Corrections in the CHARMM36 Parametrization of Chloride Interactions with Proteins, Lipids, and Alkali Cations, and Extension to Other Halide Anions.

The nonpolarizable CHARMM force field is one of the most widely used energy functions for all-atom biomolecular simulations. Chloride is the only halide ion included in the latest version, CHARMM36m, and is used widely in simulation studies, often as an electrolyte ion but also as the biological substrate of transport proteins and enzymes. Here, we find that existing parameters systematically underestimate the interaction of Cl- with proteins and lipids. Accordingly, when examined in solution, little to no Cl-association can be observed with most components of the protein, including backbone, polar side chains and aromatic rings. The strength of the interaction with cationic side chains and with alkali ions is also incongruent with experimental measurements, specifically osmotic coefficients of concentrated solutions. Consistent with these findings, a 4-μs trajectory of the Cl--specific transport protein CLC-ec1 shows irreversible Cl- dissociation from the so-called Scen binding site, even in a 150 mM NaCl buffer. To correct for these deficiencies, we formulate a series of pair-specific Lennard-Jones parameters that override those resulting from the conventional Lorentz-Berthelot combination rules. These parameters, referred to as NBFIX, are systematically calibrated against available experimental data as well as ab initio geometry optimizations and energy evaluations, for a wide set of binary and ternary Cl- complexes with protein and lipid analogs and alkali cations. Analogously, we also formulate parameter sets for the other three biological halide ions, namely, fluoride, bromide, and iodide. The resulting parameters are used to calculate the potential of mean force defining the interaction of each anion and each of the protein and lipid analogues in bulk water, revealing association free energies in the range of -0.3 to -3.3 kcal/mol, with the F- complexes being the least stable. The NBFIX corrections also preserve the Cl- occupancy of CLC-ec1 in a second 4-μs trajectory. We posit that these optimized molecular-mechanics models provide a more realistic foundation for all-atom simulation studies of processes entailing changes in hydration, recognition, or transport of halide anions.