[gmx-users] Ions close to protein

[Berk Hess]

>From: David van der Spoel <spoel at xray.bmc.uu.se>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>Subject: Re: [gmx-users] Ions close to protein
>Date: Tue, 23 May 2006 20:13:32 +0200
>
>David Mobley wrote:
>>Joanne,
>>
>>> >When I add ions to my system to neutralise it some are placed quite 
>>>close to the protein (even when using random) and in other cases during 
>>>simualtions I have seen ions move into the protein, is this normal/okay 
>>>and are there any ways of keeping these ions away from the protein?
>>
>>
>>In terms of keeping them away from the protein, not too sure. But I
>>think it is right to be nervous about ions that move into your
>>protein. These could have long correlation times.  Depending on how
>>long the correlation times are, you might need to run longer to get
>>adequate sampling. On the other hand, if your salt concentration is
>>reasonable, you might have enough ions that this behavior will
>>appropriately average out, especially if correlation times are
>>relatively short.
>>
>>Probably, the overall answer to whether this is okay depends on what
>>question you are trying to ask. My personal opinion is that ions are a
>>whole can of worms that is mostly unopened so far, but sooner or later
>>someone will have to open it and it will be messy (long correlation
>>times, bad parameters, and strongly influence certain things). But
>>perhaps I'm just being pessimistic. And, in fairness, I should say
>>that so far I've decided just to ignore the issue, because I don't
>>want to be the guy who opens the can of worms.
>
>OK, better show a glimpse of what's in the can then...
>
>We are finishing a paper about this stuff, and the conclusion is: no 
>problem as long as your simulations are reasonably long and you use PME. 
>Charged sidechains have *on average* half a counterion in our tests. A real 
>problem may be that force fields are not adapted to having realistic ionic 
>strength, and hence you might influence protein stability by introducing. 
>This is very hard to prove however.

Indeed I would not worry too much about the sampling.
Most ions should not bind very strongly to a protein
and will therefore diffuse through a large part of the system
in reasonable simulation times.

You should however worry about the combination of protein/ion/water
force fields. Depending on the type of ion and charged group
you could have serious issues when the force field (-combination)
turns out to be bad, i.e. too strong binding of charged groups.
We had issues with Na+ sticking far too strongly to an SO3- group,
but that group I had parametrized myself.
For instance Cl- should usually not cause problems.

Berk.