Early human evolution and the skulls of Dmanisi


  • Paul Craze

    1. Editor of Trends in Ecology and Evolution and co-organiser of a major symposium on human evolution to be held in 2014 (http://www.cell-symposia-humanevolution.com/)
    Search for more papers by this author


Who were our ancestors? A recently analysed skull has made headlines. Homo habilis, Homo erectus, Heidelberg Man: were they all separate species as believed until now, or variations of just one? Paul Craze takes a look at the variations between and within species that shape the controversy.

As a reader of Significance magazine, you probably have a world-view guided by careful analysis of data and objective consideration of the results. In which case you might find it odd that, according to one of the founders of evolutionary biology, our mathematical ability (and, by implication, the statistical thinking we all hold in such high esteem) can only be explained through an intervention of “the unseen universe of Spirit”. This was the belief of none other than Alfred Russel Wallace, Darwin's co-author on the first scientific paper describing the theory of evolution by natural selection1, whose death occurred exactly 100 years ago.

To be fair, it seems likely that Wallace did not hold such spiritual views at the time of publication. But later in life, when it came to humans, the existence of not only mathematics but also art and music as well as “metaphysical musings”, wit and humour were interpreted by Wallace as evidence of spiritual intervention rather than the action of those natural, biological processes that did perfectly well for other “lesser” species, thank you very much.

Is the story of human evolution to be completely rewritten? Indeed is there a clear story at all?

This belief in the specialness of humans was not restricted to Wallace or the age in which he lived; it is still held with great conviction by many. As well as revealing our inability to be truly objective when it comes to our own biology, it shows something else that is fundamental about us. We are naturally and laudably curious about our own origins and so seize excitedly upon every new discovery as if it were the key piece needed to complete the story of our beginnings. But there lies the problem. As a species, we love a complete, coherent story and mistrust the imperfect and untidy truth; and so if there are ambiguities and gaps in the evidence (and, take it from me, there are), we make short work of filling them in without too many awkward questions being asked – even if, for some, that still means recourse to Wallace's “unseen universe of Spirit”. This makes the story of our origins prone to being “completely re-written”, as the news reports put it, every time an important new fossil is uncovered. So, at a time when just such a fossil is “sending shock waves through the scientific community” and causing an “entire re-think” of human evolution as sections of the media have it, it is a good time to cast a critical statistician's eye over the fragmentary evidence: literally, in most cases, the evidence of fragments. I shall do my best to avoid speculation and stick to what the data can tell us.

The astonishing Dmanisi fossils and what they tell us

The current question of interest is whether early humans found in Africa and Asia were represented by one or by multiple species. At its core is one of the most remarkable human fossils ever discovered (Figure 1). It is a skull, unearthed by anthropologist David Lordkipanidze and his team at Dmanisi, an important site in the Republic of Georgia2. Unusually – extremely unusually – the skull is virtually complete, with even the lower jaw attached. Not only that: it is not obviously distorted by age or disease, nor by the physical and chemical pressures of fossilisation. All of this is even more remarkable when we know its age: at 1.8 million years old it is the earliest evidence of our genus, Homo, outside Africa.

Figure 1.

Dmanisi Skull 5. Courtesy of Guram Bumbiashvili, Georgian National Museum

If it stopped there the skull would still represent an iconic fossil, but it has one more astonishing feature – one that will appeal particularly to statisticians. It is the most recently unearthed of five skulls from the exact same site, differing in fossil age by no more than a few hundred years (Figure 2). All five individuals seem to have met a rather grisly end as the prey of large cats present at the same site. Knowledge, though, can be gained even from misfortune: the habit of these predators of taking their kills into sheltered dens is probably the reason we have the human fossils today. It is the largest sample in existence of early hominins from the same place and time, and so it seems very likely they are a sample from essentially the same deme – that is, a local population in which the individuals interbreed and so share the same distinct gene pool. For hominin fossil-hunters this is a one-off and is something to keep very much in mind. In most other cases, the evidence for early hominins existing in a particular place and at a particular time consists of an even smaller sample, sometimes from as little as one individual, and a very far from complete individual at that. It even used to be said that all the remains we have of early Homo could easily fit into a large suitcase. At least that is now no longer true. A sample size of five from a single, variable deme might disturb a statistician and spark worries about poor representation, but in palaeoanthropological terms, given the very early date, a sample of five is enormous.

Figure 2.

Computer generated images of Skull 5 and four other skulls from Dmanisi. Courtesy of Marcia Ponce de León and Christoph Zollikofer, University of Zurich, Switzerland

Of course, once a sample size is greater than one, it is possible to begin estimating that mainstay of statistical analysis: variation. And that is exactly what this group of researchers has been doing in the eight years since Skull 5 was discovered in 2005. By defining informative landmarks on the skulls – key points from which measurements are taken – and measuring distances between these points, they have generated a description in many variables of each individual, although none are as complete and informative as Skull 5. This has allowed them to use the techniques of geometric morphometrics with the sample. Essentially, geometric morphometrics is the application of statistical methods, most notably principal component analysis3 and cluster analysis, to the shapes and forms of the skulls. It tells us, quantitatively, the degree to which they vary from each other in shape, and where the differences and similarities lie. It tells us which shapes most resemble each other, and which form separate groups. It gives numbers – “morphological distances” – to these differences and plots them out on a graph; and so it puts us on relatively solid statistical ground.

The five Dmanisi skulls are the largest sample in existence of early hominin fossils from the same place and time

The Dmanisi skulls are probably a sample from the same population

First of all, there was the question whether the five skulls from Dmanisi really are likely to come from the same deme. What was needed was an expectation of how much variation would be seen if five individuals were sampled from a known deme. To do this the researchers carried out the same morphometric measurements on a sample of modern humans, Homo sapiens (69 individuals drawn mostly from Australia, Africa and the Americas), and samples from four demes of chimpanzee (one each of three subspecies of Pan troglodytes (N = 29, 20 and 18) and one of 18 Pan paniscus individuals). One advantage of using principal component analysis to study morphology is that it allows variation in size to be controlled for, leaving behind diagrams that capture the much more interesting variation in shape. Once these shape axes were defined, the researchers could calculate the maximum morphological distance between two individuals in the Dmanisi sample – in other words, a measure of the maximum difference in shape. To compare this to the samples of modern humans and chimpanzees, the team used resampling, drawing 1000 samples of five from each of the larger samples and measuring the 1000 maximum shape-distances this produced. The results are fairly clear, in spite of the need to reduce the sample size even further to three for one of the analyses, and some assumptions made to justify using a sample of modern humans from a very much larger area. The variation between the skulls found at Dmanisi is not consistently greater than that seen in demes of either chimpanzees or modern humans, although it is often towards the higher end. In other words, while the conclusion is not certain, this is more likely to be variation of shape within species, rather than between species. It includes variation due to age and gender as well as less systematic variation between individuals. The wide variation we see in skull morphology between people today might well have been present in the past (with the caveat that the samples of humans and chimpanzees used for this came from very much larger geographic areas).

Early Homo: many species or one?

Again, if this was all Skull 5 and its near contemporaries had shown, it would still be a major finding; but now things become even more interesting. The researchers incorporated into their analysis the same data from skulls of other recognised species of Homo, roughly contemporary (in geological terms) with those from Dmanisi but from Africa and Asia. They compared them also with other hominins dated to periods before and after. To fully appreciate the significance of what they might have found, it is useful to have at least a general idea of the timeline of human evolution. Figure 3 gives a guide. Before the genus Homo, and overlapping for some while with it in time and place, was Australopithecus, most notably A. afarensis, represented famously by the iconic Lucy skeleton, discovered in Ethiopia in 1974 and dated to 3.2 million years ago. The exact relationship between Australopithecus and the first species of Homo (probably H. habilis but also possibly a sister species, H. rudolfensis) is still obscure, but a direct evolutionary link seems likely. Fossils of H. habilis have been dated to between 2.3 and 1.4 million years ago, so that they overlapped with the next well-known Homo species, H. erectus (1.8–0.3 million years ago). The Dmanisi skulls are most probably H. erectus and, at 1.8 million years old, are the earliest record of humans outside Africa. Some consider them distinct enough to be a separate species, H. georgicus, although this is strongly debated. H. erectus was extremely successful as a species and spread from its inferred origin in Africa through the Caucasus and Central Asia to the Far East; the famous Java Man, for example, was an example of this species. H. erectus is thought to have been one outcome of a diversification in the Homo genus in Africa that produced a number of proposed species, some of them side-shoots on the evolutionary tree leading to us. The next point of some agreement is the emergence of H. heidelbergensis, from an African or Asian predecessor (possibly H. erectus) around 700 000 years ago, and the most likely direct ancestor of both modern humans, H. sapiens, and our sister species, the Neanderthals, H. neanderthalensis. Recently there have been two exciting discoveries that might take the number of very recent human species to four. These are H. floresiensis, nicknamed hobbits because of their short stature4, and the Denisovans, known from a finger bone, two teeth and a toe bone found in a cave in Siberia5. Even more recently, genetic data have given strong hints of an as yet undiscovered fifth species6. Neanderthals, Denisovans and H. floresiensis all became extinct very recently (within the last 50 000 years) leaving our own species, H. sapiens, as the sole remaining representative of Homo.

Figure 3.

Simplified timeline of human evolution. Ma = million years ago.: Some consider the Dmanisi skulls Georgian Homo erectus to be a separate species; others do not (Skull photos courtesy of Mike Morwood, Karen Baab, Peter Brown and the Kenya National Museums)4

Modern humans have skulls of widely differing shapes. The same is true of ancient humans

The additional measurements included in the geometric shape analysis were from A. afarensis; several of the possible, early African Homo species; H. neanderthalensis; and two skulls likely to represent Homo heidelbergensis and early H. Neanderthalensis respectively; the first from Kabwe (AKA Broken Hill) in Zambia and the other from Steinheim in Germany. The results are striking and are the source of the shock-waves-through-the-scientific-community headlines. There is no clear separation between any of the early Homo species found in Africa, at Dmanisi or further into Asia (Figure 4). There is variation in the size of the face and protrusion of the jaw, but not beyond that expected between individuals within a deme. Instead, with the exception of Skull 4 from Dmanisi (where disease had distorted the face), all of the hominin skulls roughly contemporary in time with those from the Georgia site form a loose cluster. There is a second cluster formed by Neanderthals, and the skulls from Kabwe and Steinheim. What distinguishes these clusters from each other, and from those of chimpanzees at one end and modern humans at the other, is the relative size of the braincase. Ordination plots have to be interpreted with care, of course, since there is a degree of subjectivity in forming clusters, but if an increase in relative neurocranial size can be taken as a proxy for evolutionary time (with chimpanzees obligingly standing in for our pre-australopithecine ancestors) the results seem to support a gradual grading of variable, single demes one into the other (Figure 5A). This raises the startling prospect that the evolutionary history of Homo might be that of a single, variable species changing over time rather than a number of distinct species that were morphologically and genetically limited and distinct, as in Figure 5B.

Figure 4.

Skull shape variation in early Homo, chimpanzees and modern humans. SC1 and SC2 are the first two shape components from geometric morphometric analysis. The two species of chimpanzee plot in the bottom left corner, while modern humans form a cluster in the top right. Between them are the Dmanisi skulls, numbered 1–5, Australopithecus africanus (stars), early Homo from Africa (triangles), H. erectus from Java (diamond), the Kabwe and Steinheim skulls (crosses) and H. neanderthalensis (plus signs)2. From Lordkipanidze et al. (2013) A complete skull from Dmanisi, Georgia, and the evolutionary biology of early Homo. Science, 342, 326–331. Reprinted with permission from AAAS

Figure 5.

Two contrasting hypotheses about the evolution of early Homo. Dots represent fossils, the white ones being the Dmanisi skulls. In hypothesis A there is wide morphological variation within the demes (d1–d4) with a single lineage evolving over time from primitive to derived. In hypothesis B (same dots) there are a number of distinct species with restricted morphological variation2. From Lordkipanidze et al. (2013) A complete skull from Dmanisi, Georgia, and the evolutionary biology of early Homo. Science, 342, 326–331. Reprinted with permission from AAAS

What the new findings mean for our understanding of human evolution

How far can the evidence be taken?

After several decades of research, the emerging prevailing view of human evolution was that shortly after the origin of Homo, there was a rapid branching of species in our East African home, almost as if nature was trying out different ways of being an early human. From this bush of species, H. erectus emerged as particularly successful and expanded its geographic range out of Africa. African H. erectus, meanwhile, produced a number of further species, one branch leading to H. heidelbergensis which gave rise to the later crop of geographically expanded Homo species, of which we are one. If the conclusions of Lordkipanidze and his group are correct, this multiple-species view of human evolution is looking decidedly shaky. The “splittters” – those anthropologists who divide hominin fossils into many species – may have to give way to the “lumpers” – those who say they are all variants of one single species.

We should be cautious, however, and question whether there really is enough evidence to confidently proclaim for one side or the other. Both interpretations are built on small amounts of data. Many putative species are represented by fragments from a small number of individuals in different places. Even the Dmanisi sample currently stands at a statistically inadequate five (and even three for some of the analyses discussed here). Geometric morphometric techniques require a degree of subjectivity, from choice of landmarks to how the data are displayed on the plots. Those sceptical of the results have pointed out that use of landmarks inevitably involves a choice of what to measure and what to leave out – and leaving out crucial diagnostic features, such as dentition, means that real differences do not appear in the results. This can lead to an overemphasis of the similarity between specimens. Also, multivariate analysis with samples of very different size needs to be done with caution as the techniques are inevitably influenced most by the larger samples, which here means the modern human and chimpanzee samples. Our real interest lies in the fossils, but the statistical techniques used do not know that. When it comes to displaying the results, the plot has been set up in a way that maximises separation on the axis representing within-deme variation. There is nothing wrong in doing this if that is what you want to emphasise in the results – but it is not clear if other plots could legitimately tell a different story of clearer separation between specimens already classified as different species. I certainly do not want to imply any deliberate biasing of the results, but some features of the analysis, in particular the subjectivity involved in multivariate analysis, mean that some possible interpretations can be emphasised over others. The stakes are enormously high in studies of human origins.

And finally, we should not forget that we as a species love a good story, and when it comes to science, there is nothing the news media like more than the overturning of a prevailing view leaving current understanding in disarray. Some of the news reports associated with this discovery have claimed exactly that. It seems to happen every week in one branch of science or another. The oft repeated words of caution from palaeoanthropologists that we need more fossils before we can draw any firm conclusions are wise counsel when weighing the evidence of our past. Also, the presence of potentially important species-specific features in the whole of the post-cranium and in details such as inner ear bone shape and tooth morphology has not been considered.

Divergence and migration, but not enough time for full species?

As well as being rightly cautious about the amount and quality of the evidence and any inherent shortcomings of the methods used to generate the results, there is a broader, biological point to consider – and it comes back to our predilection for nice, tidy, complete stories. If the Dmanisi skulls really are all of the same species, they have rightly reminded us of how variable single species can be, even in the same place between members of the same deme. It probably is true that, at least in some cases, palaeoanthropologists have been too eager to name a new find as a new species; the feud in taxonomy between the lumpers, who emphasise intra-specific variation, and the splitters, who tend to classify all new variants as new species, is a long-standing one. But that does not take away from the fact that there is systematic variation in the Homo lineage and it is useful to distinguish, for example, H. habilis from H. erectus. There is a pattern to the variation, both in time and place, and to collapse all of this down into a rather chaotic expression of intra-specific variation, with almost any skull form popping up anywhere, would be a mistake.


Darwin's finches or Galápagos finches, from Darwin's voyage of HMS Beagle, 1845. They have different beak shapes, live on different islands and are classified as belonging to different species, but they can all interbreed. (John Gould, 1804–1881)

Humans love a clear, coherent answer; the truth is that there probably isn't one

What we are probably seeing with Homo is a lineage that did become highly variable, very quickly. But at least some of that variation would have varied systematically from place to place and would have become associated with particular populations living in particular parts of Africa. As long as the individuals in those populations stayed put, they would not interbreed with other, different populations and so the result would have been a loose mosaic of demes. The question is whether these demes could be called separate species. We would like a nice, neat, coherent answer to that question and the truth is there probably is none. Nature is far messier and more pragmatic than we, as a tidy-minded, obsessively classifying species, would like it to be. It is highly likely that these demes would have diverged from each other over time and so, by looking only at morphology, we might conclude they were separate species. However, it is is possible that all humans at that time could, potentially, have interbred, rather like the interbreeding Neanderthals, Denisovans, modern humans and as yet unknown other species of much more recent times6. It can take a surprisingly long time for diverging species to lose the ability to interbreed and, during that time, the point where they can be called separate species is often under some debate. It might be useful at this stage to compare the situation in Homo to that of another famous example of evolution, Darwin's finches. These occur as a group of around 15 species distributed among the islands of the Galápagos archipelago off the west coast of South America. Although the species are generally similar, there are very clear morphological differences between them, especially in their beak shapes (remember the need to capture the important aspects of variation when defining landmarks for morphometric analysis?). Different species tend to occur on different islands, so there is a clear geographical separation. However, the species can interbreed so, in some senses, could be said to form a single, systematically variable species7. The key to this is that it is rare for two birds from geographically separate populations to meet and mate. They potentially can, and do, but it is rare compared to matings within a species. The Galápagos Islands and their resident finches have existed for only around 2–3 million years, and in this case that has not been enough time for the species to lose the ability to interbreed. The best estimates put this time at 21 million years on average for birds8. What we have is a mosaic of similar but clearly recognisable species that are not quite done with the final stages.

The situation with early Homo might well have been something similar. The average time for two diverging mammals to lose the ability to interbreed is 4 million years8, so perhaps there wasn't enough time in the whole of human history for any species to lose the ability to interbreed with any other (which means that, were even the most ancient Homo species alive today, we might have been able to breed with them). Perhaps the populations of early Homo might not have gone quite as far down the speciation road as the Galápagos finches, but the first few steps had been taken. However, within a very short time, shorter than the time needed for the populations to lose the ability to interbreed, H. erectus appears as one of the African demes and, if its expansion to Georgia and beyond is anything to go by, it got around a bit. Unlike the finches, it was not confined to small islands. Large-scale movements of individuals and the genes they carried would have prevented populations diverging too far from one another. They might still have been different but now they would have been genetically connected and so, strictly, part of a single, variable species. The more sedentary populations might also have become more similar to each other through the exchange of genes between them and so less easy to distinguish morphologically, but also more variable within themselves. At the same time, the more freely hybridising H. erectus might have become even more variable through the input of genetic material from a number of other demes; the high variability of the Dmanisi skulls might not be so surprising after all. And there you are, you see? I am doing exactly what I promised myself I would not do – filling in the gaps with speculation. Evolution-based speculation, mark you, but speculation nonetheless. We all want to know the story. None of us has the patience. We need more fossils.

Our understanding of human origins has not been rewritten; it has just become very much more interesting. Variability within species opens all kinds of new possibilities

Questioning the stories

Far from being completely rewritten and overturned, what we understand of human origins has just become very much more interesting. The re-emphasis on variability within species opens up all kinds of exciting new possibilities, questions and hypotheses to explore. The truth might well turn out to involve fewer, more variable species. Contrary to the media hype, this is not something scientists have been “forced to accept” by these new discoveries: many had already considered it possible. Our knowledge of what might have happened in our past has taken a major step forward, but only a step.

At the same time the pattern of our evolution continues to be well informed by what we see in other animals, the Galápagos finches and others, where the data are better, the desperate wish to know is less pronounced and the temptation for some to make a case for exceptionalism does not cloud any judgement. We will continue to speculate and fill in the gaps, but now with better knowledge to draw on. Amidst all this uncertainty, however, one thing is sure. Whatever new discoveries remain in Dmanisi or elsewhere, the surest way we have of giving them their proper weight and not getting carried away with our explanatory stories is to follow good statistical practice and question the validity of the assumptions, the data and the methods of analysis. Realising just how far we can push such limited data is good proof against our eager wish to know the answers, against our desire to be recognised for knowing them and against those who would like to discredit scientific explanations for our origins and replace them with something that invokes an “unseen universe of Spirit”.


The author would like to thank Chris Stringer and Jean-Jacques Hulbin for helpful corrections and comments.