought to have belonged to Neanderthals. How many other Denisovan remains have already been found, but have been misclassified? If we could identify how Denisovans might differ physically from other humans, we might be able to more effectively identify their fossils. This is a question that might possibly be answered now, thanks to an innovative new study from David Gokhman and colleagues, published this week in the journal Cell.

Gokhman and his co-authors tried to identify physical features that were unique to Denisovans using clues from their genomes. Extrapolating physical features from genomes is not at all straightforward; in fact, there are very few traits which can actually be clearly identified on the basis of genes alone. However, the authors of this study took a different and more creative approach: they focused on looking for differential gene regulation—whether or not a gene is likely to be used to make protein—in Denisovan lineages versus modern humans, Neanderthals, and nonhuman primates.

They were able to do this by searching for promoters—regions of the genome that control the regulation of genes—which are known from experimental studies to have specific effects on morphology when inactivated. Gokhman and colleagues looked for methylation at these sites, which if present would indicate that the gene was likely being silenced. They carefully excluded from their study any methylation sites known to be affected by age, sex, health status, environment, or the type of tissue they’re found in.

Based on the methylated promoters that they identified, they constructed estimates of the morphological consequences to Neanderthals, modern humans, Denisovans, and chimpanzees. Next, then compared the morphological predictions based on methylation profiles in modern humans, chimps and Neanderthals with their actual morphologies as a check on their method, finding that they “reach precision of 82.8% in reconstructing traits that separate Neanderthals and modern Homo sapiens, and 87.9% in predicting their direction of change. In the chimpanzee, we reach a similar performance, with 90.5% precision at predicting which traits are divergent and 90.9% in predicting their direction of change.”

After this validation of their method, they compiled a list of 32 traits predicted by the Denisovan methylation patterns. Many of their predicted traits were similar to those that paleontologists use to characterize Neanderthal skeletons, including robust jaws, low foreheads, and thick enamel on their teeth.  Other traits were predicted to differ from Neanderthals, including aspects of the shape of their jaws and broadening of the region of the skull between the parietal bones.

An additional–and remarkable– validation of their method occurred while the manuscript for this paper was actually in review: a jawbone from the Tibetan Plateau was identified through protein analysis as belonging to a Denisovan. The jawbone’s morphology matched seven out of the eight predictions that Gokhman and colleagues inferred from their methylation approach.

I asked Dr. Gokhman about what he sees as the significance of the team’s work, and he responded in an email to me:

“I see the significance of this work at two levels. At a more specific level, it gives us a glimpse into the morphology of Denisovans. It is far from being a complete profile: it is a qualitative, rather than precise prediction, and its accuracy is estimated at ~85%, but considering how little we know about Denisovans, I think it helps us understand them better. At a more general level (and more important in my opinion), this work suggests that looking at gene regulatory layers can teach us more about morphology than we previously appreciated. Looking at the most extreme regulatory changes, which span thousands of bases, are fixed and are in key regulatory regions, and then linking them to monogenic diseases, where the gene-phenotype link is well-established, allows us to infer which organs are expected to be affected by the regulatory change, and what is the most likely direction of phenotypic change.”

Dr. Gokhman noted that there have already been some critiques of the study on methodological grounds (see here for an example), which are “based on some misinterpretations of our goal and method. The question we tried to address, and the approach we applied are not aimed at predicting precise phenotypic information from a single sample. It is a comparative approach, the goal of which is to examine the link between the most extensive regulatory changes between human groups and the potential direction of their anatomical effect.”

Dr. Rick Smith, a postdoctoral fellow at Dartmouth and an expert on paleo-epigenetics who was not involved with the study, was impressed with it. “This paper is exciting! There are reasons to take their findings with a grain of salt, but the authors themselves carefully emphasize the fact that this is a predication and is not meant to be taken as ‘the face of a Denisovan.’ Rather, this generates predictions for aspects of Denisovan morphology in a clever way, giving us clues about their anatomy that we could not get before. This approach could help us identify other Denisovan fossils, it could help us to know what to look for. Now paleontologists and paleogeneticists can go out and test these predictions.”

The predictions the team has generated have already flagged several fossils as being good candidates for Denisovans. In particular, the predicted broad distance between parietal bones matches up with two crania dating to between 100,000 and 130,000 years ago from Xuchang in eastern China, which otherwise resemble Neanderthals. If genomes can be recovered from these remains, they would serve as a strong test of the authors’ predictive models.


Gokhman et al. 2019. Reconstructing Denisovan Anatomy Using DNA Methylation Maps. Cell 179(1): 180-192.e10.

Leave a Reply

Your email address will not be published. Required fields are marked *