Forecasting off-target drug toxicity using proteomic and genetic data: insights from Torcetrapib

In the development of new drugs, one of the leading causes of late-stage failures are off-target adverse effects, but they are difficult to predict before expensive large-scale clinical trials. Proteomic changes observed in randomized controlled trials (RCTs) and Mendelian randomization estimates of the effects of these changes can provide valuable evidence about the likely effects of drugs on health outcomes. We provide proof of principle for this approach using data from the ILLUMINATE trial of torcetrapib, a drug developed to increase high-density lipoprotein (HDL) cholesterol while reducing low-density (LDL) cholesterol, but which unexpectedly increased blood pressure and mortality. We used Mendelian randomization to estimate the causal effects of 95 proteins perturbed by 3 months of torcetrapib exposure on 19 health outcomes. Six proteins showed concordant effects with the results of the trial, including C-type mannose receptor 2 (MRC2), cGMP-specific 3';5'-cyclic phosphodiesterase (PDE5A), Spondin-1 (SPON1), and Tyrosine-protein kinase receptor Tie-1 (TIE1), which increased blood pressure in the same direction as their observed protein. Our results demonstrate a generalizable genetic-proteomic framework for predicting likely adverse drug effects, reducing potential harm to patients and drug failure costs.