Trans-ethnic meta-regression of genome-wide association studies accounting for ancestry increases power for discovery and improves fine-mapping resolution

Abstract

Trans-ethnic meta-analysis of genome-wide association studies (GWAS) across diverse populations can increase power to detect complex trait loci when the underlying causal variants are shared between ancestry groups. However, heterogeneity in allelic effects between GWAS at these loci can occur that is correlated with ancestry. Here, a novel approach is presented to detect SNP association and quantify the extent of heterogeneity in allelic effects that is correlated with ancestry. We employ trans-ethnic meta-regression to model allelic effects as a function of axes of genetic variation, derived from a matrix of mean pairwise allele frequency differences between GWAS, and implemented in the MR-MEGA software. Through detailed simulations, we demonstrate increased power to detect association for MR-MEGA over fixed- and random-effects meta-analysis across a range of scenarios of heterogeneity in allelic effects between ethnic groups. We also demonstrate improved fine-mapping resolution, in loci containing a single causal variant, compared to these meta-analysis approaches and PAINTOR, and equivalent performance to MANTRA at reduced computational cost. Application of MR-MEGA to trans-ethnic GWAS of kidney function in 71,461 individuals indicates stronger signals of association than fixed-effects meta-analysis when heterogeneity in allelic effects is correlated with ancestry. Application of MR-MEGA to fine-mapping four type 2 diabetes susceptibility loci in 22,086 cases and 42,539 controls highlights: (i) strong evidence for heterogeneity in allelic effects that is correlated with ancestry only at the index SNP for the association signal at the CDKAL1 locus; and (ii) 99% credible sets with six or fewer variants for five distinct association signals.

Transethnic meta‐analysis of metabolic syndrome in a multiethnic study

“…Han & Eskin stress in the methodological paper for RE2 that when a meta-analysis is comprised of a small number of studies, their table of pretabulated p values should be used (2011). Simulations and applications on MR-MEGA in Magi et al (2017) use at least nine studies, so we are unsure of the existing literature of MR-MEGA as to how many studies are required to produce valid results. An excess of p value = 1 in the TransMeta QQplot leads to deviations from the expected distribution and deflated λ values at the median (Figure 1).…”

“…The summary statistics from the association analyses in each sample were combined in meta-analyses separately by phenotype using FE (Stram, 2014), RE2 (Han & Eskin, 2011), TransMeta (Shi & Lee, 2016), and MR-MEGA (Magi et al, 2017). FE and RE2 were implemented using METASOFT (Han & Eskin, 2011).…”

“…The final transethnic method, MR-MEGA (Magi et al, 2017), follows a meta-regression framework that includes multidimensional scaling (MDS) axes as covariates to adjust for transethnic effects between studies. The MDS axes are found from an n n × pairwise genetic distance matrix whose ij th entry is indexes the ith GWAS study, we model (Magi et al, 2017):…”

“…The MDS axes utilized in MR MEGA are automatically calculated by the MR-MEGA software, and no alterations were specified for this calculation (Magi et al, 2017). The MDS axes utilized in MR MEGA are automatically calculated by the MR-MEGA software, and no alterations were specified for this calculation (Magi et al, 2017).…”

“…This dataset is representative of transethnic effects because it consists of samples from four diverse backgrounds: African-American, European-American, Japanese-American, and Mexican-American. We compare the two standard methods, fixed effects (FE; Stram, 2014) and random effects (RE2; Han & Eskin, 2011) meta-analysis, along with two transethnic methods, TransMeta (Shi & Lee, 2016) and MR-MEGA (Magi et al, 2017). We compare the two standard methods, fixed effects (FE; Stram, 2014) and random effects (RE2; Han & Eskin, 2011) meta-analysis, along with two transethnic methods, TransMeta (Shi & Lee, 2016) and MR-MEGA (Magi et al, 2017).…”