Can peptide folding simulations provide predictive information for aggregation propensity?

Abstract

Nonnative peptide aggregation underlies many diseases and is a major problem in the development of peptide-based therapeutics. Efforts in the past decade have revealed remarkable correlations between aggregation rates or propensities and very simple sequence metrics like hydrophobicity and charge. Here, we investigate the extent to which a molecular picture of peptide folding bears out similar relationships. Using replica exchange molecular dynamics folding simulations, we compute equilibrium conformational ensembles of 142 hexa- and decapeptide systems, of which about half readily form amyloid fibrils and half do not. The simulations are used to compute a variety of ensemble-based properties, and we investigate the extent to which these metrics provide molecular clues about fibril formation. To assess whether multiple metrics together are useful in understanding aggregation, we also develop a number of logistic regression models, some of which predict fibril formers with 70-80% accuracy and identify aggregation-prone regions in larger proteins. Importantly, these models quantify the importance of different molecular properties in aggregation driving forces; notably, they suggest that hydrophobic interactions play a dominant role.