Integrating linguistics, social structure, and geography to model genetic diversity within India

Aritra Bose, Daniel E. Platt, Laxmi Parida, Petros Drineas, Peristera PaschoubioRxiv 2017 · DOI

TL;DR

This paper introduces COGG, a novel computational framework that integrates linguistic, social, and geographic data to model genetic diversity within India. It reveals that language and social structure, particularly endogamy, are more powerful drivers of genetic substructure in India than geography alone, a finding that contrasts with patterns observed in other global populations.

What Problem Does This Paper Address?

Understanding human genetic diversity is crucial for tracing population history and migration patterns. While geography is often a primary factor in shaping genetic structure globally, India presents a unique challenge. Its intricate social stratification, primarily the caste system with its strict endogamy, and a vast diversity of spoken languages, add significant layers of complexity. Previous studies have struggled to disentangle the relative contributions of these factors. This paper aims to fill this gap by developing a model that can quantify and integrate geography, language, and social structure to explain the observed genetic substructure within the Indian subcontinent.

Methods

The core of the paper is the development of a novel framework called COGG (Correlation Optimization of Genetics and Geodemographics). COGG aims to maximize the correlation between principal components of genetic data and a "geodemographic matrix" that encodes geographic coordinates, social group affiliation, and language family.

Key methodological components include:

  • Data Integration: Merging publicly available genomic data from the Indian subcontinent with geographic and linguistic information.
  • COGG Algorithm: A quantitative framework that models genetic structure by optimizing the correlation between genetic principal components (PCs) and the geodemographic features. It uses a mathematical optimization problem to find the best linear combination of geodemographic factors that explains genetic variation.
  • Feature Selection: Employing a greedy feature selection algorithm (Orthogonal Matching Pursuit) to identify the most influential geodemographic features (language, social group, geography) driving the genetic structure.
  • Ridge Leverage Scores (RLS): A simple approach developed by the authors to identify specific ethnic groups that are most influential in capturing the observed genetic diversity within India.
  • Population Admixture Analysis: Using ADMIXTURE software to estimate ancestral components and shared ancestry between populations.
  • Mantel Tests: Used to assess the correlation between genetic (FST) and geographic distances.
  • Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA): Standard techniques for visualizing and analyzing population genetic structure.

Datasets

The study utilizes several datasets:

  • Pan-Indian Dataset: Initially compiled from publicly available sources, comprising 891 individuals from 90 well-defined sociolinguistic groups, genotyped on 47,283 autosomal SNPs.
  • Normalized Subset: A subset of 368 individuals from 33 populations, selected to ensure balanced representation across sociolinguistic groups, language families, and geographic locations. This subset was used for the COGG and subsequent feature selection analyses.
  • Eurasian/East Asian Dataset: An additional dataset of 1,323 individuals from 50 populations, merged with the normalized Indian subset to study shared ancestry between India and Eurasia/East Asia.
  • Outgroup Data: 124 samples of Yorubans (YRI) from the 1000 Genomes project were used as an outgroup for f3 statistics analysis.

Key Insights and Ideas

  • Language and Social Structure Trump Geography: The most striking finding is that spoken language and social structure (specifically endogamy within the caste system) are the dominant forces shaping genetic diversity in India, more so than geographic distance. This is a significant departure from patterns seen in many other parts of the world where geography is the primary driver.
  • Endogamy's Deep Impact: The strict endogamy imposed by the Indian caste system has created distinct genetic boundaries, overriding geographical proximity in many cases.
  • Identifying Key Drivers: The COGG framework successfully identified specific features – Austro-Asiatic (AA) and Tibeto-Burman (TB) language groups, and the "Forward Castes" (SGA) – as the most significant contributors to explaining genetic variation. These features capture a complex interplay of geographic, social, and linguistic diversity.
  • Distinct Indian Genetic Landscape: The weak correlation between genetic and geographic distances in India, as revealed by Mantel tests, underscores its unique population history compared to regions like Europe.
  • Eurasian Connections: The study reveals distinct genetic affinities between Indian populations and Eurasian groups. Indo-European and Dravidian speakers in India show shared drift with Europeans, while Tibeto-Burman speakers exhibit stronger connections with East Asians.

Key Results

  • COGG Performance: COGG significantly improved the correlation between genetic principal components and geodemographic factors. For PC1, the correlation (r²) increased from 0.6 to 0.93, and for PC2, it rose from 0.06 to 0.85. When extended to analyze multiple PCs simultaneously (COGG-CCA), the r² reached 0.94.
  • Feature Importance: The feature selection identified {AA, TB, SGA} and {AA, Latitude, SGA} as the most important feature sets, capturing over 99% of the correlation explained by all nine geodemographic features.
  • RLS Identification: The RLS statistic highlighted specific ethnic groups (e.g., Great Andamanese, Vysya, Jarawa) that are crucial for understanding India's genetic diversity, either due to isolation, founder effects, or admixture.
  • Shared Drift Patterns: Europeans showed shared genetic drift with Indo-European and Dravidian speakers of India, while East Asians had the maximum shared genetic drift with Tibeto-Burman speaking tribal groups.

Limitations and Future Work

  • Data Limitations: While the study integrated a substantial amount of data, the authors acknowledge that the pan-Indian dataset had unequal representations of language families and social groups, necessitating the creation of a normalized subset for certain analyses.
  • Interpretation of Andamanese Findings: The observed shared ancestry between Great Andamanese and Tibeto-Burman speakers is noted as a novel finding that should be interpreted with caution due to small sample sizes.
  • Generalizability of COGG: The authors suggest that the COGG framework is generalizable and can be applied to model genetic structure in other populations by correlating genomic background with geographic, social, and cultural differences. This opens avenues for understanding how human history and mating patterns translate into genomic diversity.

Summary updated

Abstract

Abstract India represents an intricate tapestry of population substructure shaped by geography, language, culture and social stratification. While geography closely correlates with genetic structure in other parts of the world, the strict endogamy imposed by the Indian caste system and the large number of spoken languages add further levels of complexity to understand Indian population structure. To date, no study has attempted to model and evaluate how these factors have interacted to shape the patterns of genetic diversity within India. We merged all publicly available data from the Indian subcontinent into a data set of 891 individuals from 90 well-defined groups. Bringing together geography, genetics and demographic factors, we developed COGG (Correlation Optimization of Genetics and Geodemographics) to build a model that explains the observed population genetic substructure. We show that shared language along with social structure have been the most powerful forces in creating paths of gene flow in the subcontinent. Furthermore, we discover the ethnic groups that best capture the diverse genetic substructure highlighted by COGG. Integrating data from India with a data set of additional 1,323 individuals from 50 populations we find that Europeans show shared genetic drift with the Indo-European and Dravidian speakers of India, whereas the East Asians have the maximum shared genetic drift with Tibeto-Burman speaking tribal groups.
BibTeX
@misc{bose2017integrating,
  author = {Aritra Bose and Daniel E. Platt and Laxmi Parida and Petros Drineas and Peristera Paschou},
  title = {{Integrating linguistics, social structure, and geography to model genetic diversity within India}},
  year = {2017},
  doi = {10.1101/164640},
  howpublished = {bioRxiv},
}
Figure