Lipid21: Complex Lipid Membrane Simulations with AMBER

Abstract

We extend the modular AMBER lipid force field to include anionic lipids, polyunsaturated fatty acid (PUFA) lipids, and sphingomyelin, allowing the simulation of realistic cell membrane lipid compositions, including raft-like domains. Head group torsion parameters are revised, resulting in improved agreement with NMR order parameters, and hydrocarbon chain parameters are updated, providing a better match with phase transition temperature. Extensive validation runs (0.9 μs per lipid type) show good agreement with experimental measurements. Furthermore, the simulation of raft-like bilayers demonstrates the perturbing effect of increasing PUFA concentrations on cholesterol molecules. The force field derivation is consistent with the AMBER philosophy, meaning it can be easily mixed with protein, small molecule, nucleic acid, and carbohydrate force fields.

Figure 1

Fit of the C–C–C angle parameter to a MP2/cc-pVTZ scan on a pentane molecule.

Figure 2

(a) Fitting of cC–oS–cA–cA and oS–cA–cA–oS torsions to QM single-point energies (MP2/cc-pVDZ) using model glyceride compounds; (b) fitting of cA–cA–cA–oT, cA–cA–oT–pA, cA–oT–pA–oT, and oT–cA–cA–nA torsions to QM single-point energies (MP2/cc-pVDZ + PCM) using model phosphatidylcholine compounds; and (c) fitting cB–cA–cA–nN torsion using model ceramide compounds using a training set (left) and performance on an equally sized test set (right).

Figure 3

Head group NMR order parameters from experiment and Lipid21 simulations for POPC, DPPC, POPS, and POPG. Values are the average over a single 300 ns simulation ± st dev. Simulation values for Lipid14 are shown as orange crosses (POPC, DPPC).

Figure 4

Tail group NMR order parameters from Lipid21 simulations and comparison to experiment for POPC and DPPC. Values are the average over a single 300 ns simulation. Error bars are not shown for clarity.

Figure 5

Tail group NMR order parameters from Lipid21 simulations and comparison to experiment for N-linked chain of PSM (16:0 sphingomyelin).

Figure 6

Small-angle X-ray scattering form factors from experiment (open black circles) and Lipid21 simulations (blue line) for POPC, DPPC,^, POPS, and POPG.^, Profiles are calculated with SIMtoEXP from average atom densities over a single 300 ns simulation.

Figure 7

(Left) Small-angle X-ray scattering and (right) small-angle neutron scattering profiles for PSM from experiment at 100% D₂O and simulation with Lipid21. The experimental data were collected at 318 K and simulation was performed at 328 K.

Figure 8

Melting point scan results for DPPC with Lipid21 parameters and 1–4 Lennard-Jones scaling factors 0.5 (SCNB = 2) or 0.167 (SCNB = 6) on the cD–cD–cD–cD torsion. Results are presented as the running average of the area per lipid, with a window size of 1 ns, as a function of temperature. The experimental melting point for DPPC is 314 K.

Figure 9

Lipid21 area per lipid results for POPS and POPG simulations using K+, Na+, or Ca2+ counterions modeled with ff99 or JC parameters. Only the K+ ff99 simulations maintain an area per lipid close to the experimental value.

Figure 10

Lipid21 area per lipid results for 1 μs simulations of DMPC and POPC with 0.15 M NaCl modeled with JC ion parameters at 303 K. Bilayers maintain the correct phase with an area per lipid close to that of the validation simulations (see Table 1).

Figure 11

(Left) Most probable cholesterol tilt angle in raft-like membranes containing increasing mol % of DAPC. (Right) Total cholesterol transit events over a combined 2 μs of simulation as a function of increasing mol % of DAPC.