Among other things, we work with peptides. We make them, assay them, and calculate NMR structures of them. Many of our peptides contain unnatural amino acids or modified amino acids such as thiazoles, oxazoles, and other non proteogenic groups. Most of the time we do these structure calculations with Xplor. Over the years there have been different versions of Xplor and the one we currently use is Xplor-NIH . Other people use Cyana or CNS, but we’ve pretty much stuck with Xplor. Xplor comes with Topology and Parameter files based on the Charmm force field, and the current distros have force fields that have evolved over the years as you can see if you read the commented out bits. One of the downsides of these parameter sets is that as soon as you stray from the 20 natural aa’s you very soon find yourself in the wonderful world of patching the force field. The current release contains a few standard patches such as LTOD to make D-amino acids, and termini such as acetyl, but beyond that you’re pretty much on your own. Now X-ray crystallographers use these tools too and often have need of HETero patches for small molecules commonly bound to proteins and so excellent resources exist (e.g. PRODRG2, XPLO2D, HICUP) to make reasonable geometries of these ligands. However, and it’s a big however, these Xray tools don’t produce topologies and parameters with all hydrogens attached, just the heavy atoms that you’re likely to see if you do an Xray structure. Not the protons that are the NMR spectroscopists bread and butter.
One of our commonly used unnatural aa’s is cyclehexylalanine, commonly called Cha. And for a long time there’s been a patch for it included in Xplor distros that is just, well, wrong. The geometries you get with this patch (CHEX) produces cyclohexane rings with awful high energy geometries. So over the years in both in-house work, and published papers, we’ve had to model our Cha’s as Leucine instead. This means that we have to throw out any nOe’s we see from the Cha epsilon and zeta protons so we do lose information doing this, but needs must. For an example of what I’m talking about, heres what happens when you take a simple compound like Ac-Cha-NH2 and energy minimise it. (I used the mmff94s forcefield and Omega2 from Openeyes,  but anything sensible will give you a similar result: A nice chair conformation of the cyclohexane ring.
Now behold the fugly twist boat you get when you use Xplor. This was using the bog standard generate_template_new.inp script.
For people who have never used Xplor you typically then take your template and do a bit of simulated annealing, and these structures are then minimised against the NMR-derived distraint constraints (you have to set a minimum of 1 distance constraint otherwise it chokes, so I just made a weak nOe from the alpha CH proton to the alpha NH). This was done 20 times with the default sa.inp script. And behold, more ugly twist boaty things. And woah, check out those overall energies!
REMARK energies: 502.279, 47.4557, 238.476, 150.718, 65.6288, 0, 0
REMARK rms-d: 3.792163E-02,4.98529,8.72467,0,0
REMARK noe, cdih
REMARK violations.: 0, 0
I had hoped that our prayers would be answered with the arrival of Automated Topology Builder (ATB), from the good folks across campus in the Mark Group. But it seems that was a tad premature, as the following picture shows, when the ATB topology and parameters are added to our standard Xplor par set:
Oh dear, well at least it made a nice cyclohexyl boat…
Clearly the impropers are all stuffed up somewhere. Of course, I could always go through the parameters one at a time and incrementally modify all the impropers to see what happens, but that way madness lies. Or make every CH2 around the ring a different atom type to remove the ambiguities that I’m sure are causing it. The search for a solution goes on…
2. Using the mmff94s_NoEstat forcefield if you’re interested (http://www.eyesopen.com)