Distinct mandibular premolar crown morphology in Homo naledi and its implications for the evolution of Homo species in southern Africa

Homo naledi displays a combination of features across the skeleton not found in any other hominin taxon, which has hindered attempts to determine its placement within the hominin clade. Using geometric morphometrics, we assess the morphology of the mandibular premolars of the species at the enamel-dentine junction (EDJ). Comparing with specimens of Paranthropus, Australopithecus and Homo (n = 97), we find that the H. naledi premolars from the Dinaledi chamber consistently display a suite of traits (e.g., tall crown, well-developed P3 and P4 metaconid, strongly developed P3 mesial marginal ridge, and a P3 > P4 size relationship) that distinguish them from known hominin groups. Premolars from a second locality, the Lesedi Chamber, are consistent with this morphology. We also find that two specimens from South Africa, SK 96 (usually attributed to Paranthropus) and Stw 80 (Homo sp.), show similarities to the species, and we discuss a potential evolutionary link between H. naledi and hominins from Sterkfontein and Swartkrans.

Scientific RepoRtS | (2020) 10:13196 | https://doi.org/10.1038/s41598-020-69993-x www.nature.com/scientificreports/ We quantitatively assess the EDJ morphology of the H. naledi mandibular premolars (P 3 and P 4 ) using geometric morphometrics (GM), and compare with specimens of Australopithecus, Paranthropus and Homo. In addition to the H. naledi teeth from the Dinaledi Chamber, our sample includes two worn teeth from the Lesedi chamber of the Rising Star Cave system 2 .
We also include specimens from southern Africa that are key to elucidating the systematic placement of H. naledi. For example, Stw 80 consists of a crushed mandible and associated teeth from Sterkfontein Member 5 West. The specimen is suggested to resemble SK 15 18 and is assigned to Homo but not given a specific designation. SKX 21204, a juvenile mandibular fragment from Swartkrans Member 1 (Lower Bank), shows a number of features that distinguish the specimen from Paranthropus 19 . However, as with Stw 80, it has been assigned to Homo but not given a species level designation. Stw 151 (Sterkfontein Member 4) preserves the skull and dentition of a juvenile that shows an overall affinity to A. africanus, but shows several derived early Homo features, particularly in cranial morphology 20 . SK 96 is a mandible fragment traditionally assigned to Paranthropus, but whose P 3 and canine differ in some respects from other P. robustus specimens [21][22][23] . Finally, the Cave of Hearths mandible, while poorly dated, is attributed to Homo and probably antedates the Dinaledi specimens of H. naledi by several hundred thousand years 24,25 . It has not been assigned to a species, and Berger and colleagues have suggested the need for comparisons with H. naledi 26 . Thus, the diversity of Homo species, and their phylogenetic relationship to both Australopithecus and Paranthropus in southern Africa remains a topic of debate and in this study we examine for the first time the taxonomic signal in mandibular premolar morphology in these specimens and Homo naledi.

Materials and methods
Study sample. The study sample is summarised in Table 1 (full details can be found in Supplementary Table 1) and consists of 97 teeth (52 P 3 s and 45 P 4 s) from a range of hominin taxa, including 11 H. naledi premolars. The sample was chosen to cover taxa endemic to southern Africa (P. robustus and Australopithecus africanus), groups that have been suggested based on other aspects of the morphology to share a close resemblance with H. naledi (early Homo), as well as later Homo (Homo neanderthalensis and Homo sapiens) to allow us to identify traits that are primitive and derived for the genus. The early Homo sample used here includes specimens from eastern Africa that have been assigned to either H. erectus (KNM-ER 992 and KNM-ER 1507) or H. habilis (KNM-ER 1802, OH7, OH13), as well as SKX 21204 from Swartkrans and Stw 80 from Sterkfontein that have been assigned to Homo but not given a specific designation. Moreover, three specimens whose taxonomic position is uncertain were here designated as indeterminate. These are Stw 151, Cave of Hearths and SK 96.
SK 96 consists of a small mandible fragment with the roots of the first deciduous molar, as well as a permanent canine and third premolar. The specimen is often assigned to Paranthropus robustus but has been the focus of taxonomic debate. The premolar is unerupted, and Tobias 23 suggested that incomplete enamel deposition, along with a crack in the crown, meant that a reported high shape index for this tooth (within the range of Homo habilis) was not accurate. The associated canine was suggested to be particularly modern in morphology, with Robinson 21 suggesting that the tooth would have been classified as Telanthropus were it not for the morphology of the premolar. Microtomography allows us to digitally remove the crack through the premolar crown and to establish that, rather than being an incomplete germ, the enamel cap is fully formed with the exception of a very small portion of the cervix. The specimen was reconstructed to account for these factors, with several alternate reconstructions tested, details of which can be found in Supplementary Note 1 and Supplementary Figure 1. Table 1. Sample summary. The hominin taxa included in the sample are listed, along with the locality in which they were found, and the number of P 3 s and P 4 s that were included in each of the geometric morphometric analyses. a The complete analysis utilises all landmarks, while the worn analysis uses only a subset corresponding to those preserved in the premolars from the Lesedi (LES) chamber-see main text for full details). A full specimen list can be found in Supplementary Image filtering. The image stacks for each premolar were filtered in order to facilitate the segmentation of enamel from dentine. Two filters were applied; a three-dimensional median filter and a mean of least variance filter, both with a kernel size of one or three. The kernel size was decided manually by assessing the level of contrast between enamel and dentine in the original scan (those with lower contrast required a kernel size of three). This process improves the homogeneity of the greyscale values for the enamel and dentine, and sharpens the boundaries at the interface between tissue types 27 , and the effect of this process on the morphology of the EDJ has previously been shown to be minimal 28 . Filters were implemented using MIA open source software 29 . tissue segmentation and landmark collection. The filtered image stacks were processed using Avizo 6.3 (https ://www.vsg3D .com) in order to produce surface models of the EDJ. Enamel and dentine were segmented using a semi-automatic process that separates voxels based on greyscale values. In some cases, tissue classes are less distinct even after image filtering, making segmentation difficult or impossible using this method. In these cases, a seed growing watershed algorithm was employed via a custom Avizo plugin to segment enamel from dentine, before being checked manually. Once enamel and dentine have been segmented, a triangle based surface model of the EDJ was produced using the unconstrained smoothing parameter in Avizo, and saved in polygon file format (.ply).
In some cases, dental wear removed dentine horn tips. Where this wear was minimal and multiple observers were confident of the dentine horns original position, dentine horns were reconstructed using surface modification tools in Geomagic Studio 2014 (https ://www.geoma gic.com). This was restricted to cases in which the wear was less than wear level 3 as defined by Molnar 30 . Landmark collection and derivation of homologous landmark sets. 3D landmarks were collected in Avizo 6.3, and homologous landmarks were derived using a software routine written by Philipp Gunz 31,32 implemented in Mathematica 10.0 (https ://www.wolfr am.com). This was done following a previously described protocol 16  Inclusion of specimens from Lesedi Chamber. Two H. naledi premolars from the Lesedi Chamber (a third and fourth premolar from the LES1 mandible) show substantial wear such that the dentine horns are almost entirely missing. Therefore, a separate analysis was run in which landmarks corresponding to the worn regions of these teeth were not included, and only the shape of the remaining portion of the EDJ ridge, as well as the cementum-enamel junction (CEJ) ridge, were included. This was done separately for P 3 s and P 4 s to reflect the slightly different patterns of wear in the two teeth. A further two worn P 3 s from the Dinaledi chamber were also included in this analysis to increase the sample size, however no further P 4 s were included.
Analysis of EDJ and CEJ shape and size. A principal components analysis (PCA) was carried out using the Procrustes coordinates of each specimen in shape space. This was completed separately for P 3 s and P 4 s, firstly utilising all landmarks (complete analysis), then subsequently with only the landmarks preserved on the Lesedi specimens (worn analysis). The specimens included in each analysis are listed in Supplementary Table 1.
The size of specimens was analysed using the natural logarithm of centroid size and visualised using boxplots. We also tested for differences between H. naledi and the other taxa using permutation tests for shape (using Procrustes coordinates) and separately for size (using the natural logarithm of centroid size). Permutation tests were carried out in Mathematica 10.0, using 100,000 permutations. A pooled Homo sp. sample was used in order to maximise sample size, and the Benjamini-Hochberg procedure was used to control the false discovery rate 33 . The permutation test, as well as the centroid size boxplot, used data from the worn analysis in order to maximise the sample size of H. naledi. Visualisation of EDJ shape variation. PCA plots of the first two principal components (PCs) were generated separately for P 3 s and P 4 s for both the worn and complete analyses, for the purpose of displaying the variation present in the sample. Further, surface warps were used to visualise the shape changes along the first two PCs in the complete analysis for P 3 s and P 4 s. Here, a reference EDJ surface was created for both the P 3 and P 4 , which was warped to display the morphology of a hypothetical specimen occupying the extreme ends of each PC, defined as two standard deviations from the mean. The surface warps were generated using Mathematica 10.0 and imaged in Avizo 6.3.  (Fig. 1a). PC1 separates modern humans, Neanderthals and the Cave of Hearths individual from earlier taxa and H. naledi. This is largely driven by the taller dentine body height seen in later taxa, the reduction of the talonid region, a reduced metaconid, and a symmetrical crown base (Fig. 1b)   naledi are the somewhat reduced talonid, mesiodistally expanded occlusal basin and buccolingually narrower crown base, relative to P. robustus. However, the specimen also has a smaller metaconid than is typical of H. naledi, which is seen in some P. robustus specimens. When looking at P 4 s, PC1 again separates modern humans and Neanderthals from other taxa (Fig. 3a), while PC2 distinguishes between modern humans and Neanderthals. Homo naledi occupies an intermediate position along PC1, which, as in the P 3 , is largely driven by dentine body height and talonid development (Fig. 3b). Although the talonid in H. naledi is relatively large, as in P. robustus, A. africanus and early Homo, the dentine body is taller than in these groups, which explains its intermediate position along the PC1. While PC2 does not distinguish between P. robustus and A. africanus, it does separate H. naledi from early Homo specimens, particularly H. erectus and SKX 21204. This is influenced by the height of the metaconid, as well as the length of the occlusal basin in the mesiodistal direction. In the case of H. naledi, the metaconid is tall, as in the P 3 , and the crown is mesiodistally expanded. This is particularly noticeable when comparing with H. erectus from Koobi Fora and SKX 21204, which have mesiodistally short crowns that are roughly circular in occlusal view. Another feature that contributes to PC2 and distinguishes between H. naledi and early Homo specimens is the buccolingually narrow shape of the H. naledi crown base.

Scientific RepoRtS
Worn analysis. As the LES1 premolars are worn beyond the stage where the dentine horns could be reconstructed, landmarks that correspond to worn regions of the crown were dropped from the analysis. Importantly, landmarks and semilandmarks for the protoconid and metaconid are excluded, meaning that the height of the dentine horns is not considered but dentine body height is captured. PCA plots for this analysis are shown in Fig. 4, where it can be seen that both LES1 premolars (P 3 and P 4 ) cluster closely with those of the Dinaledi chamber, indicating that there is little difference in premolar morphology between individuals from the two chambers. Further, the overall distribution of hominin specimens remains largely the same as in the complete analysis, albeit with minor differences, for both P 3 s and P 4 s (Figs. 1 and 3 www.nature.com/scientificreports/ Size. Specimen size was analysed using the centroid sizes calculated in the worn analysis, again to maximize sample sizes. Figure 5 shows a boxplot of these results, and the results of the permutation test for differences between each taxon and H. naledi can be found in Table 2. Homo naledi specimens are small compared to those of other taxa; the P 3 and P 4 are significantly smaller than those of P. robustus and A. africanus, while the P 4 s are also significantly smaller than the combined Homo sample. Both premolars are significantly larger than H. sapiens, and are approximately the same size as those of Neanderthals. The LES1 premolars are slightly larger than the Dinaledi specimens included here; this is more noticeable for the P 4 than the P 3 , although there are fewer Dinaledi P 4 s in the sample and it is possible that the inclusion of more specimens would change this. Interestingly, H. naledi is the only taxon included here for which the mean P 4 size is markedly smaller than the mean P 3 size, a pattern present in both Dinaledi and Lesedi premolar pairs Two H. erectus pairs from Koobi Fora, KNM-ER 992 and KNM-ER 1507, show a sub-equal size difference, while in H. habilis individuals OH7 and KNM-ER 1802 the P 4 is clearly larger. However, this pattern is also reproduced in Sterkfontein Homo specimen STW 80, while STW 151 has approximately the same size P 3 and P 4 .

Discussion
The mandibular premolars of H. naledi, particularly the P 3 , were described as showing several distinctive features 1 . We find this distinctiveness is also evident at the EDJ, with a distinct cluster of H. naledi premolars separated from all other taxa in EDJ shape, and consistently displaying a suite of distinctive features (Supplementary Figure 3). This includes a high P 3 mesial marginal ridge, a tall mesially placed P 3 metaconid, and a mesiodistally expanded P 4 crown. Further, the premolars from the Lesedi Chamber cluster closely with the Dinaledi Chamber specimens in shape space (Fig. 4). Although the centroid size of both LES1 premolars is outside the range of the Dinaledi specimens, the difference is very small for the P 3 . There is a larger difference for the P 4 , however only a small number of Dinaledi P 4 s were included, and it is possible that this difference would not be maintained with a larger sample, particularly as there is no size difference evident from linear measurements of teeth from the two chambers 1, 2 . Previous studies have suggested that H. naledi may have occupied a unique dietary niche consisting of foods with a high level of dust/grit contaminants, as reflected in the wear resistance of the molars 34 and the high level of dental chipping 35 . Although studies of the EDJ cannot address these hypotheses directly, our results would be consistent with H. naledi occupying a dietary niche distinct from that seen in other hominin groups.
Previous studies have noted low levels of variation within the H. naledi sample when compared with other hominin taxa 1,2,34,36 . We similarly find a homogenous EDJ morphology in H. naledi premolars both within the Dinaledi Chamber, and between the Dinaledi and Lesedi chambers. A previous study of the P 3 EDJ including a broader sample of fossil hominins and extant apes supports the suggestion that H. naledi is unusual in its homogeneity, with seemingly less variation in size and shape than a subspecies of chimpanzee (Pan troglodytes verus) 16 , which has itself been suggested to be less variable in dental morphology than other Pan subspecies 37 .  www.nature.com/scientificreports/ Dental morphology is thought to be highly heritable, so it is possible that this low level of dental variation reflects low genetic diversity. In African apes, it seems that population structure is important in understanding levels of dental variation; gorillas show hierarchical levels of dental variation such that there is more variation at the species than subspecies level, while in chimpanzees the subspecies and species levels of variation are roughly similar. Similarly, in modern humans there appears to be high levels of intra-population variability in premolar EDJ shape 38 . In sum, low levels of dental variability could suggest the H. naledi remains are from relatively closely related individuals in a single population. However, we also cannot rule out that there are also functional or ecological factors contributing to this uniformity, or that a larger sample of H. naledi premolars would reveal a larger degree of variation than seen in this study. A study of EDJ morphology among extant individuals of known relatedness would be useful in interpreting levels of EDJ variation in fossil taxa. The H. naledi premolars are small, overlapping with those of Neanderthals. The P 4 s, in particular, are small relative to the sample of Australopithecus, Paranthropus, H. erectus from Koobi Fora (KNM-ER 992, KNM-ER 1507), and H. habilis (OH 7 and KNM-ER 1802). The size relationship between the premolars in H. naledi is unusual amongst the hominin sample used here. Australopithecus africanus, P. robustus and H. habilis have a larger P 4 than P 3 (this pattern is evident from the median for each group as shown in Fig. 5, and is consistent among individuals for which both premolars are preserved- Supplementary Table 1), while Neanderthals and modern humans have premolars that largely overlap in size. However, in H. naledi the mean P 4 size is smaller than that of the P 3 .
Previous studies find that some African and European Pleistocene Homo groups display a P 3 > P 4 ratio when considering planimetric crown area, including H. ergaster 39,40 , we do find that the P 4 is smaller in KNM-ER 992 and KNM-ER 1507, however the difference is extremely small. Using centroid size of the partial EDJ is likely to give somewhat different results to planimetric area or linear dimensions since the thickness of the enamel is not considered, and size of the dentine crown is measured in three dimensions. Interestingly, although STW 80 has overall larger premolars than H. naledi, it is the only early Homo specimen to show a P 3 > P 4 pattern. However, relatively few specimens in the comparative sample preserve a P 3 and a P 4 , so further investigation is required to determine the consistency of this pattern within individuals. Variation in P 3 :P 4 size has been studied by a number of authors 22,[40][41][42] and may have particular taxonomic importance. Some have suggested that the anterior and posterior dentitions are somewhat independent, and that in this case the P 4 would covary with the molars, while the P 3 would covary more with the canine. In this case however, we would expect the H. naledi canines to be relatively large, which is not the case 2 . More recently, mouse models have suggested the existence of an inhibitory cascade model in which the size of a tooth is dependent on inhibition from previously developing teeth. The size of the primary postcanine teeth (deciduous molars and permanent molars) can be understood in this context 43 , however the situation is more complex for mandibular premolars as they develop after the deciduous molars, permanent canine and permanent first molar. A study considering the size of the entire tooth row would be necessary to investigate whether or not the observed tooth size patterns in H. naledi fit the expectations of this model.
Despite abundant features throughout the H. naledi skeleton that are reasonably interpreted as primitive for Homo, the premolars of H. naledi evince a number of very clear differences from the majority of the early Homo specimens included here. Moreover, we found a significant difference between H. naledi and a combined early Homo group in P 3 shape (Table 2). For both P 3 and P 4 , PC2 distinguishes between H. naledi, H. habilis and H. erectus, with H. erectus and H. naledi occupying the extremes. This axis relates to the placement of the metaconid in both premolars (mesial in H. naledi, distal in H. erectus), the development of the P 3 mesial marginal ridge (high in H. naledi, lower in H. erectus), the relative mesiodistal length of the P 4 crown (longer in H. naledi, shorter in H. erectus) and the relative buccolingual width of the crown base in both premolars (narrower in H. naledi, longer in H. erectus). In this respect, H. habilis is intermediate, and much more closely resembles the Australopithecus condition ( Figs. 1 and 3), as would be expected for a species basal to the genus Homo. Our results suggest that the two derived premolar morphologies represented by H. naledi and H. erectus could have evolved separately from a more generalised H. habilis-like ancestral condition. The alternative explanation of character transformation series that resembles H. habilis-H. erectus-H. naledi, would require reversals in a number of the features mentioned above. However, H. naledi postdates the H. erectus specimens included here by around 1.5 million years, which would be more than enough time for these changes to take place. Further, the sample used here is limited with respect to early Homo, so we should be cautious in these conclusions.
Swartkrans specimen SKX 21204 has been assigned to Homo 19 , but has not been given a specific designation. Here the P 3 and P 4 cluster more closely to eastern African H. erectus than H. habilis, with the P 4 particularly close to KNM-ER 992 in shape space (Fig. 3). Although SKX 21204 is small, similar to H. naledi when considering centroid size (Fig. 5), size appears to be variable in a number of taxa, and the shape of the premolars suggests that (1) the specimen is clearly distinguished from H. naledi and (2) among our sample, the specimen is a good match for African H. erectus. STW 151 was suggested to possibly represent an individual more derived towards Homo than other Sterkfontein A. africanus specimens 20 . Here we find that both P 3 and P 4 are close to, but not within, the A. africanus range of shape variation, and neither premolar shows any particular affinity to early Homo specimens, or those of H. naledi. However, it should be noted that the features aligning the specimen with early Homo were mostly in other areas of the dentition and the cranium, rather than the mandibular premolars. The Cave of Hearths P 3 most closely resembles Neanderthals, and is clearly distinct from H. naledi (for more details on this specimen, see refs 15,16). As with the H. erectus specimens, positing the Cave of Hearths specimen as an ancestor of H. naledi would entail a number character reversals. The Cave of Hearths P 3 is better fit as a human ancestral form than it is an ancestral form for H. naledi.
Stw 80 is a crushed mandible from Sterkfontein Member 5 West assigned to early Homo 44 , and has been suggested to show strong similarities to SK 15 from Swartkrans 18 . We find that the P 3 morphology is unusual; it has a very large talonid, which is primitive for Homo, however it also has a relatively tall dentine body, buccolingually Scientific RepoRtS | (2020) 10:13196 | https://doi.org/10.1038/s41598-020-69993-x www.nature.com/scientificreports/ narrow anterior fovea and a large accessory crest in the posterior fovea. This combination of traits is not seen in any other specimens in our sample, and the worn analysis does not show clear affinities between the P 3 and any of the hominin taxa included here. Unfortunately, the protoconid of both P 3 and P 4 are worn, precluding assessment of the relative cusp heights, which can be useful in distinguishing between taxa. The P 4 falls relatively close to H. naledi in Fig. 4, driven partly by the combination of a tall dentine body and a mesiodistally expanded crown. This combination distinguishes H. naledi from the H. habilis and H. erectus specimens in our sample, possibly representing a H. naledi apomorphy. Equally, Stw 80 shows a P 3 > P 4 pattern, which, as discussed earlier, may be a H. naledi apomorphy. The shared presence of these derived traits could suggest a phylogenetic relationship between this specimen and H. naledi, however further investigation involving the entire tooth row would be necessary to investigate this further as it is possible that the similarities we have outlined are due to homoplasy. It is important to note that while the majority of the teeth of Stw 80 are too damaged to measure, the mesiodistal length of both the canine and M 3 are larger than in H. naledi 1,44 . The taxon that falls closest to H. naledi in P 3 shape in the first two PCs is P. robustus (Fig. 1), which is driven by both taxa sharing a tall mesially placed metaconid and well-developed marginal ridges. The talonid is also somewhat expanded in both taxa, although more so in P. robustus. The taxa are separated in PC3 however (Fig. 2), which reflects in part the difference in talonid size, as well as the crown being mesiodistally longer in H. naledi. Further, there are size differences between the two taxa ( Fig. 5) and the permutation test found them significantly different in both shape and size (Table 2). Equally, the P 4 of H. naledi falls closer to A. africanus than P. robustus (Fig. 2) and is significantly smaller than both A. africanus and P. robustus (Table 2, Fig. 5). It is therefore possible that the similarities between these two taxa in P 3 morphology are due to homoplasy.
The picture is more complicated when considering SK 96 however (Fig. 6); the specimen is from Member 1 at Swartkrans, and consists of a mandible fragment, canine and P 3 . The premolar was found to have a more mesiodistally expanded crown than other P. robustus P 3 s 21 , however this was attributed to the tooth being incomplete and cracked 23 . After verifying that the crown is all-but complete, and fixing the crack, we find that the specimen is smaller in centroid size than any other P. robustus P 3 included (n = 9), instead falling within the range of the P 3 s of Neanderthals, H. habilis and slightly above the H. naledi size range. Further, the shape of the EDJ is outside the range of P. robustus, instead occupying a space between P. robustus, H. habilis, and H. naledi in the first 3 PCs (Fig. 2). A feature of Paranthropus is the very large P 3 talonid; SK 96 instead has a moderate talonid more similar to H. habilis and H. naledi. Equally however, SK 96 does not show the clearly well-developed metaconid typical of H. naledi, instead showing a smaller metaconid similar to that found in in some P. robustus and H. habilis specimens.
SK 96 also preserves a lower canine and part of the roots of a deciduous first molar. The canine was described as showing a particularly modern morphology 22 , and is the smallest P. robustus specimen in crown dimensions 22,[45][46][47] . The crown size is also smaller than H. habilis but is within the range of H. naledi. However it does not have the typical H. naledi canine morphology (Supplementary Figure 4); the SK 96 crown is less asymmetrical, more rounded, and lacks a distal accessory cuspule 1,2 . The canine of OH 7 and OH 13 are also markedly asymmetrical, and have moderately developed distal cuspules 41 , differentiating them from SK 96.
If SK 96 does belong to P. robustus, it would be extreme for the species in premolar shape (Fig. 2) and size (Fig. 5, also see refs 22,45 ), as well as canine size 22,45,46 . It would also suggest that traits considered to be characteristic www.nature.com/scientificreports/ of the species, such as expansion of the P 3 talonid, are less pronounced in some individuals. Therefore, it is possible that this specimen instead represents Homo; although there are differences in canine morphology, the SK 96 P 3 shares some features with H. habilis. Member 1 at Swartkrans contains a number of specimens assigned to Homo, including SK 27 and SK 45, and it is possible that SK 96 represents the same taxon as these specimens, as well as SK 15 from Member 2. Also worthy of note are the similarities between these specimens and H. naledi (See Supplementary Figure 5a). SK 96 resembles H. naledi in P 3 morphology more closely than any other specimen included here, while Stw 80 falls closest to H. naledi in P 4 shape, and shares the species' P 3 > P 4 pattern. This would also be consistent with the finding that two key Swartkrans Homo specimens (SK 15 and 45) share a number of molar root characteristics with H. naledi 10 . On the other hand, SKX 21204 also derives from Swartkrans Member 1 (Lower Bank) and is here found to show a number of clear differences in P 3 morphology from SK 96 (see Supplementary Figure 5a,b). SKX 21204 is a better fit for H. erectus based on premolar morphology, which could suggest the presence of multiple non-P. robustus taxa in Swartkrans Member 1. Further investigation is needed to fully assess whether these differences are sufficient to warrant species-level separation.
The similarities between SK 96 and H. naledi could be further evidence 10 for some phylogenetic link between hominins at Swartkrans and H. naledi, while the similarities with Stw 80 may suggest a similar link with Sterkfontein hominins. This is striking given that both Swartkrans Member 1 and Sterkfontein Member 5 West are suggested to predate H. naledi by more than a million years 18,24,[48][49][50] , and would suggest that H. naledi represents a long surviving lineage that split from other members of the genus Homo relatively early. In this regard, the Cave of Hearths specimen is notable because it evinces a more human-like morphology and likely antedates the dated H. naledi material by hundreds of thousands of years. However, it is also possible that the similarities between these specimens and H. naledi are due to homoplasy. It should be noted that although SK 96 and STW 80 show similarities to H. naledi individually, their P 3 morphologies are not especially similar to one another. Overall, the Sterkfontein and Swartkrans early Homo assemblage does not appear to be homogenous in premolar morphology. It is therefore important that the remaining dentition and mandibular morphologies of these specimens are also investigated and compared to H. naledi where possible to allow us to assess all available evidence.

conclusions
Overall, we find that there are a number of aspects of mandibular premolar EDJ morphology that are distinctive in H. naledi when compared to a broad sample of hominins, including a number of key early Homo specimens. The morphology of the H. naledi premolars is highly consistent and homogeneous when compared with other samples included here, and distinctive traits are displayed consistently throughout the collection including a tall well-developed metaconid in both the P 3 and P 4 , a relatively mesiodistally long P 4 crown, and strongly developed mesial marginal ridges. The worn LES1 premolars are also consistent with this morphology. This distinctive morphology may be useful in the future in identifying further specimens of H. naledi, potentially from limited and fragmentary remains. We also suggest that SK 96, previously attributed to P. robustus, differs from the hypodigm in P 3 EDJ morphology and may instead represent Homo. The specific designation of the specimen, and the relationship between this and other South African Homo specimens, including Stw 80 from Sterkfontein, requires further investigation.