Surface of active polarons: A semi-explicit solvent model for biomolecular dynamics

Water gives rise to the main driving forces behind protein binding and folding, two of the most important phenomena in the biochemistry of life. Yet a computational model to account for these forces is difficult to construct. Implicit methods such as the distance-dependent dielectric constant are often too crude to capture detailed water-solute interactions such as hydrogen bonding. Yet the potentially accurate periodic box of explicit water is often computationally too costly for biological applications. In the first part of this dissertation, a new approach to solvating biomolecules in molecular dynamics (MD) or Monte Carlo simulations is presented. The method employs a thin layer of explicit water and a set of dynamically adjusted surface charges directly added to the layer. The surface charges together form a distribution that recreates the reaction field of the external bulk, modeled as a dielectric. Short range forces modeling hydrostatic pressure and thermal fluctuations are also added. We refer to this layer of special water molecules as a Surface of Active Polarons (SOAP). Test calculations of ionic and amino acid solvation free energies are in good agreement with experiment (correlation coefficients 1.000 and 0.995 respectively). Dynamical capabilities of SOAP are illustrated through simulations of short peptides.; The second part of this dissertation shifts focus to a biological application of SOAP. We consider the protein-protein docking problem, an important unsolved problem in biophysics, in which conformations of protein-protein complexes are predicted given the structures of the independently crystallized monomers. The main difficulty stems from uncertainties in surface sidechain conformations, where one or two misplaced sidechains can completely eliminate the affinity of the binding region. We study 6 interfacial surface sidechains from 3 separate complex forming monomers (6pti, 1lza, 1bta) using MD simulation and the SOAP model. Our simulations identify highly populated conformational clusters that substantially improve ligand-receptor energies compared to the original unbound structures. This suggests that protein recognition, and induced fit, does not start from an arbitrary state but from monomers whose key sidechains are already in suitable positions. Implications for the kinetics of the binding process and the protein-protein docking problem are discussed.