From: Himanshu Khandelia (hkhandel_at_memphys.sdu.dk)
Date: Fri Nov 09 2007 - 13:57:10 CST
JC is mostly right. For pure membranes, NPT will not work with the charmm
forcefield, because it was parameterized at a time when constant surface
tension ensembles were being used for bilayers. I understand that the
CHARMM development team is working on the problem.
With a transmembrane protein, the area will still shrink, but slowly,
because the protein presents a relatively rigid barrier. The larger your
system (and the larger your protein), the slower the rate at which the
area shrinks.. If, for example, you are simulating a 100,000 atom system
for 30 ns at NPT, the area will shrink less than 5-10%, which is
acceptable in general, at least for purposes of manuscript review, as well
as for the purposes of the validity of the simulation. On the other hand,
the area shrinks 10-20% for a 30000 atom system in 40 ns, which is
significant, and is a problem.
If using NPAT, the area of the protein can be done using several ways. One
of them is to use Benoit Roux original scheme which is documented in the
CHARMM distribution (in the support/membrane section). There is a simple
script which reads in a pdb file and comes up with a cross sectional area
profile for a peptide. It can be extended to work for larger protein.
There are similar grid-based approaches one can look up in recent
As for the publications not reporting the area per lipid issue in
protein-membrane simulations using NPT charmm, my personal opinion is that:
1. either the authors are aware of the shrinking membranes in
protein-membrane simulations, but chose to push the issue under the
carpet because it is either insignificant and the protein is more
interesting, or because too much simulation time is at stake.
2. The authors are unaware of the problem completely !
If you are just starting, (do not have 6 odd months of simulation time at
stake), use NPAT, especially if your system is not too big. Just make sure
that you have a nicely equilibrated system at the right area per lipid. At
least you will be proud that you used the right methods.
Of course, all of the above discussion is redundant if you choose to use
GROMACS, where NPT is the recommended choice, and leads to the right area
per lipid for most lipids.
Good luck !
Himanshu Khandelia, PhD (Chemical Engineering),
Research Assistant Professor (Postdoc),
MEMPHYS, Center for BioMembrane Physics: www.memphys.sdu.dk
University of Southern Denmark (SDU)
Campusvej 55, Odense M 5230, Denmark
Phone: +45 6550 3510, +45 2398 7972
Fax: +45 6550 4048
On Fri, 9 Nov 2007, JC Gumbart wrote:
> Here is my thinking on the issue. Someone please correct me if I'm wrong.
> The most accurate way is to simulate at NPAT with the appropriate area per
> lipid, although, when simulating with a protein, I don't think it's obvious
> how to calculate a precise area/lipid. I think many of us get away with NPT
> because we are not interested in the behavior of the membrane anyway, as long
> as it keeps the protein "happy"; it may not be worth the trouble to find the
> optimal area. This is assuming your area stabilizes though, which sometimes,
> it may not (it can shrink a lot in some cases). Then you will certainly want
> to use constant area.
> On Nov 8, 2007, at 6:00 AM, Karol Kaszuba wrote:
> > Hello,
> > Based on the literature and previous posts I found that CHARMM forcefield
> > does not yield the correct
> > bilayer geometry in NPT ensemble, however there are many articles in which
> > we can find membrane-protein
> > simulation in NPT and no one mentions about the problems with bilayer
> > structure - so I am little
> > confused - can I use the CHARMM forcefield (release c32b1) to simulate POPC
> > membrane in NPT or
> > do I have to use NPgamma(surface tension)T or NPAT ?
> > Thank you in advance,
> > Regards,
> > Karol
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:45:30 CST