[gmx-users] pme/cutoff in membrane proteins

[Peter Tieleman]

Maybe some recent simulations we have done might be of interest for the 
current discussions on PME etc.

We (me, Berk, Mark Sansom) have done simulations of a model six-helix 
bundle that forms a water channel, comparing water-ordering with 
reaction field, cutoff, PME, with the protein fixed. I think this should 
be relevant for many membrane protein simulations. Cutoffs give very 
strong water ordering, which seems consistent with tests Berk published 
that show long-range correlations that get worse as your cutoff length 
increases. PME and reaction field give similar order in most places, but 
not in a few parts in the channel where there are long ranged 
electrostatic effects, most probably due to the strong field generated 
by the protein (6 parallel helices). I agree with some previous comments 
that reaction fields are a bad idea in membrane proteins, simply because 
they assume a homogeneous solution. This isn't all that true for 
water-soluble proteins (maybe unless they're really small), it's 
certainly not true for membrane proteins. They'd do funny things if you 
would look at ion channels with ions seperated by a distance of the 
order of your cutoff. Although there will be some papers coming out from 
me that didn't use PME or PPPM, I now think that is less than a good 
idea and is really mainly because it takes so long to finish a project.
To Erik's comment on lipids: we have some test simulations of 150 ns 
that should show converged areas and structural properties, and the 
effects of PME vs cutoff are pretty large, at least several percent in 
areas for fully hydrated DPPC. Although we have one very big membrane 
protein simulation running without PME at the moment because of the bad 
scaling, everything else is using PME.
One other anecdote: in my tests of helix bundles, I tried a simulation 
of a helix bundle in octane. Nothing says that this should be stable 
(say, the helices might dissolve in the octane), but there was an 
interesting twist: with a 1.0/1.8 nm cutoff, the bundle does dissolve. 
With PME, it doesn't in octane, both simulations were about 4 ns. Not 
conclusive, but a bit worrying. 
On small-system artefacts with PME: I compared water order inside the 
six-helix bundle channel, as this seemed very sensitive to whatever you 
change, in a bilayer of about 6.7 nm thickness and in one with 4 nm 
extra water. This made no significant difference.

Peter