A reappraisal of early hominid phylogeny

David S. Strait Doctoral Program in Anthropological Sciences, State University of New York, Stony Brook, New York, 794-4364, U.S.A. Frederick E. Grine Departments of Anthropology and Anatomy, State University of New York, Stony Brook, New York, 794-4364, U.S.A. Marc A. Moniz Department of Anthropology, Suffolk Community College, 533 College Road, Selden, New York, 784-899, U.S.A. Received 6 December 995 Revision received 6 July 996 Accepted 3 August 996 Keywords: hominid phylogeny, cladistics, Praeanthropus, Paranthropus, Australopithecus, Homo. A reappraisal of early hominid phylogeny We report here on the results of a new cladistic analysis of early hominid relationships. Ingroup taxa included Australopithecus afarensis, Australopithecus africanus, Australopithecus aethiopicus, Australopithecus robustus, Australopithecus boisei, Homo habilis, Homo rudolfensis, Homo ergaster and Homo sapiens. Outgroup taxa included Pan troglodytes and Gorilla gorilla. Sixty craniodental characters were selected for analysis. These were drawn from the trait lists of other studies and our own observations. Eight parsimony analyses were performed that differed with respect to the number of characters examined and the manner in which the characters were treated. Seven employed ordered characters, and included analyses in which () taxa that were variable with respect to a character were coded as having an intermediate state, () characters with variable states in any taxon were excluded; (3) a variable taxon was coded as having the state exhibited by the majority of its hypodigm, (4) variable taxa were coded as missing data for that character, (5) some characters were considered irreversible, (6) masticatory characters were excluded, and (7) characters whose states were unknown in some taxa were excluded. In the final analysis, (8) all characters were unordered. All analyses were performed using PAUP 3.s. Despite the fact that the eight analyses differed with respect to methodology, they produced several consistent results. All agreed that the robust australopithecines form a clade, A. afarensis is the sister taxon of all other hominids, and the genus Australopithecus, as conventionally defined, is paraphyletic. All eight also supported trees in which A. africanus is the sister taxon of a joint Homo+ robust clade, although in one analysis an equally parsimonious topology found A. africanus to be the sister of the robust species. In most analyses, the relationships of A. africanus and H. habilis were unstable, in the sense that their positions vary in trees that are marginally less parsimonious than the favored one. Trees in which robust australopithecines are paraphyletic were found to be extremely unparsimonious. 997 Academic Press Limited Journal of Human Evolution (997) 3, 7 8 Introduction Since the description of Australopithecus afarensis (Johanson et al., 978), there has been a proliferation of hypotheses concerning early hominid phylogeny (Johanson & White, 979; Tobias, 98; White et al., 98; Olson, 98, 985; Kimbel et al., 984; Skelton et al., 986; Wood & Chamberlain, 986; Chamberlain & Wood, 987). The subsequent discovery of KNM-WT 7, and its assignment to Australopithecus aethiopicus (Walker et al., 986; Kimbel et al., 988), has added to this debate. Disagreement has centered on the relationships of A. aethiopicus, Australopithecus africanus, Homo habilis and Homo rudolfensis. Initially, A. aethiopicus was reconstructed as being ancestral to Australopithecus boisei or to both A. boisei and Australopithecus robustus (Delson, 986; Kimbel et al., 988; Walker & Leakey, 988; Grine, 988a). A similar conclusion was reached by Wood (99, 99a,b) following a cladistic analysis, although he did not formally include A. aethiopicus as a member of his ingroup. In contrast, three cladistic analyses that have included this species have challenged this result. Wood (988) examined the characters described by Walker et al. (986), while Skelton & McHenry (99) employed a much larger trait list. Both studies identified the robust australopithecines as a paraphyletic group; A. aethiopicus was identified as the sister of all Correspondence to David S. Strait. 47 484/97/7+66 $5.//hu9697 997 Academic Press Limited

8 D. S. STRAIT ET AL. hominids except A. afarensis, while A. robustus and A. boisei formed a clade that is the sister of Homo. In a more recent study, Lieberman et al. (996) examined the cladistic relationships of H. habilis and H. rudolfensis. Although they did not endorse a tree, the four most parsimonious cladograms that they generated also supported robust australopithecine paraphyly. In all four, A. aethiopicus was reconstructed as being the sister of a clade that includes A. robustus, A. boisei, A. africanus and Homo. The results of these three studies suggest that A. aethiopicus represents a lineage that is distinct from that of other robust species (Skelton & McHenry, 99). Prior to the discovery of KNM-WT 7, A. africanus was widely regarded either as the ancestor of robust australopithecines (Johanson & White, 978; White et al., 98; Rak, 983; Kimbel et al., 985) or as the ancestor of all later hominids (Tobias, 98; Skelton et al., 986). Following its discovery, it was recognized that those relationships might not be tenable (Delson, 986; Walker et al., 986; Grine, 988a; Kimbel et al., 988; Walker & Leakey, 988). Recent cladistic studies have differed markedly concerning the phylogenetic relationships of A. africanus. It has been reconstructed as being the sister of robust species (Chamberlain & Wood, 987), the sister of Homo (Wood, 99, 99a), the sister of a Homo+ robust clade (Wood, 988; Skelton & McHenry, 99), or a species nested within the Homo clade (Lieberman et al., 996). In response to the growing consensus that the H. habilis sensu lato sample may represent more than one species (see Wood, 99b for review), three cladistic studies have addressed the relationships of the early Homo sample. Chamberlain & Wood (987) divided H. habilis sensu lato into two groups that correspond to geographic areas (Olduvai Gorge vs. Koobi Fora), and found evidence that Homo might be paraphyletic. However, this geographic division is not generally accepted. Subsequently, Wood (99, 99a) recognized two morphologically distinct species within H. habilis sensu lato, H. habilis sensu stricto and H. rudolfensis, and found them to be sisters. Recently, Lieberman et al. (996) concluded that they were not sisters. The present study offers a reappraisal of early hominid phylogeny. It differs from prior cladistic studies in four important respects: () the recognition of taxa and their hypodigms i.e., the construction of operational taxonomic units (OTUs), () the use of functional and structural inferences, (3) the choice of characters and the assignment of states i.e., the character analysis, and (4) the configuration of the parsimony analysis. Alpha taxonomy Hominid (ingroup) taxa recognized and employed here include A. afarensis, A. africanus, A. aethiopicus, A. robustus, A. boisei, H. habilis sensu stricto, H. rudolfensis, Homo ergaster, and Homo sapiens (Table ). The generic nomenclature employed in this study is conservative in that only two taxa are recognized, viz. Australopithecus and Homo. This is necessary so as not to presuppose evolutionary relationships among australopithecine taxa. Although we have recognized the validity of, and have employed the nomen Paranthropus elsewhere (Grine, 986, 988a,b; Jungers & Grine, 986; Grine et al., 99; Grine & Susman, 99; Grine & Daegling, 993; Grine & Strait, 994), the studies by Wood (988), Skelton & McHenry (99) and Lieberman et al. (996) have suggested that the species usually attributed to this genus are not monophyletic. If true, then Paranthropus carries with it only a grade (as opposed to a phylogenetically meaningful) connotation. Thus, the approach employed by Tobias (967) will be adopted here, in that all non-homo early hominid species are attributed to Australopithecus. However, the fundamental principle of cladistic classification is that taxonomic names should

EARLY HOMINID PHYLOGENY 9 represent monophyletic groups, and it is possible that this genus is paraphyletic. Although there has yet to be any serious doubt cast upon the monophyletic nature of the genus Homo, it too is possibly paraphyletic. Therefore, the matter of taxonomic nomenclature will be revisited following the analysis of phylogenetic relationships. There are very few instances in the African Plio-Pleistocene fossil record where postcranial remains may be assigned with confidence to a specific taxon, the holotype of which invariably is comprised by cranial, mandibular and/or dental remains. Even in those instances where isolated skeletal elements may be attributed with reasonable assurity to a particular species, it is exceedingly rare to be able to document homologous elements in more than perhaps one or two other taxa. As a result, the inclusion of postcranial skeletal features would have resulted in a data matrix with an inordinate amount of missing information. Because of this, the present analysis does not incorporate postcranial characters, other than to recognize that bipedalism, and the features associated with this mode of locomotion ultimately define the Hominidae (as traditionally recognized). Thus, in this study, only craniodental characters are examined. Similarly, it was not deemed possible to include the recently described species Ardipithecus ramidus (White et al., 994, 995), Australopithecus anamensis (Leakey et al., 995), and Australopithecus bahrelghazali (Brunet et al., 996) because they lack many of the cranial and dental elements employed in this study. The A. afarensis hypodigm employed here includes all of the cranial, mandibular and dental remains from the Laetolil Beds, Tanzania (White, 977, 98), and all of the fossils from the Sidi Hakoma, Denan Dora and Kada Hadar Members of the Hadar Formation (Kimbel et al., 98, 984, 994; White & Johanson, 98; Johanson et al., 98). Also included are the frontal fragment from Behlodelie (White, 984; Asfaw, 987), the mandible from Maka (White et al., 993), and the teeth from Fejej (Fleagle et al., 99). The fragmentary cranial vault and face from the lower part of the Tulu Bor Member of the Koobi Fora Formation (KNM-ER 6) is also accepted as representative of this species (Kimbel, 988). Although Wood (99) has argued that this specimen fits best within A. boisei or A. aethiopicus, he also has noted that it lacks an occipital marginal sinus, which appears to be characteristic of A. boisei and, more importantly, that the possibly associated dental remains have relatively thin enamel (Beynon & Wood, 986). Both A. aethiopicus and A. boisei have very thick tooth enamel (Beynon & Wood, 986; Grine & Martin, 988). Moreover, the presence of deciduous teeth in possible association with this fragmentary cranium suggests it to have been a subadult individual, which would not exclude it from A. afarensis simply because the temporal lines and nuchal ridges fail to merge. The A. africanus hypodigm comprises the specimens from Taung, Members 3 and 4 of the Makapansgat Formation, and Member 4 of the Sterkfontein Formation (Wood, 985). Specimens that have been recovered since the late 96s at Sterkfontein by the Witwatersrand University excavations of calcified Member 4 breccia from the Type Site, as well as some of the specimens recovered from decalcified breccia in the Extension Site (=West Pit) that is of presumed Member 4 equivalence have been included in this analysis. Skelton & McHenry (99) included Sts 5, a very poorly preserved neurocranium with heavily etched fragments of the cranial base, in their sample of A. africanus. It was omitted from the present study, however, because it lacks diagnostic morphology that permits its secure attribution to any of the taxa recognized here. The A. robustus hypodigm includes specimens recovered from Kromdraai B East by Broom and Brain, and by Vrba through her excavation of in situ Member 3 breccia (Grine, 988a). Although Howell (978), Grine (98, 985, 988b) and Jungers & Grine (986) have cited

D. S. STRAIT ET AL. Table Cranial and mandibular specimens included here in the hypodigms of early hominid species A. afarensis: Crania: AL 33-5, 58-, 6-8, 99-, -, 88-, 333-, 333-, 333-45, 333-5, 47-, 444- Garusi KNM-ER 6 Mandibles: AL LH 4 MAK VP-/ 8-3, 45-35, 88-, 98-, 7-3, 66-, 77-, 88-, 3-, 333w-, 333w-, 333w-6, 4-a, 47- A. africanus: Crania: Sts 5, 7,, 6, 67, 7, 5a Stw 3, 73, 5, 55 TM 5, 5 Taung MLD, 6, 9, 37/38 Mandibles: Sts 7, 36, 5b Stw 384, 44, 498, 53 MLD,,, 9, 34, 4, 45 A. aethiopicus: Crania: KNM-WT 7 L 338-y-6 Mandibles: KNM-WT 65 L 55-s-33, 86- Omo 8-967-8, 44-97-466, 57-4-968-4 A. robustus: Crania: SK, 3/4, 46, 47, 48, 49, 5, 55, 65, 79, 83, 848 SKW 8,, 9, 58 SKX 65 TM 57 Mandibles: SK 6,, 3, 34, 586 SKW 5 SKX 4446, 53 TM 57 A. boisei: Crania: OH 5 KNM-ER 45, 46, 47, 73, 733, 375, 3 KNM-WT 74 KNM-CH Omo 33-896 Mandibles: KNM-ER 43, 44, 75, 77, 78, 79, 8, 85, 8, 88, 468, 469, 483, 83, 86, 39, 33, 379, 3954, 549, 5877, 593 KNM-WT 684 L 7a-5, 74a- Natron Table continued on next page

EARLY HOMINID PHYLOGENY Table Continued from previous page H. habilis: Crania: OH 7, 3, 4, 6 KNM-ER 85, 83, 478, 3735 Sts 9 Stw 53 SK 7, 847 L 894- Mandibles: OH 7, 3 KNM-ER 5, 5, 85 SK 5, 45 H. rudolfensis: Crania: KNM-ER 47, 59, 373, 389 Mandibles: KNM-ER 89, 48, 483, 8, 8 UR 5 H. ergaster: Crania: KNM-ER 3733, 3883 KNM-WT 5 Mandibles: KNM-ER 73, 8, 99, 57 KNM-WT 5 Isolated teeth and specimens from which only dental measurements are taken were not included. features primarily subtle dental differences that may support a specific distinction between the robust australopith samples from Kromdraai B East and the Member Hanging Remnant of the Swartkrans Formation, this is certainly a minority opinion. Most authorities continue to regard these fossils as comprising a single species. This view is adopted here, principally because specimens from these two sites do not differ in the characters that were employed in the present analysis. Thus, fossils from Members, and 3 of the Swartkrans Formation are included in the A. robustus hypodigm (Grine, 988b; Grine & Daegling, 993; Grine & Strait, 994). The A. boisei hypodigm includes specimens from Olduvai Gorge Beds I and II, the Humba Formation at Lake Natron, the Chemoigut Formation at Chesowanja, the Upper Burgi, KBS and Okote Members of the Koobi Fora Formation, Members G through L of the Shungura Formation, and the Kaitio Member of the Nachukui Formation (Grine, 98; Walker & Leakey, 988; Wood, 99; Brown et al. 993; Wood et al., 994). This cranial, dental and mandibular sample conforms to that attributed by Wood et al. (994) to Paranthropus boisei sensu stricto. In particular, the Omo 33-896 cranium from Member G (G6-7) of the Shungura is here considered to belong to that taxon. The A. aethiopicus hypodigm includes the KNM-WT 7 cranium and KNM-WT 65 mandible from the Lokalalei Member of the Nachukui Formation, and the Omo 44-97- 466 mandible, the L 338-y-6 partial cranium, and the Omo 57-4-968-4 mandible from Member E (Units E-, E-3 and E-4 respectively) of the Shungura Formation. It also includes the L 86- mandible from Member F (F-) and the L 55-s-33 and Omo 8-967-8 mandibles from Member C (C-6 and C-8) among the more complete specimens. Additionally,

D. S. STRAIT ET AL. isolated teeth principally molars and premolars that have been identified by Suwa (988) as a robust australopithecine less derived than A. boisei, and by Wood et al. (994) as Paranthropus aff. P. boisei are included here in the hypodigm of A. aethiopicus. Among the isolated teeth so attributed are the deciduous molars (L 64- and L 74-) from Member D that were initially attributed to A. boisei by Grine (985). The A. aethiopicus hypodigm comprises robust australopithecine specimens that derive from strata in the Turkana Basin that predate Member G of the Shungura Formation. This attribution follows upon and is consistent with the studies by Rak & Howell (978), Walker et al. (986), Holloway (988), Suwa (988), and Wood et al. (994). In addition, Suwa (988) has identified robust australopithecine teeth that differ from those of A. boisei within the lower units of Member G. The specific attribution of the East African early Homo specimens follows that of Wood (99a). Thus, three species (H. habilis, H. rudolfensis and H. ergaster) are recognized for the Late Pliocene and Early Pleistocene gracile hominid fossils from Ethiopia, Tanzania and Kenya. Because of the controversy surrounding the attribution of the KNM-BC temporal from the Chemeron Formation (Hill et al., 99; Falk & Baker, 99; Tobias, 993) we have refrained from assigning it to any specific hypodigm and it is not considered here. The Stw 53 cranium from Member 5 of the Sterkfontein Formation is tentatively attributed to H. habilis following suggestions by Tobias (978, 99), although recent analyses indicate that it may represent a separate species (Grine et al. 993, 996). Unlike Skelton & McHenry (99), who recognized the Sts 9 basicranium from Sterkfontein as a specimen of A. africanus, we assign it to Homo following the analysis of Kimbel & Rak (993). More specifically, it is referred here to H. habilis, because that is the species to which all other Sterkfontein specimens of Homo have been attributed (Tobias, 99). This specific attribution may also require revision. Furthermore, although the SK 847 cranium from the Member Hanging Remnant of the Swartkrans Formation has been seen to represent early H. erectus (=H. ergaster) by some workers (e.g., Walker, 98; Clarke, 985), it is here tentatively attributed to H. habilis following upon the studies by Howell (978), Chamberlain (987, 989) and Grine et al. (993, 996). The outgroup taxa employed in the present analysis are Pan troglodytes and Gorilla gorilla. These species are appropriate because it is (almost universally) accepted that one or both are the closest extant relatives of hominids. Ten males and ten females of each species were sampled from the Mammalogy and Anthropology collections of the American Museum of Natural History. Although the phylogenetic relationships of Pan and Gorilla are the subject of considerable debate (e.g., Andrews & Martin, 987; Begun, 994; Goodman et al., 994; Marks, 994; Ruvolo, 994), a preponderance of genetic studies indicate that Pan is the sister taxon of humans and early hominids, and this topology is assumed here. Functional morphology in phylogenetic analysis The role of functional morphology in phylogenetic analysis has been the subject of considerable debate (e.g., Fischer, 98; Cracraft, 98; Szalay, 98a,b, 98; Skelton et al., 986; Andrews & Martin, 987; Begun, 99; Skelton & McHenry, 99). Within the framework of numerical cladistics, functional morphology is relevant because functionally related characters violate the assumption of character independence that is implicit in a cladistic analysis (Farris, 983; Kluge, 989). The bias that results from such character redundance is removed when functionally related traits are treated as a unit (i.e., as a single character or complex). Although several studies of hominoid phylogeny have appealed to this logic (Skelton et al., 986; Skelton & McHenry, 99; Begun, 99), not all of the

EARLY HOMINID PHYLOGENY 3 functional hypotheses to which they ascribed have been rigorously tested. Such an omission is risky, because the refutation of those hypotheses may undermine the validity of any phylogeny based upon them. A prudent approach, therefore, would be to test functional hypotheses prior to a phylogenetic analysis. Until this is accomplished, however, characters alleged to be functionally or structurally related should be treated as independent traits. Basicranial characters provide an excellent example. It has been suggested that A. boisei, A. robustus and members of the genus Homo share a number of presumably derived basicranial character states. These include a flexed cranial base, coronally oriented petrous bones, a horizontal foramen magnum, a deep glenoid fossa with a steep articular eminence, and a small post-glenoid process that is fused to the tympanic (e.g., DuBrul, 977; Dean & Wood, 98, 98; White et al., 98; Kimbel et al., 984; Dean, 986, 988a). These characters may indicate a sister group relationship between these species, as has been suggested by Skelton et al. (986) and Skelton & McHenry (99), or that the cranial base was characterized by considerable homoplasy during the course of hominid evolution. However, it has also been claimed that some or all of these characters are either functionally or structurally related to one of several factors, including brain or cerebellar size (e.g., Biegert, 963; Gould, 977; Dean, 986, 988a,b; Ross & Ravosa, 993), degree of facial prognathism (e.g., Scott, 958; Kimbel et al., 984), facial orientation (e.g., Cameron, 94; Enlow, 975), posture (e.g., Dart, 95; Schultz, 94, 955; DuBrul, 95; Ashton & Zuckerman, 95, 95, 956; DuBrul & Laskin, 96), the size of the masticatory apparatus (Biegert, 963), and vocalization (e.g., Laitman et al., 978, 979; Laitman & Heimbuch, 98; Lieberman, 984). If any of these hypotheses can be supported, then arguably a number of basicranial features should be treated as a single character. Support for a Homo+A. boisei+a. robustus clade would therefore be weakened. To date, few studies have tested competing hypotheses (Ross & Ravosa, 993; Ross & Henneberg, 995; Strait, 994). However, studies by Ross and colleagues refer to only a single character, namely basicranial flexion, and although the study by Strait (994) examined a broader range of basicranial characters, it was only of a preliminary nature. Thus, at present it is unclear which of the factors listed above (if any) is a primary influence on basicranial form, and which basicranial characters (if any) might be so highly related to such a factor as to evolve as a unit. Until these relationships are established, subjective attempts to group such characters into complexes for the purposes of phylogenetic analysis are likely to be either incorrect or, at the very least, the subject of considerable disagreement. Recently, Lieberman (995) and Lieberman et al. (996) have proposed another approach by which functional morphology may be incorporated into a cladistic analysis. According to this method, the only characters that should be employed in phylogeny reconstruction are those that are unlikely to be influenced by epigenetic factors, and that also produce developmentally homologous states in different taxa. These criteria, however, can be applied only when the developmental biology of a character has been established experimentally. Because this condition has not been met for many of the characters examined in this study, these two criteria were not used in character selection. Thus, because the functional, structural and developmental relationships among many cranial characters are poorly understood, this study does not attempt to conflate characters or construct complexes using functional inferences, with one exception. That exception relates to masticatory features. Skelton et al. (986) and Skelton & McHenry (99) have argued that characters related to mastication contribute disproportionately to, and thus bias reconstructions of hominid phylogeny (all of which are based upon a consideration of craniodental characters). In particular, they concluded that masticatory features unfairly link A. africanus

4 D. S. STRAIT ET AL. with the robust australopithecines (Skelton et al., 986), and unfairly supported robust australopithecine monophyly (Skelton & McHenry, 99). We are reluctant to conflate characters or construct complexes using untested functional inferences, but because of their assertions, we have undertaken an analysis in which features plausibly related to trophic adaptation were eliminated from consideration (see below). This was necessary for two reasons. In the first instance, had these characters not been omitted, we would have failed to address one of the more important conclusions reached by Skelton et al. (986) and Skelton & McHenry (99). Second, one of us (Grine, 988a) has argued in favor of robust australopithecine monophyly, and the removal of presumed masticatory features provides a conservative test of this hypothesis. Character analysis Coded character states are the raw data of any parsimony analysis. For this reason, the character analysis is arguably the most important stage in a cladistic study. Skelton & McHenry s (99) character analysis presents the most comprehensive summary to date of characters traditionally cited in studies of early hominid phylogeny. Because the characters that they employed, and many of the states that they recognized were drawn extensively from the literature, Skelton & McHenry s (99) character analysis served as a sensible starting point for the present study [as it has for other studies, e.g., Lieberman et al. (996)]. Each character employed by them, and a number drawn from other sources (e.g., Clarke, 977; Walker et al., 986; Kimbel et al., 984, 988; Wood, 99, 99a; Wood et al., 994), as well as our own observations, were examined here. One hundred and one craniodental characters were examined, of which 6 were selected for use in the present study (Table, Appendices and ). The present character analysis differs from that of Skelton & McHenry (99) in several ways. A number of characters were added that were omitted from that study. In addition, several characters that were employed by them were rejected because they failed to discriminate among taxa, or were considered to be invalid. In other instances, features employed by them were replaced by another character that described the same morphology in a very different manner. Frequently, their characters were accepted, but modified such that one or several states were changed. Furthermore, a concerted effort was made to eliminate characters that redundantly describe the same morphological feature. Such characters, which are abundant in the literature, affect a phylogenetic analysis in the same manner as functionally related characters: they unfairly increase the weight of what should be only a single trait. For instance, Kimbel et al. (984) described two features of the zygomatic (the height of the masseter origin and the shape of the zygomaticoalveolar crest) that they recognized as being necessary correlates of each other: the origin cannot be high unless the zygomatic rises superolaterally. Accordingly, the masseter origin in robust australopithecines and A. africanus is high, and these species also have a straight zygomaticoalveolar crest. In contrast, A. afarensis and Homo have a low masseter origin, which is made possible by a strongly arched crest. Kimbel et al. (984) treated these two features as a single character, because they essentially describe the same trait. Similarly, in the present study, any such group of features was expressed as a single character (Appendix ). The elimination of descriptively redundant characters is fundamentally different from the conflation of functionally related ones. For instance, cranial base flexion and the position of the foramen magnum may be functionally or structurally related to each other (e.g., DuBrul, 95;

EARLY HOMINID PHYLOGENY 5 Table Characters and the distribution of their states No. Character Pan/Gorilla A. afarensis A. aethiopicus A. africanus A. robustus A. boisei H. habilis H. rudolfensis H. ergaster H. sapiens. Projection of nasal bones above frontomaxillary suture. Inferior orbital margin rounded laterally 3. Infraorbital foramen location Projected, tapered Pan: No Gorilla: Variable High 4. Anterior pillars 5. Nasoalveolar clivus contour in coronal plane 6. Protrusion of incisor alveoli beyond bicanine line (basal view) Convex Yes 7. Nasal cavity entrance Stepped 8. Palate thickness Thin 9. Height of the masseter origin. M-L thickness of zygomatic arch at root of frontal process. Anterior projection of zygomatic bone relative to piriform aperture (dishing). Anterior palatal depth Low Thin Posterior Shallow Projected, expanded No High Convex Yes Stepped Thin Low Thin Posterior Shallow Projected, expanded Yes Low Concave (gutter) No Smooth, overlap Thick High Thick 3 Anterior (dished) Shallow 3 Variable No Variable Variable Straight Yes Stepped Thin High Thin Variable posteriorinterior Deep (shelved) Projected, expanded Yes Low Present Concave (gutter) No Smooth, overlap Thick High Thick 3 Anterior (dished) Shallow Projected, expanded No Low Concave (gutter) No Smooth, overlap Thick High Thick 3 Anterior (dished) Deep (shelved) Not projected No High Variable Straight Yes Variable Thin Low Thin Posterior Variable Not projected No Not projected No? High Straight No Stepped Thin Low Straight Yes Stepped Thin Low? Thin Intermediate Deep (shelved) Posterior Deep (shelved) Not projected No High Convex Yes 3 Smooth, no overlap Thin Low Thin Posterior Deep (shelved)

6 D. S. STRAIT ET AL. Table Continued from previous page No. Character Pan/Gorilla A. afarensis A. aethiopicus A. africanus A. robustus A. boisei H. habilis H. rudolfensis H. ergaster H. sapiens 3. Index of palate protrusion anterior to sellion (facial prognathism) 4. Masseteric position relative to sellion 5. Maxillary trigon (zygomaticomaxillary step) Prognathic At or posterior Prognathic At or anterior Prognathic At or anterior Present Variable prognathic mesognathic At or posterior Mesognathic At or anterior Present Mesognathic At or anterior Mesognathic At or posterior Mesognathic 3 Orthognathic? At or posterior 3 Orthognathic 6. Cranial capacity 3 3 4 <5 cm 3 4 5 cm 3 49 cm 3 48 55 cm 3 53 cm 3 475 53 cm 3 59 675 cm 3 75 875 cm 3 75 875 cm 3 4 7. Cerebellar morphology 8. O M sinus present in high frequency 9. Anteromedial incursion of the superior temporal lines. Sagittal crest present, at least in presumptive males. Compound T/N crest, at least in presumptive males Lateral flare, posterior protrusion No Moderate Pan: No Gorilla: Yes Extensive. Asterionic notch Present 3. Parietal overlap of occipital at asterion, at least in males 4. Squamosal suture overlap extensive, at least in males No Not extensive Lateral flare, posterior protrusion Yes Moderate Yes Extensive Present No Not extensive Lateral flare, posterior protrusion No 3 Strong Yes Extensive Present Yes Extensive Lateral flare, posterior protrusion Intermediate Moderate Yes No Not extensive Tucked Yes 3 Strong Yes Tucked Yes 3 Strong Yes? Partial No Yes? Extensive Tucked No Variable moderate weak Yes Partial Variable No Not extensive Tucked No Weak No No Not extensive Tucked At or posterior Tucked? Intermediate Weak No No Not extensive Weak No No Not extensive

EARLY HOMINID PHYLOGENY 7 Table Continued from previous page No. Character Pan/Gorilla A. afarensis A. aethiopicus A. africanus A. robustus A. boisei H. habilis H. rudolfensis H. ergaster H. sapiens 5. Lateral inflation of mastoid process relative to supramastoid crest 6. Postorbital constriction 7. Pneumatization of temporal squama Not inflated Pan: Moderate Gorilla: Marked Extensive 8. Facial hafting Low 9. Supraglenoid gutter width 3. External cranial base flexion 3. Horizontal distance between TMJ and M /M 3 3. Relative depth of mandibular fossa 33. Postglenoid process size and position 34. Configuration of tympanic 35. Medio lateral position of external auditory meatus Pan: Narrow Gorilla: Wide Flat Long Pan: Shallow Gorilla: Intermediate Large and anterior Tubular (or weak crest) Pan: Medial Gorilla: Lateral Not inflated Moderate Extensive Low Narrow Inflated Marked Extensive High Wide? Flat Long Shallow Large and anterior Tubular (or weak crest) Medial Long Shallow Intermediate Crest with vertical plate Medial Not inflated Moderate Extensive Low Narrow Moderate Long Intermediate Intermediate Crest with vertical plate Medial Inflated Marked Reduced High Wide Flexed Long Intermediate 3 Small and fused to tympanic Crest with vertical plate Lateral Inflated Marked Variable High Wide Flexed Long 3 Deep 3 Small and fused to tympanic Crest with inclined plate Lateral Variable Moderate Reduced Low Narrow Flexed Short Intermediate Variable intermediate small Crest with vertical plate Variable Not inflated Moderate Reduced Low Narrow Not inflated Moderate Reduced Low Narrow? Flexed Long Intermediate Intermediate Short Variable shallow intermediate 3 Small and fused to tympanic? Crest with vertical plate Medial Medial Not inflated Slight Reduced Low Narrow Flexed Short 3 Deep 3 Small and fused to tympanic Crest with vertical plate Medial

8 D. S. STRAIT ET AL. Table Continued from previous page No. Character Pan/Gorilla A. afarensis A. aethiopicus A. africanus A. robustus A. boisei H. habilis H. rudolfensis H. ergaster H. sapiens 36. Vaginal process Small or absent 37. Eustacian process of tympanic Present and prominent 38. Petrous orientation Sagittal 39. Heart shaped foramen magnum 4. Inclination nuchal plane 4. Position of foramen magnum relative to bi-tympanic line 4. Inclination of foramen magnum 43. Origin of digastric muscle 44. Mandibular cross-sectional area at M 45. Orientation of mandibular symphysis 46. Direction of mental foramen opening 47. Hollowing above and behind mental foramen Steeply inclined Well posterior Strongly inclined (posterior) Broad, shallow fossa Pan: Small Gorilla: Variable Receding Pan: Anterior Gorilla: Variable Present Small or absent or slight Intermediate Intermediate At bi-tympanic line Small or absent or slight Coronal Present Weakly inclined At bi-tympanic line Small or absent Present and prominent Intermediate Weakly inclined At bi-tympanic line?? Strongly inclined (posterior) Broad, shallow fossa Small Intermediate Variable Present? Broad, shallow fossa Large Vertical Lateral Small Intermediate Variable Variable Moderate to large Present and prominent Coronal Weakly inclined 3 Well anterior Roughly horizontal Deep, narrow notch Large Vertical Lateral Moderate to large or slight Coronal Present Weakly inclined 3 Well anterior Roughly horizontal Broad, shallow fossa Large Vertical Lateral Variable or slight Coronal Weakly inclined Variable at or anterior Roughly horizontal Deep, narrow notch Small Vertical Lateral? Moderate to large? or slight Coronal Coronal? Variable Weakly inclined Weakly inclined? At bi-tympanic line? Strongly inclined (anterior)? Deep, narrow notch Variable Vertical Lateral Variable Small Vertical Lateral Moderate to large or slight Coronal Weakly inclined At bi-tympanic line Roughly horizontal Deep, narrow notch Small Vertical 3 Posterior

EARLY HOMINID PHYLOGENY 9 Table Continued from previous page No. Character Pan/Gorilla A. afarensis A. aethiopicus A. africanus A. robustus A. boisei H. habilis H. rudolfensis H. ergaster H. sapiens 48. Width of mandibular extramolar sulcus 49. Mandibular deciduous canine shape Pan: Narrow Gorilla: Wide Apex central, mesial convexity, low 5. Incisal reduction No 5. Canines reduced No 5. Prominence of median lingual ridge of mandibular canine Prominent 53. Premolar crown area Pan: Smallest Gorilla: 3 54. Molar crown area Pan: Smallest Gorilla: 55. dm mesial crown profile MMR absent, protoconid anterior, fovea open 56. Distal marginal ridge ofdm Low 57. Separation of molar and premolar cusp apices 58. Frequency of well developed P 3 metaconid 59. Relative enamel thickness 6. Dental development rate Wide Thin Delayed Narrow Apex central, mesial convexity, low Moderate Somewhat Prominent Wide Variable? Apex central, mesial convexity, low Moderate Very Moderate Somewhat? Variable Wide Apex mesial, mesial convexity, high Yes Very Weak Wide Apex mesial, mesial convexity, high Yes Very Weak 4 3 5 Largest 3 Largest MMR slight, protoconid anterior, fovea open Low Wide Infrequent Thick Delayed MMR thick, protoconid even with metaconid, fovea closed 3 Largest MMR slight, protoconid anterior, fovea open? Low Narrow Frequent Hyperthick Intermediate Frequent Thick? Delayed MMR thick, protoconid even with metaconid, fovea closed High Narrow Frequent Hyperthick Accelerated MMR thick, protoconid even with metaconid, fovea closed High Narrow Frequent Hyperthick Accelerated Variable Narrow Narrow?? Apex central, mesial convexity, low Moderate Very Weak Moderate Very Weak Moderate Very Weak Narrow Apex central, mesial convexity, low Yes Very Weak Smallest Smallest?? MMR slight, protoconid anterior, fovea open Low Wide Frequent Thick Low Wide Frequent Thick? Delayed Smallest MMR slight, protoconid anterior, fovea open? Low Wide Frequent Thick Intermediate Wide Frequent Thick Intermediate

3 D. S. STRAIT ET AL. DuBrul & Laskin, 96; Kimbel et al., 984), but they do not describe the same feature. As noted above, such characters should be considered independent until relevant functional or structural hypotheses can be tested. In contrast, the identification of descriptively redundant traits does not depend upon the validity of such hypotheses. In addition, this analysis differs from those of prior studies in the procedures used to assign states to characters. With respect to qualitative characters (comprising 39 of the 6 traits employed in this study), a fossil species was here characterized as exhibiting a particular morphology only if it was present in every relevant specimen in the hypodigm. If two (or more) morphological variants were observed within a species, then it was coded as being variable for that character. This stringent definition was used so that the manner in which variable states were treated could be manipulated (see below, Analyses 4). However, this criterion did not apply to characters that are known to be highly sexually dimorphic, at least in extant hominoids. In such cases, the character was restricted to a consideration of the morphology of only one sex (e.g., Table, character : Sagittal crest present, at least in males ). Because larger samples were available for extant species, and therefore morphological outliers were more likely to be observed, P. troglodytes, G. gorilla and H. sapiens were considered variable only if two or more morphologies were present in sizable proportions (i.e., if a second variant was present in more than 5% of the sample). The quantitative characters were coded using the method of Almeida & Bisby (984). This common-sense approach assigns different states to taxa when their observed ranges are discontinuous or exhibit minimal overlap. In the event that the range of a species spanned the ranges of two relatively discontinuous groups of taxa, the former was coded as being variable. There has been considerable debate concerning methods of coding quantitative characters (Mickevich & Johnson, 976; Simon, 983; Thorpe, 984; Archie, 985; Chappill, 989; Farris, 99; Thiele, 993; Strait et al., 996), but all such methods work best with reasonable sample sizes. These are rarely obtained for fossil hominids. With respect to other cladistic analyses of early hominids, Chamberlain & Wood (987) and Wood (99) employed segment coding, Skelton & McHenry (99) assigned codes according to the rank order of taxon means (which tends to produce many states), and Lieberman et al. (996) did not state how they coded quantitative characters. The method of Almeida & Bisby (984) was used here for three reasons. First, many of the more rigorous methods (gap coding, generalized gap coding, segment coding and gap weighting) rely on arbitrary decisions made by the researcher. Second, the potential benefits of using non-arbitrary methods (homogenous subset coding and finite mixture coding) were limited by the small sample sizes present in many of the species (often a sample of one). Finally, the method of Almeida & Bisby (984) essentially treats quantitative characters in a qualitative fashion, meaning that all 6 of the characters examined here were coded in a similar manner. Of the quantitative characters, ten are represented by indices or angles (i.e., they are scale free), ten take the form of linear measurements, and one (cranial capacity) is volumetric. In general, it is desirable to avoid measurements with scale because they may merely reflect variation in body size. However, it is apparent that the linear and volumetric measurements employed here do not simply vary according to body size, because, among the taxa examined, the largest do not always exhibit the largest character state. Furthermore, australopithecine species and some species of early Homo are quite similar in average body mass (Jungers, 988; McHenry, 988, 99) and, even though they may display considerable intraspecific body size variation, on an interspecific level there is (with respect to many species) an approximation of narrow allometry.

EARLY HOMINID PHYLOGENY 3 State assignments were based on observations of original fossils, casts and descriptions in the literature. Measurements by the authors were supplemented with those recorded by other workers (e.g., White, 977, 98; Johanson et al., 98; Chamberlain, 987; Wood, 99). Parsimony analyses The parsimony analyses in this study were conducted with PAUP 3.s (Swofford, 99). Eight separate analyses were undertaken in order to determine whether tree topologies varied according to alterations in methodology. These analyses differed in their treatment of variable character states, the reversibility and ordering of characters, the presence or absence of missing data, and characters related to mastication. The eight analyses are referred to as: () VARIABLE=INTERMEDIATE, () NON-VARIABLE, (3) VARIABLE=MAJORITY, (4) VARIABLE=MISSING DATA, (5) IRREVERSIBLE, (6) NON-MASTICATORY, (7) NO MISSING DATA, and (8) UNORDERED. Analyses two through eight represent alterations of the VARIABLE=INTERMEDIATE analysis. In the first seven analyses, the most parsimonious tree was constructed using Wagner parsimony, which allows characters with ordered states to reverse freely (the exception being analysis number 5, in which some characters were held to be irreversible). Ordered characters are weighted such that a state change between morphological extremes is treated as if the character has passed through all intermediate states (e.g., a change from state to state 3 is weighted to represent three steps). In general, ordering encourages characters to change incrementally. All characters were ordered except for nasal bone projection, nasal cavity entrance, the configuration of the tympanic bone, and the direction of the mental foramen opening (Table : characters, 7, 34, 46). These four characters were unordered. In all eight analyses, each change between adjacent states (e.g., between and, between and, etc.) was counted as a single step in a tree. Skelton & McHenry (99) stated that this approach biases an analysis in favor of those characters that have many states. In an attempt to weight characters equally, they scaled their traits according to the number of states that each possessed. In other words, state changes in different characters were not weighted equally. For instance, when using scaled characters, a state change in a character with two states is weighted twice as much as a state change in a character with four states. As noted by Farris (99: p. 9), however, since in parsimony calculations the weight of a character is the numerical effect of a step, applying the idea of equal weighting in phylogenetic analyses would lead simply to attributing the same effect to steps in different characters. Consequently, equally weighted state changes are used throughout this study. Character polarity was determined by rooting the outgroup. The most parsimonious tree was obtained using the branch and bound search option. The most parsimonious tree is presented along with its length and its consistency, retention, and rescaled consistency indices. These indices are measures of the amount of homoplasy present. The consistency index (CI) is calculated as the minimum possible tree-length divided by the observed tree-length (Kluge & Farris, 969; Farris, 989). If there is no homoplasy in a tree, then its observed length equals the minimum tree-length, and the CI equals one. If homoplasy is present, then the CI is less than one. The retention index (RI) is calculated by subtracting the observed tree-length from the maximum possible tree-length, and then dividing that value by the difference between the maximum and minimum lengths (Archie, 989; Farris, 989). The rescaled consistency index

3 D. S. STRAIT ET AL. (RC) is calculated by multiplying the CI by the RI (Farris, 989). Both the RI and RC are similar in principle to the CI in that they will equal one if homoplasy is absent, and decrease in value as homoplasy increases. Although the CI is the more traditional measurement, the RI and RC have been claimed to be less sensitive to variations in maximum and minimum tree-length (Archie, 989; Farris, 989). In addition, patterns of character evolution are documented. The most parsimonious tree is presented with a reconstruction of the unambiguous character state transformations required at each node. Some state changes are ambiguous, because it is often equally parsimonious to attribute homoplasy to either parallelism or reversal. Thus, the reconstruction is not a comprehensive list of all character state transformations in the cladogram. The reconstruction of character states at nodes was performed with MacClade 3.4 (Maddison & Maddison, 99). Finally, in each analysis, a 5% majority-rule consensus tree was constructed based on the topologies of all trees within three steps of the most parsimonious cladogram. In such a figure, branching events are presented only if they occur in more than half of the trees under consideration. Thus, if the most parsimonious tree has a length of 5 steps, all trees of length 53 or less will be used to construct the consensus tree. If there are ten such trees, then a given branching event will be depicted in the consensus tree only if it is present in five or more of those ten. Each branching event is labeled to indicate the proportion of trees in which it occurred. In this way, it is possible to summarize the topologies of many trees, and to evaluate the consequences of accepting a tree that is marginally less parsimonious than the favored one. Analysis (VARIABLE=INTERMEDIATE) In this analysis, a species that was variable for a given character was assigned an intermediate character state. This is reflected in the numerical codes that correspond to character states in Table. A variable intermediate state differs from a true intermediate state, in which all specimens that comprise a hypodigm share a distinct morphology. An implicit assumption of using variable intermediate states is that characters will pass through a variable phase as they change from one state to another. This assumption may not be valid, in which case the use of variable states inflates the number of steps that are required to change between morphologies. Regrettably, there are few suitable alternatives to this procedure (see Analyses 4). Although PAUP allows the assignment of polymorphic character states, these are most appropriately applied to supraspecific taxa (Maddison & Maddison, 99). Because the OTUs in this study represent species, this option was not adopted. Variable character states are common, particularly within A. africanus (nine characters) and H. habilis ( characters). Analysis (NON-VARIABLE) As noted above, variable intermediate states require assumptions concerning character state transformation that may not be valid. Since variable character states are frequently encountered in A. africanus and H. habilis, the inferred phylogenetic relationships of these taxa could be biased. Consequently, a parsimony analysis was undertaken in which characters that exhibited variable states were excluded. A total of characters were thus excluded (Table, characters 4, 7, 3, 9,, 5, 3, 33, 35, 36, 39, 4, 44, 46 48, 5). Analysis 3 (VARIABLE=MAJORITY) Another method of addressing the problem of variable states is to assign to a species that character state displayed by the majority of the specimens in its hypodigm. This procedure

EARLY HOMINID PHYLOGENY 33 assumes that normal patterns of intraspecific variation are such that a species can be characterized as having a particular morphology even if it is not present in all specimens. The risk inherent in this methodology is that the variation present in a species may be dramatically oversimplified. This is particularly true in reference to the fossil record, where even a single individual may represent a large proportion of the species-sample. Variable states were eliminated in all characters in which the majority of the specimens in an OTU possessed a common morphology (characters, 3,,, 3,, 5, 3, 33, 35, 36, 39, 4, 46 48, 5). This necessitated a renumbering of the codes corresponding to the states of a given character (the codes presented in Table correspond to the conditions described for the VARIABLE=INTERMEDIATE analysis). If a majority state was not present (i.e., if different morphologies were present in equal numbers of specimens), then a variable intermediate state was retained. This applies to six characters (, 4, 7, 9, 44, 46). Analysis 4 (VARIABLE=MISSING DATA) Finally, a variable taxon can be assigned a code indicating that a state for the given character is unknown. This means that PAUP will assign such a species a state so that a minimum number of steps are added to the tree. The danger of this method is that the assigned state may not necessarily be one actually observed in the taxon, or it may be a state that is present in only a minority of the specimens in the hypodigm. Analysis 5 (IRREVERSIBLE) In the absence of strong evidence to the contrary, it should be assumed that all morphological characters are free to reverse their states. It might be argued, however, that evolutionary reversals are likely to occur infrequently in some of the characters examined in this study. Such characters relate to either large-scale reorganizations of cranial form, or highly complex organ systems. It seems plausible that such characters are relatively conservative, and are unlikely to reverse as frequently. Five characters were considered potentially irreversible: index of palate protrusion (a measure of facial prognathism), cranial capacity, cerebellar morphology, cranial base flexion, and petrous orientation (characters 3, 6, 7, 3, 38). The choice of these characters is subjective. Other researchers might select a different list of characters, and the results might differ accordingly. Analysis 6 (NON-MASTICATORY) Skelton et al. (986) and Skelton & McHenry (99) noted that masticatory features contribute disproportionately to the trait lists used to construct hominid phylogenies. They (Skelton & McHenry, 99) presented evidence that trophic features support robust australopithecine monophyly, and a sister group relationship between A. africanus and a robust clade. Because that topology was inconsistent with the cladogram generated from their entire trait list, and from several other functional and anatomical character complexes, they concluded that hominid evolution was characterized by a large amount of homoplasy, especially in traits related to heavy chewing and that therefore traits relating to heavy chewing are not reliable for reconstructing hominid phylogeny (99: p. 345).