Morphometric Comparison of Liang Bua 1 ( Homo floresiensis Brown et al. 2004) with Homo habilis and Australasian Homo sapiens

David Bulbeck
School of Archaeology and Anthropology
Arts, The Australian National University

Liang Bua 1, nicknamed the hobbit, may well be the most important discovery in palaeoanthropology in recent years. It is described as a new species of hominin which had survived on the island of Flores in eastern Indonesian until 18,000 years ago (Brown et al. 2004), long after Homo sapiens had crossed the seas of Wallacea to colonise Australia. In its description it is considered most likely to be a small-bodied descendant of H. erectus, but some Australian researchers have related it to H. habilis, and others have suggested it may not be a new species at all but a Homo sapiens microcephalic dwarf ( News Homo floresiensis ). A Minoan microcephalic skull has been the centre of attention in assessing the latter suggestion, but there is a logical problem with making this specimen the crucial test.

If Liang Bua 1 were a microcephalic member of our species Homo sapiens, it would specifically be a microcephalic, adult eastern Indonesian female of 18,000 years antiquity. Morphometric comparisons with microcephalic persons from contexts far away in space and time would therefore be of indirect relevance. A middle Holocene skeleton from the site of Liang Toge may seem more promising for comparison as it was found on the same island and has been described as 'pygmoid' (Jacob 1967:94). However, Liang Toge would have been no smaller than the average female adult in eastern Indonesia today (Vroklage and Lammers 1950), and would have stood head and shoulders above Liang Bua 1, a metre-high dwarf. The question we need to address is whether an 18,000 year old, eastern Indonesian adult female, afflicted by the pathological condition of microcephaly, would be likely to bear a morphological resemblance to Liang Bua 1. Unfortunately, of course, there is no such specimen for comparison (unless it is the hobbit!), so we must make do with the best comparisons we have.

Perhaps the most relevant comparative specimen is a very small female skull from Baa on the island of Rote (near Flores), measured by Michael Pietrusewsky of the University of Hawaii at Manoa, who has kindly sent me his original data. It is recent, and not quite small enough to meet the formal definition of microcephalic (cranial capacity under 700 cubic centimetres), but it is approaching what we are looking for. Another relevant specimen is an 18,000 year old, female adult skeleton from the site of Liang Lemdubu , in the Aru Islands, on the same island chain as Liang Bua 1. The Liang Toge female from Flores, probably dated to the middle Holocene, is also relevant, as is a preceramic (middle Holocene?) unsexed cranium from Leang Buidane, on Talaud Island, north of Sulawesi but still lying within Wallacea. If the hobbit’s only point in common with the H. sapiens specimens listed here is location, if we would need to travel back in time one to two million years to find the most recent common ancestor, then we would not expect these specimens to show any specific resemblances to the hobbit. If they did, we might still need to entertain the possibility of some sort of relationship between the hobbit and eastern Indonesian Homo sapiens.

Figure 1 suggests such a relationship might exist. It is certainly true that Liang Bua 1 is much closer to Homo habilis (KMM-ER 1813 and OH24) than it is to any compared Homo sapiens, both in terms of diminutive size and with regard to shape similarities. Shape features in common include relatively large measurements of the face, and a vault that is low and comparatively broad at the base (see explanation below). On the other hand, Liang Bua 1 is closer to H. sapiens than the H. habilis fossils are, and all of the specimens of Homo sapiens mentioned in the last paragraph are closer to Liang Bua 1 than are any of the recent Australasian samples included in the analysis. Further, amongst the latter, Tasmanians and eastern Indonesians are closer to Liang Bua 1 than are the Melanesian and mainland Australian samples included in the analysis ( Figure 1).

This analysis suggests we should hesitate before endorsing any attempts to disprove the microcephalic status of Liang Bua 1 based on comparisons with microcephalic skulls from faraway places and distant times. It also suggests that when ranges of morphometric variation are introduced to the analysis – a sophistication which I have not yet had time to pursue – the focus should be on H. sapiens skulls from eastern Indonesia (perhaps Tasmania too) and not from elsewhere. This analysis, however, does not address the evidence from mandibular morphology (and certain cranial anatomical traits) which, currently, would make the microcephalic hypothesis difficult to sustain ( News Homo floresiensis ). Nor would it suggest that the interpretation of Liang Bua 1 as a microcephalic H. sapiens would be any more “parsimonious” than its identification as a new, previously unknown species of Homo – after all, Liang Bua 1 would be the first documented case of a microcephalic hunter-gatherer who had survived to adulthood. How likely would this be 18,000 years ago in Flores?

Explanation of the Morphometric Analysis

Readers can replicate Figure 1 using the following procedure. (My own computations used macros I wrote in Lotus Approach and Microsoft Excel, plus a hand calculator for simple algebraic steps.) The cranial measurements include H. habilis data from Tobias (1991) and Wood (1991), Liang Bua 1 (LB1) data from Brown et al. (2004), Liang Toge data from Jacob (1967) and Storm (1995), Liang Lemdubu data from Bulbeck (in press), Leang Buidane data from Bulbeck (1981), data for the Roti female from Pietrusewsky (pers. comm.), and data for recent samples from Pietrusewsky (1985).

Michael Pietrusewsky has not (to my knowledge) published data on female Melanesian and eastern Indonesian samples, so male samples are used here. To control for sex in the analysis, with the Australian populations, both female and male samples are included. Indeed, in the shape analyses males and females always cluster; i.e., the contribution of sex to shape is minimal.

Penrose’s (1954) statistical formulae for size and shape are applied and the results expressed as square roots. By using the square roots, the size gradient from Tasmanian males (with the largest measurements overall) and Liang Bua 1 (with the smallest measurements) can then be expressed along a single dimension ( Figure 1). Treatment of the shape distances, even after expressing them as square roots, requires judicious pre-treatment (see KNM-ER 1813 shape distances and OH24 shape distances). As can be readily observed, the distances involving a single specimen tend to be larger than those involving Pietrusewsky’s samples. Populations are characterised by idiosyncracies only as a result of dramatic genetic drift, whereas individuals (of course!) often have idiosyncracies; for instance, Liang Lemdubu has a short basion-nasion length while the Roti skull is disproportionately large across its auriculare. To compensate for individual idiosyncracies, the shape distances are calibrated.

Calibration expresses all the distances as a proportion of the “expected distance” between any two specimens and/or samples, where the expected distance is the geometric average of the arithmetic averages recorded for these specimens/samples. For instance, in the analysis with KNM-ER 1813, the Liang Lemdubu and Roti skulls have a shape distance of 0.8126, which is low by the standards of either of them, but would be equivalent to a large distance between any two of Pietruswesky’s samples. The calibrated distance between Liang Lemdubu and Roti, 0.8126/(square root of [1.2636 * 1.1114]), or 0.6857, is now of the same order as the calibrated distances between the samples, and correctly identifies the shape similarities between the Liang Lemdubu and Roti specimens. To take another example, the shape distance of 1.9214 between KNM-ER 1813 and Liang Bua 1 l(LB1) ooks large, until we remember that the average distance from specimens cum samples is 2.8079 for KNM-ER1813 and 2.3767 for Liang Bua 1 (LB1). The calibrated shape distance between them, 0.7438 (1.9214/(square root of [2.8079 * 2.3767]), correctly identifies the trenchant morphological similarities between these fossils in the context of their striking differences from certified H. sapiens of Australasia.

The next step in the analysis involves production of a hierarchical tree from the calibrated distances. The nearest-neighbour technique is the precise clustering algorithm used in these analyses, as in my experience it produces better seriations than other clustering algorithms. Hierarchical clustering is not only important in clearly identifying which specimens/samples are closest to each other, but also gives an order to the overall structure of the tree. That is, when we undertake the analytical steps that separate the specimens/samples out from each other (seriation and scaling), we do this in exactly the reverse order to how the specimens/samples had clustered with each other. This procedure is not only logical but also greatly reduces the computational complexities; for instance, seriating the specimens/samples is reduced from being a barely manageable problem of n factorial complexity to one of merely n complexity (cf. Grassmann and Tremblay 1996:320-21).

Seriation of the tree is undertaken to display relationships between specimens/samples that are complementary to the relationships captured by their clustering structure. These relationships include the major morphological axis of morphological contrast implicit in the calculated distances. Consider the tree with KNM-ER 1813. The last clustering event occurred when the branch representing the KNM-ER 1813/LB1 join, clustered with the branch leading to 'everything else'. The seriated order is obvious here; one branch must go one way (left) and the other go the other way (right). Now, also, as the average distance of KNM-ER 1813 from the 'everything else' just referred to is 1.7061, but the corresponding distance for LB1 is merely 1.5588, KNM-ER 1813 obviously goes to the extreme position. With the 'everything else', Liang Toge’s branch was the last to join, so it now hives off. Its average (calibrated) distance from KNM-ER 1813 and LB1 is 1.3916, much less than the average (calibrated) distance of 1.6525 between the other samples/specimens and KNM-ER1813/LB1, so Liang Toge clearly seriates towards KNM-ER1813/LB1. The Roti and Liang Lemdubu specimens, and then the recent samples, are seriated in the same way.

Liang Lemdubu, to be precise, had actually clustered with Tasmanians and recent eastern Indonesians before this group of five clustered with the Melanesian and mainland Australian samples. Since the latter samples, after seriation, occupy the end of the spectrum furthest removed from the KNM-ER 1813…Roti group, then the seriation decision for Liang Lemdubu is whether it is closer to KNM-ER 1813…Roti or to recent Melanesians/mainland Australians (and vice versa for the four-way cluster of Tasmanians and recent eastern Indonesians). This seriation decision is readily resolved in favour of the first alternative. The average distance of Liang Lemdubu from KNM-ER 1813…Roti (1.2739) is less than the average distance of Tasmanians and eastern Indonesians from KNM-ER 1813…Roti (1.3725), while the average distance of Liang Lemdubu from recent Melanesians/mainland Australians (0.9287) is greater than the average distance of Tasmanians and eastern Indonesians from recent Melanesians/mainland Australians (0.7423). A dilemma occurs only when one of the two branches that splits off at a particular step is closer to both ends of the spectrum than the other branch is. Cases like these are resolved by choosing the seriated order that involves the least tension. For instance, after the Northern Territory branch had taken an extreme position in the tree involving KNM-ER 1813, the Melanesian samples (New Caledonia, New Britain, Mysore and Purari) and Murray Valley samples had the average distances shown below:

In a seriation, it would be impossible to represent the Murray Valley samples as being closer than the Melanesian samples to both ends. However, the greater closeness of the Murray Valley samples to the Northern Territory samples is stronger (0.0401) than their greater closeness to KNM-ER 1813…Eastern Indonesia (0.0084). Even though there are only two Northern Territory samples compared to nine KNM-ER 1813…Eastern Indonesia specimens/samples, positioning Murray Valley next to Northern Territorians still produces less tension ((2 * 0.0401 = 0.0802 )> (9 * 0.0084 = 0.0756)).

With the seriated order sorted out, the distances between the specimens and samples are now placed in a half-matrix corresponding to this order (see KNM-ER 1813 seriated shape distances and OH24 seriated shape distances). Sure enough, as we would expect after seriation, the distance between any two specimens/samples closer to each other in the seriation tends to be smaller than the distance between any two specimens/samples farther from each other in the seriation. That is, every distance closer to the diagonal tends to be equal to or less than every distance (along a row or a column) further from the diagonal. In a perfect seriation, of course, with ever step from the diagonal of the half-matrix, the distance in the cell would increase or at least stay the same. Therefore, we can produce a perfect seriation as close as possible to the seriated order we have by transposing those distances, which do not fit the perfect seriation, the least number of steps that are necessary to conform to a perfect seriation. These perfect seriations are shown in the bottom left half-matrices of KNM-ER 1813 seriated shape distances and OH24 seriated shape distances. To measure how well the seriated order shown in Figure 1 conforms to a perfect seriation, we calculate the coefficient of determination between the corresponding values in the two half-matrices in the Excel tables. These coefficients are approximately 90% and 85% (corresponding to Pearson correlation coefficients of 94.7% and 91.9%, respectively). That is, 85-90% of the variance of the shape distances in a perfect seriation of the specimens/samples considered here is explained by the seriated orders shown in Figure 1 .

The trees are scaled (as well as seriated) for two main related reasons. Branch lengths graphically represent the distance of one specimen/sample from another by the minimum distance that has to be traversed between them; and, when a specimen/sample sticks out sharply from the overall trend of the seriated order, this will tend to be shown by a lengthy branch between that specimen/sample and the node representing where it clustered with any other specimen/sample. To scale the trees, simple simultaneous equations are solved in the reverse order of clustering events. For instance, the average distance between KNM-ER1813/LB1 and Liang Toge is 1.3916, between KNM-ER1813/LB1 and the other specimens/samples is 1.6525, and between the latter and Liang Toge is 1.253. From these three values it is simple algebra to determine the lengths of the three branches meeting at a common point. To solve the length of the Roti branch (the next step), we know that the average distance of Roti from the Lembubu…Northern Territory groups is 0.9483, that the average distance of KNM-ER1813, LB1 and Liang Toge from Roti and the 13 Lemdubu …Northern Territory groups is 0.7569, and that the latter are on average 0.214 farther than Roti is from KNM-ER1813/LB1/Liang Toge. Again, three values to determine the length of the three relevant branches (and one branch segment) through simple algebra. Once the lengths between all nodes in the tree are determined, it is a simple matter to calculate the minimum distance along the tree between any pair of specimens/samples, and calculate the coefficient of determination between these distances and the (calibrated) distances that the scaled tree represents. In Figure 1 , these coefficients are circa 90%; i.e., around 90% of the variance of the calculated distances is explained by the branch lengths in Figure 1’s scaled trees.

The coefficients of determination described above compare well with the explanations of variance (commonly, 80-90%) achieved through two-dimensional plots (e.g., PCA, MDSCAL). Two-dimensional plots have the disadvantage of being widely open to inter-observer interpretation. Seriated, scaled trees do not suffer from ambiguity of interpretation – for instance, there is no question that Liang Bua 1 is closer in cranial shape (based on the compared measurements) to certified H. sapiens than the H. habilis fossils are, as determined both through seriation and branch length comparisons.

Looking at Figure 1 , we see that the hobbit’s miniature size is immediately obvious, and its shape relationships are also clear. The habilis fossils KNM-ER 1813 and OH24 cluster with Liang Bua 1 in both trees. Further, the branch between the node where the habilis fossil and Liang Bua 1 connect, and the node where the certified sapiens connect, is the longest branch in both trees. This means that their uniquely shared shape similarities, represented by this branch, explain a greater proportion of the total shape variance than any other branch in the same tree. The specific shape similarities between H. habilis and Liang Bua 1 include large facial breadths relative to braincase measurements (especially basion-bregma height), while the cranial base is broad in relation to its length (including basion-nasion length) and its height, and the frontal bone is constricted (registered by a small minimum frontal breadth). These similarities can be shown through calculating the appropriate indices (see Table 2).

At the same time, the hobbit diverges from the H. habilis pattern towards the sapient morphology, hence its position in the seriated order is less extreme in both trees, and the branch leading to the node with the hobbit is shorter than the branch leading to the H. habilis fossil. Though this trend is modest, it might not be expected if the ancestor of the hobbit had diverged from the ancestor of H. sapiens one to two million years ago. One conceivable explanation for the observed trend is that the hobbit is a H. sapiens dwarf with pathological microcephaly, as this condition is associated with a relatively large face and a vault that is low and comparatively broad at the base (Suzuki 1975). Further, the H. sapiens whose morphology grades towards the habiline/hobbit pattern include all of the ancient and recent eastern Indonesians (as well as Tasmanians). Eastern Indonesian H. sapiens, past and present, tend to have relatively large bi-asterionic and bi-auricular breadths, and relatively small cranial height and minimum frontal breadth. Cranial indices (see Table 2) show how eastern Indonesians and Tasmanians tend to faintly echo the much stronger morphological pattern observed for H. habilis and Liang Bua 1.

References

Brown, P., T. Sutikna, M.J. Morwood, R.P. Soejono, Jatmiko, E. Wahyu Saptomo and Rokus Awe Due (2004) ‘A new small-bodied hominin from the Late Pleistocene of Flores, Indonesia’, Nature 431:1055-61.

Bulbeck, F.D. (1981) Continuities in Southeast Asian Evolution since the Late Pleistocene. Unpublished MA thesis. Canberra: The Australian National University.

Bulbeck, D. (in press). The Late Glacial Maximum human burial from Liang Lemdubu in northern Sahulland. In S. O’Connor, M. Spriggs and P. Veth (eds.), The Archaeology of the Aru Islands, Eastern Indonesia, Modern Quaternary Research in Southeast Asia 19. Canberra: Pandanus Press.

Grassmann, W.K. and J.-P. Tremblay (1996) Logic and Discrete Mathematics: A Computer Science Perspective. Upper Saddle River: New Jersey.

Jacob, T. (1967) Some Problems Pertaining to the Racial History of the Indonesian Region. Utrecht: Netherlands Bureau for Technical Assistance.

Penrose, L.S. (1954) ‘Distance, size and shape’, Annals of Eugenics 18:337-43.

Storm, P. (1995) The Evolutionary Significance of the Wajak Skulls. Scripta Geologica 110.

Suzuki, H. (1975), ‘A case of microcephaly in an Aeneolithic Yayoi period population in Japan’, Bulletin of the National Science Museum, Series D (Anthropology), 1:1-10.

Tobias, P.V. (1991) Olduvai Gorge Volume 4. The Skulls, Endocasts and Teeth of Homo habilis. Cambridge: Cambridge University Press.

Vroklage, B.A.G. and H.J. Lammers (1950) De Physische Anthropologie van de Bevolking van Oost-Dawan (Nord Midden Timor). Leiden: Rijks-Universiteit.

Wood, B.A. (1991) Koobi Fora Research Project Volume 4. Hominid Cranial Remains. Oxford: Cambridge University Press.

And thanks to Debbie Argue, in my School, for help with locating Homo hablis metrical data.