Population-Matched Transcriptome Prediction Increases TWAS Discovery and Replication Rate

Abstract

Most genome-wide association studies (GWAS) and transcriptome-wide association studies (TWAS) focus on European populations; however, these results cannot always be accurately applied to non-European populations due to genetic architecture differences. Using GWAS summary statistics in the Population Architecture using Genomics and Epidemiology study, which comprises $50,000 Hispanic/Latinos, African Americans, Asians, Native Hawaiians, and Native Americans, we perform TWAS to determine gene-trait associations. We compared results using three transcriptome prediction models derived from Multi-Ethnic Study of Atherosclerosis populations: the African American and Hispanic/Latino (AFHI) model, the European (EUR) model, and the African American, Hispanic/ Latino, and European (ALL) model. We identified 240 unique significant trait-associated genes. We found more significant, colocalized genes that replicate in larger cohorts when applying the AFHI model than the EUR or ALL model. Thus, TWAS with population-matched transcriptome models have more power for discovery and replication, demonstrating the need for more transcriptome studies in diverse populations.

Transcriptome-Wide Association Study of Blood Cell Traits in African Ancestry and Hispanic/Latino Populations

“…Previous work has also suggested the value of ancestry-matched TWAS reference panels. Recent work from Geoffroy, et al [ 24 ] found more significant complex trait-associated genes in summary statistics from the diverse PAGE consortium (~50,000 HL, African American, Asian, Native Hawaiian, and Native American individuals) using models trained in African American and HL MESA participants than in European- or all-ancestry TWAS models. Work in the diverse Carolina Breast Cancer Study (CBCS) cohort also suggests that TWAS models perform poorly across race/ethnicity groups, with multiple novel discoveries for breast cancer survival found in AA women using TWAS gene expression models trained in a subset of the cohort with measured expression data [ 21 ].…”

“…Transcripts whose levels can be confidently imputed from genetic variants are then assessed for phenotype associations [ 17 ]. However, like GWAS, TWAS analyses have often included only EA populations [ 18 , 19 ], though some recent efforts have included more diverse populations [ 20 , 21 , 22 , 23 , 24 ]. Furthermore, most reference eQTL datasets used for TWAS include predominantly EA American individuals [ 25 , 26 ], limiting the predictive power of this novel method in other populations with different allele frequencies.…”

Transcriptome-wide Association Study of Blood Cell Traits in African Ancestry and Hispanic/Latino Populations

“…Previous work has also suggested the value of ancestry-matched TWAS reference panels. Recent work from Geoffroy, et al, [22] found more significant complex trait-associated genes in summary statistics from the diverse PAGE consortium (~50,000 HL, African American, Asian, Native Hawaiian, and Native American individuals) using models trained in African American and HL MESA participants than in European-or all-ancestry TWAS models. Work in the diverse Carolina Breast Cancer Study (CBCS) cohort also suggests that TWAS models perform poorly across race/ethnicity groups, with multiple novel discoveries for breast cancer survival found in AA women using TWAS gene expression models trained in a subset of the cohort with measured expression data [19].…”

“…Transcripts whose levels can be confidently imputed from genetic variants are then assessed for phenotype associations [15]. However, like GWAS, TWAS analyses have often included only EA populations [16,17], though some recent efforts have included more diverse populations [18][19][20][21][22]. Furthermore, most reference eQTL datasets used for TWAS include predominantly EA American individuals [23,24], limiting the predictive power of this novel method in other populations with different allele frequencies.…”

Multitrait transcriptome‐wide association study (TWAS) tests

“…Some of these strategies build predictors using single‐tissue expression reference panes, and others use multiple‐tissue panels (Barbeira et al, 2019, 2020; Feng et al, 2021; Hu et al, 2019). TWAS methods have also been extended to account for confounding due to colocalization or pleiotropy or differences in expression prediction across ethnicity (Barfield et al, 2018; Bhattacharya et al, 2020; Geoffroy et al, 2020; L. Liu et al, 2021; Mancuso et al, 2019; Mogil et al, 2018). As we discuss further below, several approaches to multitrait TWAS have recently been proposed (Baselmans et al, 2019; L. Liu et al, 2021; Nagpal et al, 2019).…”