Polarizable Atomic Multipole X-Ray Refinement: Particle Mesh Ewald Electrostatics for Macromolecular Crystals

Refinement of macromolecular models from X-ray crystallography experiments benefits from prior chemical knowledge at all resolutions. As the quality of the prior chemical knowledge from quantum or classical molecular physics improves, in principle so will resulting structural models. Due to limitations in computer performance and electrostatic algorithms, commonly used macromolecules X-ray crystallography refinement protocols have had limited support for rigorous molecular physics in the past. For example, electrostatics is often neglected in favor of nonbonded interactions based on a purely repulsive van der Waals potential. In this work we present advanced algorithms for desktop workstations that open the door to X-ray refinement of even the most challenging macromolecular data sets using state-of-the-art classical molecular physics. First we describe theory for particle mesh Ewald (PME) summation that consistently handles the symmetry of all 230 space groups, replicates of the unit cell such that the minimum image convention can be used with a real space cutoff of any size and the combination of space group symmetry with replicates. An implementation of symmetry accelerated PME for the polarizable atomic multipole optimized energetics for biomolecular applications (AMOEBA) force field is presented. Relative to a single CPU core performing calculations on a P1 unit cell, our AMOEBA engine called Force Field X (FFX) accelerates energy evaluations by more than a factor of 24 on an 8-core workstation with a Tesla GPU coprocessor for 30 structures that contain 240 000 atoms on average in the unit cell. The benefit of AMOEBA electrostatics evaluated with PME for macromolecular X-ray crystallography refinement is demonstrated via rerefinement of 10 crystallographic data sets that range in resolution from 1.7 to 4.5 angstrom. Beginning from structures obtained by local optimization without electrostatics, further optimization using AMOEBA with PME electrostatics improved agreement of the model with the data (R-free, was lowered by 0.5%), improved geometric features such as favorable (phi, psi) backbone conformations, and lowered the average potential energy per residue by over 10 kcal/mol. Furthermore, the MolProbity structure validation tool indicates that the geometry of these rerefined structures is consistent with X-ray crystallographic data collected up to 2.2 angstrom, which is 0.9 angstrom better than the actual mean quality (3.1 angstrom). We conclude that polarizable AMOEBA-assisted X-ray refinement offers advantages to methods that neglect electrostatics and is now efficient enough for routine use.