Mendelian Randomization integrating GWAS and eQTL data reveals genetic determinants of complex and clinical traits

Abstract

Genome-wide association studies (GWAS) identified thousands of variants associated with complex traits, but their biological interpretation often remains unclear. Most of these variants overlap with expression QTLs (eQTLs), indicating their potential involvement in the regulation of gene expression. Here, we propose an advanced summary statistics-based Mendelian Randomization approach that uses multiple SNPs jointly as instruments and multiple gene expression traits as exposures, simultaneously. When applied to 43 human phenotypes it uncovered 2,277 putative genes whose blood expression is causally associated with at least one phenotype resulting in 5,009 gene-trait associations; of note, 55% of them had no genome-wide significant SNP nearby in previous GWAS analysis. Using independent association summary statistics (UKBiobank), we confirmed that the majority of these loci were missed by conventional GWAS due to power issues. Noteworthy among these novel links are height-and intelligence-associated PEX19 and CDC42, respectively known to carry mutations leading to short stature and Takenouchi-Kosaki syndrome. We similarly unraveled novel pleiotropic causal effects suggestive of mechanistic connections, e.g. the shared genetic effects of TSPAN14 in rheumatoid arthritis, Crohn's and inflammatory bowel disease. Finally, we show that causal genes can be highly tissuespecific. Our advanced Mendelian Randomization unlocks hidden value from published GWAS through higher power in detecting associations. It better accounts for pleiotropy and unravels new biological mechanisms underlying complex and clinical traits.

A novel Mendelian randomization method identifies causal relationships between gene expression and low-density lipoprotein cholesterol levels

“…MR-link has several advantages over other recent MR methods developed to overcome bias from LD and pleiotropy 17,23 . First, MR-link can model unobserved pleiotropy, whereas sources of pleiotropy need to be specified in multivariate MR methods.…”

“…Similarly, the presence of pleiotropy cannot be excluded a priori given that the majority of variants in our genome are likely to affect one or multiple phenotypes 14–16 . While there are MR-methods 7,8,10,17–21 that extend standard MR analysis to correct for LD and pleiotropy, the application of these methods is not optimal because they require either the removal of pleiotropic IVs from the statistical model 7,20,21 or that all sources of pleiotropy are measured and incorporated into the model 22,23 . These constraints limit robust inference of the causal role of gene expression traits as there is often only a limited number of IVs (i.e.…”

On the cross-population generalizability of gene expression prediction models

“…As evidenced by application to numerous disease domains, the TWAS framework is capable of uncovering new genic associations [11][12][13][14][15][16][17]. However, the power of TWAS is inherently limited by the data used for eQTL discovery.…”

Characterising the genetic basis of immune response variation to identify causal mechanisms underlying disease susceptibility

“…Such approaches, however, are only the first step toward the identification of the causal mechanisms that underlie disease susceptibility. Validation of the proposed mechanisms, through colocalization analysis, 83 mendelian randomization approaches, 97 and randomized clinical trials, 98 are then required to formally establish causality. Finally, as whole genome sequencing is becoming increasingly common, 19 the ability to identify rare regulatory variants at known disease genes will become a crucial step toward personalized treatments of both rare and common immune disorders.…”