MR on CCs is generally a pain.  The problem is that sliding any CC along 
its own supertwist (say by one heptad) will give you a solution that 
lines up very well with the "right" model.  There are a lot of solutions 
of this type, so you basically have a multitude of models that are all 
"okay" and the "right" one is not that much better than any of them.  
Hence, the Z-score is low.  This problem gets even worse if your CC 
crystallizes end-to-end with another copy of itself (CCs like to do that).

I'd be willing to bet that you actually already have the right 
solution.  Take the "best" one from your favorite program and start 
refining it.  Do rigid body first and keep re-applying rigid-body 
refinement until the model stops moving.  This is important.  Don't just 
stop when the Rcryst flattens out.  Let the MODEL settle down.  Then 
break the CC in half.  Rigid-body each half until they stop moving.  
Keep sub-dividing like this.  Eventually, you will be refining 
individual heptads from each chain.  This is a poor-man's way to do 
"rubber body refinement" which should let you refine a bent CC.  "rubber 
body refinement" in general lets you make much bigger leaps in 
coordinate space than all-atom refinement.  Using this methodology, I 
was once able to solve an antiparallel trimer CC using a parallel trimer 
as the starting model.  (the reversed chain was apparent in the refined 
map, but not in the starting map).

It can also be useful to start with idealized coiled-coil models.  The 
PDB does not always contain what you want.  I wrote a little jiffy 
program for generating an idealized coiled-coil in PDB format here:
http://bl831.als.lbl.gov/~jamesh/scripts/supertwist.awk
The top of this file (text) contains instructions on how to use it.  The 
idealized CC MR method proved useful in solving this structure:
R. Howard et. al. Neuron 53(5), 663-75.
and the methodology is described in that paper.  Using idealized models 
allows you to "scan" over the various superhelical parameters and then 
plot the Z-score, CC, or whatever you like vs the parameter.  Once you 
have optimized each parameter, you can do a multi-dimensional "MR 
meta-search" over a small range of all the parameters.

-James Holton
MAD Scientist


Thomas Edwards wrote:
> Dear BB,
>
> I am attempting molecular replacement with a 2.8A data set from crystals of a coiled coil of about 150 residues.
> Probably p21212 but maybe p2221.
>
> So far, Phaser, MolRep, Amore, Mr Bump, have not provided a good solution as judged by Z-scores, CCs, Rfactors, and whether there is any density outside the model (there should be - I'm using a slightly shorter model to search with).
>
> One possible problem is that the coiled coil may not be straight. It may have a slight curve to it.
> I have used models from the PDB that are straight or slightly curved, so far with no success.
> I have tried a few different resolution cutoffs too.
> I have tried single helix chains or dimeric coiled coils.
>
> Is there anything special about MR with coiled coils?
> Can anybody provide any tips/hints on MR with coiled coils??
>
> Thanks
> Ed
>