- Email: [email protected]
Problems with the use of cladistic analysis in palaeoanthropology
HOMO Vol. 53/3, pp. 225–234 © 2003 Urban & Fischer Verlag http://www.urbanfischer.de/journals/homo
HOMO
Problems with the use of cladistic analysis in palaeoanthropology D. CURNOE Sterkfontein Research Unit, School of Anatomical Sciences, Medical School, University of the Witwatersrand, 7 Parktown Road, Parktown 2193, Johannesburg, South Africa. Now at: Department of Anatomy, School of Medical Sciences, The University of New South Wales, Sydney NSW 2052, Australia
Summary Cladistic analysis is a popular method for reconstructing evolutionary relationships on the human lineage. However, it has limitations and hidden assumptions that are often not considered by palaeoanthropologists. Some researchers who are opposed to its use regard cladistics as the preferred method for taxonomic «splitters» and claim it has lead to a revitalisation of typology. Typology remains a part of human evolutionary studies, regardless of the acceptance or use of cladistics. The assumption/preference for «splitting» over «lumping» in cladistics (alpha) taxonomy and the general failure to evaluate (posthoc) such taxonomies have served to reinforce this assertion. Researchers have also adopted a number of practices that are logically untenable or introduce considerable error. The evolutionary trend of human encephalisation, apparently isometric with body size, and concurrent reduction in the gut and masticatory apparatus, suggests continuous cladistic characters are biased by problems of body size. The method suffers a logical weakness, or circularity, leading to bias when characters with multiple states are used. Coding of such characters can only be done using prior criteria, and this is usually done using an existing phylogenetic scheme. Another problem with coding character states is the handling of variation within species. While this form of variation is usually ignored by palaeoanthropologists, when characters are recognised as varying, their treatment as a separate state adds considerable error to cladograms. The genetic proximity of humans, chimpanzees and gorillas has important implications for cladistic analyses. It is argued that chimpanzees and gorillas should be treated as ingroup taxa and an alternative outgroup such as orangutans should be used, or an (hypothetical) ancestral body plan developed. Making chimpanzees and gorillas ingroup taxa would considerably enhance the biological utility of anthropological cladograms. All published human cladograms fail to meet standard quality criteria indicating that none of them may be considered reliable. The continuing uncertainty over the number and composition of fossil human species is the largest single source of error for cladistics and human phylogenetic reconstruction.
Introduction Cladistic analysis is a method for reconstructing phylogenetic relationships developed more than 50 years ago by Hennig (1950). It quickly became popular in biology following the translation of Hennig’s method into English (Hennig 1966). It has been credited with having revolutionised approaches to systematics, having brought rigour to this field with its insistence on monophyly (see discussions in 0018-442X/03/53/03–225/$ 15.00/0
226
D. Curnoe
Wiley 1981; Futuyma, 1986, Ridley 1996). It is also an appealing method due to its relative simplicity and apparent objectivity. Cladistic analysis is founded on the principle that the classification of species at or above the genus category should reflect true branching (cladistic) relationships, irrespective of degree of similarity or difference among species. Cladistic classifications must be strictly monophyletic, with only species that share a most recent common ancestor grouped into genera. The concept of cladistics may be seen as reconstruction of phyletic evolution through branching (a succession of speciation «events»). Phyletic sequences are derived by comparing character states across taxa and the polarity of changes in these characters. Groups are «successively defined by sets of shared derived (synapomorphous) characters until a single derived (autaopmorphous) character, uniquely defining a lineage, is identified» (Bilsborough 1992, 13). Cladistics assumes that: (1) a group of taxa share a common evolutionary history and are closely related (by descent from a common ancestor), (2) cladogenesis is accurately represented by a pattern of bifurcation, and (3) characters change within lineages overtime. The first assumption is a general assumption of biology and is relatively uncontroversial. The second is the most controversial and appears to be contradicted by evidence from living organisms and the fossil record about how speciation occurs, for they demonstrate that temporal co-existence of parent and daughter species is common. The final assumption is the most important for cladistics because only character changes may be used to recognise different groups. This assumption is consistent with observations of neontological and fossil species and evolutionary thinking. Eldredge & Tattersall (1975) were among the first to apply cladistics to human evolution. The method has become popular and for many researchers, including the present author, has been the preferred approach to human systematics (eg, Delson et al 1977, Wood & Chamberlain 1986, Skelton et al 1986, Chamberlain & Wood 1987, Stringer 1987, Groves 1989, Wood 1991, Skelton & McHenry 1992, Rightmire 1993, Strait et al 1997, Wood & Collard 1999, Curnoe 2001, 2002). Yet, palaeoanthropology seems to be divided over the validity of this method, for example, Tobias (pers comm.) believes cladistics has lead to a revitalisation of typology. However, «modern typology» (as defined by Simpson 1975), has never left the mainstream of palaeoanthropology regardless of cladistics (eg, see approaches of Tattersall 1986, 1992, Groves 1989 Kimbel & Rak 1993). The plethora of genera and species in use today, mostly defined without the aid of cladistics, is also testimony to this fact (eg, Leakey et al 1995, 2001, White et al 1995, Bermúdez de Castro et al 1997, Hailie-Selassie 2001, Pickford & Senut 2001). There have been only a limited number of reviews of the method and its application to palaeoanthropology (eg, Trinkaus 1992, Tobias 1995, Corruccini 1994, Lieberman et al 1996, Collard & Wood, 2000). For the most part cladistics has become an accepted tool in the phylogenetic kit and with little acknowledgement of its limitations. This paper discusses some important conceptual and practical problems with the application of cladistic analysis to human phylogenetic reconstruction and offers some ways forward for their resolution.
Problems with the use of cladistic analysis
227
Cladistic characters The selection of characters and their coding are the two most important stages of cladistic analysis. The first question that should be asked when selecting characters is «to what extent do the traits employed have any biological relevance, as opposed to being merely frequently preserved and occurring in easily distinguishable character states?» (Trinkaus 1992: 5). Unfortunately, this question seems to be asked infrequently, yet should occupy much of our attention when undertaking systematics in general. Care must be taken to ensure that each character is independent. This is a basic assumption which if violated adds redundancy, causing particular anatomical complexes to be over-represented or characters to be inadvertently duplicated. This has, and continues to be, a source of confusion and potential error in palaeoanthropology (eg, Trinkaus 1992, Strait et al 1997, Skelton & McHenry 1998, Strait & Grine 1998, McCollum 1999, Curnoe 2001). It seems that independence of characters is an unrealistic assumption given the complexity of genetic expression of the phenotype via common phenomena like pleiotropy, epistasis, coadapted gene complexes and polygenic traits. For example, studies of genetic architecture of the mouse mandible by Klingenberg et al (2001) have demonstrated the complexity of its development and evolution. They found that at least 12 quantitative trait loci (QTL) affect size and at least 25 QTL affect shape, however, they suggest that a large number of single loci may have joint effects on both size and shape and found that additive and dominance effects of each QTL on shape are markedly different and «can even affect entirely separate parts of the mandible» (Klingenberg et al 2001, 799). Two types of characters may be used, continuous and discontinuous (or discrete), and they may also introduce error if used inappropriately. Continuous characters are measurable using a continuous scale (eg, millimeters, cubic centimeters, kilograms). The problem with many of these characters relates to the fact that they are very often correlated with body size and should be avoided for cladistics. Some researchers (eg, Strait et al 1997) have argued that the use of some of these features is valid asserting that «among the taxa examined, the largest do not always exhibit the largest character state» and that «australopithecine species and some species of early Homo are quite similar in average body mass» (p 30). The present author has also used the same argument (Curnoe 2001). Under scrutiny these two assumptions do not hold up. Firstly, it appears that brain size increased isometrically with body size over the past 2 million years or so of human evolution (Henneberg 1998), making cranial capacity, which is a popular cladistic character, size dependent. Secondly, the reduction in the gut and masticatory apparatus, which also occurred during this period (Aiello & Wheeler 1995), accounts for a «sizeable portion of the variation in total body size» on the human lineage (Henneberg 1998: 748). This means that masticatory characters are also correlated with body size and in ways not yet understood. Another important problem with such claims is the large errors associated with estimates of fossil human body size making assessment of size dependency a complicated task (De Miguel & Henneberg 1999).
228
D. Curnoe
Discrete characters generally avoid bias in character state determination and some of the problems associated with body size. Those characters that are presence or absence, and having only two states, may be coded relatively objectively and in general require fewer assumptions to be made about how they evolved. Cladistic characters are generally selected more or less arbitrarily and then coded (e.g., «0», «1», «2» etc.) according to whether they appear to be advanced (apomorphic) or ancestral (plesiomorphic) compared to an «outgroup» species. There are two principles that govern the selection of an outgroup (Trueman, pers comm): (1) select a taxon that can be coded for characters that are informative about ingroup relationships, and (2) select a taxon that is «firmly» and definitely an outgroup. Most cladistic studies in human evolution use chimpanzees as the outgroup because they are thought to be closest (morphologically) to the last common ancestor of humans and chimpanzees and are assumed to be informative about ingroup relationships. Given the genetic proximity of humans, chimpanzees and gorillas and strength of the evidence in favour of their congeneric classification (Goodman et al 1989, 1990, 2001, Castresana 2001, Watson et al 2001, Curnoe & Thorne 2002), chimpanzees and gorillas realistically should be ingroup taxa in anthropological cladistics. Because our purpose for doing cladistic analysis is to group species into monophyletic groups at the supraspecific level chimpanzees and gorillas cannot sensibly be considered outgroup taxa. Making them ingroup taxa would considerably enhance the biological meaning of cladograms. An alternative outgroup such as living or fossil orangutans should be used or a hypothetical ancestral ‹body plan› constructed (Wiley 1981). Irrespective of which outgroup is used, the method suffers a logical weakness, or inherent circularity, that leads to bias when characters with multiple states are used. When coding characters, derived states may only be determined as advanced using prior criteria. This is usually done with an existing phylogenetic scheme in mind. For example, the ‹robust› australopithecine taxa all share a suite of features thought to result from diet. These include aspects of the teeth and jaws, musculature and bony cranial attachment points, and aspects of size and shape of the face. They are certainly different to the inferred «ancestral» features as seen in the chimpanzee outgroup, but how can we know whether they are close to or a long way from the ancestral state? That is, how ancestral (plesiomorphic) or how advanced (apomorphic) are they? If the outgroup is coded «0» for a particular state, do we code the states in the ‹robusts› «1», «2» or «3» to take account of the states seen in other taxa? The use of a prior phylogenetic hypothesis adds considerable bias to cladistic analysis because the final tree produced is the result of codes we assign to the characters (combined with sorting of homologies using parsimony). The other problem with this approach is that in determining polarity an assumption is made about how characters evolved between species. Must an advanced character go through intermediate states during its evolution? Or can an advanced character state evolve directly from the ancestral condition? While there are ways of handling such assumptions when building trees, the prior coding of characters makes assumptions about their evolution without knowledge of their actual evolution. Such assumptions seem never to be scrutinised by palaeoanthropologists. Another problem with coding character states is the way variation within species is handled. This form of variation is usually ignored, for one need only examine the
Problems with the use of cladistic analysis
229
data matrices used in palaeoanthropology to see that many of the characters included do vary within species, yet appear as only a single state. Some characters are recognised in some matrices as varying, but the way these are handled is inappropriate and introduces considerable error. Variable characters are sometimes coded as a single character state, often even as the most derived state (e.g., character states are «0» to «4» with variable species coded «5»). For example, Strait et al (1997) studied 60 characters, of which 43 (more than two-thirds) had multiple states and 19 had variable characters coded as a single state. Cladistic analysis then treats «variation» as a separate state, although, it is in fact multiple states which may and usually do include most of the states seen among in-group taxa. Some characters have even been coded as the most derived feature for a character (eg, Strait et al 1997) and treating variation in this way adds considerable error to cladistic analyses. Cladograms are trees of character evolution, therefore, the coding of characters is one of the largest sources of error in cladistics. With around one-third of their characters handled in this way it is difficult to place much confidence in the topologies and inferred relationships of Strait et al (1997). The present author has also made this error (Curnoe 2001, 2002).
Parsimony Trees are usually constructed in cladistic analysis using the principle of parsimony. Parsimony assumes that the best (most accurate) tree is the one with the lowest number of steps or character state changes. It also assumes homologous characters are more common than analogous characters, therefore, the cladogram that requires the fewest steps (character state changes) to build is likely to minimise the number of analogous states. This assumption is violated regularly with a large number of potential homoplasies reported in palaeoanthropology (Wood & Chamberlain 1986, Kimbel et al 1988, Wood 1992, Corruccini 1994, Lieberman et al 1996). These characters reduce the reliability of cladograms and dramatically lower the probability of obtaining correct results. This error may also be compounded «because homoplastic characters often correlate with each other, thereby supporting many alternative false trees from the same set of characters» (Lieberman et al 1996, 99, also, see Sober 1988).
Ordered versus unordered Multistate characters may be treated as ordered or unordered in cladistic analyses. Unordered characters have no plausible evolutionary sequence, and change between any pair of states is counted as one step. Ordered characters follow a plausible evolutionary sequence in which changes between adjacent states are counted as one step (eg, «0» to «1» = 1 step), and changes between non-adjacent states are counted as one+No. skipped states (e.g., «0» to «3» = 4 steps). Human cladograms are sensitive to decisions about treating characters as ordered or unordered (Curnoe 2001). Different treatments result in very different
230
D. Curnoe
trees (hypothetical relationships among species) and the number of more or less equally parsimonious trees.
How reliable are human cladograms? Collard & Wood (2000) also posed this question. They compared phylogenetic trees of craniodental and molecular data for humans, the great apes and a species of Old World monkey. They found the two sets of data produced conflicting cladograms and concluded that «little confidence can be placed in phylogenies generated solely from higher primate craniodental evidence» (2000: 5003). As Curnoe & Thorne (2002) showed, this finding largely reconfirms earlier studies of the decoupling of morphological and molecular evolution. Recent research has also shown that identical genes in humans and apes may exhibit species-specific expression patterns producing different proteins and phenotypes (Enard et al 2002). These findings are consistent with the view expressed by some geneticists about the poor reliability of morphological characters for reconstructing evolutionary history (eg, Nei & Kumar 2000). It should be noted that this line of reasoning extends also to cladistic alpha taxonomy. It has been shown many times that there is no direct relationship between speciation and adaptation, or a discernible change in morphology. Speciation is not the passive consequence of accumulated morphological change (Tattersall 1994), but it involves the disruption of reproductive coherence through the development of isolating mechanisms. Alpha taxonomy is fundamental to cladistics as species must be correctly identified in order that the correct species morphological «patterns» are used for cladistics. Yet, morphology is unreliable for identifying species. Palaeoanthropologists should explore alternative approaches to identifying species and testing current taxonomies (eg, Curnoe & Thorne 2002). Another approach to assessing the reliability of human cladograms is the application of quality criteria. Lieberman et al (1996) have provided four conditions to be satisfied in order that cladistics can be used to provide reliable reconstruction of human evolutionary relationships: (1) Taxonomic units (species) must be correctly identified so that samples contain only one species, (2) Characters used must be heritable, with clear links to the genome, (3) Character states must vary more between taxa than they do within taxa, and (4) Analogous characters (homoplasies) must be sorted from homologous characters by phylogenetic analyses. At present every fossil human species is the subject of controversy. There is also little agreement about which species concept is valid and which method of recognising species from potentially mixed samples is the most appropriate. This means that the first condition is violated in all human cladistic studies. Realistically, this condition will probably never be met because it actually requires prior knowledge about which species have been correctly identified (or actually existed). Some researchers also claim the best (safest) strategy for phylogenetic reconstruction is to recognise too many species rather than too few, that is, «it is better to be a ‹splitter› than a ‹lumper›» (Wiley 1981, Lieberman et al 1996: 99). The jus-
Problems with the use of cladistic analysis
231
tification given is that lower taxonomic units (subspecies) may be recombined to make higher ones after their phylogeny has been resolved (Wiley 1981, Lieberman et al 1996). The problem for palaeoanthropology is that this never seems to be done, with most workers accepting their own alpha taxonomy (or that of others) as reflecting true species diversity. Curnoe (2002) has used topology-dependent permutation probability (T-PTP) tests to examine prior hypotheses about possible monophyly of species. Statistical support was found for the prior hypothesis that the robustus-group and boiseigroup are monophyletic (using both unordered and ordered treatments). In other words, the assumption that they are separate species was refuted by the data matrix used. T-PTP tests provide a useful way of testing prior taxonomic hypotheses and should be used more often in palaeoanthropology. The second criterion seems a sensible condition to meet for it would allow for the selection of characters believed to be largely free from the problems of analogy. It is, however, presently impossible to meet. The central problem of modern genetics is the relationship of genes and characters (Marks 1992), and the fact is «that while genes affect characters, one is equally hard pressed to name a single character in eukaryotic organisms whose genetic basis is understood» (Marks 1992: 237). It is now a decade since Marks made this statement and despite the great progress made in genetics his observation remains relevant. Additionally, genes do not exert their effects in isolation but work via biochemical reactions, thus their effects depend on the chemical and physical milieu in which these reactions take place. Environmental conditions also affect the rate of biochemical reactions, as organisms are constructed of materials derived from their environment. Phenotypic plasticity is an important «problem» for the study of human systematics and one that has been greatly underestimated (Curnoe & Thorne 2002). About the third criterion, it cannot be satisfied without the first and second conditions having been met. It is also, unfortunately, circular in its logic for it depends completely upon the first condition being met. The coding of intraspecific variation as a single character state (see above) also violates this criterion. The fourth criterion is met by using a method such as parsimony analysis, but as Lieberman et al (1996) themselves have shown parsimony is unreliable for this purpose (see above).
Conclusions Cladistic analysis is likely to remain a popular method for reconstructing human evolutionary relationships. It is, however, a method with major limitations and strict assumptions that have not been met in many influential studies in palaeoanthropology. Researchers have also adopted a number of practices in human phylogenetic reconstruction that are logically untenable or introduce considerable error into cladograms. Published cladistic studies of the human lineage have failed to meet standard quality criteria advanced by palaeoanthropologists themselves (eg, Lieberman et al 1996). It can only be concluded therefore that most, if not all, published human cladograms are erroneous. Most importantly, continuing debate over the number and composition of species on the human lineage is the largest single source of uncertainty surrounding
232
D. Curnoe
the use of cladistics. A recent test of morphological taxonomies using molecular data (Curnoe & Thorne 2002) suggests there may have been only five species in a single genus during human evolution. Should this conclusion prove correct cladistics would have little to offer human phylogenetic reconstruction. It may, however, continue to offer opportunities to study character evolution.
Acknowledgements I wish to thank two anonymous referees for critical comments about this paper. I am grateful also to Phillip Tobias and Beverly Kramer for hosting a Post-Doctoral visit to the University of the Witwatersrand. The generous financial support of the University of the Witwatersrand is gratefully acknowledged.
References Aiello LC, Wheeler, P (1995) The expensive-tissue hypothesis. Curr Anthropol 36: 199–221. Bermúdez de Castro JM, Arsuaga JL, Carbonell E, Rosas A, Martínez I, Mosqurera M (1997) A hominid from the Lower Pleistocene of Atapuerca, Spain: possible ancestor to neandertals and modern humans. Science 276: 1392–1394. Bilsborough, A (1992) Human Evolution. Blackie Academic & Professional, UK. Castresana, J (2001) Cytochrome b phylogeny and the taxonomy of great apes and mammals. Mol Biol Evol 18:465–471. Chamberlain, AT, Wood, BA (1987) Early hominid phylogeny. J Hum Evol 16: 119–133. Collard, M, Wood, B (2000) How reliable are human phylogenetic hypotheses? Proc Nat Acad Sci USA 97: 5003–5006. Corruccini, RS (1994) How certain are hominid phylogenies? The role of confidence intervals in cladistics. In Corruccini, RS, Ciochon, RL (eds) Integrative Paths to the Past. Prentice-Hall, New York, 167–183. Curnoe, D (2001) Early Homo from southern Africa: a cladistic perspective. S Afr J Sci 97: 186–190. Curnoe, D (2002) Fossil human taxonomic hypotheses examined with Topology-dependent Permutation Tail Probability (T-PTP) tests. S Afr J Sci 98: 310–312. Curnoe D, Thorne, A (2003) Number of ancestral human species: a molecular perspective. Homo 53: 201–224. De Miguel C, Henneberg M (1999) Variation in hominid body size estimates: do we know how big our ancestors were? Perspect Hum Biol 4: 65–80. Delson E, Eldredge N, Tattersall I (1977) Reconstruction of hominid phylogeny: a testable framework based on cladistic analysis. J Hum Evol 6: 263–278. Eldredge N, Tattersall I (1975) Evolutionary models, phylogenetic reconstruction, and another look at hominid phylogeny. Contra Primat 5: 218–242. Enard W, Khaitovich P, Klose J, Zollner S, Heissig F, Giavlisco P, Nieselt-Struwe K, Muchmore E, Varki A, Ravid R, Doxiadis GM, Bontrop RE, Paabo S (2002) Intra- and interspecific variation in primate gene expression patterns. Science 296: 340. Futuyma, DJ (1986) Evolutionary Biology. Sinauer Associates, Sunderland, Mass. Goodman M, Koop BF, Czelusniak J, Fitch DHA, Tagle DA, Slightom JL (1989) Molecular phylogeny of the family of apes and humans. Genome 31: 316–335. Goodman M, Tagle DA, Fitch DHA, Bailey W, Czelusniak J, Koop BF, Benson P, Slightom JL (1990) Primate evolution at the DNA level and a classification of hominoids. J Mol Evol 30: 260–266. Goodman M, Czelusniak J, Page S, Meireles C (2001) Where DNA sequences place Homo sapiens in a phylogenetic classification of primates. In Tobias PV, Raath MA, Moggi-Cecchi J, Doyle GA (eds) Humanity from African Naissance to Ccoming Millennia. Firenze University Press and Witwatersrand University Press, 279–290. Groves CP (1989) A theory of human and primate evolution. Clarendon Press, Oxford.
Problems with the use of cladistic analysis
233
Hailie-Selassie Y (2001) Late Miocene hominids from the Middle Awash, Ethiopia. Nature 412: 178–181. Henneberg M (1998) Evolution of the human brain: is bigger better? Clin Exper Pharmacol Physiol 25: 745–749. Hennig W (1950) Grundzüge einer Theorie von Phylogenetischer Systematik. Deutscher Zentralverlag, Berlin. Hennig W (1966) Phylogenetic Systematics. University of Illinois Press, Urbana. Kimbel WH, Rak Y (1993) The importance of species taxa in palaeoanthropology and an argument for the phylogenetic concept of the species category. In Kimbel WH, Martin LB (eds) Species, Species Concepts and Primate Evolution. Plenum Press, New York, 461–484. Kimbel WH, White TD, Johanson DC (1988) Implications of KNM-WT 17000 for the evolution of ‘robust’ Australopithecus. In Grine FE (ed) Evolutionary History of the ‘Robust’ Australopithecines. Aldine de Gruyter, New York, 259–268. Klingenberg CP, Leamy LJ, Routman EJ, Cheverud JM (2001) Genetic architecture of mandible shape in mice: effects of quantitative trait loci analysed by geometric morphometrics. Genetics 157: 785–802. Leakey MG, Feibel CS, McDougall I, Walker A (1995) New four-million-year-old hominid species from Kanapoi and Allia Bay, Kenya. Nature 376: 565–571. Leakey MG, Spoor F, Brown FH, Gathogo PN, Kiarie C, Leakey LN, McDougall I (2001) New hominin genus from eastern Africa shows diverse middle Pliocene lineages. Nature 410: 433–440. Lieberman DE, Wood BA, Pilbeam DR (1996) Homoplasy and early Homo: An analysis of the evolutionary relationships of H. habilis sensu stricto and H. rudolfensis. J Hum Evol 30: 97–120. Marks J (1992) Beads and string: the genome in evolutionary theory. In Devor EJ (ed) Molecular Applications in Biological Anthropology. Cambridge University Press, 234–254. McCollum MA (1999) The robust australopithecine face: a morphogenetic perspective. Science 284: 301–305. Nei M, Kumura S (2000) Molecular Evolution and Phylogenetics. Oxford University Press. Pickford M, Senut B (2001) ‘Millennium Ancestor’, a 6-million-year-old bipedal hominid from Kenya. S Afr J Sci 97: 22–27. Ridley M (1996) Evolution. Blackwell Scientific, Boston. Rightmire GP (1993) Variation among early Homo crania from Olduvai Gorge and the Koobi Fora region. Am J Phys Anthrop 90: 1–33. Simpson GG (1975) Recent advances in methods of phylogenetic inference. In Luckett WP, Szalay FS (eds) Phylogeny of the Primates. Plenum Press, London, 2–19. Skelton RR, McHenry HM Drawhorn GM (1986) Phylogenetic analysis of early hominids. Curr Anthrop 27: 21–43. Skelton RR, McHenry HM (1992) Evolutionary relationships among early hominids. J Hum Evol 23: 309–349. Skelton RR, McHenry HM (1998) Trait list bias and a reappraisal of early hominid phylogeny. J Hum Evol 34:109–113. Sober E (1988) Reconstructing the Past: Parsimony, Evolution, and Inference. MIT Press, Cambridge MA. Strait DS, Grine FE, Moniz MA (1997) A reappraisal of early hominid phylogeny. J Hum Evol 32: 17–82. Strait DS, Grine FE (1998) Trait list bias? A reply to Skelton and McHenry. J Hum Evol 34: 115–118. Stringer CB (1987) A numerical cladistic analysis for the genus Homo. J Hum Evol 16: 135–146. Tattersall I (1986) Species recognition in human paleontology. J Hum Evol 15: 165–175. Tattersall I (1992) Species concepts and species identification in human evolution. J Hum Evol 22: 341–349. Tattersall I (1994) How does evolution work? Evol Anthrop 3: 2–3. Tobias PV (1995) The place of Homo erectus in nature with a critique of the cladistic approach. In Bower JRF, Sartono S (eds) Human Evolution in its Ecological Context. Volume 1, Palaeo-anthropology. Pithecanthropus Centennial Foundation, Leiden, 31–46. Trinkaus E (1992) Cladistics and later Pleistocene human evolution. In Bräuer G, Smith FH (eds) Continuity or Replacement? AA Balkema, Rotterdam, 1–7. Watson E, Esteal S, Penny D (2001) Homo Genus: a review of the classification of humans and the great apes. In Tobias PV, Raath MA, Moggi-Cecchi J, Doyle GA (eds) Humanity from African Naissance to Coming Millennia. Firenze University Press and Witwatersrand University Press, 307–318.
234
D. Curnoe
White TD, Suwa G, Asfaw B (1995) Australopithecus ramidus, a new species of early hominid from Aramis, Ethiopia. Nature 371: 306–312. Wiley EO (1981) Phylogenetics. Wiley, New York. Wood BA (1991) Hominid cranial remains. In Leakey REF (ed) Koobi Fora Research Project. Vol. 4. Clarendon Press, Oxford. Wood BA (1992) Early hominid species and speciation. J Hum Evol 22: 351–365. Wood BA, Chamberlain AT (1986) Australopithecus: grade or clade? In Wood BA, Martin L, Andrews P (eds) Major Topics in Primate and Human Evolution. Cambridge University Press, Cambridge, 220–248. Wood BA, Collard M (1999) The human genus. Science 284: 65–71.
Author’s address: Dr. DARREN CURNOE, Sterkfontein Research Unit, School of Anatomical Sciences, Medical School, University of the Witwatersrand, 7 Parktown Road, Parktown 2193, Johannesburg, South Africa. Now at: Department of Anatomy, School of Medical Sciences, The University of New South Wales, Sydney NSW 2052, Australia; e-mail: [email protected] Ms received 26.2.02, accepted 5.5.02, resubmitted 9.9.02
HOMO
Problems with the use of cladistic analysis in palaeoanthropology D. CURNOE Sterkfontein Research Unit, School of Anatomical Sciences, Medical School, University of the Witwatersrand, 7 Parktown Road, Parktown 2193, Johannesburg, South Africa. Now at: Department of Anatomy, School of Medical Sciences, The University of New South Wales, Sydney NSW 2052, Australia
Summary Cladistic analysis is a popular method for reconstructing evolutionary relationships on the human lineage. However, it has limitations and hidden assumptions that are often not considered by palaeoanthropologists. Some researchers who are opposed to its use regard cladistics as the preferred method for taxonomic «splitters» and claim it has lead to a revitalisation of typology. Typology remains a part of human evolutionary studies, regardless of the acceptance or use of cladistics. The assumption/preference for «splitting» over «lumping» in cladistics (alpha) taxonomy and the general failure to evaluate (posthoc) such taxonomies have served to reinforce this assertion. Researchers have also adopted a number of practices that are logically untenable or introduce considerable error. The evolutionary trend of human encephalisation, apparently isometric with body size, and concurrent reduction in the gut and masticatory apparatus, suggests continuous cladistic characters are biased by problems of body size. The method suffers a logical weakness, or circularity, leading to bias when characters with multiple states are used. Coding of such characters can only be done using prior criteria, and this is usually done using an existing phylogenetic scheme. Another problem with coding character states is the handling of variation within species. While this form of variation is usually ignored by palaeoanthropologists, when characters are recognised as varying, their treatment as a separate state adds considerable error to cladograms. The genetic proximity of humans, chimpanzees and gorillas has important implications for cladistic analyses. It is argued that chimpanzees and gorillas should be treated as ingroup taxa and an alternative outgroup such as orangutans should be used, or an (hypothetical) ancestral body plan developed. Making chimpanzees and gorillas ingroup taxa would considerably enhance the biological utility of anthropological cladograms. All published human cladograms fail to meet standard quality criteria indicating that none of them may be considered reliable. The continuing uncertainty over the number and composition of fossil human species is the largest single source of error for cladistics and human phylogenetic reconstruction.
Introduction Cladistic analysis is a method for reconstructing phylogenetic relationships developed more than 50 years ago by Hennig (1950). It quickly became popular in biology following the translation of Hennig’s method into English (Hennig 1966). It has been credited with having revolutionised approaches to systematics, having brought rigour to this field with its insistence on monophyly (see discussions in 0018-442X/03/53/03–225/$ 15.00/0
226
D. Curnoe
Wiley 1981; Futuyma, 1986, Ridley 1996). It is also an appealing method due to its relative simplicity and apparent objectivity. Cladistic analysis is founded on the principle that the classification of species at or above the genus category should reflect true branching (cladistic) relationships, irrespective of degree of similarity or difference among species. Cladistic classifications must be strictly monophyletic, with only species that share a most recent common ancestor grouped into genera. The concept of cladistics may be seen as reconstruction of phyletic evolution through branching (a succession of speciation «events»). Phyletic sequences are derived by comparing character states across taxa and the polarity of changes in these characters. Groups are «successively defined by sets of shared derived (synapomorphous) characters until a single derived (autaopmorphous) character, uniquely defining a lineage, is identified» (Bilsborough 1992, 13). Cladistics assumes that: (1) a group of taxa share a common evolutionary history and are closely related (by descent from a common ancestor), (2) cladogenesis is accurately represented by a pattern of bifurcation, and (3) characters change within lineages overtime. The first assumption is a general assumption of biology and is relatively uncontroversial. The second is the most controversial and appears to be contradicted by evidence from living organisms and the fossil record about how speciation occurs, for they demonstrate that temporal co-existence of parent and daughter species is common. The final assumption is the most important for cladistics because only character changes may be used to recognise different groups. This assumption is consistent with observations of neontological and fossil species and evolutionary thinking. Eldredge & Tattersall (1975) were among the first to apply cladistics to human evolution. The method has become popular and for many researchers, including the present author, has been the preferred approach to human systematics (eg, Delson et al 1977, Wood & Chamberlain 1986, Skelton et al 1986, Chamberlain & Wood 1987, Stringer 1987, Groves 1989, Wood 1991, Skelton & McHenry 1992, Rightmire 1993, Strait et al 1997, Wood & Collard 1999, Curnoe 2001, 2002). Yet, palaeoanthropology seems to be divided over the validity of this method, for example, Tobias (pers comm.) believes cladistics has lead to a revitalisation of typology. However, «modern typology» (as defined by Simpson 1975), has never left the mainstream of palaeoanthropology regardless of cladistics (eg, see approaches of Tattersall 1986, 1992, Groves 1989 Kimbel & Rak 1993). The plethora of genera and species in use today, mostly defined without the aid of cladistics, is also testimony to this fact (eg, Leakey et al 1995, 2001, White et al 1995, Bermúdez de Castro et al 1997, Hailie-Selassie 2001, Pickford & Senut 2001). There have been only a limited number of reviews of the method and its application to palaeoanthropology (eg, Trinkaus 1992, Tobias 1995, Corruccini 1994, Lieberman et al 1996, Collard & Wood, 2000). For the most part cladistics has become an accepted tool in the phylogenetic kit and with little acknowledgement of its limitations. This paper discusses some important conceptual and practical problems with the application of cladistic analysis to human phylogenetic reconstruction and offers some ways forward for their resolution.
Problems with the use of cladistic analysis
227
Cladistic characters The selection of characters and their coding are the two most important stages of cladistic analysis. The first question that should be asked when selecting characters is «to what extent do the traits employed have any biological relevance, as opposed to being merely frequently preserved and occurring in easily distinguishable character states?» (Trinkaus 1992: 5). Unfortunately, this question seems to be asked infrequently, yet should occupy much of our attention when undertaking systematics in general. Care must be taken to ensure that each character is independent. This is a basic assumption which if violated adds redundancy, causing particular anatomical complexes to be over-represented or characters to be inadvertently duplicated. This has, and continues to be, a source of confusion and potential error in palaeoanthropology (eg, Trinkaus 1992, Strait et al 1997, Skelton & McHenry 1998, Strait & Grine 1998, McCollum 1999, Curnoe 2001). It seems that independence of characters is an unrealistic assumption given the complexity of genetic expression of the phenotype via common phenomena like pleiotropy, epistasis, coadapted gene complexes and polygenic traits. For example, studies of genetic architecture of the mouse mandible by Klingenberg et al (2001) have demonstrated the complexity of its development and evolution. They found that at least 12 quantitative trait loci (QTL) affect size and at least 25 QTL affect shape, however, they suggest that a large number of single loci may have joint effects on both size and shape and found that additive and dominance effects of each QTL on shape are markedly different and «can even affect entirely separate parts of the mandible» (Klingenberg et al 2001, 799). Two types of characters may be used, continuous and discontinuous (or discrete), and they may also introduce error if used inappropriately. Continuous characters are measurable using a continuous scale (eg, millimeters, cubic centimeters, kilograms). The problem with many of these characters relates to the fact that they are very often correlated with body size and should be avoided for cladistics. Some researchers (eg, Strait et al 1997) have argued that the use of some of these features is valid asserting that «among the taxa examined, the largest do not always exhibit the largest character state» and that «australopithecine species and some species of early Homo are quite similar in average body mass» (p 30). The present author has also used the same argument (Curnoe 2001). Under scrutiny these two assumptions do not hold up. Firstly, it appears that brain size increased isometrically with body size over the past 2 million years or so of human evolution (Henneberg 1998), making cranial capacity, which is a popular cladistic character, size dependent. Secondly, the reduction in the gut and masticatory apparatus, which also occurred during this period (Aiello & Wheeler 1995), accounts for a «sizeable portion of the variation in total body size» on the human lineage (Henneberg 1998: 748). This means that masticatory characters are also correlated with body size and in ways not yet understood. Another important problem with such claims is the large errors associated with estimates of fossil human body size making assessment of size dependency a complicated task (De Miguel & Henneberg 1999).
228
D. Curnoe
Discrete characters generally avoid bias in character state determination and some of the problems associated with body size. Those characters that are presence or absence, and having only two states, may be coded relatively objectively and in general require fewer assumptions to be made about how they evolved. Cladistic characters are generally selected more or less arbitrarily and then coded (e.g., «0», «1», «2» etc.) according to whether they appear to be advanced (apomorphic) or ancestral (plesiomorphic) compared to an «outgroup» species. There are two principles that govern the selection of an outgroup (Trueman, pers comm): (1) select a taxon that can be coded for characters that are informative about ingroup relationships, and (2) select a taxon that is «firmly» and definitely an outgroup. Most cladistic studies in human evolution use chimpanzees as the outgroup because they are thought to be closest (morphologically) to the last common ancestor of humans and chimpanzees and are assumed to be informative about ingroup relationships. Given the genetic proximity of humans, chimpanzees and gorillas and strength of the evidence in favour of their congeneric classification (Goodman et al 1989, 1990, 2001, Castresana 2001, Watson et al 2001, Curnoe & Thorne 2002), chimpanzees and gorillas realistically should be ingroup taxa in anthropological cladistics. Because our purpose for doing cladistic analysis is to group species into monophyletic groups at the supraspecific level chimpanzees and gorillas cannot sensibly be considered outgroup taxa. Making them ingroup taxa would considerably enhance the biological meaning of cladograms. An alternative outgroup such as living or fossil orangutans should be used or a hypothetical ancestral ‹body plan› constructed (Wiley 1981). Irrespective of which outgroup is used, the method suffers a logical weakness, or inherent circularity, that leads to bias when characters with multiple states are used. When coding characters, derived states may only be determined as advanced using prior criteria. This is usually done with an existing phylogenetic scheme in mind. For example, the ‹robust› australopithecine taxa all share a suite of features thought to result from diet. These include aspects of the teeth and jaws, musculature and bony cranial attachment points, and aspects of size and shape of the face. They are certainly different to the inferred «ancestral» features as seen in the chimpanzee outgroup, but how can we know whether they are close to or a long way from the ancestral state? That is, how ancestral (plesiomorphic) or how advanced (apomorphic) are they? If the outgroup is coded «0» for a particular state, do we code the states in the ‹robusts› «1», «2» or «3» to take account of the states seen in other taxa? The use of a prior phylogenetic hypothesis adds considerable bias to cladistic analysis because the final tree produced is the result of codes we assign to the characters (combined with sorting of homologies using parsimony). The other problem with this approach is that in determining polarity an assumption is made about how characters evolved between species. Must an advanced character go through intermediate states during its evolution? Or can an advanced character state evolve directly from the ancestral condition? While there are ways of handling such assumptions when building trees, the prior coding of characters makes assumptions about their evolution without knowledge of their actual evolution. Such assumptions seem never to be scrutinised by palaeoanthropologists. Another problem with coding character states is the way variation within species is handled. This form of variation is usually ignored, for one need only examine the
Problems with the use of cladistic analysis
229
data matrices used in palaeoanthropology to see that many of the characters included do vary within species, yet appear as only a single state. Some characters are recognised in some matrices as varying, but the way these are handled is inappropriate and introduces considerable error. Variable characters are sometimes coded as a single character state, often even as the most derived state (e.g., character states are «0» to «4» with variable species coded «5»). For example, Strait et al (1997) studied 60 characters, of which 43 (more than two-thirds) had multiple states and 19 had variable characters coded as a single state. Cladistic analysis then treats «variation» as a separate state, although, it is in fact multiple states which may and usually do include most of the states seen among in-group taxa. Some characters have even been coded as the most derived feature for a character (eg, Strait et al 1997) and treating variation in this way adds considerable error to cladistic analyses. Cladograms are trees of character evolution, therefore, the coding of characters is one of the largest sources of error in cladistics. With around one-third of their characters handled in this way it is difficult to place much confidence in the topologies and inferred relationships of Strait et al (1997). The present author has also made this error (Curnoe 2001, 2002).
Parsimony Trees are usually constructed in cladistic analysis using the principle of parsimony. Parsimony assumes that the best (most accurate) tree is the one with the lowest number of steps or character state changes. It also assumes homologous characters are more common than analogous characters, therefore, the cladogram that requires the fewest steps (character state changes) to build is likely to minimise the number of analogous states. This assumption is violated regularly with a large number of potential homoplasies reported in palaeoanthropology (Wood & Chamberlain 1986, Kimbel et al 1988, Wood 1992, Corruccini 1994, Lieberman et al 1996). These characters reduce the reliability of cladograms and dramatically lower the probability of obtaining correct results. This error may also be compounded «because homoplastic characters often correlate with each other, thereby supporting many alternative false trees from the same set of characters» (Lieberman et al 1996, 99, also, see Sober 1988).
Ordered versus unordered Multistate characters may be treated as ordered or unordered in cladistic analyses. Unordered characters have no plausible evolutionary sequence, and change between any pair of states is counted as one step. Ordered characters follow a plausible evolutionary sequence in which changes between adjacent states are counted as one step (eg, «0» to «1» = 1 step), and changes between non-adjacent states are counted as one+No. skipped states (e.g., «0» to «3» = 4 steps). Human cladograms are sensitive to decisions about treating characters as ordered or unordered (Curnoe 2001). Different treatments result in very different
230
D. Curnoe
trees (hypothetical relationships among species) and the number of more or less equally parsimonious trees.
How reliable are human cladograms? Collard & Wood (2000) also posed this question. They compared phylogenetic trees of craniodental and molecular data for humans, the great apes and a species of Old World monkey. They found the two sets of data produced conflicting cladograms and concluded that «little confidence can be placed in phylogenies generated solely from higher primate craniodental evidence» (2000: 5003). As Curnoe & Thorne (2002) showed, this finding largely reconfirms earlier studies of the decoupling of morphological and molecular evolution. Recent research has also shown that identical genes in humans and apes may exhibit species-specific expression patterns producing different proteins and phenotypes (Enard et al 2002). These findings are consistent with the view expressed by some geneticists about the poor reliability of morphological characters for reconstructing evolutionary history (eg, Nei & Kumar 2000). It should be noted that this line of reasoning extends also to cladistic alpha taxonomy. It has been shown many times that there is no direct relationship between speciation and adaptation, or a discernible change in morphology. Speciation is not the passive consequence of accumulated morphological change (Tattersall 1994), but it involves the disruption of reproductive coherence through the development of isolating mechanisms. Alpha taxonomy is fundamental to cladistics as species must be correctly identified in order that the correct species morphological «patterns» are used for cladistics. Yet, morphology is unreliable for identifying species. Palaeoanthropologists should explore alternative approaches to identifying species and testing current taxonomies (eg, Curnoe & Thorne 2002). Another approach to assessing the reliability of human cladograms is the application of quality criteria. Lieberman et al (1996) have provided four conditions to be satisfied in order that cladistics can be used to provide reliable reconstruction of human evolutionary relationships: (1) Taxonomic units (species) must be correctly identified so that samples contain only one species, (2) Characters used must be heritable, with clear links to the genome, (3) Character states must vary more between taxa than they do within taxa, and (4) Analogous characters (homoplasies) must be sorted from homologous characters by phylogenetic analyses. At present every fossil human species is the subject of controversy. There is also little agreement about which species concept is valid and which method of recognising species from potentially mixed samples is the most appropriate. This means that the first condition is violated in all human cladistic studies. Realistically, this condition will probably never be met because it actually requires prior knowledge about which species have been correctly identified (or actually existed). Some researchers also claim the best (safest) strategy for phylogenetic reconstruction is to recognise too many species rather than too few, that is, «it is better to be a ‹splitter› than a ‹lumper›» (Wiley 1981, Lieberman et al 1996: 99). The jus-
Problems with the use of cladistic analysis
231
tification given is that lower taxonomic units (subspecies) may be recombined to make higher ones after their phylogeny has been resolved (Wiley 1981, Lieberman et al 1996). The problem for palaeoanthropology is that this never seems to be done, with most workers accepting their own alpha taxonomy (or that of others) as reflecting true species diversity. Curnoe (2002) has used topology-dependent permutation probability (T-PTP) tests to examine prior hypotheses about possible monophyly of species. Statistical support was found for the prior hypothesis that the robustus-group and boiseigroup are monophyletic (using both unordered and ordered treatments). In other words, the assumption that they are separate species was refuted by the data matrix used. T-PTP tests provide a useful way of testing prior taxonomic hypotheses and should be used more often in palaeoanthropology. The second criterion seems a sensible condition to meet for it would allow for the selection of characters believed to be largely free from the problems of analogy. It is, however, presently impossible to meet. The central problem of modern genetics is the relationship of genes and characters (Marks 1992), and the fact is «that while genes affect characters, one is equally hard pressed to name a single character in eukaryotic organisms whose genetic basis is understood» (Marks 1992: 237). It is now a decade since Marks made this statement and despite the great progress made in genetics his observation remains relevant. Additionally, genes do not exert their effects in isolation but work via biochemical reactions, thus their effects depend on the chemical and physical milieu in which these reactions take place. Environmental conditions also affect the rate of biochemical reactions, as organisms are constructed of materials derived from their environment. Phenotypic plasticity is an important «problem» for the study of human systematics and one that has been greatly underestimated (Curnoe & Thorne 2002). About the third criterion, it cannot be satisfied without the first and second conditions having been met. It is also, unfortunately, circular in its logic for it depends completely upon the first condition being met. The coding of intraspecific variation as a single character state (see above) also violates this criterion. The fourth criterion is met by using a method such as parsimony analysis, but as Lieberman et al (1996) themselves have shown parsimony is unreliable for this purpose (see above).
Conclusions Cladistic analysis is likely to remain a popular method for reconstructing human evolutionary relationships. It is, however, a method with major limitations and strict assumptions that have not been met in many influential studies in palaeoanthropology. Researchers have also adopted a number of practices in human phylogenetic reconstruction that are logically untenable or introduce considerable error into cladograms. Published cladistic studies of the human lineage have failed to meet standard quality criteria advanced by palaeoanthropologists themselves (eg, Lieberman et al 1996). It can only be concluded therefore that most, if not all, published human cladograms are erroneous. Most importantly, continuing debate over the number and composition of species on the human lineage is the largest single source of uncertainty surrounding
232
D. Curnoe
the use of cladistics. A recent test of morphological taxonomies using molecular data (Curnoe & Thorne 2002) suggests there may have been only five species in a single genus during human evolution. Should this conclusion prove correct cladistics would have little to offer human phylogenetic reconstruction. It may, however, continue to offer opportunities to study character evolution.
Acknowledgements I wish to thank two anonymous referees for critical comments about this paper. I am grateful also to Phillip Tobias and Beverly Kramer for hosting a Post-Doctoral visit to the University of the Witwatersrand. The generous financial support of the University of the Witwatersrand is gratefully acknowledged.
References Aiello LC, Wheeler, P (1995) The expensive-tissue hypothesis. Curr Anthropol 36: 199–221. Bermúdez de Castro JM, Arsuaga JL, Carbonell E, Rosas A, Martínez I, Mosqurera M (1997) A hominid from the Lower Pleistocene of Atapuerca, Spain: possible ancestor to neandertals and modern humans. Science 276: 1392–1394. Bilsborough, A (1992) Human Evolution. Blackie Academic & Professional, UK. Castresana, J (2001) Cytochrome b phylogeny and the taxonomy of great apes and mammals. Mol Biol Evol 18:465–471. Chamberlain, AT, Wood, BA (1987) Early hominid phylogeny. J Hum Evol 16: 119–133. Collard, M, Wood, B (2000) How reliable are human phylogenetic hypotheses? Proc Nat Acad Sci USA 97: 5003–5006. Corruccini, RS (1994) How certain are hominid phylogenies? The role of confidence intervals in cladistics. In Corruccini, RS, Ciochon, RL (eds) Integrative Paths to the Past. Prentice-Hall, New York, 167–183. Curnoe, D (2001) Early Homo from southern Africa: a cladistic perspective. S Afr J Sci 97: 186–190. Curnoe, D (2002) Fossil human taxonomic hypotheses examined with Topology-dependent Permutation Tail Probability (T-PTP) tests. S Afr J Sci 98: 310–312. Curnoe D, Thorne, A (2003) Number of ancestral human species: a molecular perspective. Homo 53: 201–224. De Miguel C, Henneberg M (1999) Variation in hominid body size estimates: do we know how big our ancestors were? Perspect Hum Biol 4: 65–80. Delson E, Eldredge N, Tattersall I (1977) Reconstruction of hominid phylogeny: a testable framework based on cladistic analysis. J Hum Evol 6: 263–278. Eldredge N, Tattersall I (1975) Evolutionary models, phylogenetic reconstruction, and another look at hominid phylogeny. Contra Primat 5: 218–242. Enard W, Khaitovich P, Klose J, Zollner S, Heissig F, Giavlisco P, Nieselt-Struwe K, Muchmore E, Varki A, Ravid R, Doxiadis GM, Bontrop RE, Paabo S (2002) Intra- and interspecific variation in primate gene expression patterns. Science 296: 340. Futuyma, DJ (1986) Evolutionary Biology. Sinauer Associates, Sunderland, Mass. Goodman M, Koop BF, Czelusniak J, Fitch DHA, Tagle DA, Slightom JL (1989) Molecular phylogeny of the family of apes and humans. Genome 31: 316–335. Goodman M, Tagle DA, Fitch DHA, Bailey W, Czelusniak J, Koop BF, Benson P, Slightom JL (1990) Primate evolution at the DNA level and a classification of hominoids. J Mol Evol 30: 260–266. Goodman M, Czelusniak J, Page S, Meireles C (2001) Where DNA sequences place Homo sapiens in a phylogenetic classification of primates. In Tobias PV, Raath MA, Moggi-Cecchi J, Doyle GA (eds) Humanity from African Naissance to Ccoming Millennia. Firenze University Press and Witwatersrand University Press, 279–290. Groves CP (1989) A theory of human and primate evolution. Clarendon Press, Oxford.
Problems with the use of cladistic analysis
233
Hailie-Selassie Y (2001) Late Miocene hominids from the Middle Awash, Ethiopia. Nature 412: 178–181. Henneberg M (1998) Evolution of the human brain: is bigger better? Clin Exper Pharmacol Physiol 25: 745–749. Hennig W (1950) Grundzüge einer Theorie von Phylogenetischer Systematik. Deutscher Zentralverlag, Berlin. Hennig W (1966) Phylogenetic Systematics. University of Illinois Press, Urbana. Kimbel WH, Rak Y (1993) The importance of species taxa in palaeoanthropology and an argument for the phylogenetic concept of the species category. In Kimbel WH, Martin LB (eds) Species, Species Concepts and Primate Evolution. Plenum Press, New York, 461–484. Kimbel WH, White TD, Johanson DC (1988) Implications of KNM-WT 17000 for the evolution of ‘robust’ Australopithecus. In Grine FE (ed) Evolutionary History of the ‘Robust’ Australopithecines. Aldine de Gruyter, New York, 259–268. Klingenberg CP, Leamy LJ, Routman EJ, Cheverud JM (2001) Genetic architecture of mandible shape in mice: effects of quantitative trait loci analysed by geometric morphometrics. Genetics 157: 785–802. Leakey MG, Feibel CS, McDougall I, Walker A (1995) New four-million-year-old hominid species from Kanapoi and Allia Bay, Kenya. Nature 376: 565–571. Leakey MG, Spoor F, Brown FH, Gathogo PN, Kiarie C, Leakey LN, McDougall I (2001) New hominin genus from eastern Africa shows diverse middle Pliocene lineages. Nature 410: 433–440. Lieberman DE, Wood BA, Pilbeam DR (1996) Homoplasy and early Homo: An analysis of the evolutionary relationships of H. habilis sensu stricto and H. rudolfensis. J Hum Evol 30: 97–120. Marks J (1992) Beads and string: the genome in evolutionary theory. In Devor EJ (ed) Molecular Applications in Biological Anthropology. Cambridge University Press, 234–254. McCollum MA (1999) The robust australopithecine face: a morphogenetic perspective. Science 284: 301–305. Nei M, Kumura S (2000) Molecular Evolution and Phylogenetics. Oxford University Press. Pickford M, Senut B (2001) ‘Millennium Ancestor’, a 6-million-year-old bipedal hominid from Kenya. S Afr J Sci 97: 22–27. Ridley M (1996) Evolution. Blackwell Scientific, Boston. Rightmire GP (1993) Variation among early Homo crania from Olduvai Gorge and the Koobi Fora region. Am J Phys Anthrop 90: 1–33. Simpson GG (1975) Recent advances in methods of phylogenetic inference. In Luckett WP, Szalay FS (eds) Phylogeny of the Primates. Plenum Press, London, 2–19. Skelton RR, McHenry HM Drawhorn GM (1986) Phylogenetic analysis of early hominids. Curr Anthrop 27: 21–43. Skelton RR, McHenry HM (1992) Evolutionary relationships among early hominids. J Hum Evol 23: 309–349. Skelton RR, McHenry HM (1998) Trait list bias and a reappraisal of early hominid phylogeny. J Hum Evol 34:109–113. Sober E (1988) Reconstructing the Past: Parsimony, Evolution, and Inference. MIT Press, Cambridge MA. Strait DS, Grine FE, Moniz MA (1997) A reappraisal of early hominid phylogeny. J Hum Evol 32: 17–82. Strait DS, Grine FE (1998) Trait list bias? A reply to Skelton and McHenry. J Hum Evol 34: 115–118. Stringer CB (1987) A numerical cladistic analysis for the genus Homo. J Hum Evol 16: 135–146. Tattersall I (1986) Species recognition in human paleontology. J Hum Evol 15: 165–175. Tattersall I (1992) Species concepts and species identification in human evolution. J Hum Evol 22: 341–349. Tattersall I (1994) How does evolution work? Evol Anthrop 3: 2–3. Tobias PV (1995) The place of Homo erectus in nature with a critique of the cladistic approach. In Bower JRF, Sartono S (eds) Human Evolution in its Ecological Context. Volume 1, Palaeo-anthropology. Pithecanthropus Centennial Foundation, Leiden, 31–46. Trinkaus E (1992) Cladistics and later Pleistocene human evolution. In Bräuer G, Smith FH (eds) Continuity or Replacement? AA Balkema, Rotterdam, 1–7. Watson E, Esteal S, Penny D (2001) Homo Genus: a review of the classification of humans and the great apes. In Tobias PV, Raath MA, Moggi-Cecchi J, Doyle GA (eds) Humanity from African Naissance to Coming Millennia. Firenze University Press and Witwatersrand University Press, 307–318.
234
D. Curnoe
White TD, Suwa G, Asfaw B (1995) Australopithecus ramidus, a new species of early hominid from Aramis, Ethiopia. Nature 371: 306–312. Wiley EO (1981) Phylogenetics. Wiley, New York. Wood BA (1991) Hominid cranial remains. In Leakey REF (ed) Koobi Fora Research Project. Vol. 4. Clarendon Press, Oxford. Wood BA (1992) Early hominid species and speciation. J Hum Evol 22: 351–365. Wood BA, Chamberlain AT (1986) Australopithecus: grade or clade? In Wood BA, Martin L, Andrews P (eds) Major Topics in Primate and Human Evolution. Cambridge University Press, Cambridge, 220–248. Wood BA, Collard M (1999) The human genus. Science 284: 65–71.
Author’s address: Dr. DARREN CURNOE, Sterkfontein Research Unit, School of Anatomical Sciences, Medical School, University of the Witwatersrand, 7 Parktown Road, Parktown 2193, Johannesburg, South Africa. Now at: Department of Anatomy, School of Medical Sciences, The University of New South Wales, Sydney NSW 2052, Australia; e-mail: [email protected] Ms received 26.2.02, accepted 5.5.02, resubmitted 9.9.02