Post

Ever wondered which molecular forcefield you should use for MD simulations of lipid membranes? The three most popular all-atom parameter sets, AMBER Lipid21, CHARMM36, and Slipids-2020 are quite similar in equilibrium properties. But what about under deformation? www.biorxiv.org/content/10.6...

Pore formation and lipid rearrangement are important features of lipid membranes, important for understanding the fundamental biology of cells and the mechanics of electroporation and membrane-active molecules in synthetic biology applications.

This project was inspired by noticing that the Chain Coordinate for pore formation gave very different results if you used Lipid21 compared with CHARMM36 to form pores in DMPC membranes. Turns out the three behave quite differently when you start deforming the membranes.

Pore formation energy barriers computed using the three force fields for different lipids revealed a consistent pattern. Lipid21 had a higher energy barrier than the other two. This appears to correlate with tilt modulus (k_θ) - how easy it is to rotate a lipid against the membrane normal.

Further, these pores had very short lifetimes when the biasing potential was dropped. It should be noted, however, that experimental measurements of pore lifetimes under electrical potentials give much longer lifetimes than observed in MD simulations.

Morphology of formed pores followed from the observed energy barriers and metastability. CHARMM36 and Slipids rearranged into toroidal pores, while Lipid21 had a core composed of pure water and resisted formation the entire way.

Astute readers might have noticed that I haven't mentioned the right panels of the figures covering DMPG (negatively charged) membranes.That's where things get weird. Lipid21 is known to freeze with the state-of-the-art JC ion set. We found that Slipids has the same problem, to a lesser extent.

Thus, for those simulations, we used the Åqvist 1990 parameters with potassium instead of JC parameters with sodium, which appears to facilitate pore formation in both Lipid21 and Slipids.

We also studied the Highly Mobile Membrane Mimetic (HMMM) Model across force fields. Aside: Calling it the "HMMM Model" instead of just "HMMMM" was a crime against fun acronyms by the original developers. The HMMM floats shortened lipids on an organic solvent layer to encourage rearrangement.

The HMMM is notably unstable. The developers recommend Z-position restraints to maintain a membrane conformation, a revision of their original suggestion of a constant area (NPAT) simulation. We find that for CHARMM36 and Slipids, but not Lipid21, the position restraints are necessary at 1 µs scale.

So what does this mean for future simulations? For pore formation simulations with the Chain Coordinate, Slipids appears to be the better AMBER-family forcefield, with lower energy barriers and longer metastability, though CHARMM still remains undefeated.

For the HMMM, this resistance to remodeling actually makes Lipid21 advantageous. It maintains a bilayer conformation for significantly longer, allowing unrestrained simulation and more accurate surface interaction dynamics. It will still break up between 1 and 2 µs, though.