(156d) Replica Exchange Molecular Dynamics-Based Methods for Predicting Peptide Self-Assembly and Aggregation

Peptides in solution can assemble into a variety of complex nanoscale morphologies. On one hand, their aggregation is implicated in several famous amyloid diseases, including diabetes and Alzheimer's, and remains a leading problem in the development of bio-therapeutics. On the other, these systems have become versatile platforms for new nanomaterials that self-assemble in response to cues spanning temperature, pH, salt, solvent, and chemical additives. Despite considerable ongoing work in these areas, it has been challenging to develop general, quantitative theories that can capture sequence-specific effects on self-assembly. Such methods could guide experimental design of peptide materials and aggregation-resistant drugs.

Here, we discuss two computational methods for understanding and predicting self-assembly behavior at a level that is sensitive to single-residue mutations. These methods are based entirely on physical atomic interactions and canonical sampling at physiological temperature. Thus they provide realistic molecular driving forces and are, in principle, broadly applicable to both natural and synthetic peptides in materials and medicine.

The first method is a ?one-body? approach that begins with the prediction of the equilibrium conformational ensemble of a single peptide in solution, using extensive replica exchange molecular dynamics (REMD) simulations. From these simulations, molecular physical properties like solvent accessible surface area, helicity, and backbone entropies are computed. These properties are then correlated with aggregation propensity in a regression model using a large library of experimentally-characterized peptides. The method identifies dominant aggregation driving forces in specific sequences and is able to achieve 80% classification success for identifying fibril forming sequences when applied to peptides not included in the regression dataset. Application of the model to larger proteins shows that it successfully predicts aggregation ?hot spots? in their sequences, when compared to results of scanning mutagenesis experiments.

The second method is a ?two-body? approach that computes the potential of mean force (PMF) between two peptides interacting in solution. This technique combines REMD simulations with umbrella sampling techniques to surmount the substantial challenges in exploring the intra- and intermolecular degrees of freedom. Through long runs and multiple trials, we demonstrate that the approach has good convergence properties and can compute PMFs to high accuracy. We use this method to study several peptide systems that exhibit self-assembly transitions in response to environmental cues. The computed PMFs and secondary structure profiles of this class of peptides show signatures of their transition behavior, when compared to those of non-transitioning peptides.