Indian genetic heritage in Southeast Asian populations

Abstract

The great ethnolinguistic diversity found today in mainland Southeast Asia (MSEA) reflects multiple migration waves of people in the past. Maritime trading between MSEA and India was established at the latest 300 BCE, and the formation of early states in Southeast Asia during the first millennium CE was strongly influenced by Indian culture, a cultural influence that is still prominent today. Several ancient Indian-influenced states were located in present-day Thailand, and various populations in the country are likely to be descendants of people from those states. To systematically explore Indian genetic heritage in MSEA populations, we generated genome-wide SNP data (using the Affymetrix Human Origins array) for 119 present-day individuals belonging to 10 ethnic groups from Thailand and co-analyzed them with published data using PCA, ADMIXTURE, and methods relying on f-statistics and on autosomal haplotypes. We found low levels of South Asian admixture in various MSEA populations for whom there is evidence of historical connections with the ancient Indian-influenced states but failed to find this genetic component in present-day hunter-gatherer groups and relatively isolated groups from the highlands of Northern Thailand. The results suggest that migration of Indian populations to MSEA may have been responsible for the spread of Indian culture in the region. Our results also support close genetic affinity between Kra-Dai-speaking (also known as Tai-Kadai) and Austronesian-speaking populations, which fits a linguistic hypothesis suggesting cladality of the two language families.

Prehistoric human migration between Sundaland and South Asia was driven by sea-level rise

“…Thus, we included the Dai population in the MSMC analysis as mainland Southeast Asians although the population was from Xishuangbanna in China, the area along the border of Burma and Laos. We show that this Dai population in our dataset has a similar composition of ancestries to most Thai populations by analysing them together with the comprehensive Thai population dataset 65 . The analysis results are shown in Supplementary Fig.…”

Performance of qpAdm-based screens for genetic admixture on admixture-graph-shaped histories and stepping-stone landscapes

“…Models with targets pre-dating proxy sources are encountered in high-throughput qpAdm screens, but do not constitute a majority of models (Narasimhan et al 2019, Librado et al 2021, Allentoft et al 2022, Bergström et al 2022, Lazaridis et al 2022, Taylor et al 2023). We also explore FDR of the proximal non-rotating (Harney et al 2018, van de Loosdrecht et al 2018, Narasimhan et al 2019, Prendergast et al 2019, Wang et al 2020, Calhoff et al 2021, Wang et al 2021, Zhang et al 2021, Changmai et al 2022a, Changmai et al 2022b, Maróti et al 2022, Brielle et al 2023, Lee et al 2023), distal rotating (Narasimhan et al 2019, Librado et al 2021, Allentoft et al 2022, Bergström et al 2022, Lazaridis et al 2022, Taylor et al 2023), and distal non-rotating protocols (Haak et al 2015, Mathieson et al 2015, Lazaridis et al 2016, Antonio et al 2019, Mathieson et al 2018, Marcus et al 2020, Yang et al 2020, Papac et al 2021, Yaka et al 2021, Patterson et al 2022) ( Tables 1 and 2 ). In the distal protocols, only qpAdm models where target group’s sampling date is strictly contemporaneous with or post-dates sampling of both proxy sources were considered.…”

“…In many publications (e.g., Haak et al 2015, Mathieson et al 2015, Antonio et al 2019, Mathieson et al 2018, Prendergast et al 2019, Marcus et al 2020, Papac et al 2021, Wang et al 2021, Yaka et al 2021, Changmai et al 2022a, 2022b, Patterson et al 2022) this first non-rotating step remains the only qpAdm protocol used (in its distal or proximal forms). In a model competition protocol, subsequent analysis is focused on targets for whom two or more alternative qpAdm models emerge as feasible at the first step.…”

“…In contrast, in our study all simulated individuals were co-analyzed for calculating PCs since this analysis was based on high-quality data only. Unsupervised ADMIXTURE analyses in the literature are usually performed on worldwide or continent-wide panels of populations that often overlap just partially with population sets used for qpAdm analyses (see, for instance, Rasmussen et al 2010, Haak et al 2015, Harney et al 2018, Moreno-Mayar et al 2018a, Zhang et al 2021, Changmai et al 2022a, Brielle et al 2023), while in our study identical population sets were used for qpAdm , PCA, and ADMIXTURE analyses.…”

“…Eight years later we find qpAdm -based protocols routinely employed in large-scale screens of ancient human or animal populations for admixture (often between closely related sources) and used as formal tests for admixture (see Lazaridis et al 2016, Skoglund et al 2017, Harney et al 2018, Mathieson et al 2018, Antonio et al 2019, Narasimhan et al 2019, Fernandes et al 2020, Marcus et al 2020, Ning et al 2020, Wang et al 2020, Yang et al 2020, Calhoff et al 2021, Papac et al 2021, Librado et al 2021, Sirak et al 2021, Wang et al 2021, Yaka et al 2021, Zhang et al 2021, Allentoft et al 2022, Bergström et al 2022, Changmai et al 2022a, Changmai et al 2022b, Gnecchi-Ruscone et al 2022, Lazaridis et al 2022, Maróti et al 2022, Oliveira et al 2022, Patterson et al 2022, Brielle et al 2023, Lee et al 2023, Taylor et al 2023 for examples). qpAdm fits admixture models to a matrix of f 4 -statistics of the form f 4 (“left” group i , “left” group j ; “right” group i , “right” group j ), which in the absence of missing data at the group level can be reduced to a smaller matrix f 4 (target group, “left” group j ; “right” group 1 , “right” group j ), considering algebraic relationships between different f 4 -statistics (Peter 2016).…”