Replica Exchange Molecular Dynamics with Hydrogen Mass Repartitioning Improves Sampling of Peptide Binding to a Lipid Bilayer

Abstract

We compared the performance of two mass repartitioning models, HMR3 and HMR2, with tripled and doubled hydrogen masses, against the model with standard masses (SM). Due to heavier hydrogens, HMR3 and HMR2 afford longer integration steps of 4 and 3.5 fs, respectively, whereas SM used a 1 fs time step as a reference. All-atom replica exchange molecular dynamics simulations of the antimicrobial peptide PGLa binding to an anionic DMPC/DMPG bilayer were used as a case study. Across all conditions, HMR3 and HMR2 simulations each collected 36 μs of sampling, which is 50% more than our previous SM simulations. Therefore, the motivation for our work was to evaluate the HMR performance in complex biomolecular systems coupled with an advanced sampling algorithm. Our investigations led to three main conclusions. First, by affording longer simulations, HMR3 and HMR2 models provide better equilibration of PGLa peptide binding to the lipid bilayer. Specifically, they eliminated the metastable surface-bound (SB) state observed in SM. In the mass repartitioned models, PGLa exclusively sampled the inserted state (I), which also appeared as a dominant state in SM but alongside the SB state. Importantly, this better equilibration of peptide binding improved the consistency with experimental data. Second, HMR3 and HMR2 closely reproduce a broad range of structural properties previously reported for SM after correcting them for equilibration. Additionally, HMR3 and HMR2 demonstrate excellent agreement between themselves in sampling the conformational ensemble. These findings indicate that the mass repartitioning models preserve structural properties in this complex biomolecular system. Third, the actual acceleration of sampling by HMR3 is about 3.5-fold, which is lower than the theoretical 4-fold gain. This slight underperformance may be due to the slow dynamics of heavy hydrogens. HMR2 shows qualitatively similar results. An increase in the integration step in SM to 2 fs or adjustments in the computations of long-range interactions may reduce the computational gains offered by mass repartitioning, but do not negate their advantages in efficiency. HMR2, and particularly HMR3, are excellent options for accelerating conformational sampling in complex biomolecular systems.