Causal effects on complex traits are similar across segments of different continental ancestries within admixed individuals

Abstract

Individuals of admixed ancestries (e.g., African Americans) inherit a mosaic of ancestry segments (local ancestry) originating from multiple continental ancestral populations. Their genomic diversity offers the unique opportunity of investigating genetic effects on disease across multiple ancestries within the same population. Quantifying the similarity in causal effects across local ancestries is paramount to studying genetic basis of diseases in admixed individuals. Such similarity can be defined as the genetic correlation of causal effects (radmix) across African and European local ancestry backgrounds. Existing studies investigating causal effects variability across ancestries focused on cross-continental comparisons; however, such differences could be due to heterogeneities in the definition of environment/phenotype across continental ancestries. Studying genetic effects within admixed individuals avoids these confounding factors, because the genetic effects are compared across local ancestries within the same individuals. Here, we introduce a new method that models polygenic architecture of complex traits to quantify radmix across local ancestries. We model genome-wide causal effects that are allowed to vary by ancestry and estimate radmix by inferring variance components of local ancestry-aware genetic relationship matrices. Our method is accurate and robust across a range of simulations. We analyze 38 complex traits in individuals of African and European admixed ancestries (N = 53K) from: Population Architecture using Genomics and Epidemiology (PAGE), UK Biobank (UKBB) and All of Us (AoU). We observe a high similarity in causal effects by ancestry in meta-analyses across traits, with estimated radmix=0.95 (95% credible interval [0.93, 0.97]), much higher than correlation in causal effects across continental ancestries. High estimated radmix is also observed consistently for each individual trait. We replicate the high correlation in causal effects using regression-based methods from marginal GWAS summary statistics. We also report realistic scenarios where regression-based methods yield inflated estimates of heterogeneity-by-ancestry due to local ancestry-specific tagging of causal variants, and/or polygenicity. Among regression-based methods, only Deming regression is robust enough for estimation of correlation in causal effects by ancestry. In summary, causal effects on complex traits are highly similar across local ancestries and motivate genetic analyses that assume minimal heterogeneity in causal effects by ancestry.