TWAS were performed using a trans-ancestry method. In brief, this method fits a series of meta-regression models including the first four axes of genetic variation (MDS components), similar to that of our multi-ancestry meta-analysis model minus the random-effect term. Genetic effect estimates from these four models were then used to estimate phenotypic effects of each variant. Together, with variant weights taken from PrediXcan55 based on 49 tissues from GTEx10 release version 8 (which includes up to 15% of individuals of non-EUR ancestry), the phenotypic effect estimates were used to construct a single TWAS statistic for each MDS component. A minimum P value approach56 was then applied to combine all four TWAS statistic P values. Finally, we used a Cauchy combination test57 to combine P values across all available tissues for each gene. The final, combined P value was subjected to a Bonferroni correction for 22,121 genes in 49 tissues. We present our TWAS results based on per gene P values combined across all available tissues, resulting in a 5 (phenotype) × 22,121 (gene) matrix of P values. Pathway enrichment was
