Predicting Toxicity of Insect Venom-Derived Antimicrobial Peptides Using MD Simulations: A Comparative Study of Multi-Component and Realistic Mammalian Membrane Models.

Abstract

Insect venom-derived antimicrobial peptides (AMPs) hold significant therapeutic promise, but their application is constrained by mammalian cell toxicity. Toxicity assays are rapid and high-throughput, but screening large peptide libraries remains resource-intensive due to the requirements for peptide synthesis, purification, and testing. Alternatively, molecular dynamics (MD) simulations using mammalian membrane models provide an efficient and robust method for preliminary toxicity prediction. To benchmark the optimal model, two distinct mammalian membrane systems with diverse lipid compositions were evaluated for a set of sixteen toxic and fourteen non-toxic AMP analogs from five distinct insect AMP families, viz. anoplin, polybia, halictine, hyline, and macropin. In this study, a total of 25 µs of MD simulation time was generated. The analysis of MD trajectories, each spanning 500 ns for each of the 30 peptides, revealed significant variations in structural stability and membrane permeability between toxic and non-toxic AMPs, which aligned with the experimental results. Root Mean Square Deviation (RMSD) of the peptides during the last 100 ns of the simulation period successfully distinguished toxic from non-toxic AMPs with 90% accuracy when using realistic membrane models. The well-cited multicomponent mammalian membrane model failed to effectively predict mammalian toxicity. These findings underscore the efficacy of MD simulations in predicting the toxicity of venom-derived AMPs, thereby opening avenues for the accelerated development of safer antimicrobial therapies.