One hominin taxon or two at Malapa Cave? Implications for the origins of Homo

AFFILIATIONS: 1Department of Anatomy and Anthropology, Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel 2Institute of Human Origins and School of Human Evolution and Social Change, Arizona State University, Tempe, Arizona, USA 3Department of Zoology, Tel Aviv University, Tel Aviv, Israel 4Department of Evolutionary Anthropology, Duke University, Durham, North Carolina, USA 5Sports Therapy Department, Ono Academic College, Kiryat Ono, Israel


Introduction
The proposal of a new hominin species, Australopithecus sediba, announced and described by Berger 1 , Berger et al. 2 , and de Ruiter et al. 3 , is based primarily on the analysis of two partial skeletons, MH1 and MH2. The taxon is claimed to exhibit many features that suggest that it represents an intermediate species between Australopithecus africanus and Homo. This assertion was recently reiterated in a special issue of PaleoAnthropology dedicated to Australopithecus sediba. 4,5 However, a careful assessment of the mandibular remains leads us to conclude that the proposed Au. sediba species actually encompasses two species representing separate genera -Australopithecus and Homo -and as such cannot play a role in the origin of the latter. The discovery of two hominin species at one site is not unheard of in South Africa.
The two mandibles from Malapa plainly exhibit different patterns of ramal morphology: MH1 resembles australopith morphology, and MH2 displays the generalised morphology exhibited by Homo sapiens and other Homo species.
The morphology of the ascending ramus of the mandible in hominins has been found to be a diagnostic character 6 (note that Wolpoff and Frayer 7 claim that the upper part of the ramus is not diagnostic enough to distinguish between H. sapiens and H. neanderthalensis, but they cannot refute our argument because they have not applied our method to their sample); as such, the ramal morphology clearly distinguishes between Australopithecus and H. sapiens 8 . In the latter, the condylar and coronoid processes are relatively slender in a lateral view, they are similar in size, and they are separated by a broad, scooped out mandibular (sigmoid) notch, whose deepest point lies about halfway between the tips of the two processes ( Figure 1). This configuration lends the notch a somewhat symmetrical appearance. In Australopithecus, on the other hand, the coronoid process is tall and broad, occupying about three-fourths of the ramal breadth. The process's superior end is rather flat, with a hook-like profile, and overhangs the relatively small mandibular notch, which is shallow in relation to the mandibular condyle. As a result, the outline of the notch is confined and asymmetric.
Similarly, H. sapiens and australopith rami seem to differ vis-à-vis the preangular notch. In H. sapiens, as in many other primates (i.e. the generalised configuration), the concave anterior margin of the ramus forms this notch, which is also present in MH2 (Figure 1). In the australopiths, as in MH1, the anterior margin of the ramus usually slopes diagonally in a straight line until it meets the mandibular body. Some exceptions to this dichotomy can be noted -for example, the presence of the preangular notch on MLD 40, Sts 52, Sts 7 and SKW 5, despite their assignment as Au. africanus. These exceptions somewhat diminish the diagnostic power of the preangular notch.
Because H. sapiens shares its ramal morphology with many other primates (for example, chimpanzees, orangutans, vervets and colobines), that morphology is clearly the primitive one, whereas the Australopithecus ramal configuration is derived -a synapomorphic character that combines Au. robustus, Au. africanus and Au. afarensis (and possibly other australopiths, such as Au. anamensis and Au. boisei, neither of which has a ramus that is sufficiently preserved to permit study) into what seems to be a monophyletic group. To suggest that the derived configuration, that of Australopithecus, evolved into the modern human configuration violates the principle of parsimony.
Although we are convinced that the discrepancies that we have observed in ramal morphology stem from profound biomechanical differences, elucidation of the functionality at play (of the derived configuration) is a major project and beyond the scope of this study. Because the morphological differences are manifested in very young individuals 9,10 , as described later, one can be certain that these morphologies are embedded in the genome and not generated by some activity during an individual's lifetime. In any case, the functional issue has no bearing on the taxonomic question treated here.
In this paper, our aim is not to determine which species of Australopithecus or Homo the Malapa mandibles belong to, but to determine how the two mandibles differ and what those differences mean. To accomplish these goals, we show that the differences are beyond what is expected in a trait's normal range of distribution in a given population. Our null hypothesis is that the two mandibles of Au. sediba represent a single taxon (as claimed, for example, by Berger 1 ; Berger et al. 2 ; de Ruiter et al. 3 ; de Ruiter et al. 5 ; Ritzman et al. 9 ; and Williams et al. 4 ). The alternative hypothesis is that the mandibles of Au. sediba represent a mix of taxa; in this scenario, a statistical analysis would classify one mandible with the Australopithecus cluster (but not provide any species assignment), and the other mandible with the generalised cluster (bearing a shared morphology). Indeed, our evidence supports this alternative hypothesis.

Materials and methods
Our sample includes 115 mandibles from mature extant primates, both male and female (Supplementary table 1 The juvenile specimens help increase the sample and were added after it became apparent that no ontogenetic change occurs in ramal morphology 9,10 ( Figures 2-4). Another young individual, DIK11, from the Ethiopian Dikika site, exhibits the same ramal morphology, as seen on a photograph of the specimen (no cast has been available to us as yet). In addition, one Ardipithecus ramidus ramus, specimen GWM5sw/P56 11 (Figure 3), was included as an unknown. Five Homo fossils (three H. erectus specimens from Choukoutien, restored by Franz Weidenreich 12 ; KNM-WT 15000 -H. ergaster; and ATD696, a mandible from Gran Dolina, Spain) were also analysed, although they proved to be of limited value (see Discussion).
Gorillas were excluded from our analysis. It was demonstrated in a 2007 study that the ramal morphology of gorillas is similar -although not identical -to that of Australopithecus. 8 As noted in that study: given a phylogeny in which chimpanzees and modern humans are sister groups, parsimony dictates that we view the similarity in ramal morphology between Australopithecus afarensis [in fact, all the australopiths that provide ramal evidence] and gorillas as a homoplastic character, a character that appears independently and as such has no phylogenetic value. 8(p.6570) The similarity between the gorilla ramus and that of Au. robustus may well stem from the very tall ramus in both groups.
Regarding reconstruction, MH1 requires none. In MH2, the tip of the coronoid process is damaged; nevertheless, its reconstruction is straightforward, as seen in Figure 5. The three dotted white lines on the superimposed images were added by us. The lines demonstrate that there is no way to reconstruct the coronoid process in MH2 to resemble the robust configuration.   . Note that the upper part of the ramal morphology is the same in the juvenile specimen and its corresponding mature specimen in both pairs. In the Australopithecus mandibles, the coronoid process is taller than the condylar process; the mandibular notch between them is confined and asymmetric; and its deepest point is very close to the condylar process. This configuration is quite different from that of H. sapiens, in which the two processes are the same size in the juvenile and the adult; the mandibular notch is wide; and the deepest point of the notch is midway between the condylar and coronoid processes.    We define variable T as the maximum horizontal distance between the condyle and two-thirds of the anterior ramal margin's height. In this way, we accentuate the most diagnostic part of the ramal outline (A-T). Note that the use of the point defining T (or any other point on the ramus) does not affect the height measurement of the coronoid process in the mandibles under study.
The intersection of each contour with a vertical line and a horizontal line (i.e. coordinates) is assigned a value representing the distance of the intersection point from the zero horizontal line (for example, 10, 20 or 30) ( Figure 6). These are the numerical values used for the statistics. Note that as long as all the contours are on the same grid, units of measure are irrelevant, as is the distance between the lines (provided that it is constant).
We chose the same orientation for the posterior margin of all the rami in our sample because that orientation seems to be fixed in relation to the base of the skull, the Frankfurt horizontal, and the zygomatic arch (indicating functional significance), as demonstrated in Figure 7.
Alternatively, positioning all the mandibles with a horizontal orientation of the occlusal plane or of the base of the mandibular body would introduce variation in the shape of the mandibular notch.
The 20 (AT) variables served as independent variables in a general discriminant analysis to classify two unknown fossil specimens (MH1 and MH2). We used Jump version 15 software for all analyses. General discriminant analysis applies the general linear model approach to discriminant analysis and can use both continuous and categorical independent variables. Our reference classes consisted of Australopithecus, Pan, Pongo pygmaeus and H. sapiens mandibles.
The prior probability of classification was set as equal. The key assumption in discriminant analysis is that the variables used are not completely redundant.
To reduce dimensionality and eliminate the dependence between the variables, we used two independent approaches. First was the best-subset approach. Of 1 048 556 possible models, we inspected the 100 models that accounted for most of the variation (i.e. that exhibited the lowest misclassification rate) (Supplementary figure 1). Out of those models, we selected the one with the least number of parameters and used its functions to predict the state of unknowns. This approach considerably reduces the number of variables in the analysis and keeps the power of classification nearly the same. Thus, the retained variables are those that are most multidimensionally informative for the classification.
Our second approach was a principal component analysis to accommodate the effects of collinearity among the variables. All 20 variables were collapsed into principal components. Five components -those with an eigenvalue greater than or equal to 1 (after varimax rotation) -were retained. We used the components' scores as independent variables in the general discriminant analysis, and the resulting discriminant functions served to classify the unknown fossil specimens. For cross-validation, we applied the leave-one-out procedure. Through the two approaches just described, we classified the unknown fossil specimens. We also reran the analysis under a two-class model: Australopithecus and taxa with a generalised ramus (Pan, Pongo and Homo).

Results
Even in the absence of the coronoid process on MH2, the differences between it and MH1 are readily visible, as was shown in 2010 by Berger et al. 2 themselves (reproduced and modified here in Figure 5).
The unreconstructed outline of the mandibular (sigmoid) notch in MH2 diverges quite clearly from the comparable area in MH1. When the two specimens are adjusted to the same scale ( Figure 6), the deepest part of the notch in MH2 is situated much more anteriorly than in MH1 and descends much farther relative to the zero point, i.e. the mandibular condyle, as in the generalised configuration. The statistical analysis tells us that the difference between the height of the two coronoid processes, reconstructed or not, is of less importance than the outline of the mandibular notch itself and has little effect on the results.
The best-subset approach yielded a classification success ranging from 94.6% to 92.5%. Most of these models share variables A, G, H, I, P, R and T ( Figure 8). The smallest subset model consists of eight effects (A, F, G, I, O, P, R and T; Figure 8), which correctly classify 93.3% of the cases (the leave-one-out cross-validation classification success is 87.4%).
According to the posterior probabilities (p(k i )) from the smallest subset model, MH2 falls in the generalised group (assigned as most likely an orangutan, with p(k i ) = 0.76), whereas MH1 is assigned as most likely Australopithecus (p(k i ) = 1.00). Finally, when only Australopithecus and the generalised ramus group are considered, the two best models consist of a single variable (I or J) that correctly classifies all cases (100%) ( Figure 6). This variable corresponds to the deepest point of the notch in the generalised group; compare with the position of the homologous point in the specialised group ( Figure 6). According to this model, MH1 is assigned to the Australopithecus cluster (p(k i ) = 0.98) and MH2 to the generalised ramus group (p(k i ) = 1.00). The ramus of the Ar. ramidus mandible 11 ( Figure 3) falls in the generalised cluster, with a probability of 0.98.
Note that the size of the H. sapiens sample (41 individuals) is not what counts; rather, the statistical analysis regards the entire generalised sample, consisting of 115 individuals, as one group, because the real issue is whether the Sediba mandibles fall in the generalised cluster, the specialised one, or both.
In the principal component approach, the four factors with an eigenvalue greater than or equal to 1 together accounted for 91.7% of the variation in the data. (The first four principal components accounted for 50%, 17%, 17% and 8% of the variance, respectively, totalling 92%.) The eigenvalues of the factors were 9.9 (49.7%), 3.4 (17.2%), 3.4 (17.2%) and 1.5 (7.6%). All four factors were significantly different from each other (Bartlett test, p < 0.0001 for each of the factors).
The general discriminant analysis correctly classifies 74.2% of cases (with the leave-one-out crossvalidation classification success at 69.7%).
Posterior probabilities of this model assign MH1 as Australopithecus (p(k i ) = 1.0) and MH2 as most likely an orangutan (p(k i ) = 0.755). The latter is in contrast to a probability of 0 as Australopithecus. Ar. ramidus is classified as most likely a chimpanzee (p(k i ) = 0.49) or orangutan (p(k i ) = 0.49). Finally, when only Australopithecus and the generalised ramus group are considered, MH2 and Ar. ramidus are assigned to the generalised ramus group (p(k i ) = 1.00 in both cases) and MH1 is classified as Australopithecus (p(k i ) = 1.00).   The morphological overlap between the comparative taxa is high only in the groups displaying the generalised morphology (not surprisingly, given that 'generalised', by definition, is shared). On the other hand, there is no overlap whatsoever between the generalised group and the derived one ( Figure 6 and Figure 8). Note that no attempt has been made to assign MH1 and MH2 to particular species (because one of these configurations is synapomorphic and the other is generalised). The fact that MH2 is classified as most likely an orangutan is of little relevance, nor does it come as a surprise. What counts is that MH2's generalised configuration puts it in the generalised cluster.

Discussion
The data, as seen in both the distribution of the actual contours ( Figure 6) and the plot of the discriminant analysis (Figure 8) Although we were limited to specimens that are complete enough to be included in our analysis, other, more fragmentary, specimens of Australopithecus (A.L. 333100, A.L. 333w15, A.L. 333n1, A.L. 333108, A.L. 4381g, A.L. 288-1i and DNH 8) exhibit what is undoubtedly the derived configuration of the ramus. Although not a single ramus of Au. africanus is complete enough to be included in the analyses, one can clearly see the derived morphology on a forgotten fragment, Sts 7, that is still embedded in matrix (Supplementary figure 2). (Note that Kimbel and Rak's study 13 of the face of MH1 led them to conclude that the specimen is Au. africanus.) Not surprisingly, the fossil Homo specimens that were included in the sample fall in the same cluster as the generalised hominoids ( Figure 8). Nevertheless, the Homo fossils are of little value to the analysis because in order to serve as an outgroup, they must be assigned to a branch that predates the emergence of the so-called Au. sediba species (i.e. fossils that are nested between the H. sapiens branch and the Australopithecus clade are of no use in this context) -a scenario that we find hard to accept. Ar. ramidus is the only hominin that is helpful in this respect; indeed, like chimpanzees and orangutans, Ar. ramidus displays a generalised configuration of the ramus (Figure 8).
A recent study 9 examines a claim that has been presented in several forums [14][15][16]  The reason that Ritzman et al. 9 cannot clearly distinguish between the gorilla cluster and that of modern humans, for example, nor between MH1 and MH2, is rather simple: in their study's method, a large percentage of the variables (semi-landmarks) are identical in humans, gorillas, and all the other groups in their sample because of the straight anterior outline of the rami in all. In other words, only a small percentage of the semilandmarks are of diagnostic value and worth comparing. The Ritzman et al. 9 analysis is thus incompatible with our analysis, which takes into consideration only the relevant, more diagnostic, part of the anatomy.
Furthermore, the absence of an australopith sample in Ritzman et al.'s 9 study seriously detracts from their conclusions. Indeed, the inclusion of australopiths as a distinct known group in our statistical analysis demonstrates clearly that MH1, when treated as unknown, falls in the australopith cluster, whereas MH2, when treated as unknown, falls in the generalised group, as noted earlier ( Figure 8). (An additional factor affecting the analysis by Ritzman et al. 9 is their inclusion of gorillas, due to the different goals of their study.) The discovery of a hypodigm represented by only two individuals whose morphologies fall at the extreme opposite ends of the range of a population with a normal distribution is highly unlikely. Is it possible that, in keeping with the common primate pattern, the specialised mandible (MH1) represents a male, exhibiting more specialised cranial anatomy, and MH2, with its generalised mandible, represents a female? This scenario is also unlikely, given that primate mandibles usually do not exhibit sexual dimorphism to such magnitude in characters other than size (see also Ritzman et al. 9 ). Even species with pronounced sexual dimorphism, such as the gorilla, demonstrate no sexual dimorphism in ramal shape. The specialised cluster in our study includes specimens that are clearly female (for example, A.L. 822), as well as young individuals, which, although expected to display the generalised anatomy, are nevertheless characterised by a specialised morphology. Most important -even if MH1 and MH2 do represent one highly dimorphic, transitional species -the presence of a specialised mandible (as commonly defined) in that hypodigm would be sufficient to invalidate Berger et al.'s 3 original claim that the proposed species is part of our ancestry. Hence, the credibility of the MH2 reconstruction is of no relevance to the phylogenetic issue once we recognise that the complete MH1 mandible is specialised. (See also Du and Alemseged 17 .) Regarding the question of the Homo species at play, note that the main concern is the distinction between Australopithecus at large and Homo. Hence, we do not deal with nomenclature on a species level; it is sufficient to demonstrate that one of the Malapa mandibles belongs to the genus Homo and the other to the genus Australopithecus -that is, that representatives of both genera existed at Malapa.
The presence at one site of two hominin species of two genera ought not to be surprising. The coexistence of Homo and Australopithecus at South African cave sites has already been documented. In the mid-20th century, a Homo mandible with small teeth was discovered in Swartkrans Cave, which has yielded specimens that are mostly Au. robustus. 18 Later, Clarke and Howell 19 recognised that the Swartkrans specimens SK 80 and SK 47 actually constitute one Homo specimen, SK 847. In the nearby Sterkfontein site, the presence of both Homo and Au. africanus was acknowledged with the discovery of the Homo specimen StW 53 20 , although admittedly, the latter specimen may be several hundred thousand years younger than the Au. africanus specimens from Sterkfontein. The StW 53 skull includes a neglected fragment of a badly preserved ramus, whose morphology supports the claim by Hughes and Tobias 20 that the specimen represents Homo. However, Sts 19, a controversial specimen that emanates from the Sterkfontein type site itself along with numerous Au. africanus specimens, has been identified by some researchers as Homo rather than Australopithecus. [21][22][23][24] The Sterkfontein group of Homo specimens probably includes the immature individual StW 151 as well. 25 Another South African site, Drimolen, provides additional evidence of the coexistence of these two taxa. 26,27 For a meticulous discussion of the chronology of the hominin-bearing layers in South Africa, see Herries et al. 28 and Herries and Shaw 29 .

Conclusions
The phylogenetic scenario that Berger 1 and Berger et al. 2 propose, in which Au. sediba is a link in our evolutionary chain, ought to be ruled out because one of the Malapa mandibles is too derived to be positioned in the human lineage. Furthermore, one need not dismiss in limine the presence of Homo at sites contemporary with Malapa or even earlier, or question their geological age, as does Berger in his discussion of a 2.4-million-year-old Homo specimen par excellence, A.L. 666, from the Afar in Ethiopia. 1 Malapa itself clearly contains a generalised specimen, and Homo, given the size and shape of the fossil, is the only candidate. The split between Homo and the robust clade must have occurred earlier than the occupation of Malapa.
Hence, Au. sediba is not a species that 'shares more derived features with early Homo than any other australopith species', as claimed by Berger et al. 2 In fact, their Au. sediba seems to represent a mixture of two hominin taxa, leading Berger et al. to refer to the new species as a transitional one ( Figure 5). Moreover, by viewing Au. sediba as an ideal link between Au. africanus and H. habilis, they ignore all the synapomorphic features that the former already shares with the robust clade and to which attention was drawn many years ago (for example, by Aguirre 30  The right tree depicts our proposal, in which MH1 lies on the robust clade and MH2 is on the Homo lineage. In addition, an analysis of the MH1 skull indicates that there is no reason to exclude it from the Au. africanus hypodigm. 23 The Au. afarensis mandibular ramus is too derived to allow us to place it on the Homo lineage. We have placed Au. africanus on the robust lineage, based on facial synapomorphies that it shares with the robust australopiths. 32 All the australopiths on which the relevant ramal morphology is preserved (Au. afarensis; Au. africanus, including the Australopithecus specimen at Malapa; and certainly Au. robustus) are actually too derived to play the role of a H. sapiens ancestor. Given that Malapa already contains representatives of two hominin branches, one of which appears to be Homo, we must seek the latter's origin in geological layers that are earlier than those at Malapa, which are dated at approximately 2 million years before present. 33 Support for such a scenario can be found in earlier Ethiopian fossils attributed to the genus Homo: A.L. 666, dated at 2.4 million years 34 , and LD 3501, dated at 2.8 million years 35 .