Paper short abstract:

We report a comprehensive analysis of genomic variation in Island Southeast Asia, from which we infer a primarily common pre-Neolithic ancestry for Taiwan and ISEA populations, but also two minor Neolithic dispersals, from China/Taiwan and Mainland Southeast Asia respectively.

Paper long abstract:

There are two very different interpretations of the prehistory of Island Southeast Asia (ISEA), with genetic evidence invoked in support of both to this day. The default “out-of-Taiwan” model was inspired by linguistic reconstructions and proposes a major expansion of Neolithic Austronesian speakers from Taiwan within the last 5000 years (5 ka). A more recent model has its roots in palaeo-climatic reconstructions, and proposes that Late Glacial and postglacial sea-level rises triggered largely autochthonous dispersals. However, it can be argued that this second model fails to account adequately for the Austronesian language dispersal. Combining novel mitochondrial DNA (mtDNA) and Y-chromosome data with published genome-wide data, we carried out the most comprehensive analysis of the region to date, obtaining consistent results across all three systems. We infer a primarily common ancestry for Taiwan and ISEA populations that was already established before the Neolithic, supporting the second model. Nevertheless, we also detected clear signals of two minor migrations that took place within the last 5 ka, one probably representing a Neolithic input from Mainland Southeast Asia that mostly impacted on southern ISEA, and one from South China, via Taiwan, primarily affecting the Philippines in the north. In particular, the phylogeographic signal from mtDNA haplogroup M7c3c shows strong correspondence with the putative “out-of-Taiwan” dispersal, in sharp contrast with postglacial haplogroups B4a1a1a and E. This may therefore have mediated the spread of Austronesian languages, thus implicating small-scale migration and language shift for their spread, rather than the large demographic event of the first model.