On the adjustment for covariates in genetic association analysis: a novel, simple principle to infer direct causal effects

Abstract

In genetic association studies, different complex phenotypes are often associated with the same marker. Such associations can be indicative of pleiotropy (i.e. common genetic causes), of indirect genetic effects via one of these phenotypes, or can be solely attributable to non-genetic/environmental links between the traits. To identify the phenotypes with the inducing genetic association, statistical methodology is needed that is able to distinguish between the different causes of the genetic associations. Here, we propose a simple, general adjustment principle that can be incorporated into many standard genetic association tests which are then able to infer whether an SNP has a direct biological influence on a given trait other than through the SNP’s influence on another correlated phenotype. Using simulation studies, we show that, in the presence of a non-marker related link between phenotypes, standard association tests without the proposed adjustment can be biased. In contrast to that, the proposed methodology remains unbiased. Its achieved power levels are identical to those of standard adjustment methods, making the adjustment principle universally applicable in genetic association studies. The principle is illustrated by an application to three genome-wide association analyses.

Common Coding Variants of the HNF1A Gene Are Associated With Multiple Cardiovascular Risk Phenotypes in Community-Based Samples of Younger and Older European-American Adults

“…Together these observations support the complex genetic regulatory relationships that likely underlie the direction of some of the observed HNF1A genotype – CVD phenotypic associations, as well as the observation that by adjusting our fibrinogen analysis for CRP, we were able to uncover an HNF1A genotype-fibrinogen relationship that was distinct from the observed effect of HNF1A genotype on CRP. While multiple phenotypic associations may suggest a common genetic cause (pleiotropy), it is also important to point out that such a scenario may also represent indirect genetic effects primarily with a subset of these phenotypes and/or complex non-genetic/environmental correlations between CVD traits [43]. Ultimately, the application of more complex multivariate statistical methods along with molecular functional studies will be required distinguish among these possibilities.…”

Pleiotropy in complex traits: challenges and strategies

“…If the association persists (that is, if the variant is associated with the target phenotype even when the intermediate phenotype is not present), then the CP effect is probably not fully mediated. However, this approach can produce biased results when the phenotypes share a that is influenced by the genetic variant 69 . To address this shortcoming, approaches using causal inference methodology have been developed to test whether a genetic variant influences the target phenotype through a path that does not involve the intermediate phenotype 69–71 .…”

“…However, this approach can produce biased results when the phenotypes share a that is influenced by the genetic variant 69 . To address this shortcoming, approaches using causal inference methodology have been developed to test whether a genetic variant influences the target phenotype through a path that does not involve the intermediate phenotype 69–71 . Such an approach demonstrates that the association between SNPs at 15q25.1 with both smoking and lung cancer mostly reflects direct effects on each phenotype, rather than mediated pleiotropy 72 .…”

Methods for Analyzing Multivariate Phenotypes in Genetic Association Studies

“…The detect association can affect some of the phenotypes and/or mediate through these phenotypes to affect the other phenotypes. Vansteelandt et al [38] illustrated potential confounding mechanism between the genotype of a genetic marker and a phenotype using a causal diagram (Figure 1): the association between the genotype, denoted as G , and the response phenotype Y can occur through the paths connecting the two variables along all unbroken sequences of edges regardless of the direction of the arrows, given that there are no colliders (i.e., variables in which two arrows converge, e.g., variables K and L in Figure 1) in the sequence [39]. …”

“…To overcome the limitation of the two commonly adapted approaches, Vansteelandt et al [38] proposed a least squared regression model to estimate the direct effect size of K on Y . This regression model includes the suspected intermediate phenotype, the score of the genetic marker genotype, X ( G ), and other common risk factors between the two phenotypes as regressors: $$Efalse(Y_{i}false)={\gamma }_0+{\gamma }_1{K}_i+{\gamma }_2{X}_i+{\gamma }_3{L}_i.$$…”