Different pleiotropic scenarios determine joint distribution of human metabolite levels
Adding covariates can either increase or decrease the power of finding an effect of a predictor on the outcome of interest. In a recent publication in GigaScience we used knowledge of biochemical networks to select covariates for conditional genetic association analysis. We demonstrated that the power depends on whether genes and environment play along or not, and provide examples of different pleiotropic scenarios1.
The Cheverud conjecture2 implies that genetic and environmental correlations normally have the same sign and the same magnitude. This means that on average, genes and environment are expected to “play along”. We showed that under the “Cheverud scenario” inclusion of a covariate that is genetically correlated with the outcome may lead to loss of power. However, under an anti-Cheverud scenario, when the gene variant does not play along with the residual sources of covariance, the power will be increased.
Even more interestingly, we were able to provide clear sequence variant-level examples of Cheverud and anti-Cheverud behavior of variants affecting metabolite levels (see Figure). The specific examples that we find for acylcarnitines and lysophosphatidylcholines are supported by mechanistic knowledge of biochemical regulation of these metabolites.
This work was done in collaboration with Novosibirsk State University and Helmholtz Center Munich.