Membranes Matter: Predicting Drug Toxicity

Abstract

It remains a challenge to predict whether a new drug candidate will have undesirable side effects. Here we explore a general mechanism that may cause side effects, namely that many biologically active molecules, including drugs and drug-leads, are amphiphiles that partition into lipid bilayers. This may alter bilayer physical properties, which may cause undesirable changes in cell function that, if large enough, may cause toxicity. Thus, it may be possible to predict whether a compound will have important off-target effects based on the compound's bilayer-modifying potential. Using a gramicidin-based fluorescence assay (GBFA), which reports how a compound alters the gramicidin monomer <-> dimer equilibrium, we have shown that many drugs and drug-leads alter lipid bilayer properties at the concentrations where these compounds become indiscriminate modifiers of membrane protein function and are likely to have off target effects. We pursued this question in a blinded study on a library of 488 compounds (289 non-toxic, 199 toxic) that had been tested for cytotoxicity with “high-content” screening. We found that the GBFA can be used to predict cellular toxicity, with a false positive rate of 5%. We are currently investigating why these false positives arise. We hypothesize that their toxicity is due to direct protein interactions and are examining the impact of albumin on their bilayer perturbing effects. We also explore a computational approach to gain insight into which physicochemical parameters drive a compound's bilayer-perturbing propensity. We can predict which compounds will produce bilayer-modifying effects at either extreme (i.e. little membrane effect or significant membrane perturbation) with 90% accuracy. Our results support a mechanism by which amphiphiles exert their toxicity, namely by altering lipid bilayer physical properties and that this in vitro measurement could be used as a warning sign for off-target biological effects in drug discovery efforts.