Craniometrics Reveal “Two Layers” of Prehistoric Human Dispersal in Eastern Eurasia

This cranio-morphometric study emphasizes a “two-layer model” for eastern Eurasian anatomically modern human (AMH) populations, based on large datasets of 89 population samples including findings directly from ancient archaeological contexts. Results suggest that an initial “first layer” of AMH had related closely to ancestral Andaman, Australian, Papuan, and Jomon groups who likely entered this region via the Southeast Asian landmass, prior to 65–50 kya. A later “second layer” shared strong cranial affinities with Siberians, implying a Northeast Asian source, evidenced by 9 kya in central China and then followed by expansions of descendant groups into Southeast Asia after 4 kya. These two populations shared limited initial exchange, and the second layer grew at a faster rate and in greater numbers, linked with contexts of farming that may have supported increased population densities. Clear dichotomization between the two layers implies a temporally deep divergence of distinct migration routes for AMH through both southern and northern Eurasia.


Results
Two major populations are discerned in the cranial affinities, as expressed through analysis of Q-mode correlation coefficients, based on 16 cranio-morphometric datasets recorded from a total of 89 population samples (Fig. 1, Tables 1 and 2, see Materials and Method section). The results are depicted in a Neighbor Net Split map (Fig. 2), here termed the 'Phoenix' tree, due to the shape reminiscent of the mythical bird with large wings.
The 'Phoenix' tree shows a straightforward dichotomization in two major clusters. (1) The 'head' cluster (upper left side) includes Northeast and East Asians (blue circle), as well as Southeast Asians, for the most part referring to early farming and later populations. (2). The 'tail' (lower right side) cluster includes Australo-Papuans and late Pleistocene/early Holocene East/Southeast Asians (red circle), strongly corresponding with pre-farming and Hoabinhian contexts.
Within the overall clustering patterns, naturally some overlap or exchange can be seen in a closer examination, as an expected outcome of small-scale admixture. For example, the data points for Austroasiatic-speaking farmers are branched from the East Asian cluster, slightly toward the side of the red circle cluster that primarily would refer to Australo-Papuan groups. Similarly, the sub-cluster for Austronesian-speaking groups in Island SEA deviates somewhat from the East Asian cluster and instead branches toward the Australo-Papuan affinity. Deviating from the Australo-Papuan cluster, a few samples such as from Gaomiao, Zengpiyan, and the Andaman Islands appear to share a slight affinity with the NEA populations.

Discussion
If the original AMH populations across eastern Eurasia during the Pleistocene possessed mostly Australo-Papuan affinities, then how and when did these groups diminish while distinct East Asian affinities became widespread more recently? In order to address this issue, we examined a series of human skeletal remains from archaeological sites in China, Japan, Russia, and Southeast Asia that derive from multiple pre-farming, early farming, and later contexts of the Late Pleistocene through AD 300.
Human skeletal remains and fossils sites of the last major ice age (Pleistocene) were crucial in this research. A range of Late Pleistocene crania from the Upper Cave at Zhoukoudian (northern China), as well as from sites at Liujiang (southern China), Minatogawa and Shiraho-Saonetabaru (Japan), Tam Pa Ling (Laos), Moh Khiew (Thailand), Tabon (Philippines), Niah (Malaysia), Wajak (Indonesia) and others, have been dated within the range of 47 kya to 16 kya 3,20 . Preservation of measurable characteristics was a major concern, yet in total these specimens supplied multiple confident cranio-morphometric measurements.
The Phoenix map reveals close cranial affinities between the archaeological samples from the Upper Cave at Zhoukoudian and those from the Liujiang and Wajak sites, as well as with the larger Australo-Papuan and Veddha-Andaman groupings. This result suggests that the Late Pleistocene people who lived at these sites shared genetic ancestry with AMH settlers across much of eastern Eurasia, including as far to the east as modern-day Australia and New Guinea. Dispersal of AMH at this time coincided with Pleistocene glacial conditions when significantly lower global sea levels had created vast land masses and shorter water crossings from East Asia through Mainland and Island SEA and as far as Australia and New Guinea.
We documented a continuation of the "first layer" AMH in southern China on the basis of hunter-gatherer sites that were dated between ca. 14 kya and 5 kya (Fig. 2). These study sites included Dalongtan, Zengpiyan, Huiyaotian, and Liyupo in Guangxi Province, Gaomiao in Hunan Province, Qihedong in Fujian Province, and Liangdao in the Taiwan Strait. Although some site contexts within this group chronologically coincided with the earliest known rice and millet farming in Yellow and Yangtze River regions, hunter-gatherer groups still had occupied southern areas. From those hunter-gatherer sites, diagnostic features of skeletal remains included the presence of dolichocephalic calvaria, large zygomatic bones, remarkably prominent glabellae and superciliary arches, concave nasal roots, and low and wide faces 1,[29][30][31][32][33][34][35] . Notably, ancient Japanese Jomon hunter-gatherers belonged to this same grouping.
In addition to the samples from China, pre-Neolithic SEA hunter-gatherer groups were represented in this analysis mostly by archaeological samples from cave sites that contained pebble-tool complex of "Hoabinhian" associations [36][37][38] . Our Phoenix map (Fig. 2) reveals that all of the analyzed Hoabinhian remains from Vietnam and Malaysia shared cranial characteristics with Australo-Papuans. These traits were retained into later post-Hoabinhian hunter-gatherer contexts, including the shell midden site of Con Co Ngua (Vietnam), dated around 6.5 kya 39 . Likewise, the remains of hunter-gatherers recovered from the ca. 5 kya Gua Harimau site (Sumatra, Indonesia) share close affinities with Australo-Papuans 40 .
The "second layer" population identified in this study is associated with present-day NEA people, including all Siberian ethnic groups. The tight clustering of cranial morphologies reflects strong inter-group homogeneity that can be explained most parsimoniously via the single shared origin of a flat and long face and comparatively short head. These definitive characteristics may have originated among people who lived in cold conditions and adapted by reducing their total body surface.
The early hunter-gatherer communities gave way to populations with northern morphometric affinities, seen at Neolithic and Bronze-Iron Age population samples in eastern Eurasia. The prevailing hypothesis for the origin of the "second layer" and the spread of its descendants across much of East Asia and SEA implies a key role for rice and millet agriculture in China in promoting population growth and expansion. Such farming traditions now are traced confidently to 9 kya within the Yellow and Yangtze River area [21][22][23] . Between 7 kya and 5 kya, rice and millet agriculture supported a number of large settlements encompassing an expanding geographical range across China, and several of the resident groups developed complex social, political, economic, and religious systems 22,23 .
The early Chinese farming groups represented here from the archaeological sites of Jiahu, Baligan, Xipo (Henan Province), Hemudu, Weidun (Zhejiang Province), Xitou, and Tanshishan (Fujian Province) all exhibit close affinities with their NEA Siberian counterparts (Fig. 2). With these results, we infer that the "second layer" of population was associated with the earliest occurrences of farming in this region. Moreover, we interpret that the "second layer" of population had been affected by NEA-associated gene flow from the north, demonstrably differentiated from pre-existing Australo-Papuan traits seen in our older Chinese and SEA samples.
Previous research utilizing archaeological evidence and language history has demonstrated that a remarkable cultural transition took place in SEA between 4.5 and 4 kya [24][25][26][27] . This conclusion now is reinforced by the "second layer" identified here on the basis of skeletal remains, specifically from the sites of Man Bac and An Son (Vietnam), Tam Hang (Laos), and Ban Chiang, Khok Phanom Di, Ban Non Wat, and Non Nok Tha (Thailand) (Fig. 2). This cross-regional archaeological signature reflects the geographic expansion of a "Neolithic" horizon of advanced pottery and stone tool traditions, farming economies, and residential settlement structures that can be traced ultimately to the Yangtze River Valley (e.g. Hemudu in Zhejiang Province in Fig. 2 The cranio-morphological datasets in this study consistently confirm affinities with NEA-derived "second layer" populations at "Neolithic" sites dated between 5 kya and 4 kya in southern China and SEA and slightly later in Oceania. This relationship is corroborated by the fact that burial traditions at these locations involved extended-position interment, in contrast to the older flexed-position formats 39,41 . The findings are most striking in China and SEA, where archaeological records reveal "first layer" affinities for thousands of years in duration that suddenly were replaced across large geographic scales by groups of the "second layer". The interface between different populations appears to have been more complex in some regions of SEA where the "first layer" occupants were well established and somewhat diversified for long periods of time before the notably late arrivals of the "second layer" after 2.4 kya. Such late transitions during Bronze or Iron Ages occurred at Hoa Diem (Vietnam) and in the upper layer at Gua Harimau (Sumatra, Indonesia) (Fig. 2). Those close cranial affinities with Bunun (Taiwan), Sumatra, and the Moluccas, Philippines, and Celebes Islanders suggest vigorous human movements, trade networks, and other exchanges crossing through South China. Besides, their clustering with the Neolithic Xitou (Fujian Province) in southern China, later arrival Liangdao 2 (Matzu Island in the Taiwan Strait) implies their remote homeland somewhere in southern China. In Liangdao and Gua Harimau (see: Liangdao 1, Gua Harimau 1 in Fig. 2), provide an extensive evidence for a replacement between local indigenous populations with extremely deep prehistorical roots from Pleistocene, and secondary movements of migrants from the north across Southern China Sea. Linguistic evidence 28 equates these ancient expansions with the Austronesian language family in Taiwan, parts of Mainland SEA, most of Island SEA, and into Oceania, as well as with the Austroasiatic language family in Mainland SEA.
The overall heterogeneity seen in this sub-cluster suggests regional variations in the degrees of genetic admixture between first and second layer populations, although the NEA features are very dominant. The clustering with Non-Austronesian groups (Fig. 2, Thai, Myanmar, Cambodia, and Laos) can be explained by the results of  variable intermixture ratios between the descendants of early indigenous groups and later immigrants. Such a patterning cannot be reconciled with a single origin or regional continuity model of all populations in total. Aeta and Semang Negritos, despite possesing phenotypically different features from surrounding people (small body size and dark skin color), in particular appear to have interbred with the surrounding populations. Given the greater heterogeneity among SEA samples, in contrast to the homogeneity of NEA samples, the most probable scenario had involved a strongly homogeneous genetic input from NEA population flows into the diverse SEA region. Our data reveal a clear dichotomy between first and second population layers that remains consistent across large geographic scales and implies a shared genetic origin for the emergence of the second layer as well as its spread across eastern Eurasia. This degree of cross-regional consistency points to a strongly unified "second layer" of population, rather than the much less likely coincidental convergence of the same outcome at multiple sites due to the effects of climate, diet, nutrition, or other localized factors that might have influenced cranial morphology.
These findings from cranial measurements find extra support from non-metric dental morphology 42 , generally believed under strong genetic control and free of environmental influence, pointing to the same two layers of populations. One grouping is apparent in Australo-Papuan and early SEA teeth, consistent with the "first layer". Another grouping is apparent in NEA and American natives, consistent with the "second layer. " Future research may consider the deeper relation between NEA and American populations, likely involving a shared ancestry through Siberia during the Pleistocene.
Our findings are congruent with the emerging picture in genome data [43][44][45][46] . One key point has been a deep population divergence in AMH, suggesting a branching event prior to the diversification of present-day east Eurasian populations. Traces of such a deep divergence were found in samples from Vietnam, Philippine Negritos, and Jomon hunter-gatherers in Japan. The genome study 45 found Denisova admixture most notably in the Philippine Negritos and to variable extent in other Australo-Papuan populations, thus indicating a distinctive ancient contribution in the "first layer" of the SEA landmass and extending into the ancient Pleistocene continent of Sahul. Meanwhile, Denisova ancestry was absent in mainland Asian populations 18 . The missing Denisova In terms of the deeper origins of the apparently homogenous NEA population, we may consider the more ancient homelands and migratory routes, prior to the entrance into the Yellow and Yangtze River areas by 9 kya but potentially much earlier. In one possible scenario, ancient people perhaps of the "first layer" with Australo-Papuan features moved into Siberia and subsequently adapted to the extremely cold climate during the Last Glacial Maximum (LGM) of 24-16 kya. Another possibility may have involved a Western Asia or European origin, wherein people migrated from western to eastern Siberia across northern Eurasia. In any case, this issue is unresolved, because the ancestral morphology of NEA people so far has been undefined in the scarce skeletal material from the Pleistocene of Siberia. In the Siberian regional samples, so far not enough cranial measurements can refer to the ancient periods pre-dating the cold climate adaptations such as facial flattening. Until these and other issues can be resolved, our study cannot expand to compare substantively with similar-age cranial data from the western hemisphere.
Among the few known pre-40 kya Siberian AMH samples, the DNA analysis of the Ust'-Ishim specimen dated to 45 kya offered a high-quality genome sequence 19 , wherein this AMH individual derived from the basal population of northern Eurasia. This individual had shared ancestors in common with present-day east Eurasian populations and pre-farming west Eurasian populations, with a trace of Neanderthal gene. Another DNA analysis has been possible with the 40 kya AMH in Tianyuan Cave near Beijing 47 , revealing a close genetic relationship with present-day East Asians and evidently different from the diagnostic DNA markers in current European people, therefore suggesting a divergence between European and Asian populations at least in this case. Interpretations may yet be modified with future findings in more cross-regional samples from these ancient time frames.
Taking all of the evidence into consideration, two populations of AMH in eastern Eurasia reflect a deep divergence that most likely accorded with separate migrating events and routes, as expressed in our "two layer" model. The two migration contexts may have been separated by the natural boundary of the Himalaya mountainous zone, posing a barrier between southern and northern routes. The southern route would concur with views of AMH following the coastal rim of the Indian sub-continent and continuing through SEA and onward into the ancient Pleistocene continent of Sahul, at least as early as 50 kya, linked with the later emergence of the archaeologically defined Hoabinhian stone tool complex of this region. So far, the northernmost trace of this "first layer" group has been verified in the Zhoukoudian Upper Cave series in our craniometrics analysis.
The northern route of AMH is less clear in terms of paleo-geographic mapping. In our hypothetical scenario, the NEA ancestral groups had migrated across Siberia from western Eurasia around 45 kya 1,4,19 , and their archaeological signatures involved microblade traditions 23 . Their descendants later developed a quite different cultural trajectory in China after 9 kya, with domesticated millet and rice, and their even later descendants expanded to occupy the larger region while bringing variations of farming economies with them (Fig. 3).
Our "two layer" model in eastern Eurasia may contribute to discussions of global-scale population dispersals and interactions. Our findings are compatible with the AMH dispersal model in west Eurasia, advocated by genome data 48 . Most present-day Europeans derived from later arrivals, along with farming dispersals from the Near East, admixed with a pre-existing base of indigenous hunter-gatherer Eurasian population.
The results of this study are congruent with the archaeological signature of a geographic expansion of "Neolithic" groups, as an added layer flowing through pre-existing populations. In our particular study case, the "second layer" groups can be defined not only by their cranio-morphometric features but also by their pottery traditions, extended-position burials, residential settlements, and farming economies. These groups brought Austroasiatic languages to the mainland and Austronesian languages to the islands from Taiwan southward. Independently confirming our interpretation, other studies of ancient genome analysis 43,44,49 and nonmetric dental traits 42,50 have demonstrated the rapid contribution of NEA genes into SEA, explained by large-scale population expansions of farming groups.

Materials and Methods
The materials used in this study are archaeological and modern cranial series from Northeast Asia, Southeast Asia, and Pacific Oceania, listed in Tables 1 and 2, and all localities are summarized in Fig. 1. The dataset includes samples from contexts of Late Pleistocene, Early to Middle Holocene, Neolithic, Bronze Age, Iron Age, Proto-Historic, Historic, and Modern. Space precludes a review of each sample in the dataset, while the references in Table 1 provide details of the majority of the primary sources. The chronological category 'Neolithic' is assigned to communities with clear evidence for agricultural subsistence without metal, regardless of pottery manufacture, according to the current professional standards in this region 26,39 .
Geographic terminology is of crucial importance in this study. "Northeast Asia (NEA)" encompasses modern China, North and South Korea, Japan, Mongolia, and the Russian Far East including Sibera. "Southeast Asia (SEA)" refers to modern Myanmar (Burma), Thailand, Vietnam, Laos, Cambodia, Malaysia, Singapore, Indonesia, Brunei, the Philippines, Taiwan and the Andaman and Nicobar Islands. "Eastern Eurasia" refers the area encompassing to both the NEA and SEA.
In  Tables 1 and 2, raw data and group averages of cranial measurements are given in the separate files of Tables S1 and S2.
Among the cited data, discrepancies are evident in the measurement systems of upper facial height and orbital breadth between Howell's data 52 and the procedures of other researchers. Howell's upper facial height (NPH) was measured at the anatomical point of the prosthion, while others used the alveolar point according to Martin's method (M48). As for the orbital breadth, Howell used the dacryon (OBB), while most others used the maxillofrontale (M51). Pietrusewsky 53,54 (cited in Tables 1 and 2) adopted Howell's method for these measurements. For those incompatible measurements, as well as missing measurement items, data were recorded in the present study in accordance with the first authors using applicable cranial specimens (see: data source in Table 1, remark in Table 2).
The craniometric affinities of comparative samples were assessed with Q-mode correlation coefficients 55 , using the standardized 16 measurements of group averages. Then standard deviation data was used from the Thai sample which provides the largest sample size among the comparative groups. Concerning the ancient archaeological samples in this study, the available data often required working with single specimens as representative of their sites, especially for those sites of late Pleistocene and early Holocene contexts. So far, the cranial affinity can be assessed at the individual level in most cases. As this study does not rely on statistical significance tests, the potential error is negligible in the use of averaged data.
In order to aid our interpretation of phenotypic affinities between the samples, Neighbor Net Split tree diagrams were generated using the software Splits Tree Version 4.0 56 , applied to the distance (1-r) matrix of Q-mode correlation coefficients (r) in a separate file (Table S3).

Data Availability
Supplementary data are available in the online version of this work.