Toward Atomistic Resolution Structure of Phosphatidylcholine Headgroup and Glycerol Backbone at Different Ambient Conditions

Abstract

Phospholipids are essential building blocks of biological membranes. Despite a vast amount of very accurate experimental data, the atomistic resolution structures sampled by the glycerol backbone and choline headgroup in phoshatidylcholine bilayers are not known. Atomistic resolution molecular dynamics simulations have the potential to resolve the structures, and to give an arrestingly intuitive interpretation of the experimental data, but only if the simulations reproduce the data within experimental accuracy. In the present work, we simulated phosphatidylcholine (PC) lipid bilayers with 13 different atomistic models, and compared simulations with NMR experiments in terms of the highly structurally sensitive C–H bond vector order parameters. Focusing on the glycerol backbone and choline headgroups, we showed that the order parameter comparison can be used to judge the atomistic resolution structural accuracy of the models. Accurate models, in turn, allow molecular dynamics simulations to be used as an interpretation tool that translates these NMR data into a dynamic three-dimensional representation of biomolecules in biologically relevant conditions. In addition to lipid bilayers in fully hydrated conditions, we reviewed previous experimental data for dehydrated bilayers and cholesterol-containing bilayers, and interpreted them with simulations. Although none of the existing models reached experimental accuracy, by critically comparing them we were able to distill relevant chemical information: (1) increase of choline order parameters indicates the P–N vector tilting more parallel to the membrane, and (2) cholesterol induces only minor changes to the PC (glycerol backbone) structure. This work has been done as a fully open collaboration, using as a communication platform; all the scientific contributions were made publicly on this blog. During the open research process, the repository holding our simulation trajectories and files () has become the most extensive publicly available collection of molecular dynamics simulation trajectories of lipid bilayers.

Improving the accuracy of molecular dynamics simulations: parameterisation of interaction potentials for small molecules

“…The sources of error include an algorithmic implementation (Section 3.1.1), appropriate choice of constant temperature scheme (Section 3. 105 and Pluhackova et al 106 were unable to reproduce previously published results using the GROMOS 54A7 lipid force field with recent versions of the GROMACS simulation package (versions 4.5, 4.6 and 5.1.2) 69,107 , suggesting that this model may not reproduce the experimentally observed phase properties of phosphatidylcholine lipid bilayers modelled under certain hydration and temperature conditions. Specifically, it was found that the area per lipid was lower and the degree of acyl chain order greater than expected.…”

“…This suggested that either the original parameterisation was incorrect or that results obtained using recent versions of GROMACS differed significantly from those obtained using previous versions of the same code (versions 3.3.3 and 4.0.7). 105,[108][109][110] In MD simulations, the evaluation of long-range pairwise interactions is one of the most resourcedemanding aspects of the algorithm and has been a major focus of efforts to accelerate simulations from the earliest studies. [92][93] Interactions within a system occur over a range of time and length scales.…”

“…Clearly the failure of Botan et al 105 and Pluhackova et al 106 to obtain appropriate values for AL and other properties of a fluid-phase DPPC bilayer at 323 K using the GROMOS 54A7 lipid force field is not due to errors in the original parameterisation, but changes in the integration algorithm within GROMACS. Specifically, with the change in the multiple-time-step algorithm, it is no longer possible to update the long-range forces every 10 fs without severely affecting the outcomes of the simulations.…”

“…In the supplementary information accompanying Botan et al 105 , it was noted that the values of AL published by Poger et al 108 could be reproduced using version 4.0.X and earlier and that the differences most likely arose due to the use of the multiple-time-step algorithm based on the Trotter decomposition. A claim was also made, after discussions with the developers of GROMACS, that this problem had been resolved in version 4.6.6.…”

“…A claim was also made, after discussions with the developers of GROMACS, that this problem had been resolved in version 4.6.6. However, Botan et al 105 Table 4.1. However, the results obtained do not match the reference conditions (rows 1-7, Table 4.1).…”