Comparing atomistic molecular mechanics force fields for a difficult target: a case study on the Alzheimer’s amyloid β-peptide

Abstract

Macromolecular function arises from structure, and many diseases are associated with misfolding of proteins. Molecular simulation methods can augment experimental techniques to understand misfolding and aggregation pathways with atomistic resolution, but the reliability of these predictions is a function of the parameters used for the simulation. There are many biomolecular force fields available, but most are validated using stably folded structures. Here, we present the results of molecular dynamics simulations on the intrinsically disordered amyloid β-peptide (Aβ), whose misfolding and aggregation give rise to the symptoms of Alzheimer’s disease. Because of the link between secondary structure changes and pathology, being able to accurately model the structure of Aβ would greatly improve our understanding of this disease, and it may facilitate application of modeling approaches to other protein misfolding disorders. To this end, we compared five popular atomistic force fields (AMBER03, CHARMM22 + CMAP, GROMOS96 53A6, GROMOS96 54A7, and OPLS-AA) to determine which could best model the structure of Aβ. By comparing secondary structure content, NMR shifts, and radius of gyration to available experimental data, we conclude that AMBER03 and CHARMM22 + CMAP over-stabilize helical structure within Aβ, with CHARMM22 + CMAP also producing elongated Aβ structures, in conflict with experimental findings. OPLS-AA, GROMOS96 53A6, and GROMOS96 54A7 produce very similar results in terms of helical and β-strand content, calculated NMR shifts, and radii of gyration that agree well with experimental data.

Keywords:

• molecular dynamics
• protein folding
• molecular mechanics
• force field
• simulation

Acknowledgments

The authors thank Prof Michael Zagorski at Case Western Reserve University for providing the experimental NMR chemical shifts, David Rosenman, and Prof Chunyu Wang at Rensselaer Polytechnic Institute for providing J-coupling data, and the administrators of Virginia Tech Advanced Research Computing for providing computing time on the SystemX and HokieOne supercomputers. J.A.L. and D.R.B. designed research. S.R.G. performed simulations. S.R.G., J.A.L., and A.M.B. analyzed data. J.A.L., S.R.G., A.M.B., and D.R.B. wrote the paper.