Simulating Bacterial Membrane Models at the Atomistic Level: A Force Field Comparison.

ABSTRACT

Molecular dynamics (MD) simulations are currently an indispensable tool to understand both the dynamic and nanoscale organization of cell membrane models. A large number of quantitative parameters can be extracted from these simulations, but their reliability is determined by the quality of the employed force field and the simulation parameters. Much of the work on parametrizing and optimizing force fields for biomembrane modeling has been focused on homogeneous bilayers with a single phospholipid type. However, these may not perform effectively or could even be unsuitable for lipid mixtures commonly employed in membrane models. This work aims to fill this gap by comparing MD simulation results of several bacterial membrane models using different force fields and simulation parameters, namely, CHARMM36, Slipids, and GROMOS-CKP. Furthermore, the hydrogen isotope exchange (HIE) method, combined with GROMOS-CKP (GROMOS-H2Q), was also tested to check for the impact of this acceleration strategy on the performance of the force field. A common set of simulation parameters was employed for all of the force fields in addition to those corresponding to the original parametrization of each of them. Furthermore, new experimental order parameter values determined from NMR of several lipid mixtures are also reported to compare them with those determined from MD simulations. Our results reveal that most of the calculated physical properties of bacterial membrane models from MD simulations are substantially force field and lipid composition dependent. Some lipid mixtures exhibit nearly ideal behaviors, while the interaction of different lipid types in other mixtures is highly synergistic. None of the employed force fields seem to be clearly superior to the other three, each having its own strengths and weaknesses. Slipids are notably effective at replicating the order parameters for all acyl chains, including those in lipid mixtures, but they offer the least accurate results for headgroup parameters. Conversely, CHARMM provides almost perfect estimates for the order parameters of the headgroups but tends to overestimate those of the lipid tails. The GROMOS parametrizations deliver reasonable order parameters for entire lipid molecules, including multicomponent bilayers, although they do not reach the accuracy of Slipids for tails or CHARMM for headgroups. Importantly, GROMOS-H2Q stands out for its computational efficiency, being at least 3 times faster than GROMOS, which is already faster than both CHARMM and Slipids. In turn, GROMOS-H2Q yields much higher compressibilities compared to all other parametrizations.