Human Remains from the Pleistocene-Holocene Transition of Southwest China Suggest a Complex Evolutionary History for East Asians

Background Later Pleistocene human evolution in East Asia remains poorly understood owing to a scarcity of well described, reliably classified and accurately dated fossils. Southwest China has been identified from genetic research as a hotspot of human diversity, containing ancient mtDNA and Y-DNA lineages, and has yielded a number of human remains thought to derive from Pleistocene deposits. We have prepared, reconstructed, described and dated a new partial skull from a consolidated sediment block collected in 1979 from the site of Longlin Cave (Guangxi Province). We also undertook new excavations at Maludong (Yunnan Province) to clarify the stratigraphy and dating of a large sample of mostly undescribed human remains from the site. Methodology/Principal Findings We undertook a detailed comparison of cranial, including a virtual endocast for the Maludong calotte, mandibular and dental remains from these two localities. Both samples probably derive from the same population, exhibiting an unusual mixture of modern human traits, characters probably plesiomorphic for later Homo, and some unusual features. We dated charcoal with AMS radiocarbon dating and speleothem with the Uranium-series technique and the results show both samples to be from the Pleistocene-Holocene transition: ∼14.3-11.5 ka. Conclusions/Significance Our analysis suggests two plausible explanations for the morphology sampled at Longlin Cave and Maludong. First, it may represent a late-surviving archaic population, perhaps paralleling the situation seen in North Africa as indicated by remains from Dar-es-Soltane and Temara, and maybe also in southern China at Zhirendong. Alternatively, East Asia may have been colonised during multiple waves during the Pleistocene, with the Longlin-Maludong morphology possibly reflecting deep population substructure in Africa prior to modern humans dispersing into Eurasia.


Introduction
Research about the evolution of modern humans has historically focused on the fossil records of Europe and Africa as well as the Levantine corridor connecting them. As a result, the role of the vast Asian continent in this evolutionary episode remains largely unknown. Human remains from the Upper Pleistocene of South Asia are scarce, being confined to just two sites possibly within the 33-25 thousand years (or ka) range [1]. In East Asia, human fossils are more numerous [2], but their significance has been difficult to assess due to poor knowledge of their geological context and inadequate dating [1,[2][3]. For clarity, we consider East Asia to comprise the geographic region bordered by the Ural Mountains in the west, the Himalayan Plateau in the southwest, Bering Strait in the northeast, and extending into island southeast Asia.
One widely discussed candidate for the oldest modern human in East Asia is the skeleton from Liujiang, southern China [2]. Yet, the geological age of this individual has ''been an everlasting dispute since the discovery of the fossils in 1958'' as ''there is no documentation on the exact stratigraphic position of the human remains'' [4, p. 62]. As a result, its estimated age lies within the broad range of .153-30 ka [2,4]. The age of the Upper Cave (Zhoukoudian) remains is similarly problematic and has been a major source of uncertainty since their discovery in the 1930s, with estimates ranging from ,33-10 ka [2,4]. Furthermore, the Niah Cave child from East Malaysia possesses uncertain provenience [5]. However, a recent field and lab program aiming to assess the stratigraphy and dating of the deposits at the site has proposed an age of ,45-39 ka for this cranium [5].
Most other candidates for the earliest modern humans in East Asia are similarly problematic. Among the human remains recovered from Tabon Cave, Philippines, the only taxonomically diagnostic specimen is a frontal bone assigned to H. sapiens [6], and dated to 16.562 ka [7]. Moreover, the oldest specimen from the site -directly dated to 47+11/210 ka [7] -might be from an orangutan [6]. At Callao Cave, Luzon, a hominin metatarsal has been directly dated to an estimated 66.761 ka [8]. This specimen is, however, difficult to classify reliably, making its assignment to H. sapiens uncertain [8]. A recently described individual from Tianyuan Cave near Zhoukoudian town, northeast China, is estimated to be ,42-39 ka [9]. The Tianyuan partial skeleton comprises 34 pieces apparently from the same individual, its femur being directly dated to 40,3286816 cal. yr BP [9]. This specimen seems to provide the best candidate for the earliest modern human in East Asia, but is significantly younger (.20 kya) than genetic clock estimates for colonisation of the region (see below). Finally, a mandibular fragment from Zhirendong, southern China, has been dated on stratigraphic grounds to .100 ka [10]. Unfortunately, the specimen is fragmentary and possesses a mosaic of archaic and modern characters also making its taxonomic status unclear [10][11].
Given ongoing uncertainty surrounding the human fossil record, palaeoanthropologists have come to rely on the results of genetic sequencing of samples from living populations to reconstruct the origins of modern humans in East Asia. Genetic research suggests that the earliest humans dispersed into Eurasia from Africa around 70-60 ka, rapidly colonising Southeast Asia and Australasia after this time [12][13][14][15]. This seems to have been followed by a later migration within Eurasia after 40-30 ka, adding the founding lineages of modern Northeast Asians and Europeans [15]. Several later migrations seem to have occurred within the region, some associated with the Neolithic [12][13][14]. Finally, DNA extracted from a .50 ka hominin fossil from Denisova Cave in Central Asia belonging within the Neandertal lineage shares features exclusively with Aboriginal Southeast Asians and Australasians [16][17][18]. This has been interpreted as: 1) evidence for interbreeding between the 'Denisovans' and the earliest modern humans to colonise the region; and 2) implying occupation of Southeast Asia by this archaic population during the Upper Pleistocene [16][17].
Given the central importance of the East Asian fossil record to testing regional and global scenarios of human evolution, in 2008 we began a collaborative research project with the aim of determining the age and providing detailed comparisons of possible Pleistocene human remains from southwest China. This paper focuses on human remains from two localities: Longlin Cave (Longlin or LL) and Malu Cave (Maludong or MLDG) ( Figure 1).
The Longlin human remains were discovered opportunistically in 1979 by a petroleum geologist (Li Changqing) in a cave near De'e, Longlin County, Guangxi Zhuang Autonomous Region, Guangxi Province ( Figure 1). A block of consolidated fine-grained sediment containing the human remains, unidentifiable animal bones, charcoal and burnt clay fragments was removed from the cave and taken to Kunming in neighbouring Yunnan Province shortly after its discovery. A partial mandible and some fragments of postcranial bone were prepared from the block at this time [2], although, the remainder of the skull and other postcranial bones were only prepared from the sediment by our team during 2010. During preparation we recovered a thin flowstone adhering to the surface of the vault of the partial LL 1 skeleton, while charcoal fragments were collected from sediment within its endocranial cavity. Association of cranium, mandible and postcranial elements with similar preservation from within a small (,1 m 3 ) block of sediment suggests that postdepositional disturbance was limited. The cave has been closed to the public and we have so far been unable to undertake research to clarify the stratigraphy and geological context of the human remains. Maludong is a partially mined cave fill located near the city of Mengzi, Honghe Prefecture, Hani and Yi Autonomous Region, southeast Yunnan Province [2,19] (Figure 1). The site was originally excavated in 1989 by a Chinese team including one of us (BZ), and most of the fossil and archaeological materials were recovered at that time [19]. Excavations during 2008 by several of the present authors (DC, JX, AH, BK, BZ, ZY & LY) allowed for a re-evaluation of the remaining stratigraphic section and the collection of a large number of samples for dating and archaeomagnetic analysis. Additional human remains were discovered during the current study, both during our small-scale excavation (506506370 cm) for stratigraphic analysis and from unstudied and unsorted fossil collections recovered during the 1989 field season.

Dating analyses
Radiocarbon dating of charcoal from sediment removed from within the endocranial cavity of LL 1 provided an age of 11,5106255 cal. yr BP (OZM369) ( Table 1; see also, Table S1). Three U-Th age determinations were attempted on ,25 mg subsamples of the flowstone attached to the LL 1 vault ( Table 2). Two of these were too contaminated with detrital Th from cave sediments to allow calculation of useable age estimates, but were able to be used to derive a robust estimate of initial 230 Th/ 232 Th activity in this contaminating phase (0.8260. 20). The remaining less-contaminated age determination has been corrected using this figure to provide an absolute age of 7.860.5 ka (UMB03650) for the flowstone ( Table 2). The flowstone must have formed after the skeleton was deposited, but its dating confirms the Pleistocene-Holocene transition age based on radiocarbon dating of charcoal.
During the original excavation at Maludong three stratigraphic layers were identified [19]. However, in our recent research on the remaining ,3.7 m section at the site we identified 11 distinct stratigraphic aggregates ( Figure 2). AMS radiocarbon ages from 14 charcoal samples were used to determine an age versus depth profile (Table 1, Figure 2; see also, Table S1), providing unambiguous absolute ages for the stratigraphic units recognised at Maludong. Radiocarbon dating of bone was unsuccessful due to a lack of preserved collagen. A magnetic susceptibility record corroborates the stratigraphically coherent and internally consistent radiocarbon-based chronology for the site, indicating that the dated charcoal was deposited at the same time as its enveloping sediments ( Figure 2; see also, Text S1).

Morphological description and comparison
A full list of human remains recovered from Longlin and Maludong is provided in Table 3. Here we describe and compare the cranial, mandibular and dental remains, as they are the most informative with respect to evolution and systematics. Details of comparative samples are provided in Table 4 (see also data  Figure 3) preserves a mostly complete frontal squama with left supraorbital margin, but lacks the right lateral supraorbital part and zygomatic process. The superior section of the nasal bones and superomedial orbital walls are present, as are the left and right frontal processes of the maxillae. Most of the left facial skeleton is present and comprises a nearly complete zygomatic process, alveolar process from mid-line to M 1 , and a largely intact left zygomatic. The right maxilla is incomplete save much of the lateral margin of the piriform aperture and alveolar process. What remains has been rotated ,45u from the median sagittal plane owing to post-burial compression. The left side (MSP to lateral) is largely free of distortion, with the landmarks prosthion and nasospinale readily identifiable. The tip of the anterior nasal spine is broken away, but its base is easily discerned. The bony palate lacks most of the left and right palatine processes. The morphology of the preserved left maxillary tuberosity is consistent with M 3 agenesis. Parts of the sphenoids, anterior occipital, including anterior margin of the foramen magnum, partial left occipital condyle and basioccipital clivus remain. The temporal fragment ( Figure 4) preserves a section of the squama, the base of the mastoid process (tip broken away), tympanic part with a damaged external acoustic meatus, mostly complete and pathologically unmodified mandibular fossa, base of the styloid process, vaginal process and stylomastoid foramen, large carotid canal, preserved foramen lacerum, foramen ovale and foramen spinosum, and a largely intact petrous part.
The MLDG 1704 calotte ( Figure 5) comprises mostly complete frontal and paired parietal bones, but lacks its occipital, temporals and most of the sphenoids, as well as the entire viscerocranium. Evidently the specimen lost its base and facial skeleton owing to anthropogenic alteration, with cut-marks seen along the walls of the vault and on the zygomatic process. Its preserved morphology is unaffected by this alteration. The specimen is free from postdeposition distortion as indicated by visual inspection and scrutiny of CT-scans.
The LL 1 mandible ( Figure 6) and two partial mandibles recovered from Maludong (MLDG 1679 and MLDG 1706: Figure 7) are also compared. Specimen LL 1 comprises a largely complete body, but is missing its left ramus save the root and  coronoid process, and lacks the entire right ramus. The position of the take-off of the left ramus relative to M 3 makes clear that a retromolar space would have been present (M 3 being uncovered [21]). The external surface of the symphysis has been displaced superiorly such that the bone is out of alignment with the alveolar process. This makes accurate assessment of chin development problematic. The left alveolar part retains the roots and crowns of I 1 , canine, P 3 , partial P 4 , M 2 and partial M 3 . The first molar is missing and the alveolar bone shows signs of ante-mortem tooth loss with resorption and new bone growth/remodelling. Much of the right body is preserved and retains the mental foramen, I 1 -P 4 and M 2 roots and crowns. The right M 1 seems again to have been lost ante-mortem, with signs of remodelling of the alveolar bone. The transverse tori are somewhat thickened such that the internal surface of the symphysis is not vertical, a small internal plane being present.
Specimen MLDG 1679 ( Figure 7) comprises a right mandibular body fragment preserved from just anterior to M 2 , with intact M 2 -M 3 crowns and roots, and including a complete ramus in two pieces. The internal morphology of the body and ramus is well preserved including the mandibular foramen, pterygoid surface, and coronoid and condylar parts (the former having been modified somewhat by osteoarthritis).
Specimen MLDG 1706 ( Figure 7) is a right hemi-mandible, broken just lateral to the symphysis (left side), through to a mostly complete ramus. The body is damaged (abraded) along its inferior border, while scoring marks the external surface of the symphysis. No dental crowns were recovered with the specimen, but all of the alveoli are open and lack signs of bony remodelling, indicating a full (adult) set of dentition would probably have been present about the time of death. Externally, the mental foramen is present and well preserved. Internally, the morphology of the surface of the body and ramus is clear: the symphysis expresses enlarged tori, the mandibular foramen is clear, and the pterygoid surface and coronoid and condylar parts are well preserved (the former being modified slightly by osteoarthritis). Supraorbital region. The supraorbital part of LL 1 is conspicuous, with a well-developed glabella. It lacks the obvious signs of division typically seen among modern humans (i.e. lacks a dividing sulcus between medial and lateral parts), but it does thin in the vertical dimension mediolaterally. The supraorbital torus of MLDG 1704 is marked by a strongly developed glabella and superciliary ridges, but also thins laterally. Its supraorbital part is, however, divided into medial and lateral components by a distinct sulcus, being bipartite in form. The presence of a supraorbital torus is a condition rarely seen in recent humans [22], but is more frequent among Pleistocene H. sapiens [23]. A bipartite supraorbital like that seen in MLDG 1704 is characteristic of H. sapiens and distinguishes it from archaic taxa [23]. Table 5 compares supraorbital projection and vertical thickness dimensions. Supraorbital projection at the medial location [24] is moderate in LL 1 (11 mm), but high in MLDG 1704 ( [24], supraorbital projection is comparatively weak in LL 1 (11 mm; EUEHS z-1.58), but moderate in MLDG 1704 (16 mm), and identical to the EUEHS mean (Table 5). In contrast, mid-orbit projection [24] is strong in WAEHS (20 mm) and NEAND (2262 mm; MLDG 1704 z-2.97, p0.002; LL 1 z-5.44, p,0.0001). At the lateral position [24], projection in LL 1 is similar to EUEHS (2063 mm; z-0.32), while in MLDG 1704 it is strong (,23 mm; EUEHS z0.95), the specimen closely resembling WAEHS (24 mm, n4) and NEAND (2462 mm; z-0.49; LL 1 z-2.47, p0.009).
The ratio of parietal/frontal chord and parietal/frontal arc distinguishes samples of Eurasian H. sapiens (means: chord 104-107%, arc 99-107%) from NEAND (chord 9868%, arc 96610%). The shortened parietals of AFEHS (chord 98%, arc 85%) are, in contrast, a putative ancestral trait shared with NEAND. Thus, parietal sagittal expansion is characteristic of Pleistocene Eurasian H. sapiens. For these indices, MLDG 1704 (both 92%) is highly distinct from H. sapiens (EAEHS z2.24, p0.03;  A commonly deployed index of postorbital width is the frontal constriction index, or ratio of minimum/maximum frontal breadth. Its value for MLDG 1704 (76%) is unusually low, and while it sits (just) within the lower part of the range of EAEHS (76-89%), it is most similar to the mean for ERECT (8265%, MLDG z-1.17). In contrast, MLDG 1704 is distinct from mean values for EUEHS (8262%; z-2.91, p0.005), WAEHS (8363%; z-3.04, p0.01) and NEAND (8665%; z1.93, p0.03). Facial skeleton. The facial skeleton of LL 1 is unusual compared with early H. sapiens in exhibiting strong alveolar prognathism. The mid-face is flat, both at the nasal root and piriform aperture, and zygomatic process of the maxilla. The specimen lacks a canine fossa, but possesses a deep sulcus maxillaris. The left zygomatic arch is laterally flared. The zygomatic bone is strongly angled such that its inferior margin sits well lateral to the superior part. The zygomatic tubercle is small and sits lateral to a vertical line projected from the orbital pillar. The anterior section of the masseter attachment is marked by a broad and deep sulcus, but the zygomatic tubercle is small. The anterior wall of the zygomaticoalveolar root is in an anterior position (above P 4 /M 1 ). The lateral orbital margin (pillar) exhibits strong transverse incurvation when viewed in lateral aspect. In most of these features, LL 1 displays the putative plesiomorphic condition for later hominins, being highly distinguishable from the modal condition of H. sapiens.
Tables 9-10 compare standard measurements and indices of the facial skeleton for LL 1 and a single measurement for MLDG 1704 with Pleistocene modern human (Table 9) and archaic (Table 10) samples. Data for superior facial breadth are unavailable for AFEHS. However, the narrower upper face of  The facial skeleton of LL 1 is broad. Bizygomatic breadth is estimated to be wide (c144 mm), strongly distinguishing it from EAEHS, the value for LL 1 being outside of (slightly above) its range. Its bizygomatic most closely resembles NEAND (14568 mm; z-0.12), and is similar also to AFEHS (142 mm). A second index of postorbital constriction is the ratio minimum frontal breadth/bizygomatic breadth, providing a more direct measure of the relative size of the temporal fossa. The value for LL 1 is large (66%) by later hominin standards. While it is equal to the minimum value for EUEHS and WAEHS, its value is distant from their means (EUEHS (7364%; z-1.67, p0.06; WAEHS 70%). It shortening is also seen in archaic EAMPH (74 mm), but not to the great extent characterising late Pleistocene H. sapiens (but more so than the AFEHS specimen from Herto). The facial index (superior height/bizygomatic breadth) for LL 1 (44%) shows its bony face to be very short relative to breadth. The value in the specimen is not especially close to any sample mean and is significantly different to EAEHS (5062%; z2.81, p0.01) and NEAND (5962%; z-6.94, p0.0004).
While the left orbit of LL 1 is broad (45 mm), being identical to the WAEHS mean, this measurement has little discriminating power, as LL 1's value lies within one standard deviation unit of  (7) 10165 (17) 10365 (6) 112 (1) MFB   Table 11 summarises the results of principal components analysis of size-adjusted [27] variables for three sets of analyses comparing LL 1 or MLDG 1704 to fossil specimens. The first analysis included LL 1 ( Figure 8) and 22 other crania, employed 9 variables (Table 11), and generated three principal components. For principal component (PC) 1, the highest loading variables were frontal chord and frontal arc, and these were contrasted mostly with measures of facial height (orbit height and upper facial height) and breadth (chiefly nasal breadth) (Table 11). For PC 2, facial breadth (orbit breadth and bimaxillary breadth) accounted for most of the variance (Table 11), while PC 3 contrasted maximum frontal breadth with bizygomatic breadth (Table 11). A bivariate plot of object scores for PC 1 and PC 2 ( Figure 8A) Figure 8B) shows the third principal component to distinguish ERECT from all other taxa. In this plot, LL 1 sits in a unique position, well away from all crania owing to a combination of a high positive score for PC 2 and moderate score for PC 3. Z-tests of object scores indicate that the difference between LL 1 and the H. sapiens mean is not significant for all PCs (PC 1 z-0.89; PC 2 z0.95; approaching significance for PC 3 z1.70, p0.05).
The second analysis included MLDG 1704 and 23 other crania, employed eight variables (Table 11), and generated three principal components. For PC 1, the highest loading variables were parietal chord and parietal arc, and these were contrasted mostly with measures of vault width (biparietal breadth and superior facial breadth) (Table 11). For PC 2, frontal chord and frontal arc accounted for most of the variance (Table 11), while PC 3 was mostly explained by maximum frontal breadth (Table 11). A bivariate plot of object scores for PC 1 and PC 2 ( Figure 9A) shows PC 1 to separate crania belonging to H. sapiens from those of NEAND, H. heidelbergensis sensu stricto and ERECT. Specimen MLDG 1704 sits just outside of the convex hull for H. sapiens, but clusters close to Cro Magnon 1 and 3 ( Figure 9A). A plot of PC 2 versus PC 3 ( Figure 9B) shows that these principal components do not discriminate well among taxa. Principal component 3 does, however, distinguish MLDG 1704 from all other crania, with its high positive score. For PC 3, its score is outside of the range of all crania, exceeding the next highest score by 0.29 eigenfactor units (almost double the difference between the H. sapiens and NEAND means). Z-tests of object scores indicate that the difference between LL 1 and the H. sapiens mean is not significant for all PCs [PC 1 z0.75; PC 2 z1.51; approaching significance for PC 3 (z1.62, p0.06)]. In contrast, the mean difference for NEAND is significant for PC 1 (z-6.89, p0.001), but not for PC 2 (z1.70) or PC 3 (z1.22).
The final analysis included MLDG 1704 and 43 other crania, employed 6 variables (Table 11), and generated two principal components. For PC 1, parietal chord and parietal arc explained most of the variance (Table 11). For PC 2, frontal arc was contrasted with maximum frontal breadth (Table 11). A bivariate plot of object scores for PC 1 and PC 2 ( Figure 10) shows good separation between H. sapiens on the one hand, and NEAND and ERECT on the other. Specimen MLDG 1704 sits just within the H. sapiens convex hull, but near the edge of the ERECT range. It also sits close to the Nazlet Khater 2 cranium from Egypt, a late Pleistocene specimen also possessing a mix of modern and archaic characters [28]. The object score of MLDG 1704 for PC 1 is closest to the archaic Petralona (0.08 eigenfactor units difference) and NEAND Amud 1 (0.13) crania. Moreover, z-tests indicate that for PC 1, the difference is significant between MLDG 1704 and the H. sapiens mean (excluding Nazlet Khater 2) (z2.00, p0.03), but not for NEAND (z-0.01) or ERECT (z-1.04). For PC 2, the difference is significant between MLDG 1704 and NEAND (z2.19, p0.03), but not for H. sapiens (excluding Nazlet Khater 2: z0.42) or ERECT (z-0.03).
In contrast, the parietal lobes of MLDG 1704 are very short (99 mm), contrasting with the long parietal lobes of EUEHS (12267 mm; z-3.00, p0.01), recent Chinese (10664 mm; z-1.72, p0.04) and recent Japanese (10767 mm; z-1.13). The parietal chord length for NEAND is also moderate (106 mm), like Liujiang (107 mm) and Minatogawa 1 (103 mm). While the parietal lobes of MLDG 1704 are short, much shorter even than Kabwe (104 mm), they are longer than ERECT (8767 mm, z1.64). Figure 11E is a bivariate plot of the breadth of the frontal lobes versus frontal height. It confirms the modern size and shape of the frontal of MLDG 1704, its value sitting well within the range of EUEHS (i.e. Predmost crania) and recent Japanese. Figure 11F compares frontal chord and parietal chord dimensions of the endocast and shows MLDG 1704 to be just within the range of recent Chinese, outside of the range of fossil H. sapiens, and very close to the ERECT specimen Zhoukoudian 10.
Mandibles. Table 16 compares a range of commonly employed mandibular characters and Table 17 body metrics for distinguishing among later Pleistocene hominins. While the symphyseal region of LL 1 has been damaged, in our judgement, it would probably have possessed a chin of Rank 3  [30]. The chin of MLDG 1706 is Rank 3 [30], and while relatively common among Eurasian early H. sapiens (29.2-49.5%), the Chinese Tianyaun 1 and Zhirendong 3 mandibles possess Rank 4 chins. Specimens LL 1 and MLDG 1706 lack a vertical keel and lateral tubercles, features which form the major components of the modern human 'inverted-T' form chin [31]. In inferior view, the anterior border of the body (beneath the symphysis) is rounded in both LL 1 and MLDG 1706, more like the condition seen in archaic hominins [31]. The mental foramen is located below P 4 / M 1 in LL 1 and MLDG 1706. Location of this foramen mesial to M 1 is characteristic of early H. sapiens (88-100% presence versus 12.2% NEAND). Mandibular foramen bridging is absent in MLDG 1679, but present in MLDG 1706 (cannot be scored on LL 1). Absence of bridging is common in NEAND (57.1% presence), but rare in Eurasian early H. sapiens (0-7% presence).
Both Maludong mandibles show asymmetry of the mandibular notch. However, the coronoid process of MLDG 1679 is disproportionately large, a feature common among NEAND, while in MLDG 1706 it is greatly reduced, the H. sapiens condition. In LL 1, the coronoid process is large, but its proportions cannot be assessed owing to an absence of the notch and condylar process. Specimens LL 1 and MLDG 1679 possess a retromolar space (M 3 is uncovered [21]), a common characteristic of NEAND (presence: 75%, versus 32.9-40% in early H. sapiens). In MLDG 1706, the M 3 is partially covered; scored here as absence of a retromolar space. While the medial pterygoid attachment area is strongly scarred in both Maludong mandibles, they lack a prominent superior pterygoid tubercle (present: NEAND 81.2%, Eurasian early H. sapiens 10-76.7%). Finally, the crest of the mandibular notch meets the condyle laterally in MLDG 1679, but it is more medially located in MLDG 1706. Medial placement of the crest is found frequently in NEAND (63% presence, versus 100% absence in Western and European H. sapiens) and characterises the Dar-es-Soltane 5 mandible with its apparent archaic affinities [1,32].
Internally, the alveolar plane of LL 1 and MLDG 1706 is posteriorly inclined and the transverse tori are thickened. This is a common feature among archaic later Pleistocene hominins such as Témara 1 (North Africa), but is largely absent from early H. sapiens [32]. Externally, the symphysis is somewhat undercut in lateral aspect, and its anterior symphyseal angle is low (77u), a value closest to NEAND (80.867.3u; z-0.51) and the Témara 1 mandible (80u). In contrast, Pleistocene H. sapiens angles are more acute (means 86.6-96.5u; z-1.34 to -3.02), as seen also in the East Asian mandibles Tianyuan 1 (,96u) and Zhirendong 3 (91u). Body height (26.9 mm) and thickness (13.3 mm) at the level of the mental foramen in MLDG 1706 are comparatively low, showing the specimen to be similar to modern humans in its size. Its body height sits just outside of the range of Pleistocene East Asian H. sapiens mandibles (range 27.4-33.7 mm), but body thickness is comfortably within their range (11.3-14.4 mm). Body height (28 mm) and thickness (14 mm) measured slightly posterior to the mental foramen in LL 1 is similar to the Maludong specimen (Table 16). Dentition. The mostly well preserved, but worn, dental crowns of LL 1 and MLDG 1706 ( Figure 6-7), and an isolated maxillary third molar (MLDG 1747: Figure 12), also reveal important information about their morphology and affinities. Buccolingual (BL) crown diameters and descriptive statistics for comparative samples are provided in Table 18 (mandibular dentition) and Table 19 (maxillary dentition).
Measurements made on CT-scans of the in situ M 3 of MLDG 1679 (not given) indicate that this tooth is taurodont (Taurodontism index [33] 26.1%, or hypotaurodont). Additionally, MLDG 1747 is also taurodont (Figure 12), its three roots being fused for most of their course. Taurodontism is rare among recent and EUEHS humans [34][35], but is commonly considered a distinguishing feature of NEAND [35][36].

Discussion
The partial human skull from Longlin Cave and the human calotte, partial mandibles and teeth from Maludong both present a range of individual features and a composite of characters not seen among Pleistocene or recent populations of H. sapiens. It is clear that they share no particular affinity with either Pleistocene East Asians, such as Liujiang or Upper Cave 101, or recent East Asians. These features belong to multiple developmental-functional complexes [37], spanning the neurocranial vault, including endocranium, cranial base, facial skeleton, mandible and dentition. Where they can be assessed, metrical dimensions involved are characterised mostly by moderate to high heritability [38][39][40][41]. Given their morphological similarity, close geographical proximity (,300 km apart) and young geological age (i.e., Pleistocene-Holocene transition), it seems likely that both samples belong to the same population. Multivariate analysis of vault shape, a method shown to track neutral genetic distances [42], indicates a somewhat mixed picture with respect to the phenetic affinities of LL 1 and MLDG 1704. The dominant phenetic signal in these analyses, as indicated by the first principal component (accounting for 45-46% of total variance), shows LL 1 and MLDG 1704 to be at the edge of variation within Pleistocene H. sapiens, and in some analyses, on the edge also of H. erectus variability. A weaker phenetic signal, revealed particularly by principal component 3 (,12-14% of total variance), shows them to exhibit a unique cranial shape among all later Pleistocene hominins.
A range of features support the conclusion that these remains show affinities to H. sapiens: N Neurocranium: moderately projecting and laterally thin supraorbital part, which has the bipartite form in MLDG 1704; frontal bone with a moderate chord and arc length, but broad maximum width; and an endocast with long, broad and tall frontal lobes.  N Viscerocranium (LL 1 only): strong alveolar prognathism; flat mid-face, both at the nasal root and aperture and zygomatic process of the maxilla; broad facial skeleton (interorbital, bizygomatic and bimaxillary); very narrow nasal bones; broad piriform aperture; absence of a canine fossa, and possessing a deep sulcus maxillaris; zygomatic arch is laterally flared; zygomatic strongly angled such that its inferior margin sits well lateral to the superior part; zygomatic tubercle is small and sits lateral to a line project from the orbital pillar (anterior aspect); the anterior masseter attachment area is marked by a broad and deep sulcus; strong transverse incurvation of lateral orbital pillar (lateral aspect); and anterior placement of the anterior wall of the zygomaticoalveolar root (above P 4 /M 1 ).  The finding of human remains with such a combination of modern (H. sapiens) and archaic (putative plesiomorphic) characters is unusual, especially in Eurasia. In Africa, there are several Pleistocene remains that also combine modern features with putative later Homo plesiomorphies: from Klasies River Mouth Cave [43][44] and Hofmeyr [45] (South Africa), Iwo Eleru (Nigeria) [46], Nazlet Khater (Egypt), and Dar-es-Soltane and Témara (Morocco) [1,28,45]. Most of them are, however, much older than Longlin and Maludong: Dar-es-Soltane and Témara are undated,  28]. Some of them, such as from Skhul and Qafzeh (Israel) and Pestera cu Oase (Romania) have been included in our analyses, and overall seem to be metrically well within the range of Pleistocene H. sapiens (e.g. Figures 8-9). The former (Levantine) samples do, however, show some similarities to LL 1 and MLDG 1704 in univariate comparisons.
How might the presence of this unusual morphology during the Pleistocene-Holocene transition of East Asia be explained? The remains from Longlin and Maludong could represent very robust individuals within a previously unknown Epipalaeolithic popula- tion in southwest China. We consider this to be an unsatisfactory explanation because of the presence of several apparently unique features combined with an unusual mixture of modern and archaic features is seen in several specimens and spans multiple developmental-functional complexes (as noted above). Moreover, this hypothesis could also be invoked to explain the morphology of remains from Klasies River Mouth Cave, Hofmeyr, Iwo Eleru, Nazlet Khater, Dar-es-Soltane, Témara and Zhirendong, but has not because many of their archaic features are rare or absent among H. sapiens. The same situation applies to the Longlin and Maludong remains, as shown strongly here.
In our opinion, there are more plausible explanations. One possibility is that the Longlin and Maludong remains represent a late surviving archaic population, perhaps similar to that sampled at Dar-es-Soltane and Témara [1,28,32]. Unfortunately, little is known of the morphology of these North African remains, and their affinities and taxonomy are unclear [1,28,32]. Within East Asia, the recently described mandibular fragment from Zhirendong also possesses a mosaic of modern and plesiomorphic characters making its taxonomic status problematic [3,[10][11]. It has, although, been dated on stratigraphic grounds to .100 ka [10], similar in age to the North African Aterian assemblage, but much older than Longlin and Maludong. Another recently described specimen from the site of Salkhit (Mongolia) has also been described as belonging to an unspecified archaic taxon [50]. Dating is uncertain, although, a preliminary date of ,20 ka has apparently been reported [3]. Moreover, doubts about its archaic affinities have been expressed [3] (Note: we have been unable to include this specimen in our analyses as we found errors in the measurements of this and other specimens included in Table 1 of Coppens et al. [50]).
Another possible explanation is that the unusual morphology of the Longlin and Maludong remains results from the retention of a large number of ancestral polymorphisms in a population of H. sapiens. The concept of incomplete lineage sorting is commonly invoked to explain morphologically mixed groups where the features of interest are present also in allopatric populations belonging to the same taxon [51]. Related to this, recent morphological studies have suggested that Pleistocene H. sapiens was deeply geographically subdivided within Africa prior to its dispersal into Eurasia [52]. This explanation has also been invoked to explain the unusual morphology of the Iwo Eleru calvaria [46]. The morphology documented at Longlin and Maludong might be interpreted as consistent with this hypothesis, the Chinese remains perhaps sampling a previously unknown human population (or migration?) that may not have contributed genetically to recent East Asians. Ancient DNA could allow for a test of this idea, however, our ongoing attempts to extract DNA from a specimen from Maludong have so far proven unsuccessful owing to a lack of recoverable genetic material. Either way, the presence of the unusual morphology sampled at Longlin and Maludong during the Pleistocene-Holocene transition indicates that the evolutionary history of humans in East Asia is more complex than has been understood until now. It further highlights the need for much more research in the region as a matter of priority.

Radiocarbon dating
Fifteen charcoal samples for AMS radiocarbon assay were prepared and measured at the ANTARES-STAR Accelerator Mass Spectrometry Facility at the Australian Nuclear Science and Technology Organisation described in [60]. All samples were pre-treated and converted to graphite following methods described by [61]. The external surfaces of charcoal pieces selected for assay were scraped with a cleaned scalpel to remove sediment and soil attached to charcoals. The samples were then cut into smaller pieces to increase surface areas for more efficient chemical pre-treatment. Each sample was then treated with an acid-base-acid sequence as follows: N 2 M HCl at 60uC for 2 hours to remove carbonate and any infiltrated fulvic acid contaminants, N 0.5-4% NaOH at 60uC for 10 hours to remove infiltrated fulvic and humic acid contaminants. This treatment is commenced with a very weak alkali solution of 0.5% NaOH  The cleaned charcoal pieces are finally placed into an oven at 60uC for 2-3 days to dry and then taken for combustion using routine methods for conversion of charcoal to graphite [60]. A portion of each graphite sample was used to determine d 13 C for mass fractionation correction from the graphitisation process. Measured AMS 14 C/ 13 C ratios are converted to conventional radiocarbon ages after background subtraction and d 13 C fractional correction. Radiocarbon ages (see Table 1) are given with 1 standard deviation (1s) precisions ranging from 60.3 to 0.5. All radiocarbon ages were converted to calibrated calendar ages BP (before-present, 1950) using the CALIB 6.02 calibration software and the IntCal09 data sets [62]. All calendar age errors quoted in this paper are given as 2 standard deviation errors (2s). Table 1 provides radiocarbon ages and calibrated calendar ages for each charcoal sample measured by AMA. Table S1 provides ancillary information pertaining to sample pretreatment, graphite AMS mass and d 13 C values used to correct AMS radiocarbon data from Maludong.

Archaeomagnetics
Detailed theories and methods related to the use of magnetic measurements for reconstructing palaeoclimate and anthropogenic alteration are outlined in [63][64][65]. Bulk sediment samples were taken from every single excavated stratigraphic unit during excavation. These bulk samples were divided into sub-samples, air dried and sieved to remove any large non-magnetic particles (i.e. limestone clasts). The sieved bulk sub-samples were then subjected to a range of mineral magnetic measurements. Low temperature and room temperature dual frequency magnetic susceptibility measurements were undertaken on a Bartington MS2 system. Isothermal remanent magnetisation (IRM) acquisition and backfield curves, hysteresis loops and thermomagnetic curves were run on a Magnetic Measurements Variable field Translation Balance (MM-VFTB).

Endocast rendering and volume estimation
A virtual endocast of MLDG 1704 was generated from computed tomography (CT) data in Mimics (Ver. 13.02) by: 1. Segmenting out extraneous material and generating a mask for MLDG 1704, 2. Generating a cutting plane and converting this mask into a 3D object, 3. Positioning the 3D object such that it closed the open region of the cranium, 4. Generating a mask from the repositioned 3D of the cutting plane, 5. Combining the mask of MLDG 1704 with that of the cutting plane, 6. Using the 'cavity fill' tool to create a partial endocast from this combined mask, 7. A 3D surface mesh was then generated from this mask of the endocast and imported into Strand7 (ver. 2.4), and 8. A solid mesh of the partial endocast was then created in Strand7 and the volume taken from the model summary.
Six endocasts and their respective volumes were generated from CT scans of complete Holocene age southern African San crania using this same general approach ( Figure S1). In these instances 'holes' in the masks of the crania representing nerves and blood vessels were filled before applying the 'cavity fill' tool to produce the endocasts.
A template of Type I, Type II and Type III [66] landmark points was created to capture the whole surface morphology of the six modern human endocasts ( Figure S2). Warping of cranial exterior surface morphology using a mixture of landmark points and slid semi-landmarks has been shown to be highly effective at reproducing target cranial shape [67]. Here we apply a similar methodology, utilising landmarks, pseudo-landmarks and slid semi-landmarks, to these endocrania. The landmark template was designed to capture as much as possible of the endocranial shape that was common to all six modern humans.
Our landmark template consisted of 715 landmarks. We used 33 single points (Type I and II landmarks), 9 curves (the beginning and end of the curves were defined by Type II landmarks, with 8 additional Type III semilandmarks slid between these across the endocranial surface) and 12 polygon regions (9 user defined Type II landmarks with additional slid semilandmarks). The polygon regions were used to capture the morphology of the different lobes of the brain. Four of the polygons were defined by 100 landmarks (9 Type II, 91 slid semilandmarks), with the remaining 8 polygons defined by 25 landmarks (9 Type II, 16 slid semilandmarks). Once the landmarks were placed on all of the crania, Template Optimisation was used to create the 'mean' endocranial whole surface shape of these six humans ( Figure S3). Template Optimisation has been shown to be accurate in reproducing the target mesh shapes [68].
The mean endocranial shape was registered with NMB 1204 using an Iterative Closest Point (ICP) registration algorithm [69][70][71], to place it in a 3D space relevant to that of the other human endocrania. The modern human endocrania and that of MLDG 1704 were ICP registered with the mean endocranium to minimise any orientation differences between endocranial specimens and the mean shape ( Figure S4). STLs of the registered 'mean' the San and MLDG 1704 endocasts were imported into Mimics and a cutting plane was generated and positioned as above, with the endocast of MLDG 1704 superimposed ( Figure S4). This was used to separate that part of the mean San endocast that was not preserved in MLDG 1704. The volume of this separated portion amounted to 39% of the original total brain volume for the mean San endocast. The total brain volume of MLDG 1704 is estimated to be 1327 cm 3 assuming similarity in proportions between the two.  Figure S2), B) same landmark template applied to specimen BW 1240, C) and mean modern (San) endocranial shape generated from the average landmark configuration of the 6 modern human specimens. (TIF)  Text S1 Archaeomagnetics -results.  "Sample abbreviations and compositions see Table 4 and [58][59]; m6s(n); ztest results do not employ Bonferroni correction as per [56]. {Data compiled by the authors from literature (see Table 4 and Text S2). {Comparative sample statistics from Trinkaus et al. [58][59]. doi:10.1371/journal.pone.0031918.t019

Supporting Information
in 2008, and Deng Yamin (First Hospital Affiliated with Kunming Medical College) for undertaking CT-scans of human fossils. The staff of the University of Liverpool Geomagnetism Laboratory, particularly Mimi Hill and John Shaw, are thanked for assistance with undertaking some of the archaeomagnetic analyses. We thank James Brink for permission to use CT-Scans of Holocene San crania. Finally, we wish to thank two anonymous reviewers whose comments improved our work.