Fig. 1: Overview of the three-sample multivariable Mendelian randomization (3S-MVMR) design to quantify mediation of complex traits through DNA methylation (DNAm) and transcripts.
Genetic instruments (SNPs) are selected to be directly and significantly associated (dashed arrows) with either the exposure (DNAm, green) or any mediator k (transcript in cis, red). The total effect θT (green-blue dotted arrow) of the exposure on the outcome (complex trait, blue) is estimated in a univariable MR analysis based on exposure-associated SNPs only. The direct effect θD (green-blue arrow) is estimated in an MVMR analysis on all valid instruments. The mediation effect θM (green-red and red-blue arrows) results from the difference between θT and θD, and allows to calculate the mediation proportion (MP). The genetic effect sizes β on the exposure, mediator and outcome come from m/eQTL and GWAS summary statistics, respectively. Transcripts were required to be causally associated to the DNAm-exposure to be included as mediators.
