Sacrum morphology supports taxonomic heterogeneity of “Australopithecus africanus” at Sterkfontein Member 4

The presence of multiple Australopithecus species at Sterkfontein Member 4, South Africa (2.07–2.61 Ma), is highly contentious, and quantitative assessments of craniodental and postcranial variability remain inconclusive. Using geometric morphometrics, we compared the sacrum of the small-bodied, presumed female subadult Australopithecus africanus skeleton Sts 14 to the large, alleged male adult StW 431 against a geographically diverse sample of modern humans, and two species of Pan, Gorilla, and Pongo. The probabilities of sampling morphologies as distinct as Sts 14 and StW 431 from a single species ranged from 1.3 to 2.5% for the human sample, and from 0.0 to 4.5% for the great apes, depending on the species and the analysis. Sexual dimorphism and developmental or geologic age could not adequately explain the differences between StW 431 and Sts 14, suggesting that they are unlikely to be conspecific. This supports earlier claims of taxonomic heterogeneity at Sterkfontein Member 4.


Main
Traditionally, all early hominin fossils from Taung, Sterkfontein Member 4 and Makapansgat have been attributed to Australopithecus africanus, although several authors noted the remarkably high morphological variability within these assemblages [1][2][3][4][5][6][7][8][9][10] . This led Clarke [1][2][3]11 to propose the presence of a second Australopithecus species at Sterkfontein Member 4 and Makapansgat, which he named A. prometheus after 12 . Recent studies of the StW 573 "Little foot" skeleton from Sterkfontein Member 2 renewed debates on the functional biology and taxonomy of the Plio-Pleistocene hominins from South Africa 11,13−15 . Nevertheless, the presence of two closely related Australopithecus taxa at Sterkfontein Member 4 is not widely accepted 16,17 because of the ambiguous research evidence and the fragmentary preservation of the fossils. Moreover, quantitative studies focused mainly on the craniodental remains and rarely considered morphological variation within a broad comparative setting.  and StW 431 [21][22][23][24][25] are the best-preserved partial skeletons from Sterkfontein Member 4. The unfused apophyses of the iliac crests and ischial tuberosities and the partially fused epiphyseal plates of the sacral alae of Sts 14 suggest an age of about 15 years compared to modern human standards 23,26,27 . Its acetabulum, which already reached adult dimensions, predicts a body size of 25.4 kg 28 , suggesting a female sex. StW 431, an adult individual possibly of male sex based on its greater robusticity and large body size, probably weighed considerably more than 40 kg 28,29 . Both Sts 14 and StW 431 are traditionally attributed to A. africanus. However, the StW 431 sacrum is narrower and more elongated while that of Sts 14 is rather gracile and wide relative to its small sacral body 25,30 (Fig. 1). The superior tips of the transverse processes of the sacrum are prominent in both StW 431 and Sts 14, though they project superiorly in StW 431 while they are laterally oriented in Sts 14, resulting in a smooth and elongated superior aspect of the sacral wings 23 . This contrasts to the sacrum of A.L. 288-1 (Australopithecus afarensis), which lacks well-developed tips of the transverse processes 31 .
Here, we used different geometric morphometric approaches based on 113 3D landmarks to compare the morphological differences between the StW 431 and Sts 14 sacra to all pairwise differences within our sample of extant species, including a geographically diverse sample of juvenile and adult modern humans (N = 74) and an extensive sample of extant great ape species (N = 94). Additionally, the A. afarensis A.L. 288-1 'Lucy' was considered. Because the size and shape of the sacral wings varied strongly and largely independently from that of the sacral body, we used two different morphometric approaches, one in which we registered the landmark con gurations by a Generalized Procrustes Analysis (GPA) of all 113 landmarks, and one in which all landmarks were registered based on a GPA of the landmarks on the body of the rst sacral vertebra only. Both datasets were analyzed separately in shape space and form space 32,33 .

Results And Discussion
In the principal component analysis (PCA) of the shape coordinates based on the full GPA, Sts 14 and StW 431 plotted at the opposite sides of the modern human distribution, whereas all extant great apes clearly separated from modern humans along PC1 (accounting for 50% of the total variance) ( Fig. 2 and Supplementary data le S1). StW 431 resembled the great apes in having a considerably narrower and more elongated sacrum compared to that of humans, whereas Sts 14 possessed relatively wider and superiorly atter alae. Nevertheless, StW 431 was overall more similar to modern humans than great apes (see the quadratic discriminant analysis in Supplementary Note 1). The two reconstructions of the A.L.
288-1 sacrum were close to Sts 14. Shape variation along PC2 (11%) was driven by the supero-inferior orientation of the sacral alae relative to the sacral body. StW 431 differed along PC2 from Sts 14 and A.L.
288-1 for its caudally directed sacral portion of the linea terminalis. PC3 (6%) re ected changes in the antero-posterior orientation of the sacral alae and did not add considerably to the differences between StW 431 and Sts 14. Some Pongo specimens were intermediate between the human and great ape clusters; one Pongo pygmaeus female even overlapped with the Homo cluster because of its relatively broad shape. Pongo and Gorilla separated from one another quite well, whereas Pan displayed considerable overlap with the two other ape genera.
The different species within the genera Gorilla, Pan, and Pongo did not separate based on sacral morphology (Fig. 3A), nor did the different modern human populations (Fig. 3B). The sacral shape of the subadult modern humans-with ages comparable to that of Sts 14-tended to separate from adult modern humans along PC2, while Sts 14 and StW 431 separated along PC1. Even though modern human males and females overlapped in the PCA plot, the mean shapes differed signi cantly in full shape space (p=0.0005). Size-related shape variation accounted for 12.3% of total sacral shape variation within the full sample but for only 3.3% within the hominin sample (see also Supplementary Fig. 1).
Since sacrum morphology of Australopithecus more closely resembled that of Homo than that of great apes, we repeated the geometric morphometric analysis by excluding the great apes (Fig. 3B). In this reduced sample, the main axis of variation was driven by the supero-inferior orientation of the alae, which again distinguished StW 431 from Sts 14. The A.L. 288-1 sacrum deviated from modern humans and the other Australopithecus specimens for its more posteriorly oriented alae and the weakly developed superior tips of the transverse processes (PC2). The Procrustes distance (a measure of shape difference) between StW 431 and Sts 14 was among the largest observed distances within the modern human sample, including both adults and subadults (Table 1, Fig. 4). Only 2.5% of the Procrustes shape distances between all pairwise male-female comparisons within modern humans were higher than the Procrustes distance between Sts 14 and StW 431. A very similar result (2.4%) was obtained when all pairwise Procrustes distances of modern humans, regardless of their sex, were considered. Procrustes distances in form space (comprising differences in both shape and size; see Supplementary Fig. 2) resulted in slightly higher percentages (3.4% and 3.3%, respectively). Within great apes, a greater percentage of pairwise Procrustes distances exceeded that between Sts 14 and StW 431 (Table 1).
To explore the contribution of sexual dimorphism to sacral shape, we also performed a PCA for modern human adults and the two fossils Sts 14 and StW 431 only ( Supplementary Fig. 3). Sexual dimorphism was represented by PC2 and PC3, despite a large overlap between males and females, whereas both sexes showed the same range of variation along PC1. Importantly, the differences between Sts 14 and StW 431 along PC2 and PC3 resembled the average sexual dimorphism in humans, both in pattern and magnitude. However, the two fossils mainly differed along PC1, which is unrelated to sexual dimorphism. Along this PC1, 1.0% of all pairwise differences and 1.1% of male-female pairwise differences exceeded the distance between Sts 14 and StW 431.
Additionally, with the aim to better capture the relative dimensions of the transverse processes, we compared the sacral shape of StW 431 and Sts 14 to those of the recent species after the registration of the landmark con gurations based on the corpus landmarks only. The resulting PCA plot (Supplementary Fig. 4) was similar to that presented above in Fig. 2 for the separation of hominins from great apes along PC1, but showed a more pronounced overlap between the great apes. The percentages of pairwise Procrustes distances in Homo exceeding the distance between Sts 14 and StW 431 ranged from 2.1% to 2.6%. Importantly, the corresponding percentages in the great ape species were considerably lower than for the full GPA, ranging from 0% in P. troglodytes to 4.5% in G. gorilla for male-female pairs (Table 1).
Our study uses a comprehensive geometric morphometric approach to quantitatively assess the morphological variation within the Sterkfontein Member 4 Australopithecus sacrum sample against the background of a large, worldwide sample of modern humans and six great ape species. We restricted our comparative sample to specimens without lumbosacral transitional vertebrae because Sts 14 and StW 431 do not patently show sacral segmentation anomalies. However, since these fossils preserve only the rst two and a half sacral vertebrae, it is impossible to completely rule out transitional vertebrae in either of these specimens, hence this aspect should be investigated further. In fact, Sts 14 shows a segmentation anomaly at the thoracolumbar transition. Border shifts of the thoracolumbar junction are frequent in hominin fossils and are often associated with border shifts at the lumbosacral junction 34 .
The morphological differences between the partial sacra of StW 431 and Sts 14 are remarkable when compared to the range of variation in modern humans and great apes. Although these morphological discrepancies can be partly explained by individual variation, further exploration of the impact of sexual dimorphism, allometry, different individual and geological age, and, eventually, taxonomic heterogeneity are necessary.
Sexual dimorphism, allometry, and individual age. In modern humans, sexual dimorphism of the sacrum is low compared to that of other parts of the pelvis 35 . Although we found a signi cant shape dimorphism in the human sacrum, which is also re ected by the shape differences between StW 431 and Sts 14, the very large overall shape difference between the two fossils remained very high when compared to the morphological distances of all possible male-female pairwise combinations. As expected, we did not observe considerable sexual dimorphism in the great ape sacrum shape. In addition, size-related shape differences within hominins were of small magnitude (3.3% allometry shape variation) and, hence, are unlikely to account for the differences in sacral shape between StW 431 and Sts 14. It is also unlikely that the shape differences between the fossils result from differences in individual age because in our PCA analysis the vector between StW 431 and Sts 14 was almost perpendicular to the vector of average  23 . This distribution does not conform to that of a simple talus cone resulting from bones and debris falling into a cave from an above opening as suggested previously 40,41 . Rather, it indicates the presence of deep crevasses into which the fragments have been dislocated. This complex taphonomy makes it challenging to date the StW 431 skeleton, but suggests a rather young age compared to other Member 4 fossils.
However, a possible difference in geological age between StW 431 and Sts 14 can only partially, if at all, account for the differences in sacrum shape. The human populations in our sample completely overlapped in the PCA of sacrum shape, even though they diverged at least 260 to 350 ka years ago 42 .
Similarly, P. pygmaeus and P. abelii diverged about 400,000 years ago 43 but were indistinguishable in the PCA. Even Pan troglodytes and P. paniscus were comparable in sacrum shape although they diverged 1.6 Ma ago, with extensive gene ow until at least 200 ka ago 44 . Gorilla gorilla and G. beringei diverged 1.75 Ma ago, with substantial gene ow until at least 20 ka ago 45 , but also largely overlapped in sacrum shape. Thus, sacrum shape appears to be evolutionary conserved, presumably due to stabilizing selection imposed by biomechanical and obstetric demands e.g., 46 .
In conclusion, our results show that, under the assumption that the A. africanus sacrum was as variable in shape as that of modern humans and extant great ape species, it is possible but unlikely that Sts 14 and StW 431 belong to a single species, supporting earlier claims of taxonomic heterogeneity at Sterkfontein Member 4. Yet, as neither Sts 14 nor StW 431 is associated with craniodental remains, it is at the moment impossible to infer which one of these partial skeletons, if any, belongs to A. africanus.
3D surface models of the sacra were generated via a high-resolution optical 3D-surface scanner (QT Sculptor PT-M4, Polymetrics, Darmstadt, Germany) 49 . In instances where CT data were available, the meshes were generated by segmentation using Amira software (www.fei.com). The skeletal collections and the source of the surface models are listed in Table 2.

Virtual reconstruction of the Australopithecus specimens
The StW 431 sacrum preserves most of the rst two and a half sacral vertebrae on the left side, while the right side is more damaged. The most inferior portion of the left auricular surface is also broken off and was restored based on the shape of the auricular surface of the ilium (Fig. 1). This resulted in an approximately 2 mm longer auricular surface than the one originally preserved. Afterwards, the restored left side of the fossil was mirrored with respect to the mid-sagittal plane to replace the incomplete right side.
The Sts 14 sacrum preserves the left side of the rst two sacral vertebrae. We reconstructed it virtually by removing the restoration by Robinson 19 in plaster of Paris on the right side, and mirroring the originally preserved left side with respect to the mid-sagittal plane.
Since the A.L. 288-1 sacrum is taphonomically distorted, we used the same virtual protocol to obtain two symmetrised versions of the fossil, one for the mirrored right side and the other for the mirrored left side.

Geometric morphometric analyses
The landmark con guration (Supplementary Fig. S5) included 29 anatomical landmarks, 36 semilandmarks on ve curves, and 48 surface semilandmarks. In particular, the curves described the margins of the superior sacral surface and of the auricular surfaces, the anterior aspect and the superior aspect of the alae, and the posterior corner of the alae. The anterior and superior aspects of the upper portion of the sacrum were represented by surface semilandmarks. The posterior surface of the sacrum is highly variable and heavily reconstructed in Sts 14. Therefore, only few landmarks and curve semilandmarks were gathered on the dorsal aspect. The analysis was repeated both with and without the surface semilandmarks to explore the contribution of the inter-landmark surface patches. Since the anterior aspect of the sacrum consists of a rather smooth surface between the chosen landmarks, we did not obtain relevant differences in the outcomes and thus presented the results for the complete landmark con guration.
Standard geometric morphometric analyses were performed using principal component analysis (PCA) of the Procrustes shape coordinates both in shape space and form space by augmenting the shape coordinates with the natural logarithm of centroid size (lnCS) 32,50-52 . The landmarks were collected in Viewbox 4 (www.dhal.com), and analysed in the Evan Toolbox (www.evan-society.org) and R software package 53 . PAST software was used for generating the PCA plots. The morphological differences between StW 431 and Sts 14 were evaluated with respect to the rest of the sample by running a pairwise comparison of the Procrustes distances (i.e., the square root of the summed squared differences between the corresponding shape coordinates of two landmark con gurations) after the full GPA analysis in both shape and full space, and after the Procrustes t based on the subset of 13 landmarks on the body of the rst sacral vertebra. The percentages of higher Procrustes distances with respect to that observed between StW 431 and Sts 14 were computed for all pairwise comparisons and separately also for malefemale pairs only, both on the genus and species level, if at least 20 individuals could be included. A comparative evaluation of size was performed by visualization of the lnCS using box plots R software package, 53 ( Supplementary Fig. 2). The in uence of size on shape (i.e., allometry) was evaluated by regressing the Procrustes shape variables on lnCS. The male and female group mean differences were tested for the modern humans using a permutation test of the Procrustes distances (10,000 random permutations. Since StW 431 plotted between modern humans and Pongo in the PCA analysis after full GPA for the entire sample, the likelihood of classifying StW 431 to either groups was evaluated using Quadratic Discriminant Analysis.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.