Thanks for your answer. It's a bit complicated for me to describe. I realize I forgot to mention some important points. My colleague did the molecular replacement in the first place. She did not know that my protein is that large and calculated Matthews with the shorter sequence (62%). The model is of exactly the same length. So it is basically my protein without flexible residues. Like this she got 12 molecules in the ASU and a water content of about 45 %.

When I tried to do the MR I calculated Matthews and got 8 molecules per ASU. For 12 molecules the water content would be 27 %. When I tried to use fewer molecules (8,9,..) in the ASU there was always a lot of "green" density, which looked like a missing molecule and I did not manage to fill the gap with the residual flexible residues. Especially because the density was always concentrated at one place and not distributed equally in relation to the other molecules. So it did not look like the missing parts.

It is absolutely ok for me, when the flexible residues are still flexible. I just do not understand how everything can fit in this ASU?? My colleague said, that flexible residues need less space than folded residues in a crystal. Is that right? From my intuition I would say it is the other way round.

Could it be that my model is not good enough? Because some parts have undergone conformational changes, for example? Can phase bias be that severe? I always thought that this issue would be overcome by using difference maps.

Have you looked at your native Patterson maps? They should help you figure out how many molecules you have in the ASU, as well as potential symmetry between them.

I can answer a part of your above queries. that flexible residues needs a lesser space in that crystal and your worry might be true in the second case as you can try to use the different search model (If any) for Molecular Replacement.

I would possibly also suggest calculating some omit maps and checking whether these fit your molecules... Maybe just one or two in the ASU have terribly bad omit maps.. indicating that they might not really be there. Also if you say 29% Rfree... how far are you in your refinement and where do you want to go.. also I assume the R is not too far off??

Hi Annika,

At your resolution, you should be able to try automated methods like ARP-wARP or Phenix autobuilding. These have a lot of advantages, as they are usually quite good at deleting regions of your protein that you think look right, but which are in fact incorrect. Also, you have to do a lot less work to sort out the problem :-).

Another method that worked well for me once in a tricky MR case was to perform quite aggressive density modification of the result from MR. This highlighted that some of the model was in the wrong conformation, and broke me out of the minimum.

At 2.3 A, phase bias can be quite sufficient to prevent successful structure solution, at least by hand. Obviously, the better the model, the more likely you are to be successful.

At the end of the day, if you are having this many problems, it will probably be less work to get experimental phases than to try and rescue the MR. Halide soaking works in nearly 50% of cases (especially with recent improvements in SHELX), and selenomethionine labelling is usually very robust in E. coli. Sometimes you also get better resolution with the phasing sample. If you phase experimentally, you will be certain of your answer (even if it is that the loops are mobile), whereas with MR, you might always be concerned that there was a problem with the model....

Prior to the native Patterson map suggested above (which makes sense) I would calculate a self rotation map from the native diffraction data set (no coordinates necessary). This can be done e.g. with "polarrfn" from the CCP4 suite or with Liang Tong's nice program "glrf". Whenever there is space in the ASU for more than one copy of the protein a combination of self rotation map (providing rotational non-crystallographic symmetry) and native Patterson (providing translational non-crystallographic symmetry) makes sense and give valuable hints about the number of protomers per ASU.

You can confirm regions of flexibility using HDX then if you choose redesign the protein construct to eliminate those regions and it's likely you will get very good quality crystals for a better structure. Then rebuild the flexible regions if you choose. Usually highly flexible and disordered regions are involved in protein interactions. Based on what you know from the structure so far is there evidence that the protein is part of a family of proteins that are known to interact with other proteins or even DNA? You may find these references helpful.

http://www.ncbi.nlm.nih.gov/pubmed/14715906

http://www.ncbi.nlm.nih.gov/pubmed/15557262

Before you do anything you may want to check if your protein contains "intrinsically disordered regions". If it does, there's no reason for trying to find the structure of your loops. I'd recommend MeDor for this: http://www.vazymolo.org/MeDor/

I personally feel that the 27% which is in non crystal form probably has interacting groups such as polar or charged amino acids or stretch of similar amino acids like observed in the proteins of Plasmodium sp which therefore, do not occupy the space and gives you falls crystals

Further, I think you should consider to compare the protein structure with proteins having repeated amino acid stretch specially polar or charged amino acid proteins.

Great suggestions from the other commenters. Run your data through phenix.xtriage which will help diagnose problems such as twinning. What are your space group and unit cell dimensions?

Annika,

I feel your protein is not crystallized and hence it is better to think and model accordingly.To validate the modeling result you can follow one one of the best technique is Small Angle X-ray Scattering (SAXS) at nano order structural length.

Thank you all for your great suggestions. I really appreciate all of them and I am happy that you agree that this is a problematic issue.

Personally, I would prefer to go for selenomethionine labelling and experimental phasing, just to be sure. But I will also try ARP/wARP.

For the Patterson maps and space group determination, and everything else concerning the original reflection data, I will ask my colleague.