The new skull from Dmanisi

19 minute read

It smells like ashes. Holding it and examining it is really not like the other fossil crania I’ve studied. The other Dmanisi crania strike me as being very like some Neandertals in their preservation – not with the heavy mineralization or “shine” of more fossilized crania. But this new skull is earthier. It has the feel of some Bronze Age crania. The preservation is remarkable.

When you look at the cranial base of D4500, you realize something truly special is there. The number of well-preserved basicrania from Homo erectus is very small, none as intact as this one.

D4500 basal view
D4500, basal view

This and the other photos are from the new paper by David Lordkipanidze and colleagues, in Science this week.

I saw the skull this summer when I visited Georgia. For people interested in the Human Evolution Past and Future, the course includes some incredible interview footage with Lordkipanidze at the site, and features the Dmanisi crania prominently. I had an incredible time there, and am very grateful to Lordkipanidze, Reid Ferring, and Ani Margvelashvili for their kindness during my visit.

While I did not study D4500 in detail, I did spend enough time with the specimen to get a very good feel for its anatomy relative to the African record. I need to work carefully through the paper to think about the description and non-metric traits. I got an overall feeling that this basicranium is built very much like a human – a very robust human with a small braincase – but a human nonetheless. It is remarkable to have a skull with so little need for reconstruction, because it gives some insight not only into the anatomy of the skull but into how that anatomy must have been built.

Two days before I saw D4500, I was in the Ditsong Museum in Pretoria, and just before that I was in the Malapa lab with Lee Berger and Steve Churchill looking at the MH 1 anatomy in some detail. So I had the anatomy of Australopithecus sediba and the South African examples of early Homo fresh in my mind. I was struck by the many similarities between A. sediba and the Dmanisi crania, and with D4500 specifically.

It is hard not to focus on the similarities, because D4500 is more comparable to earlier hominins in many ways. Many of the later African crania have larger brain sizes, so they look like balloons next to the much smaller skull of MH 1. D4500, despite its great robusticity, has a volume of only 546 ml. That’s bigger than MH 1 by nearly 30 percent, but small enough to make the overall anatomy more comparable. That small size of D4500, coupled with the large mandible, give it some australopithecine-like proportions. But as this paper shows, it is overall put together like other crania of early Homo – and indeed is more humanlike in its configuration than similarly-sized crania of Homo habilis like KNM-ER 1813 or OH 24.

Lordkipanidze and colleagues reflect on the fact that many of the comparable crania are subadults, like the Nariokotome skull KNM-WT 15000 and the adolescent D2700 from Dmanisi.

The current perception of skull morphology of early Homo is thus strongly influenced by adolescent representatives (and the less complete and taphonomically distorted adult cranium KNM-ER 1813) that exhibit comparatively orthognathic faces with lightly built superstructures. The architecture of skull 5 now indicates that small-brained, large-faced, remarkably prognathic and robust morphologies all belonged to the normal range of variation of early Homo.

With that in mind, I think that this paper should have included a comparison with MH 1. It is an adolescent, and yet it is so closely similar in many ways to the adolescent D2700. D4500 is a mature, robust adult, yet with a substantially smaller brain than D2700. Could the contrast between these two Dmanisi crania give a hint to the adult MH 1 would have grown into? There are so many other African crania here in the supplement that the lack of A. sediba is a notable omission. The paper does include a comparison with the purported early Homo specimen from Hadar, AL 666-1. I think that most of the shared anatomy with that specimen would also be shared with MH 1, and indeed with a broader range of South African australopithecines.

Swartkrans thoughts

When I was with the specimen I also had this comparison closely in mind:

Dmanisi D4500 compared to SK 847 from Swartkrans
Dmanisi D4500 (bottom) compared to SK 847 from Swartkrans (top). From figure S4 of Lordkipanidze et al. 2013. The SK 847 reconstruction here involves a symmetrical flip of part of the cranium.

The SK 80/847 composite skull from Swartkrans rivals the Dmanisi sample for the title of the earliest representatives of Homo erectus anywhere in the world. There is much less of the specimen than the Dmanisi crania – and in particular not nearly enough of the vault to examine the characteristic vault morphology of H. erectus. But D4500 clarifies much about the SK 847 comparison. For one thing, the endocranial volume of D4500 at 546 ml is approximately the same as those of large male australopithecines. Yet despite the small brain we have little trouble recognizing the strong H. erectus-like morphology of D4500 in the face and masticatory muscle configuration. SK 847 lacks most of the vault but has those H. erectus similarities in the face.

Some of those facial characters are also characteristic of Australopithecus robustus. The SK 847 upper face is not terribly different from SK 48, the iconic A. robustus specimen, and the dental arcade and temporal bones show some similarities as well (among many differences). Many have pointed at the possibility that A. robustus and Homo may share an evolutionary history. D4500 shows a way to be morphologically robust outside the specializations of robust australoipithecines. Whether those alternative pathways to robusticity are reflected by non-dental similarities, or whether the non-dental similarities reflect common ancestry, remain unclear.

The single species hypothesis for early Homo

D4500, oblique view

Lordkipanidze and colleagues write forcefully in favor of a single-lineage interpretation of early Homo. This part of the paper has attracted the most attention from the press. I’ve seen articles that have described this as “throwing the story of human evolution into disarray” or some such nonsense. I think the paper is very straightforward: here is what they wrote:

Given the scattered and fragmentary fossil record in Africa that predates Dmanisi, questions of earliest African Homo phylogenetics and classification remain unresolved. It remains to be tested whether all of the fossils currently allocated to the taxa H. habilis and H. rudolfensis belong to a single evolving Homo lineage. Although we regard this null hypothesis as parsimonious and fully compatible with new evidence from Dmanisi, alternative scenarios exist. Given the range of variation seen in the Dmanisi paleodeme, there is no convincing signature, at present, of early Homo cladogenesis. The African fossils that post-date the Dmanisi ensemble show brain size increase and correlated change in craniofacial morphology within the evolving lineage of H. erectus. Moreover, it is likely that both the underpinning of the East Asian dispersal of H. erectus, as well as the roots of subsequent H. erectus evolution in Africa (for example, OH 9, Daka), shared greater craniofacial robusticity.

Here’s a press quote from Christoph Zollikofer:

"Everything that lived at the time of the Dmanisi was probably just Homo erectus," said Prof Zollikofer. "We are not saying that palaeoanthropologists did things wrong in Africa, but they didn't have the reference we have. Part of the community will like it, but for another part it will be shocking news."

“Everything that lived at the time of Dmanisi” includes all of the Homo habilis sample from Olduvai and the later part of the Homo habilis sample from the Turkana Basin. The supplementary data in the paper include a several CT reconstructions of African Homo habilis specimens – they could easily have been two or three separate short papers. The paper does not include all the images that would illustrate such a set of comparisons, but does have this really useful graphic:

Figure 4, shape variation from Lordkipanidze et al 2013
Figure 4 from Lordkipanidze et al. 2013. Original caption: Shape variation of the Dmanisi crania in comparative context. Dmanisi crania [2:D2282 (skull 2), 3:D2700 (skull 3 and picture), 4:D3444 (skull 4), 5: D4500 (skull 5 and picture)]; African early Homo (triangles with specimen names); A. africanus (asterisks); H. erectus Java (diamond); Kabwe and Steinheim (crosses); H. neanderthalensis (plus signs); H. sapiens (dots and pictures; populations from Africa, Australia, and America); P. troglodytes: P. t. troglodytes (solid squares and pictures), P. t. verus (vertical rectangles), P. t. schweinfurthii (horizontal rectangles); P. paniscus (open squares). Black symbols indicate adult individuals; gray symbols indicate subadult individuals. Large symbols indicate male and female averages. Shape component SC1 captures within-group cranial variation from large-faced/prognathic to small-faced/orthognathic individuals; SC2 captures shape change associated with grade shifts in neurocranial size between taxa.

The essential claim: the shape variation among these four Dmanisi crania is approximately the same as the shape variation among all East African early Homo. Moreover, measured on the same axes, the shape variation among all these early Homo specimens is approximately equivalent to the shape variation among living humans, or the shape variation within living chimpanzees.

One could point out that there is a sampling issue – the total extent of variation within this early Homo sample may look about the same as that within living humans, but we can look only at a small sample of early Homo and a very large sample of humans, so the comparison is not statistically fair. We should repeatedly simulate small samples of living humans to see if the early Homo level of variation is found in samples of equivalent number. But as you can see from the graphic, actually those modern human and chimpanzee samples are pretty small already. We could really benefit from looking at much larger samples of living humans. And the early Homo sample covers hundreds of thousands of years – giving it another reason for variability. No, I think that this comparison is pretty strong.

Stronger: Chimpanzees and bonobos overlap very little in this shape comparison. There would be hardly any chance of sampling four bonobos that graded across nearly the entire chimpanzee pattern of shape variation. Yet the four Dmanisi crania extend across nearly the entire early Homo shape variation, and the four Homo habilis and Homo rudolfensis crania likewise extend across nearly the entire early Homo shape variation.

Adam Van Arsdale and Milford Wolpoff (2012) tested this hypothesis of a single lineage of early Homo using a different method. They reasoned that an evolving lineage would show a similar pattern of variation at any single time, but would have much greater variation when lumped together across its entire timespan. That is, the entire sample of early Homo might look quite variable, but that variability reflects a lower level of variability in any given time slice, coupled with directional evolution. In their case, they had a stronger comparison because they explicitly considered a multiple-lineage case including A. boisei alongside the possible single-lineage case of early Homo, showing a clear statistical outcome. As they conclude:

Our findings [about early Homo] are not consistent with a pattern of sister species evolving away from a recent common ancestor. The notion of parallel evolving lineages, in turn, is both less parsimonious than that of a single evolving lineage and also fails to fit the data (see Figs. 1 and 2).

Lordkipanidze and colleagues really go a step further. Not only does early Homo look like a single lineage when we consider how it changes across time – the entire early Homo sample has barely more variability than is found within the single Dmanisi site, at a single time. But both comparisons arrive at the same conclusion: early Homo does not look like a group of radiating species, it looks like a cohesive lineage.

The changing reality of human population structure

So what do I really think about this issue? For the last three years I’ve been looking hard at the origin of early Homo, and I’ve now examined many of the essential specimens first-hand. My thoughts are not yet complete, and I’ve considered much more seriously the problems posed by Australopithecus sediba that will demand more first-hand comparisons. But at midstream I can give some impressions of where I have come so far.

Let’s look harder at this concept of “parsimony”. The basic principle is that plurality should not be posited unnecessarily. But what kind of plurality?

A simple answer is that we should adopt the null hypothesis that a sample of fossils represent a single species. Multiple species would be an unnecessary plurality, unless we discover that the sample has statistical properties that are rarely or never found in known samples of a single species. This is the conclusion of Van Arsdale and Wolpoff, and of Lordkipanidze and colleagues, and indeed it is my own attitude. A single species is a good null hypothesis, and the data considered here don’t reject it.

Still, saying that the data don’t reject a null hypothesis is not the same as saying that the null hypothesis is true. Maybe the data have little power to reject any hypothesis. Parsimony may guide us toward the choice of a null hypothesis, but that’s no reason for us to promote it, if the data are indecisive.

I’ve been thinking very hard about what “plurality” means in the context of fossil lineages. Van Arsdale and Wolpoff note in their paper that a single lineage may nonetheless be very complex. Species evolve in fits and starts, with different patterns of change at different times, continual diversification of regional populations, and occasional crashes and replacements. Explaining a pattern of evolution within a single species requires a plurality of evolutionary mechanisms.

From a certain point of view, assigning a series of fossils to different species is simpler: Species bifurcating with a simple pattern of divergence are the most simplistic model of evolutionary change. It may seem perverse, but many anthropologists really believe that extreme splitting is more parsimonious than lumping. A handful of anthropologists believe that the five Dmanisi crania, all found within a few meters of each other, may represent three different species. This outcome is indefensible – each pair among these five crania share features not shared uniformly by the others. Like many other instances in the fossil record, these crania represent a diverse population, but there is no justification for calling them multiple species. But there it is – if we adopt a null hypothesis that will not admit of diversity, we arrive at absurdity.

The Neandertal genome was a major breakthrough for understanding human population dynamics, and early Homo is the next frontier. Consider what we have learned from Neandertal and Denisovan DNA:

  1. There were ancient populations not yet attested by fossil morphology.

  2. Modern humans have more than 90% of their genetic makeup from Africans of the late Middle Pleistocene, comprising one major layer of genetic similarity worldwide. This major layer (whether by massive gene flow or direct population movement) overlies a small contribution from earlier Neandertals, Denisovans, and one or more ancient African populations.

  3. Neandertals and Denisovans themselves represent a second major layer of genetic similarity, sharing a common ancestral population with Middle Stone Age Africans sometime between 300,000 and 600,000 years ago, depending on the mutation rate and population model. The diversification of these populations was less than or equal to that in living subspecies of chimpanzees.

  4. The Denisova mtDNA, which diverged from those found in living humans and Neandertals around a million years ago, seems to represent yet a third layer of genetic similarity more recent than the initial dispersal of H. erectus worldwide. The Denisovans themselves may have mixed with humans from this earlier layer.

  5. It is not yet clear how much the initial dispersal of H. erectus was reflected in the ancestry of archaic humans – or whether that ancestry may have been greater in some regions with Middle Pleistocene Homo erectus samples, such as China or Java.

  6. The “out of Africa” layer of genetic similarity is the last global one, but later regional layers of similarity – reflecting large-scale migration or rapid gene flow – also happened. One of these may have eliminated most traces of Denisovan ancestry on the Asian mainland. Two more are evidenced by the rapid genetic turnover, and repeated turnover, in Neolithic Europe. Another is the rapid replacement of Dorset Culture people in the Arctic.

  7. Within the Neandertals there were diverse populations, large-scale movements, and partial replacements – probably with mixture. They had population dynamics much like those of modern humans across the same timeframe.

These discoveries have clear implications for early Homo:

  1. The half-million year period beginning at 2 million years ago likely had just as much complexity as the equivalent period from 600,000 to 100,000 years ago.

  2. We should expect large-scale movement or gene flow regionally or globally within timespans of 100,000 years.

  3. Some groups representing more ancient layers of similarity may have persisted alongside those experiencing greater movement or gene flow – that is, the evolutionary trajectory need not have been homogeneous across the entire human range.

  4. Populations with different morphological characters – more different, even, than Neandertals and living people – almost certainly mixed.

  5. Long-distance movement and rapid gene flow, conditioned by selection, could affect widespread populations without requiring synchronous evolutionary changes.

From my point of view, the population structure of Neandertals, Denisovans and MSA Africans is the null hypothesis for early Homo. This is more or less consistent with the population structure of chimpanzees, of gorillas, and what we know from genetics about the common ancestors of those living hominoids.

Humans today are a poor model for understanding early Homo, because the last major layer of shared ancestry has left us too genetically uniform. The last major out-of-Africa series of events do not provide a good model for the first out-of-Africa dispersal, but both may have been conditioned by population structure within Africa more than anything else. Looking at these times of significant dispersal causes me to refocus upon the importance of dynamics within Africa.

The model I outline is complex. It is not a simple bifurcation of species, branching without reticulation over time. But the model I outline is necessary for Neandertals, Denisovans and MSA Africans, it is necessary within MSA Africa, it is necessary for living chimpanzees, and for living gorillas, and it is necessary during the last 50,000 years of our evolution. These are the mix of demographic and selective forces that characterize every close model we have for the evolution of early Homo.

What population structure characterized the African ancestors of the Dmanisi hominins? If we look to the MSA African model, the structure would be one of multiple populations, strongly differentiated, that had existed for hundreds of thousands of years. They may have had adaptations to local ecological conditions, but they were not isolated – they shared genes and one might occasionally replace another, only to re-differentiate as climate fluctuated. The African populations that existed at 1.8 million years ago were probably a modified subset of those that existed 2 million or 2.2 million years ago. Some of these populations would have been morphologically distinctive enough that paleontologists might call them different species. Some of the remixture between them would have been slight, on the scale of Neandertal mixture into today’s human populations. But those cases were at one end of a continuum that included larger amounts of genetic exchange and more rapid turnover. It was a braided stream, in which some of the rivulets were long, but others were short.

You can see where populations like those represented by the Malapa hominins might be important to our evolution in this scenario. The reality that the Malapa sample shares some features with early Homo while contemporary East African specimens share different similarities with early Homo may be a feature of this population structure, not a bug.

If we look later in time, as early Homo existed within and outside Africa, we see the findings outlined by Lordkipanidze and colleagues, and by Van Arsdale and Wolpoff. This pattern of similarities cannot be easily explained under a model of bifurcating change. It implies some degree of continued dispersal and mixture. I expect that the dispersal and mixture were biased in direction, probably more moving out of Africa than back into Africa. The dispersal and mixture happened episodically, not continuously. But viewed within the time-averaged record of 500,000 years, the overall appearance was one of gradual concerted evolutionary change of a highly diverse metapopulation. It should go without saying that this pattern is multiregional evolution; but it is multiregional evolution with an especially episodic rather than continuous nature.

I think that this population structure makes sense of the period from 1.2 million to 800,000 years ago as well. When we look at the classic “Asia versus Africa” question of variation in Homo erectus, the widespread sharing of anatomical variation, the parallel changes in brain size across the entire Homo erectus range, it seems we need dynamics to explain the global pattern of change in concert with substantial regional differentiation and largely overlapping variability.

D4500 and D2600, lateral view in CT reconstruction
D4500 and D2600, lateral view in CT reconstruction

This is my null hypothesis. I have no objection if people want to use names for fossil samples, like Homo habilis or Homo rudolfensis. I just don’t think we should pretend those names predict very much. The names get in the way more than they help.

Four of the five Dmanisi crania, good Homo erectus crania in most of their anatomy, have smaller endocranial volumes than KNM-ER 1470, the type specimen of Homo rudolfensis. The brain size of D4500, the most robust and male-like of the Dmanisi crania, is fully a third smaller than that of KNM-ER 1470. The matching D2600 jaw is almost the size of the Peninj mandible of Australopithecus boisei. The teeth are not robust australopithecine teeth, but they are extraordinarily large for Homo erectus, at least in some ways. That’s not an outlier for Homo erectus, other mandibles, like Sangiran 6, have similar robusticity.

We simply don’t get much mileage out of pretending that these populations evolved in a bifurcating, non-mixing fashion. That simplistic population model requires us to posit far more species than can credibly have existed as independent entities.

So we have learned the alternative: Our evolution was complicated. And now we can get to the business of understanding how and when important things happened.

See also: Adam Van Arsdale’s own thoughts on the skull: “The new (wonderful) Dmanisi skull”.


David Lordkipanidze, Marcia S. Ponce de Len, Ann Margvelashvili, Yoel Rak, G. Philip Rightmire, Abesalom Vekua, and Christoph P. E. Zollikofer. 2013. A Complete Skull from Dmanisi, Georgia, and the Evolutionary Biology of Early Homo. Science 18 October 2013: 342 (6156), 326-331. URL

Van Arsdale, A. P., Wolpoff, M. H., 2013. A single lineage in Early Pleistocene Homo: Size variation continuity in Early Pleistocene Homo crania from East Africa and Georgia. Evolution 67 (3), 841-850. URL