Advancing GIST-Based Solvent Functionals through Multiobjective Optimization of Solvent Enthalpy and Entropy Scoring Terms.

Abstract

Water molecules and their impact on the enthalpy and entropy of protein-ligand binding are of considerable interest in drug discovery. In this contribution, we use multiobjective optimization to fit the solvent enthalpy and entropy scoring terms of grid inhomogeneous solvation theory (GIST)-based solvent functionals to measured isothermal titration calorimetry (ITC) data of protein-ligand-binding reactions for ligand pairs of the protein thrombin. For the investigated ligand pairs, the overwhelming contribution to the relative binding affinity difference is assumed to be attributed to the contribution of water molecules. We present different implementations of the solvent functionals and then proceed by analyzing the most successful one in more detail through error assessment and presentation of the scoring regions in the binding pocket and the unbound ligands of selected examples. We find overall good agreement between calculated and experimental data and, although physically not fully justified, the ligand-desolvation score increases binding affinity, thus suggesting that the solvent molecules on the surface of the unbound ligand constitute a proxy for interactions gained through the protein. Furthermore, we find limited transferability of the parameters even between similar protein targets, thus suggesting refitting for each new protein target. Possible reasons for the limited transferability may arise through the initial assumption of dominating water contributions to binding affinity. Nonetheless, overall our study demonstrates a consistent approach to assign thermodynamic quantities to water molecules that is sensible to measured thermodynamic signatures and enables bridging the gap between experimentally determined water positions in protein-ligand complexes and measured thermodynamic data.