Evaluation and refinement of all-atom force fields for reproducing collagen structure and dynamics.

Abstract

Collageneous protein domains, characterized by the XYGly sequence repeat motif, trimerize and fibrilize to serve as the molecular skeleton of extracellular matrices and their mutations are frequently associated with disease. Because of experimental challenges in studying the effect of mutations on the properties of collagen, accurate atomistic Molecular Dynamics (MD) simulations are an invaluable tool. We evaluate the accuracy of state of the art MD force fields using recent experiments on model peptide homotrimers composed of Proline-4(R)-Hydroxyproline-Glycine (POG) repeats: the stabilizing POG motif appears with high frequency in several types of collagen. POG-repeats are used as templates to explore the role of amino acid substitutions in modulating collagen structure. We have compared the structure and dynamics of collagen POG10 homotrimers with various force fields from the CHARMM, AMBER, and GROMOS families together with various water models to aggregated crystal structure data, NMR data, and SAXS form factors. Of the tested force fields, we find those from AMBER and CHARMM give an acceptable description of collagen structure. AMBER force fields accurately reproduce collagen dihedrals, side chain torsions, amide spin relaxations, and SAXS data. CHARMM force fields were found to systematically shift backbone ϕ and ψ dihedrals, adopt incorrect side chain torsional angles, and overstructure POG10, increasing the persistence length relative to POG10 in AMBER force fields. However, by scaling the CHARMM36 CMAP terms of all dihedrals in POG10, we are able to capture a level of accuracy relative to experiment similar to that for the AMBER force fields. We suggest the use of AMBER ff99sb force fields or CHARMM36 with CMAP terms involving Pro, Hyp, and Gly rescaled by a factor of 1/2 (which we term CHARMM36mGP) for modeling collagen-like peptides.