Dear Bonsor,
I fully second James suggestions but have a few additional comments:
If you get a solution in P6522 with one molecule, you should get the same
solution in P65 with 2 molecules. One of the crystallographic symmetry
operators would then be non-crystallographic.
The current version of Refmac will test all possible twinning operations, so
there is no need to do it yourself (provided of course that you get a molecular
replacement solution).
I would also try your rebuilt model with extended helix as a model for MR.
I suspect that the dimer which has formed is asymmetric and that it may be
randomly packed in your crystal. If the helix is a small compared to the
complete protein, it may not show up in twinning tests.
Good luck!
Herman
-Ursprüngliche Nachricht-
Von: CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] Im Auftrag von James
Holton
Gesendet: Sonntag, 15. Dezember 2013 23:29
An: CCP4BB@JISCMAIL.AC.UK
Betreff: Re: [ccp4bb] Wrong Space Group?
Its possible you are in a lower space group, perhaps with some twinning, but
your search model is different enough to only find a solution when things are
over-merged.
Try refining your P6522 model against data merged in P65. If the other copy
(symmetry mate in P6522) does not show up, you may be in trouble (wrong MR
solution). I'd also try refinement/building in the other triogonal/hexagonal
space groups, but again, start with the PDB file that you got for P6522. Just
change the space group in the header, and switch out the MTZ file. You will
need to merge your data in each space group and also check the a-b flip
re-indexing for most of them. Have a look at the CCP4 reindexing list for the
h,k,l operators to try:
http://www.ccp4.ac.uk/html/reindexing.html
note how similar they are to the twinning operators:
http://www.ccp4.ac.uk/html/twinning.html
If I have counted right, that means you have 36 jobs to run.
I'd also recommend turning the TWIN option in refmac off and on for each of
these cases. This will always give you a lower R factor, because of the dynamic
range compression you get with twinning, but if one particular combination of
twinning with a particular space group and axis reindexing is markedly better
than all the others, then you have just found your right space group. So, now
we are up to 72 jobs, but hardly a lot of work compared to growing the crystals
in the first place.
You might also want to try being clever and generating the symmetry mates of
your P6522 model and refine these partners as separate molecules as you reduce
the symmetry of the data. It's tricky, but think of it as an exercise. Which
real-space operator becomes what reciprocal-space operator? You can check your
answer by loading it up in coot and seeing if symmetry mates clash with the
input coordinates.
Yes, its a lot of work to try all these combinations, but that's the annoying
thing about twinning, it opens up a lot of ambiguities.
Good luck!
-James Holton
MAD Scientist
On 12/14/2013 6:44 AM, D Bonsor wrote:
Dear all,
I have collected ~160 degrees of data on a new crystal form of a protein
which has already been solved. Data was processed with XDS and reindex,
scaled and truncated with Aimless. Both XDS and Pointless suggested a Laue
group of P6/mmm with a possible space group of P6122 or P6522. Stats showed
an overall Rmerge of 0.131 but an Rpim of 0.041 (multiplicity/redundancy of
19.1), a completeness of 99.1% and resolution of 2.8Ang.
With cell dimensions of 63.1 63.1 243, only one protein chain can be found in
the asymmetric unit (two copies would leave a solvent content of 8%). I ran
phaser with all alternative space groups and a single solution in P6522 with
a TFZ of 10.0.
I then performed 20 Refmac cycles ending up with an R/Rfree of 35.5/45.5. I
open the structure and map in Coot and could see that there was a large
conformational change of helix-turn-helix actually becoming just a long helix
(https://www.dropbox.com/s/4s6g8apatsi5xcg/Before_Building.png) and then
dimerizing through the long helix with one of the symmetry mates.
This section was rebuilt
(https://www.dropbox.com/s/5j7tv0i5yq3mxxx/After_Building.png) and ran
through Refmac again resulting in an R/Rfree of 35.5/44.3. Looking through
the rest of the structure I see nothing else really to be modeled. Nothing
that could bring the Rfactors down to a reasonable range.
I have therefore tried several things. I ran the structure through Zanuda
server to look at other space group possibilities. The server suggested I was
in the correct space group. However I did reprocess the data to P6, P3, P312,
P321, C2221, P2 and C2, and reran phaser in search all alternative space
groups using the original search model but found no solutions. I did
reprocess the data in P1, though I did not collect enough data.
Twinning tests show no twinning. Although that does not mean there is no
twinning, I can