Fig. 1: Conceptual framework of the study.

Illustration of identifying putative factors that influence the original observations. a Where (gx) is the SNP–exposure effect, (xy) is the exposure–outcome effect as estimated through MR analysis from the non-outlier SNPs, (gp) is the SNP–candidate trait effect and (py) is the causal effect of the candidate trait on the outcome. b The open circles represent valid instruments and the slope of the dotted line represents the causal effect estimate of the exposure on the outcome. The closed red circle represents an outlier SNP which influences the outcome through two independent pathways, P and X. c One way in which the red SNP can exhibit a larger influence on the outcome than expected given its effect on the exposure is if it influences the outcome additionally through another pathway (horizontal pleiotropy). d Using the MR-Base database of GWAS summary data for hundreds of traits we can search for ‘candidate traits’ with which the outlier SNP has an association. e Instruments excluding the original outlier SNP are obtained for each candidate trait, LASSO-based multivariable MR is used to prune the candidate traits to avoid redundancy, and the causal influence of each of those independent candidate traits on the outcome can subsequently be estimated. This allows us to identify alternative traits that putatively influence the outcome and adjust the SNP–outcome associations for pleiotropic pathways in the original exposure–outcome model.