I have often talked on this blog about the fallacy of deriving single average conformations for molecules from average NMR data. How can a molecule with ten rotatable bonds (and frequently more) have any single or dominant conformation in solution? And yet, one repeatedly comes across publications dealing with single average conformations derived for organic molecules.
Peptides, being favourite targets for solution conformational analysis, are often subjected to such analysis. One comes across many instances where people want an alpha helix, or a 3/10 helix, or a gamma turn. They then design a peptide which they think is constrained to form that motif. They get their average NMR data (coupling constants and distances from NOE data), and then do a constrained molecular dynamics run, and see their favourite motif on the screen. But of course, what you put in is what you get out. If you are constraining the molecule to fit the NMR data in the first place, it's not going to stray away from it, and not surprisingly you see what you want to see (Isn't this a general theme and trap for so many of us...to be tempted to see what we want to see?).
In general, molecular dynamics is not a very efficient method of searching conformational space. Plus, every MD program uses some force field, and force fields are well-known to give fairly accurate geometries if well parametrized, but not good energies. So there is no guarantee that all the "low-energy" conformations of any molecule will be sampled. But in theory, if one could actually search all of conformational space with a good force field, then one would explore the whole ensemble of structure that a molecule adopts in solution. Even with faster computers though, that goal is going to unrealizable for a general situation.
However, MD can be a good way of exploring conformational space if the molecule is highly constrained in the first place. In such a case, it can serve as a good way of deconvoluting the average NMR data. And this is prcisely what Nikiforovich and Marshall, two researchers who do seem aware of the general problem of assigning multiple conformations, have done (J. Med. Chem, DOI: 10.1021/jm070084n). They have taken a cyclic pentapeptide, Pro-Ala-Ala-Ala-Ala, and done long 100 nanosecond (which is a huge amount of time for a MD run) on the molecule. Most importantly, their MD runs are unconstrained. After running the MD with two different force fields and different starting velocities, they consistently see the same results; a major conformer without a gamma turn, and a minor one (about 6%) with a gamma turn, a usually forbidden motif. They also gratifyingly note that the average distances from their ensemble of MD structures match well with the NMR data.
This analysis was feasible and realistic because of a couple of reasons; firstly, the molecule is small (although it still has 8-10 rotatable bonds) and one can actually do a 100 ns MD run on this. Secondly, they used many different conditions that converged to the same two conformational profiles, and thirdly they used unconstrained MD to remove any bias. I am not completely satisfied with it though, because the molecule still does have 10 rotatable bonds and is quite flexible. But it is a ring, and some of the bonds are not as flexible, so in general it is constrained and I think the conclusions are quite believable.
For once, it gave me a satisfied feeling to see an analysis that gave results that are quite likely to be true, complete, and sweet.
P.S. Garland Marshall also has one of the more memorable quotes in science that I have heard:
"Nature does not shave with Ockham's Razor!"