South Asian medical cohorts reveal strong founder effects and high rates of homozygosity

Abstract

The benefits of large-scale genetic studies for healthcare of the populations studied are well documented, but these genetic studies have traditionally ignored people from some parts of the world, such as South Asia. Here we describe whole genome sequence (WGS) data from 4806 individuals recruited from the healthcare delivery systems of Pakistan, India and Bangladesh, combined with WGS from 927 individuals from isolated South Asian populations. We characterize population structure in South Asia and describe a genotyping array (SARGAM) and imputation reference panel that are optimized for South Asian genomes. We find evidence for high rates of reproductive isolation, endogamy and consanguinity that vary across the subcontinent and that lead to levels of rare homozygotes that reach 100 times that seen in outbred populations. Founder effects increase the power to associate functional variants with disease processes and make South Asia a uniquely powerful place for population-scale genetic studies.

Fig. 1. Fine-scale population structure in the healthcare delivery system reflects the geographical locations of the sample sources.

UMAP was run on all samples using the first 15 principal components. a In the South Asian subset, samples cluster into three major groups by sample origins: Pakistan, South India, and West Bengal and Bangladesh. The X-axis (UMAP1) was flipped so that the similarity between the graphical position of the three populations and the map of South Asia was apparent. b, c Samples with detailed locations or self-reported group memberships are shown to segregate within Pakistan and South India clusters. Among the samples from Pakistan and South India, some segregate with recent immigrants (e.g., Bengalis and Gujaratis) and historical immigrants (e.g., Lambadas), reflecting the metropolitan nature of the recruitment centers. d Samples from Birbhum District, West Bengal, have detailed self-reported group membership information. Upper castes, scheduled castes, and scheduled tribes clearly segregate, reflecting the historical reproductive isolation between these groups. Bayen and Santhal are two notable population isolates. e ADMIXTURE analysis of samples from the Birbhum District shows four major components. Labels are self-reported group identities with “general” denoting a lack of specified identity. PKN Pakistan, BLR Bangalore, MAA Chennai, COI Coimbatore, BAN Bangladesh, BRB Birbhum District, West Bengal, LAM Lambada. For other 3-letter codes, see Supplementary Data 1.

Fig. 2. Homozygosity and inbreeding across different cohorts.

a Observed/expected proportions of rare homozygotes, stratified by minor allele frequency and population. The expected values assume random mating. b Stacked bar chart showing the estimated degree of inbreeding for individuals in the South Asian medical cohorts. c Same as in panel a but for “inbred” individuals (whose parents are estimated to be third-degree relatives or more closely related) only. d Same as in panel a but for “outbred” individuals (whose parents are estimated to be sixth-degree relatives or more distantly related) only. e Ridgeplots showing the distribution across individuals of the total (genetic) length of the genome contained in ROHs that are at least 1 cM in length. f Ridgeplots showing the stratification of panel e’s PKN plot into groups with different estimated degrees of inbreeding.

Fig. 3. Loss of function mutations.

a Number of high confidence loss of function genes found at a minimum of 0.1% MAF in their relative population for overall non-Finnish European (NFE), NFE and not in SAS (NFE Unique), NFE and SAS (NFE and SAS), SAS and not NFE (SAS Unique) and overall for SAS. b Loss of function gene space by population. Each square represents a distinct gene and is colored by its maximum AF within the relative group. Genes are separated by groups in which they are found (from top to bottom and then left to right): NFE unique, NFE and SAS, PKN unique, PKN and SOI, PKN and BNG, all of SAS (PKN and SOI and BNG), SOI unique, BNG unique, and BNG and SAS. c Effects of pLoF variants on blood lipid markers replicated the known biology: PCSK9 pLoFs associated with decreased LDL, ANGPTL3 pLoFs associated with decreased triglycerides, and CETP pLoF associated with increased HDL. Only samples from South India (Bangalore and Chennai) were included. P values were calculated using the Wilcoxon rank-sum test. Box shows median and middle 50% of the distribution; whiskers show values within 1.5 times the interquartile range from the first and third quartiles. d Mean number of homozygous pLoF variants per individual, stratified by population and estimated degree of inbreeding. e APOC3 p.Arg19Ter alleles are found at a high frequency among Balochi and Sindhi individuals from Southern Pakistan. Three of the self-reported Balochis and Sindhis were heterozygous carriers, but a larger number of carriers without self-reported identity were mapped to the same region on the UMAP plot.

Fig. 4. Improved genotyping of South Asian genomes.

a Gene space plot of all protein-altering alleles that are directly genotyped using either the SARGAM or the Illumina GSA3 arrays. Protein-coding genes of the human genome are depicted as an array of 19,600 squares. Genes whose variants are genotyped are colored to indicate the number of gene-specific variants that are genotyped. b Accuracy of non-reference allele imputation expressed as the concordance rate and plotted versus South Asian minor allele frequency. Array genotypes were modeled by down-sampling from an independent dataset of 30× WGS data. Missing genotypes were imputed using the indicated reference panels and the variant site accuracy of non-reference alleles was calculated and graphed for variants imputed from the two indicated model array datasets. c Impact of imputation on polygenic risk score (PRS) calculation. PRS were calculated using imputed genotypes from a CAD case–control cohort of 2963 South Asian individuals genotyped using the Illumina GSA3 array and using a SAS PRS model. The individuals were divided into 10 groups based on deciles of PRS and odds ratios were calculated from the case–control status of the individuals in each group. For comparison, a case–control cohort of white Britons, matched for age and gender with the SAS cohort, was selected from the UK Biobank dataset. PRS was calculated using a European model; point estimates of the odds ratios are displayed as solid lines for each PRS, and the corresponding 95% confidence intervals (using the empirical variance based on the case/control counts in each decile) are shown as a shaded area.