Identifying and sampling conformational states of SARS-CoV-2 mpro for flexible receptor docking.

Abstract

One major limitation of structure-based drug discovery is the reliance on static protein structures from X-ray crystallography or homology modeling. As macromolecules floating in aqueous environments or buried in lipid membranes, proteins exist in ensembles of conformational states, with probabilities related to the Gibbs free energy of each state. Achieving the desired functional effect in a drug discovery campaign often relies on finding molecules that bind to one state preferentially over others. Unfortunately, molecular docking software packages such as DOCK that are designed for large chemical libraries treat the protein receptor as static, ignoring how each ligand interacts with the different states. The flexible receptor (FlexRec) extension to DOCK allows the user to define different states of the receptor with different equilibrium probabilities for each. The user can then dock each ligand in the library to each state, penalize rarer states’ DOCK scores, and predict both the overall probability of binding and the propensity to bind to one state over the others. We are testing the ability of FlexRec DOCK to prospectively find ligands that bind to different states of a protein using the SARS-CoV-2 main protease (Mpro) as a target. This chymotrypsin-like cysteine protease is essential in the viral life cycle, and there have already been over 1200 published crystal structures of this protein in various experimental systems, including with different ligands bound. We have identified different states of Mpro by clustering these structures and shown that docking predictions for a set of known ligands change depending on the state used. We are now using molecular dynamics (MD) to sample the free energy differences between the states for a large-scale FlexRec DOCK campaign, which we will later test experimentally.