Stephen,
 
Stephen,
 
Although your peptide is smaller than the one I once worked on, here are some thoughts that might be applicable.
 
1. Check and play with the radius of integration in the molecular replacement. The default is probably not appropriate for your case but the value should be much smaller than the default.
2. Take a model - any model - of your peptide and make an arbitrary artificial dataset by rotating the model around an arbitrary angle, and put it in an arbitrary unit cell (say P1 for simplicity). Then use the original non-rotated model and see what parameters will give you the best solution and use those as starting parameters for your search.
3. Consider that your model may be very asymmetric, i.e. much longer in one direction than the other. In theory, you want the radius of integration to be such that it covers one copy of the model but not more than one. If the peptide is much longer than it is wide (which is somewhat likely), you might run into the situation where the correct radius for the length would incorporate multiple copies in the other direction(s). If this is the case, I am not sure you can fix it. In my humble opinion, which might be very out-of-date, this might be one of the reasons why MR does not work well on small molecules.
4. I think that it is possible that your crystal could be built from multiple copies of randomly oriented copies of the peptide, which are similar in their nature, but not exactly the same. This sounds odd but I convinced myself a long time ago that such a crystal could be made.
 
A long time ago I worked on a sea anemone toxin that had similar properties. At the time I could not make step 2 above work, that is, I convinced myself that I was unable to find parameters that did the job. That was enough for my mentor to tell me that I should not pursue the project... Of course such molecules are easily(*) resolved with NMR.
 
Mark
(*) Relative statement, and not by me.
 
 
-----Original Message-----
From: Stephen Campbell <[log in to unmask]>
To: CCP4BB <[log in to unmask]>
Sent: Tue, Apr 16, 2013 7:39 pm
Subject: [ccp4bb] Difficult data

Hello,

I am having a few issues with a data set I have been working on recently, and was hoping to get some ideas on how to deal with it, if anyone is in the mood.

I have been working with a very small bacterocin (about 3 kDa) and set up some crystal trays in hope of getting some high diffracting crystals.  I failed, but did manage to get a data set of reasonable quality to about 4A from crystals that reproduce very poorly.  Now, this sounds horrendous, but the MR model I have available is of a similar bacteriocin, whose structure is predicted to be essentially identical (different ORF, but almost identical sequence).  I was thinking it would be done in a day.  The bravais lattice is P4, and 118 x 118 x 165 (which seems HUGE for such a small protein...indeed calling it a protein is generous...peptide).  The data seems reasonably nice (nice spots, no visible overlap within 4A, but is very mosaic, about 1.5 degrees.

The data integrates and scales nicely, with very good chi2, R-factors and % rejected reflections.  It is hard to predict the correct space group, since all the tetragonal options have the same stats.  The systematic absences seem to predict p41212 (as does pointless), but it wouldn't be the first time I screwed that up.

I can't get a solution, no matter what I try.  Is this the nature of such a small peptide in such a large unit cell (placing the first model is difficult for MR since there may be many copies in the AU), or are there some tricks?  Is it likely that the unit cell is wrong?  Self Rotation functions give 8 peaks, but this is considering a peak fairly generously (approx. 15-20% of origin).  Are there some blatantly obvious red flags that I may be missing?  Any advice would be great - even if that advice is that it may be time to move on.

I should note that I have considered the idea that the peptide may be forming some sort of oligomeric structure such as a coiled coil, but it fails coiled coil predicting software, and there is no evidence that this should be the case.  The homologous structure is very rigid, and I would say fairly confidently that mine is likely the same - I just want to confirm it. 

Thanks so much,

Stephen Campbell
Post Doctoral Fellow
Department of Biochemistry
McGill University