Evaluating off-target drug effects seen in RCTs using proteomics and mendelian randomisation

Abstract Introduction Serious adverse drug effects are often not detected until late in drug development after thousands of patients have been exposed to an investigational drug in long and costly clinical trials. More efficient approaches are desirable. Proteomics-informed Mendelian randomisation (MR) can identify potential molecular mechanisms underlying off-target effects of pharmaceuticals and drug repurposing opportunities. Purpose As a demonstration of this approach, we used findings from a previous plasma proteomics study nested within a clinical trial of torcetrapib, which was terminated early because of an increased risk of cardiac events and mortality, in which 200 plasma proteins significantly increased (n=68) or decreased (n=132) after three months of torcetrapib use. We surveyed potential plasma protein-mediated causal effecs of torcetrapib on a wide array of relevant phenotypes in three broad domains: cardiovascular(n=10), immunological(n=9) and cancer(n=4), representing major categories of adverse outcomes that were more common in the treatment arm. Methods We performed three-sample MR on torcetrapib-associated proteins. First, we identified independent cis-pQTL instruments for these proteins in the Atherosclerosis Risk in Communities (ARIC) Study. Then, we used pQTL effect estimates from the deCODE study, to reduce the potential for bias due to winner’s curse and to increase precision. Lastly, we conducted MR analyses on disease-relevant phenotypes using a debiased-IVW estimator and summary results from the largest available GWAS on these phenotypes: coronary artery disease, ischaemic stroke, blood pressure, lipids, heart rate, QT interval, eGFR, serious infections, blood cell counts, immune cell fractions, and several cancers. The significance level for causal protein-phenotype associations was p=1.85E-5 (0.05/(104*26)), correcting for the 104 proteins with valid instruments and the 26 tested outcomes. Results Thirty-three proteins had evidence of a causal association with at least one phenotype. Seventeen proteins were associated with cardiovascular outcomes, including 8 for blood pressure and 9 for lipid levels. We identified more proteins, 25, associated with immune-related phenotypes. One protein was associated with cancer. Of note, 9 proteins had pleiotropic effects across several domains. Conclusions Our approach of using Mendelian randomisation to investigate the causal effects of plasma proteins impacted by medication with known off-target adverse effects revealed several novel causal relationships, highlighting the role of immune function in the development of cardiovascular disease. These results support the use of intermediate omics evaluation to identify potential biomarkers for safety monitoring for drug development and reveal avenues of investigation for drug repurposing.