On Using Atomistic Solvent Layers in Hybrid All-Atom/Coarse-Grained Molecular Dynamics Simulations

Hybrid all-atom/coarse-grained (AA-CG) simulations in which AA solutes are embedded in a CG environment can provide a significant computational speed-up over conventional fully atomistic simulations and thus alleviate the current length and time scale limitations of molecular dynamics (MD) simulations of large biomolecular systems. On one hand, coarse graining the solvent is particularly appealing, since it typically constitutes the largest part of the simulation system and thus dominates computational cost. On the other hand, retaining atomic-level solvent layers around the solute is desirable for a realistic description of hydrogen bonds and other local solvation effects. Here, we devise and systematically validate fixed resolution AA-CG schemes, both with and without atomistic water layers. To quantify the accuracy and diagnose possible pitfalls, Gibbs free energies of solvation of amino acid side chain analogues were calculated, and the influence of the nature of the CG solvent surrounding (polarizable vs nonpolarizable CG water) and the size of the AA solvent region was investigated. We show that distance restraints to keep the AA solvent around the solute lead to too high of a density in the inner shell. Together with a long-ranged effect due to orientational ordering of water molecules at the AA-CG boundary, this affects solvation free energies. Shifting the onset of the distance restraints slightly away from the central solute significantly improves solvation free energies, down to mean unsigned errors with respect to experiment of 2.3 and 2.6 kJ/mol for the polarizable and nonpolarizable CG water surrounding, respectively. The speed-up of the nonpolarizable model renders it computationally more attractive. The present work thus highlights challenges, and outlines possible solutions, involved with modeling the boundary between different levels of resolution in hybrid AA-CG simulations.