We show how Hamiltonian replica exchange molecular dynamics simulations can be used to construct a ensemble of ligand binding locations. We restrain the ligand in the vicinity of the protein and let the ligand sample all available conformational space by applying a variety of Hamiltonian exchange, Monte Carlo, and other sampling techniques to escape local binding minima. Because of the rigorous nature of the statistical mechanical sampling, we can also extract binding free energy estimates at all putative binding sites. In the majority of simulations, we use a GBSA implicit solvent, though we compare in some cases to explicit solvent results. For these implicit solvent simulations, the process takes only 3-5 days, a time scale that is potentially feasible in a drug design pipeline. From these simulations, we can construct a ligand binding density, and attempt to predict the binding location(s) from this density.For the T4 lysozyme L99A model system, we find that this binding density gives discrete clusters, is robust over multiple simulation runs and that we can distinguish binders from non-binders, with free energies consistent with experiment. For simulations of the ASTEX diverse data set, over 40% of the time the most populated cluster corresponds to the experimental binding site, and over 80% of the time one of the top three most populated clusters corresponds to the experimental binding site.We propose that these binding ensembles may approximate ligand binding ensembles actually occurring in many protein-ligand complexes. We present theoretical arguments that such solution phase ensembles are not inconsistent with crystal structures showing only single bound configurations, and discuss the implications of these results for fragment-based drug design.

Exploring the drug binding ensemble using Hamiltonian replica exchange simulations

247th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.