Linking protein to phenotype with Mendelian Randomization detects 38 proteins with causal roles in human diseases and traits

To efficiently transform genetic associations into drug targets requires evidence that a particular gene, and its encoded protein, contribute causally to a disease. To achieve this, we employ a three-step proteome-by-phenome Mendelian Randomization (MR) approach. In step one, 154 protein quantitative trait loci (pQTLs) were identified and independently replicated. From these pQTLs, 64 replicated locally-acting variants were used as instrumental variables for proteome-by-phenome MR across 846 traits (step two). When its assumptions are met, proteome-by-phenome MR, is equivalent to simultaneously running many randomized controlled trials. Step 2 yielded 38 proteins that significantly predicted variation in traits and diseases in 509 instances. Step 3 revealed that amongst the 271 instances from GeneAtlas (UK Biobank), 77 showed little evidence of pleiotropy (HEIDI), and 92 evidence of colocalization (eCAVIAR). Results were wide ranging: including, for example, new evidence for a causal role of tyrosine-protein phosphatase non-receptor type substrate 1 (SHPS1; SIRPA) in schizophrenia, and a new finding that intestinal fatty acid binding protein (FABP2) abundance contributes to the pathogenesis of cardiovascular disease. We also demonstrated confirmatory evidence for the causal role of four further proteins (FGF5, IL6R, LPL, LTA) in cardiovascular disease risk.Author summary: The targets of most medications prescribed today are proteins. For many common diseases our understanding of the underlying causes is often incomplete, and our ability to predict whether new drugs will be effective is remarkably poor. Attempts to use genetics to identify drug targets have an important limitation: standard study designs link disease risk to DNA but do not explain how the genotype leads to disease. In our study, we made robust statistical links between DNA variants and blood levels of 249 proteins, in two separate groups of Europeans. We then used this information to predict protein levels in large genetic studies. In many cases, this second step gives us evidence that high or low levels of a given protein play a role in causing a given disease. Among dozens of high-confidence links, we found new evidence for a causal role of a protein called SHPS1 in schizophrenia, and of another protein (FABP2) in heart disease. Our method takes advantage of information from large numbers of existing genetic studies to prioritize specific proteins as drug targets.