To the CCP4 community,
I received a number of very helpful replies to my lengthy question last week about twinning and/or problems with indexing. I am very grateful for all the suggestions and I have learned more about what to look for and what to try. You will find the replies (chronological order) below and my original question at the bottom.
Thank you,
-Andy Torelli
Daniel Schlieper wrote:
In addition to the twinning tests, it might be useful to look at the reflections in pseudo-precession images (eg, hklview). Check if the systematically absent reflections behave like predicted.
Eleanor Dodson wrote:
You can read the documentation on twinning http://www.ccp4.ac.uk/dist/html/twinning.html
You can have twinning in I23 and the best indicator is that table of <I^2> /<I>^2 That looks significantly < 2 so it is likely you do have twinned data. The twinning would make hkl nearly equal to k h -l, (a symmetry pair for I423) so you would not get very good stats in I423
Seeing it is likely to be I23 - not I423 - you also have to be careful when merging crystals since there are two equally valid ways to index
any crystal (see http://www.ccp4.ac.uk/dist/html/reindexing.html fo some discussion)
There is a program pointless written by Phil Evans in the latest CCP4 release. That allows you to give a master data set - I woud choose the least twinned - and suggested reindexing all new crystals to the first chosen convention. It will also tell you how strong the "4-fold" symmetry that is generated by the twinning) and give you a preliminary indication of the amount of twinning. Perfect "4-fold" = perfect twinning ( 0.5 twin factor)
You can use it with mosflm output, scalepack unmerged output or xds ftp://ftp.mrc-lmb.cam.ac.uk/pub/pre/pointless-1.2.9.tar.gz and pointless.html for documentation.
Peter Zwart suggested downloading Phenix and running phenix.xtriage.
Graeme Winter wrote:
I would start at the beginning - how sure are you of the lattice?
You say that the indexing solutions are poor. A very useful thing I find is to compare the statistics from postrefinement with the asserted lattice applied and ignored - roughly speaking refine the cell in I23 / I4 and separately in P1 and see if the stats (r.m.s. deviations etc) look much better in P1. If they do, I would take a closer look, perhaps testing lower symmetry lattices. This is something that xia2 does, so perhaps you could give this a go (had to give it a plug somewhere!)
If the stats come out about the same (this test is more sensitive to cell angles than cell lengths, so is unlikely to distinguish between your two candidate spacegroups. What I would then to is take the integrated but unscaled reflection file and put this into something like pointless or xtriage and see what symmetry elements they find - this should give you a pretty solid (twinning notwithstanding) determination of the correct pointgroup.
From here the twinning tests should be more helpful.
Miroslav Papiz wrote:
Irrespective of twinning issues. Have you looked at I432 ? if you are convinced that the 4 and 3 folds are present (i.e similar Rsym in I422 and I23) then this is also a possible spacegroup. You should at least try this before looking for other explanations.
Bill Scott wrote:
I can say that phenix is very good at handling both twinning and RNA.
phenix.xtriage will also give you useful information about twinning (applies several tests, etc) and the phenix authors have been incredibly helpful and are in fact looking for RNA test cases, so it would be easy to seek professional advice.
Dominika Borek wrote:
Answers to your questions depend a lot on other things:
Twinning alone:
1. Are you going use molecular replacement or rather heavy
atoms-based phasing?
2. How many copies of RNA molecule do you expect in your ASU
(quite large BTW)?
3. If you plan to phase with HA - which HA are you using or
going to use?
Data alone:
1. Did you collect at home or at synchrotron? I expect that with such large unit cell you collected at synchrotron. If this is the case and if you have in your crystallization solution (as it happens for RNA) potassium, manganese, rubidium etc. radiation damage even at home source will be the issue. I do not know what software did you used. If it was scalepack and you used option "fit B" during scaling, log file will tell you quite well how bad was radiation damage. If fitted B-factor is higher than 5-6 for the last images that already indicates significant radiation damage. Another indication of radiation damage is exactly what you described - high R-merge, a lot of rejections etc.
Unfortunately, to tell more I would have to see .log file from scaling/merging software.
I personally think that it is quite possible that you do not have twinning just lower symmetry (completeness issue suggests it). Lack of birefringence happens in tetragonal (and other) space groups if symmetry axes are properly oriented.
Roman Hillig wrote:
Just a quick comment: If it really was an I-centered tetragonal space group with hemihedral twinning, then the true space group would be I4, or I4(1). The space group I422 would only be the apparent space group (triggered by the twinning). So you would actually have to process your data in I4 and then run the twinning tests etc.
I had such a case some time ago. Twinning became also apparent from a careful analysis of the selfrotation function, and comparison of the self
rotation functions from with data sets from different crystals with varying twin fractions. If you are interested, you will find the details here:
Hillig RC, Renault L. Detecting and overcoming hemihedral twinning during the MIR structure determination of Rna1p. Acta Crystallogr D Biol
Crystallogr. 2006 Jul;62(Pt 7):750-65
Ethan Merritt wrote:
There is a tendency these days to blame everything on twinning, but I don't follow the logic in your case. Failing to recognize twinning would not explain either poor Rsym nor poor completeness. Nor would it explain presence/absence of a 3-fold axis.
You say you are considering either I23 or I422. One of these has a 3-fold but no 4-fold. The other has a 4-fold but no 3-fold. Why not process in P1 and look in the self-rotation function to see which is true?
I would not bother worrying about twinning until you have the space group pinned down.
Original question(s):
I have collected data from an RNA molecule that extends to 2.9 angstroms, exhibit mosaicity less than 0.9 degrees and generally show nice, round spots. The crystals look cubic and are not birefringent (suggesting a cubic lattice). However, the data index poorly with the best solutions being either I23 (a=b=c=141.68) or I422 (a=b=141.74; c=141.57). Predicting reflections using each of these indexing solutions appears to confirm each as a valid indexing solution. However regardless of which space group I select to process the data, the final result after scaling reveals relatively high Rsym values overall and for individual batches (10% - 20%). Also, a large wedge of data are required to achieve nearly 100% completeness (>60 degrees; if the lattice was truly cubic I would expect much less to be required for high completeness).
These discrepancies led a colleague to suggest twinning might be a problem. The UCLA twinning server (Yeates method) finds the following:
For the data processed as I422 in the partial twinning test: No merohedral twinning laws found for that space group
For the data processed as I23 in the partial twinning test: Twin fraction of 0.408
For the data processed as I23 in the perfect merohedral twinning test:
Resolution ; <I2> / <I>2
16.147 ; 1.99 (n = 404)
7.224 ; 1.57 (n = 404)
6.067 ; 1.40 (n = 404)
5.434 ; 1.33 (n = 404)
5.006 ; 1.37 (n = 404)
4.687 ; 1.44 (n = 404)
4.440 ; 1.35 (n = 404)
4.240 ; 1.34 (n = 404)
4.070 ; 1.88 (n = 404)
3.924 ; 1.54 (n = 404)
3.799 ; 1.61 (n = 404)
3.689 ; 2.03 (n = 404)
3.590 ; 1.74 (n = 404)
3.501 ; 2.00 (n = 404)
3.421 ; 2.08 (n = 404)
3.347 ; 2.17 (n = 404)
3.278 ; 2.15 (n = 404)
3.216 ; 1.42 (n = 404)
3.158 ; 1.58 (n = 404)
3.104 ; 1.69 (n = 404)
3.054 ; 1.59 (n = 404)
3.007 ; 1.48 (n = 404)
2.962 ; 1.80 (n = 404)
2.920 ; 5.42 (n = 404)
This is my first experience with twinning (hmmm...I feel like I'm being initiated), and I have several questions that I have not been able to answer yet from researching the literature or CCP4bb archives. I should mention that several data sets from several similar crystals all behave the same in terms of the difficulties in data reduction and even the apparent twinning fraction (the same to within a few %). I know the first advice will be to try new conditions, but I wonder if I can work with these data since I already collected data sets for several derivatives and also anomalous. Any advice or literature references are greatly appreciated to any or all of these questions:
1.) How should I go about assigning/identifying the correct space group? Does the apparent presence of merohedral twinning for the I23 processed data, but not I422, indicated that I do not have a cubic lattice?
2.) How is it possible that the data processed in the lower symmetry I422 space group are not also found to be twinned? I can't visualize how the same merohedrally twinned lattice could be described without conflict in the lower symmetry space group.
3.) I looked at the original T. Yeates paper in Meth. Enz. regarding twinning. There is an example of data from plastocyanin which are perfectly twinned. The reported plot of <I squared> / <I>squared as a function of resolution show a fluctuation around 1.5 that looks similar to the values I reported above as output from the perfect twinning test. How does one determine from those plots whether or not you have perfect merohedral twinning? Should I consider the average value, the lowest value, the distribution, or is my apparent partial twinning fraction sufficiently far from 50% to be sure that I don't?
4.) I tried running the perfect- and partial-merohedral detwinning scripts in CNS for the data processed as the I23 space group. The result of the perfect-merohedral detwinning script resulted in generally higher values of <I squared> / <I>squared, but it's not clear to me what that means or how it is possible to detwin perfect merohedral twinned data. After the partial-merohedral detwinning script however, the twinning fraction dropped to 17%. Is that informative with regards to what space group I'm dealing with or whether or not I have partial vs. perfect twinning?
5.) The last questions are about how to proceed with solving the structure. As I mentioned, I have collected data that I hope to use for MIR, potentially including anomalous. With a twinning fraction of 17% after detwinning, is it possible/appropriate to solve the structure by MIR or SIRAS (I'm guessing differences in the twinning will just diminish my signal to noise for finding the heavy atom peaks)? I also understand that it is possible to solve a perfectly merohedrally twinned data set by molecular replacement. I have a partial MR solution using the I23 data that appears to have unique phase information. I know there are several refinement programs that could be used for twinned data. Can anyone recommend one that handles RNA well?
