Dear Arthur:
This is why trigonal space groups are such fun.
CNS (and CCP4 programs for that matter) will arbitrarily assign a* and b*,
which, in trigonal cells, are not identical. For that reason, you have to
reindex (use ccp4's reindex program for this) and try both possible
index settings for refinement, empirically assessing which one is the
right one. Statistically, you should get it right 50%
of the time (right being defined arbitrarily as what works), but for me, the
odds were always 90% against. I was always having to reindex. You quickly
find that it is best to have two data sets to work on in parallel, and
then you probably need the enantiomorph of each.
HTH,
Bill
On Mon, 27 Mar 2006, Arthur Glasfeld wrote:
Sorry for this long message...
We've been working with data from a crystal form of an unligated protein (140
residues) that diffracts to 2.8 ?. The data index as P3(i)21 with an Rmerge
of 4.6% overall (vs. 4.4% in P3). The unit cell dimensions are 42x42x149,
which is consistent with single subunit in the asymmetric unit of P3(i)21.
The protein is active as a homodimer, and as expected the best molecular
replacement hits put the molecular two-fold along the crystallographic
two-fold. P3(2)21 does not give as good results. (The search model is the
same protein molecule bound to its ligand).
When rigid body refinement (to 3.5 ?) is performed on this molecular
replacement solution, the R/Rfree is in the 35% range, but simulated
annealing (with data to 2.8 ?) causes immediate problems, with the working R
dropping to about 32% as the Rfree goes to above 40%. Electron density maps
are poor throughout the model, and efforts at rebuilding have not improved
things. I have tried to use other space groups, including P3(1) and C2
(dimensions 73 x 42 x 149, 90 x 90.1 x 90) in hopes of reducing symmetry
constraints on the refinement, but the problem remains.
An obvious issue that sprung to mind is twinning, but I saw no evidence of
twinning when I use Todd Yeates' server, the CNS & CCP4 utilities, or when I
look at intensity distribution in the Truncate plot. A recent thread on
this bulletin board described a problem similar to mine (my thanks to Michael
Hothorn and the several responders). Folks recommended using the
Padilla-Yeates analysis in dataman. I've done that now, and have come up
with an odd result that I don't understand - the data (either integrated as
P3 or P321) fall well below the curve expected for either twinned or
untwinned data, with <|L|> = 0.592 and <L^2>=0.439. The plot from dataman is
available at:
http://www.reed.edu/~glasfeld/PY_p321.pdf
If someone can help interpret this plot I'd be grateful.
The self-rotation maps and native pattersons don't reveal anything that I can
readily interpret, though there are some weak peaks in both that might have
meaning and I'm just oblivious to their importance. The self-rotation map
(from GLRF) has two-folds where it ought (at around 11 sigma) and several
weaker peaks (around 4 sigma):
http://www.reed.edu/~glasfeld/k180_xyk_p321.pdf
The peaks above 5 sigma in the native patterson are listed below (calc'd in
CNS) and a link to the Harker section is given below as well.
peak no. interpolated (x,y,z,height) grid point (x,y,z,rho) (fractional
coordinates)
1 0.0667 0.0000 0.0000 8.3871 0.0667 0.0000 0.0000
8.3871
2 0.6667 0.3333 0.9891 7.2202 0.6667 0.3333 0.9867
6.8977
3 0.5520 0.1026 0.9059 6.0885 0.5556 0.1111 0.9067
6.0000
4 0.6667 0.3333 0.7735 5.7798 0.6667 0.3333 0.7733
5.7790
5 0.0000 0.0000 0.0662 5.7358 0.0000 0.0000 0.0667
5.7241
6 0.0000 0.0000 0.0344 5.0269 0.0000 0.0000 0.0333
4.9362
http://www.reed.edu/~glasfeld/patterson_p321.pdf
I've read that having a molecular two-fold aligned with a crystallographic
axis can make twinning difficult to detect, but given the above results I'm
not sure if that's really what's going on, nor how to work around the
potential problem. I feel like I must be missing something obvious, but
after several weeks of struggling, I'm hoping someone can enlighten me.
Thanks in advance,
Arthur Glasfeld
Department of Chemistry
Reed College
3203 SE Woodstock Blvd.
Portland, OR 97202
USA
