Ancestry-aware mixed model gwas of major depression charts a path for inclusive and diverse genetic research

Genome-wide association studies (GWAS) in individuals of European ancestry have identified >200 loci for major depression (MD). There has been a slow increase in ancestral diversity of participants. Most previous studies assigned subjects to ancestry groups based on genetic similarity to reference samples. Individuals who could not be assigned, especially those with mixed ancestry, were routinely excluded, leading to ethical concerns and missed scientific opportunities. To address this, we conducted a large GWAS meta-analysis of MD in samples from mixed-ancestry populations using an ancestry-aware mixed model. This project included 12 cohorts which had recruited individuals with diverse ethnicity, including the UK Biobank (UKB), WHI, HCHS/SOL, GERA, IHS, Army-STARRS, eMERGE, QIMRB, MESA, BHRCS, HELIUS and the MVP. Results reported here were based on a subset of 6 cohorts including 14,001 cases and 44,122 controls. Individuals of European ancestry were excluded, if they had already been part of previous studies. PC-Air was used for robust population structure inference in the presence of known or cryptic relatedness, and PC-Relate for accurate genetic relatedness inference in samples with complex population structure, enabling a joint analysis of samples from diverse ancestry groups in a single analysis. Fixed effect meta-analysis was conducted to combine results across studies. PC-relate uncovered incorrect relatedness assignments by methods not designed for admixed populations. For example, 1503 pairs classified as relatives in the official UK Biobank release using KING were shown to be unrelated by PC-Relate. We first analysed each cohort separately and compared results to the conventional approach based on separating samples by ancestry, analysing them separately, followed by meta-analysis. In the UKB cohort, the mixed ancestry GWAS identified a novel locus at 7q32.1 (P = 4.37 × 10–8), whereas it failed to reach the genome-wide significance threshold when using the conventional approach. The nearest protein coding gene, TSPAN33, is involved in pore complex assembly. A nearby variant, rs34534607, was previously reported to have a suggestive association with MD (P = 6 × 10–6) (Coleman et al., Mol Psychiatry, 2020). A meta-analysis combining data from the cohorts analysed to date did not identify any novel associations. We found relatedness misclassification by conventional approaches in cohorts with individuals of admixed ancestry. Our results demonstrate the potential increase in power when samples with admixed ancestry are included, highlighting the scientific opportunities alongside the ethical importance of more inclusive analytic approaches.