Cross-ancestry polygenic risk scores for psychiatric disorders in indian populations

Psychiatric disorders, while globally prevalent, exhibit heterogeneity in manifestation and risk across diverse populations. This variability can be attributed to a complex interplay of genetic and environmental factors. India's genetic diversity, due to unique demographic events and population substructure, presents a rich but under-explored backdrop for psychiatric genetics research. To bridge this gap, we analyzed approx. 3000 samples from two cohorts from India: cVEDA, a multi-centric pan-India neurodevelopmental cohort; and a clinic-based sample of Bipolar Affective Disorder (BPAD) and Obsessive-Compulsive Disorder (OCD) patients from NIMHANS, India. We utilized Illumina's Global Screening Array (GSA) for genotyping and performed quality control (QC) procedures to maintain dataset integrity. Post-QC, imputation accuracy of our data was assessed against three reference panels: 1000 Genomes (1000G), Genome Asia, and Haplotype Reference Consortium (HRC), with HRC showing superior INFO scores. Principal Component Analysis (PCA) addressed genetic structure and population stratification at global, continental, and regional levels. Ancestry of our samples was further delineated using a Support Vector Machines (SVM) model trained on labeled data from the reference panels (1000 Genome Phase 3, Human Genome Diversity Project and Genome Asia Pilot data.). We conducted unsupervised admixture analysis to explore potential substructure within our Indian samples and estimated the fixation index (Fst) to gauge genetic diversity. The main goal of our analysis was testing the cross-ancestry predictive accuracy of Polygenic Risk Scores (PRS) for Bipolar Disorder (BPAD) and Obsessive-Compulsive Disorder (OCD) derived from European GWAS summary statistics on Indian samples. We employed the Bayesian PRS-CS-auto method with a European LD reference panel and applied continuous shrinkage to refine the estimation of SNP effect sizes, thus enhancing the predictive power of the PRS. The PRS were computed using summary statistics from the PGC's 2017 meta-analysis for OCD and a 2021 meta-analysis for BPAD. The global PCA depicted our Indian samples overlapping with South Asian (SAS) samples from the 1000 Genomes Project. A more detailed sub-continental PCA, leveraging the GenomeAsia dataset, highlighted the extension of the Indian peninsula from SAS towards Oceania, South-East Asia, and North-East Asia. The SVM models predicted the continental labels better than the sub-continental labels. Admixture analysis suggested a 2-ancestry solution, reflecting genetic diversity within Indian populations, comparable to the genomic diversity observed across Europe. The BPAD PRS demonstrated significant predictive power for the disorder [Nagelkerke R2=0.0314, 10th decile OR (95%CI)= 3.08 (1.8, 5.0), N=1993 (Cases: 472, Controls: 1,521)], whereas the OCD PRS did not [N=2078 (Cases: 557, Controls: 1,521)]. The differential PRS prediction accuracy could be due to the smaller sample size of the OCD discovery GWAS and variations in allele frequencies and effect estimates across ancestries. Our results underscore the need for a comprehensive understanding of ancestry's role in psychiatric risk prediction, especially in genetically diverse populations like India. Future genetic research should aim to involve larger and more diverse samples to improve the accuracy and inclusivity of genetic risk predictions for psychiatric disorders across various ancestries.