Exploring Binding Ensembles of Protein-Ligand Interactions in Explicit Solvent by Expanded Ensemble Simulation

Binding interactions to proteins are important to consider in design of small molecules as putative drugs. We present a method for computationally generating and grouping sets of representative bounds structures of ligands and apply it to a model protein, T4-lysozyme L99A.

Expanded ensemble simulations using explicitly solvated structures provides increased accuracy relative to continuum solvent representation while taking advantage of the enhanced sampling due to alchemical modification of the ligand. Enhanced sampling reduces the time needed for the ligand to experience many transition events and therefore sample most physically accessible locations around the protein. Alchemical modification further allows us to estimate binding free energies without significantly increasing simulation run times.

We study four ligands with a range of known experimental binding affinities to identify alternative binding locations, even those of low ligand affinity. We present binding free energies to each location and to the protein overall using the multistate Bennett acceptance ratio method to estimate free energy differences between alchemical states.

We find that contrary to expectations implicit solvent results were better able to estimate binding free energies for benzene both to the protein overall and to the experimental binding location. Explicit solvent simulations predict free energies that are not within error of the experimental measurements for benzene and phenol. Identified binding locations do not consistently match between implicit and explicit methods. Several issues identified during the simulations likely contributed to the disagreement, and we explain some ways that these issues can be addressed.