A Hominin Femur with Archaic Affinities from the Late Pleistocene of Southwest China

The number of Late Pleistocene hominin species and the timing of their extinction are issues receiving renewed attention following genomic evidence for interbreeding between the ancestors of some living humans and archaic taxa. Yet, major gaps in the fossil record and uncertainties surrounding the age of key fossils have meant that these questions remain poorly understood. Here we describe and compare a highly unusual femur from Late Pleistocene sediments at Maludong (Yunnan), Southwest China, recovered along with cranial remains that exhibit a mixture of anatomically modern human and archaic traits. Our studies show that the Maludong femur has affinities to archaic hominins, especially Lower Pleistocene femora. However, the scarcity of later Middle and Late Pleistocene archaic remains in East Asia makes an assessment of systematically relevant character states difficult, warranting caution in assigning the specimen to a species at this time. The Maludong fossil probably samples an archaic population that survived until around 14,000 years ago in the biogeographically complex region of Southwest China.


Introduction
The number of Late Pleistocene archaic hominin species, their biogeographical distribution, and the timing of their extinction remain important but intractable questions in palaeoanthropology. These issues have been brought into sharp relief over the last half-decade following the accumulation of genomic evidence for interbreeding between the Pleistocene ancestors of living humans and archaic hominins across much of the Old World [1][2][3][4]. Yet, in many regions, archaic remains are sparse or absent from the Late Pleistocene fossil record. The few exceptions include: 1) the Neanderthals (H. neanderthalensis) in Europe, West Asia, and extending at times into southern Siberia, the youngest example being from Mezmaiskaya dating~39 ka [5]; 2) possibly the North African specimens Dar-es-Soltane and Témara from Morocco [6], with a reported age range of~110 to 40-20 ka for the associated Aterian lithic

Materials and Methods
Measurements for a sample of Pleistocene femora were obtained from the literature [27][28][29]. Comparative fossils were grouped according to chronology, in the case of Lower Pleistocene Homo (LPHO) and Middle Pleistocene Homo (MPHO); an informal grouping that represents a taxonomic unit, in the case of the Neanderthals (H. neanderthalensis, herein NEAN); and two samples of AMH: Middle Pleistocene Modern Humans (MPMH) and Early Upper-Late Upper Palaeolithic humans (EULU). A list of the specimens included in these samples is provided in S1 Table. The only specimen studied directly in the present research was MLDG 1678, which is stored at the Mengzi Institute of Cultural Relics, Mengzi, Yunnan Province, China. Permission to study it was provided by the Yunnan Provincial Culture Bureau in accordance with a memorandum of understanding between the Bureau and the University of New South Wales Australia, and by co-author Zhende Bao, Director of the Mengzi Institute of Cultural Relics, who recovered the femur during excavations in 1989.
CT-scans were made of MLDG 1678 using a GE LightSpeed VCT9 scanner, with helical rotation, 0.625 mm slice thickness, KVP 140 kV and current 400 mA, at the Honghe Prefecture First Hospital during July 2012. The specimen was aligned to the relative AP and ML axes for scanning. The external and internal shaft had already been freed of matrix during curation so threshold-based segmentation was not required. However, some Plaster of Paris was added during restoration and we highlight it where it is visible in Fig 1. Measurements were selected on account of their availability on MLDG 1678. Subtrochanteric (ST) measurements were taken 10 mm below the distal edge of the lesser trochanter [30,31], while the region for midshaft (MS) measurements was selected following identification of a pilaster [29]. All measurements of MLDG 1678 were made by D. Curnoe on CT-scan slices following standard definitions [32] and were subsequently checked on the original specimen. Because the head is missing, neck-shaft angle was estimated using four approaches: 1) a goniometer on the original specimen; 2) a goniometer on a hardcopy (printed) photograph of the specimen; 3) the angle tool on a digital photo using the program ImageJ; and 4) the angle tool on a CT-scan slice at the approximate mid-coronal plane using the program OsiriX v6.5. We measured anteroposterior (AP) and mediolateral (ML) external diameters and cortical thickness dimensions on CT-scan slices using OsiriX v6.5. Total area (TA) and cortical area (CA) at the ST and MS regions were calculated using the "EEM_Macro" provided by Christopher Ruff (downloaded from: www.hopkinsmedicine.org/fae/cbr.htm).
Body mass for MLDG 1678 was estimated using the product of the ST AP and ST ML diameters and formulae from Ref [33]. Body mass was originally estimated for the comparative femora in S1 Table using head diameters [28], but these values may not be directly comparable to our estimate for MLDG 1678. Therefore, we reconstructed body mass again for the femora in our comparative samples using the same method as employed for MLDG 1678 and published AP and ML values. Size-adjustment was also undertaken on MS areas using reconstructed body mass, following the recommendations and method of Ref [34].
In scatterplots, we compared MLDG 1678 with the median, 95% confidence interval (CI) for the median and 1.5 x the interquartile range (1.5IQR) for each comparative sample. Sample 95% CIs for the median were calculated using the formula m AE 1:58 Ã IQR ffiffi n p [35]. Doing so allowed us to visually assess for possible significance of median differences among samples [35]. We supplemented them with Kruskal-Wallis tests to assess overall significance combined with post hoc Mann-Whitney tests to assess pairs of samples. The inclusion of 1.5IQR for all samples, equivalent to 99.3% coverage (or 2.7 standard deviations) for normally distributed data [35], also allowed us to visually assess the affinities of MLDG 1678 in addition to its location relative to sample medians.
We undertook principal component analysis (PCA) and neighbour joining analysis using the out-group method (NJA) with the PAST program [36]. In multivariate studies of fossils a balance must be struck between information content (number of variables) and within-group variation of comparative samples in order to account for the various sources of intrasample variation (number of objects). Moreover, arbitrary decisions often need to be made about the variables to be employed. Here, we emphasised information content and undertook analyses that employed metric only and combined metric and discrete data. Variables were selected on account of their ability to distinguish between modern and archaic femora, as revealed through our morphological and univariate comparisons. Continuous variables were logged prior to analysis, while discrete traits (presence, absence) were scored on a binary scale. For NJA studies using only continuous variables we employed Euclidean distances, but for our mixed datasets, combining continuous and binary data, we used the Gower distance. Resampling scores were calculated using 100,000 replicates in all NJA investigations.

Description and univariate comparison
MLDG 1678 consists of a 175 mm long proximal segment of an adult right femur broken slightly inferior to the MS region (Fig 1). It preserves the greater trochanter and a mostly complete neck. An isolated partial femoral head was also recovered from Maludong (MLDG 1717), but it is unclear whether is belongs to MLDG 1678. The biomechanical neck length is long, and we estimate that it must originally have measured~70 mm; assuming MLDG 1717 provides a reasonable indication of the size of the head of MLDG 1678. Thus, it would have resembled LPHO, MPHO and NEAN, with their long biomechanical necks, rather than AMH (MPMH and EULU), with their short ones [37,38]. The horizontal AP diameter is~27 mm, a value that would sit comfortably within the range of both archaic hominins and AMH [39].
Anteriorly, although the intertrochanteric line is somewhat obscured by plaster, owing to breakage in this region, it must have been faint. Buttressing of the proximal medial diaphysis can be seen in anterior, posterior and medial views, and is pronounced (Fig 1). The greater trochanter is a large [ML broad and superoinferiorly (SI) tall] structure that bulges beyond the lateral wall of the diaphysis. The superior surface of the neck is partly missing along its medial course, and what remains of it has been partially obscured by plaster during restoration of the specimen (Fig 1). Posteriorly, the proximal diaphysis is dominated by a massive lesser trochanter, situated on the approximate mid-plane of the shaft, and projecting slightly medially ( Fig  1). CT-scan slices also show the massive size of this feature, which is composed almost entirely of cancellous bone (Fig 1). In AMH, the lesser trochanter most frequently faces strongly medially. While anteversion of the femur resulting in a more posteriorly located lesser trochanter is Size-adjusted total area 1 8.8 Size-adjusted cortical area 1 6.8 Pilastric index (%) 107.8 Mid-shaft/ST area (%) 90.0 seen in recent humans [40], the combination of its posterior orientation and massive size in MLDG 1678 would be rare (if present at all) in AMH. The pectineal line is a crest-like buttress running SI along the median plane of the posterior diaphysis from the base of the lesser trochanter inferiorly (Fig 1). The hypotrochanteric fossa is extensive SI and ML, merging across a wide area with the gluteal crest (S1 Fig), rather than being a distinct and narrow fossa that distinguishes the lateral wall of the femur from the weakly developed proximo-lateral buttress, as is typically seen when this feature is developed in AMH femora. It closely resembles the condition seen in NEAN in its morphology, which is probably autapomorphic for this group [41]. The intertrochanteric crest is mostly missing (not preserved), and is moderately developed inferiorly. The quadrate tubercle is not preserved.
In Table 1 we present metric data for MLDG 1678. Values for the Maludong femur are compared with summary statistics for comparative samples in Figs 2 and 3. At the ST level, the AP diameter failed to show significant median differences among groups (Fig 2A). The value for MLDG 1678 is small (22.2 mm), being closest to the LPHO median, but sits within 1.5IQR of all samples except MPMH (Fig 2A). The ML diameter was also characterised by an absence of significant median differences ( Fig 2B). MLDG 1678 possesses a small ML diameter (28.3 mm), again most similar to the LPHO median, but outside of the small 1.5IQR of this sample, and within 1.5IQR of all others ( Fig 2B). TA at the ST level lacked significance when a Kruskal-Wallis test was performed, but post hoc Mann-Whitney tests revealed a significant difference between NEAN and EULU medians (p0.038). TA for MLDG 1678 (493.4 mm 2 ) is unsurprisingly small, and once again is most similar to the LPHO median. However, its value sits with 1.5IQR of all samples except NEAN (Fig 2C). Median differences for CA also lacked significance with a Kruskal-Wallis test, but post hoc Mann-Whitney tests revealed a significant difference between LPHO and NEAN (p0.019), and NEAN and EULU (p0.032). CA for the Maludong femur (322.6 mm 2 ) is also very small, again being most similar to the LPHO median, and with 1.5IQR only of LPHO and EULU (Fig 2D).
For the index %-CA (CA/TA x 100), median differences were found to be non-significant. Moreover, the %-CA for MLDG 1678 (65.4%) was most similar to the LPHO and EULU medians, but sat within 1.5IQR of all samples (Fig 2E). In terms of the platymeric index, a Kruskal-Wallis test revealed significant differences among sample medians (p0.040): post hoc comparisons revealed significant differences between MPHO and NEAN (p0.016), and NEAN and EULU (p0.014) (Fig 2F). With a high platymeric index (78.4%), the shaft of the Maludong femur is reasonably circular at the ST level, being most similar to the median for NEAN ( Fig  2F). While the neck-shaft angle also lacked significance with a Kruskal-Wallis test, post hoc Mann-Whitney tests revealed significant median differences between MPHO and MPMH (p0.035), NEAN and MPMH (p0.008), and NEAN and EULU (p0.035) (Fig 2G). The neckshaft angle of MLDG 1678 is estimated to lie within the narrow range of 116-120°, and taking the median of this (c118°), it was found to be within 1.5IQR of MPHO and EULU, and equal to the negative 1.5IQR value for NEAN (Fig 2G).
We reconstructed body mass for MLDG 1678 to be low (50 kg), a value very similar to the LPHO and MPHO medians, but within 1.5IQR of all samples; although, it is clearly small by Pleistocene AMH standards (Fig 2H). Kruskal-Wallis and post hoc Mann-Whitney tests failed to reveal any significant differences among sample medians.
The MS AP diameter was characterised by significant median differences (Kruskal Wallis: p0.002). Specifically, post hoc tests revealed important differences between all archaic samples and MPMH (Mann Whitney: p0.022-0.0067), between LPHO and MPHO (p0.012), and LPHO and EULU (p0.011) (Fig 3A). Thus, this variable serves to distinguish archaic femora from Pleistocene AMH ones. The MLDG 1678 value is small (24.7 mm) and found to be most similar to the medians of LPHO and MPHO (Fig 3A). It did, however, sit within 1.5IQR of all   samples except MPMH (Fig 3A). The ML diameter was also found to show significant differences among sample medians (p0.018), only this time EULU was distinguishable from MPHO and NEAN (p0.024 & p0.002) (Fig 3B). The Maludong ML diameter (22.9 mm) is very small and was not especially similar to any sample median, but did lie within 1.5IQR of LPHO and EULU ( Fig 3B). As at the MS region, TA lacked significance in a Kruskal-Wallis test. The small area for MLDG 1678 (444.2 mm 2 ) was most similar to the LPHO median, but did sit within 1.5IQR of all samples (Fig 3C). While MS CA also lacked overall significance, post hoc Mann-Whitney tests revealed a significant difference between NEAN and EULU medians (p0.039) (Fig 3D). The Maludong femur CA is very small (337.0 mm 2 ), but sat within 1.5IQR of all samples except MPMH (Fig 3D). %-CA was found to exhibit significant median differences among samples (p0.005), with post hoc tests showing this to be the case between LPHO and MPHO (p0.025), MPHO and NEAN (p0.049), MPHO and MPMH (p0.009), and MPHO and EULU (p0.001) (Fig 3E). %-CA in MLDG 1678 (75.9%) was virtually identical to the LPHO median, but it did sit within 1.5IQR of all samples (Fig 3E).
We also calculated size-adjusted TA and CA using reconstructed body mass. Size-adjusted TA lacked overall significance, but post hoc Mann-Whitney tests revealed a significant difference between MPMH and EULU (p0.016) (Fig 3F). The very small value for MLDG 1678 (8.8) was most similar to the LPHO median, although, not especially close to it. Its value did, however, sit within 1.5IQR of all samples except NEAN (Fig 3F). Significant median differences were found for size-adjusted CA (p0.044), with post hoc tests showing important differences between MPHO and EULU (p0.029), and MPMH and EULU (p0.037) (Fig 3G). The small value for the Maludong femur (6.7) was most similar to the EULU median, but it did lie within 1.5IQR of all samples except MPMH (Fig 3G).

Multivariate analyses
Although the metric values for MLDG 1678 most often showed strongest affinity to the LPHO median, the individual values for the specimen were often found to lie within 1.5IQR of most comparative samples. Thus, its affinities are difficult to resolve satisfactorily using univariate methods alone, necessitating deployment of a multivariate strategy (see Materials and Methods; S1 and S2 Appendices).
The results of PCA employing 10 continuous variables and 26 Pleistocene femora are summarised as an object plot for principal component (PC) 1 versus PC2 in Fig 4 (see also, S2 and S3 Tables). PC1, accounting for 52.86% of total variance, showed extensive overlap of AMH (MPMH and EULU) and all archaic femora. Highest loading variables for PC1 were ST TA and ST CA. Thus, this PC highlighted strong similarities between AMH and archaic femora, especially at the ST level, which are likely to reflect common developmental responses to individual differences in femoral neck orientation and pelvic proportions [28]. In contrast, PC2 (28.92% of variance) clearly distinguished modern and archaic femora (Fig 4), the highest loading variable being MS relative cortical area (%-cortical area). This index expresses disparities between subperiosteal deposition and endosteal resorption of bone during development and is thought to be a measure of both differential developmental and aging processes as well as structural reinforcement of the diaphysis [28]. The results of PCA suggest a fundamental difference between archaic and AMH femora in this regard, as noted above in univariate comparisons, and also suggested in some previous investigations of Pleistocene femora [28]. A minimal spanning tree indicated that the shortest distance was between MLDG 1678 and KNM-ER Late Archaic Hominin from Southwest China 1481A, a specimen belonging to African early Homo [42,43], followed by Tabun 1, which is likely to belong in NEAN (Fig 4).
We performed NJA on the same dataset employing KNM-ER 1472 (also early Homo [43]) as an out-group. The resulting tree also showed a clear distinction between archaic and modern human femora, although, the division did not receive strong support in bootstrap analysis ( Fig  5). MLDG 1678 clustered in a branch along with KNM-ER 1481A, which was found in 100% of trees during 100,000 replicates in our resampling results (Fig 5). Thus, this tree confirmed the phenetic and phylogenetic affinities of the Maludong femur to early Homo. The final NJA we performed included 15 variables (binary and continuous) and employed comparative sample medians (see Materials and Methods; Table 2; S2 Appendix). This time, LPHO was used as the out-group. The resulting tree placed MLDG 1678 as the most basal branch separate to a branch containing MPHO, MPMH and EULU (Fig 6). The branch containing MLDG 1678 was found in 100% of trees during 100,000 bootstrap replicates (Fig 6). The results showed the Maludong femur to be highly plesiomorphic and morphologically and phylogenetically distinct from Middle and Late Pleistocene hominins.

Discussion and Conclusions
In light of the absence of archaic hominins from mainland Eurasia younger than about 39 ka [5,11,17,20] it would be reasonable to expect MLDG 1678 to sample an Upper Palaeolithic individual, given its young geological age (i.e.~14.0 ka). Instead, our comparisons with Pleistocene femora reveal the Maludong specimen to be morphologically distinct from AMH. Moreover, of the 15 traits summarised in Table 2, eight are shared with archaic hominins to the exclusion of Pleistocene AMH femora, and a further two are unusual in MLDG 1678 by Late Pleistocene hominin standards. The others failed to distinguish AMH from one or more archaic samples, so have little bearing directly on the affinities of MLDG 1678 (Table 2). It is striking that for the five discrete traits compared, the Maludong femur shares states only with archaic hominins, differentiating them all from AMH (Table 2). Noteworthy, overall, is its: 1) small ST TA and CA, the latter being suggestive of low resistance to axial loads [28]; 2) relatively high platymeric index, indicating its shaft to be rather circular at the ST level; 3) small MS AP and ML diameters, which could reflect relatively low levels, although, the level of buttressing seen in the Maludong femur suggests otherwise and perhaps instead results from its small body size [34]; and 4) moderate pilastric index value, associated with the presence of a Sample compositions, see S1 Table. doi:10.1371/journal.pone.0143332.t002 Late Archaic Hominin from Southwest China weakly developed femoral pilaster, a feature often taken to distinguish AMH from archaic hominins [44] (Table 2).
Putting the results of our analyses together, MLDG 1678 shows morphological and phylogenetic affinities to archaic femora, particularly Lower Pleistocene early Homo. We conclude, therefore, that the Maludong femur represents an individual that probably belonged to an archaic taxon rather than AMH. Just which taxon it represents is difficult to determine, however, because of the scarcity of Late Pleistocene hominin femora from East Asia. It also remains unclear whether Late Pleistocene archaic groups from western Eurasia provide the most appropriate model for East Asian hominins during this period. In the case of AMH, though, we did include femora from East Asia, including the four from Minatogawa, and MLDG 1678 was found to be very distinct from them.
It is intriguing that such a plesiomorphic hominin could have survived at Maludong until near the end of the Pleistocene. Yet, this finding applies also to H. floresiensis, with its apparent minimum geological age only slightly older than MLDG 1678 [13]. Homo floresiensis has only been found on the island of Flores in eastern Indonesia though, and its occurrence has been explained by island biogeography [45,46]. Moreover, the Maludong femur is distinct from the highly unusual femora of this species, with its unique mosaic of traits including resemblances to Australopithecus taxa [47]. One possible explanation is that the Maludong femur samples the population presently known only from Denisova Cave in the Altai region and dubbed the  Table]). doi:10.1371/journal.pone.0143332.g006 Late Archaic Hominin from Southwest China "Denisovans" [2,3]. Another candidate is the presently unnamed taxon represented by the Xujiayao fossils [9][10][11]. However, the absence of femora from both of these groups makes these scenarios impossible to test at present. Besides, the similarities of MLDG 1678 to Lower Pleistocene hominins implies that other possibilities should be considered, such as a late surviving descendent of a Lower Pleistocene East Asian group [48] or even the Dmanisi hominins [49].
Yunnan Province is characterised by complex topography associated with Himalayan uplift and extrusion of the Indochina block resulting in vicarious biogeographic divisions [50]. It is one of 20 floristic endemic centres in China, comprising subtropical evergreen broad-leaved and sclerophyllous forests, and contains high levels of palaeoendemism [50][51][52]. The region around Maludong is also biogeographically on the northern edge of tropical Southeast Asia [50]. Thus, the Maludong femur might represent a relic, tropically adapted, archaic population that survived relatively late in the biogeographically complex region of Southwest China.
Supporting Information S1 Appendix. Ten variable matrix employed in PCA and NJA of individual fossils.