I believe the picture Harry is thinking of is Fig 2 from this paper: http://dx.doi.org/10.1080/0889311X.2010.527964 Where they show better diffraction by isolating a corner of a crystal, albeit with a smaller and denser beam. But I would stop short at concluding the crystal GREW that way. A lot of horrible things happened to it before it finally found its way into the beam. The harvesting and mounting of protein crystals involves extremely small surface-to-volume ratios in combination with a complex mixture of volatile compounds (including water) and a "payload" that is perhaps the the world's most sensitive detector of solution conditions: a protein crystal. If you have ever tried mounting a crystal out of a drop containing alcohols you may have noticed the "dancing crystal problem". Loss of volatiles from the surface is fast enough to set up eddy currents inside the drop, and they can carry crystals with them. This should give you an idea of the evaporation rates and the gradients of concentration that result. Yes, water has a lower vapor pressure than ethanol, but only by a factor of two. We have known for quite some time (Magdoff and Crick 1955 : http://dx.doi.org/10.1107/S0365110X55001461) that small changes in solvent content generate very large changes in structure factors. Once you have your crystal in a loop and out in the air, the surface-to-volume goes way up. By the time you get your loop into liquid nitrogen the outside of the drop in your loop has already dehydrated significantly, and the core has not had time to catch up. A crystal that spans this hydration gradient, or even sticks out of the loop will have experienced very different solution conditions before cryo-cooling, and even the cooling rate itself can be different at the leading edge of the droplet vs the back. Sometimes this can be a good thing! Dehydration can improve diffraction, as can faster or slower cooling rates. You never know until you try. That said, something that can't happen during cryo-cooling is twinning. Crystals have to grow that way, and I have now had more than two observations of needle or plate crystals that were actually two triangular crystals stuck together, forming a twin. Not always merohedral either. In one case the unit cell of one domain had "stretched" to match the rotated version of itself in the other domain every 5th spot. That one was a pain. I mention this because it is possible the OP is experiencing a similar "stretched twin". If that is the case, then then you have the opportunity for a novel kind of de-twinning: plot a given structure factor vs the "translation axis", and extrapolate to just off the tip. You have to do this for each hkl, however, so you might want to write a script. It is also possible that a hydration gradient or thermal gradient generated during sample prep has made the needle self-non-isomorphous. If so, then that is good news: it means that by optimizing your sample prep you may be able to get homogeneous crystals. My favorite tricks are working under oil, and trying to keep the crystal "floating" in the middle of a loop. Yes, you get more background than you would from a denuded crystal, but once the thickness of the loop/drop is less than twice that of the crystal itself you start getting diminishing returns on background by reducing it further. I recommend going for 2x the crystal size as the "buffer zone" around it. Rad dam is also a leading cause of "non-isomorphism". We can correct for changes in overall scale and B factor, but beyond that is what I call "residual non-isomorphism". Seems to accumulate at about 1%/MGy, but the resolution dependence has not been studied. Zero-dose extrapolation (ZDE) is formally the way to correct for it, and Kay has written more than one paper about that: http://dx.doi.org/10.1107/S0907444903006516 http://dx.doi.org/10.1107/S0907444905031537 There is an option in XDS to turn it on. However, ZDE is still far from a push-button solution, it depends critically on knowing what the relative dose is for each observation, and if you crystal is moving in and out of the beam then the default of assigning relative dose to frame number is not very accurate. XDS doesn't "know" the 3D shape of how your crystal intersects the beam. You need to think about how your crystal and the beam line up to get this right. A program called RADDOSE-3D http://raddo.se/ can help with this. The good news is that since you are "translating" you don't need to re-discover the indexing at every wedge. Since your one-wedge indexing seems to be unstable, I would definitely try giving all of the data to the indexing step. You may want to resort to just using the first image from each wedge to avoid cells that have expanded with radiation damage. Once you have an orientation (auto.mat from mosflm, or XPARM.XDS from xds) you want to propagate that orientation to all of the wedges. You generally have to do this manually. Again, you need to write your own scripts for this. My Elves front-end to mosflm will do orientation-propagation for you by default, but for the multi-wedge indexing you may need to re-organize the image files. Depends on how you named them. As usual, the more difficult the problem you are trying to solve the more you need to understand the nuts and bolts. The problem of combining incomplete and damage-ridden data from lots of potentially non-isomorphous crystals is an active field of research. If you want to play around with test data, I have a simulation a situation very similar to the OP's here: http://bl831a.als.lbl.gov/~jamesh/challenge/microfocus/ it is of crystals in P212121, but with a and b cell edges very close to the same value. In actuality, all the unit cells are the same size and the data are prefectly isomorphous, but you would never guess that from processing them. The only difference between the wedges is the crystal's orientation. Rad dam is calibrated for 5 micron crystals and the "right answer" is 1glc. I'd be interested if anyone out there has an automation pipeline that you can just feed these images and get a structure out. -James Holton MAD Scientist On 7/19/2015 10:50 AM, Harry Powell wrote: > hi Christopher > > I wasn't suggesting that Pointless might give a different space group > - I was suggesting its use for combining multiple MTZ files in a > consistently indexed way. > > Be that as it may, Kay has pointed out that P212121 has no indexing > ambiguity, so that's not the problem. > > It often happens that crystals are not homogeneous (I remember a > picture of a crystal in a micro focus beam with multiple sweet (and > sour...) spots. I don't remember who had the slide in their talk, I'm > afraid). This also happens with long needle-like crystals - often the > ends are substantially different from the point of nucleation, which > is usually near the middle. > > What happens to your processing stats etc if you try combining partial > data from different bits of the crystal (I.e. rejecting some parts of > the data)? > > Harry > -- > Dr Harry Powell, MRC Laboratory of Molecular Biology, Francis Crick > Avenue, Cambridge Biomedical Campus, Cambridge, CB2 0QH > Chairman of International Union of Crystallography Commission on > Crystallographic Computing > Chairman of European Crystallographic Association SIG9 > (Crystallographic Computing) > > On 19 Jul 2015, at 17:15, Christopher Barnes <[log in to unmask] > <mailto:[log in to unmask]>> wrote: > >> Harry- Yes we have used pointless when merging multiple sets, and we >> always come back with p212121. >> >> Tim - Yes i am confident that the values are not a by-product of >> generally little data per wedge, as some wedges containing more data >> (because of thicker regions of the crystal) produce the I/sigI or Isa >> values that I mentioned in my previous message. >> >> Kay - As for the indexing ambiguity, two of the axes are pretty close >> (within 5 Angstroms of each other), and so we do sometimes see >> programs index in higher symmetry (P4) spacegroup. However, I have >> collected multiple low dose datasets from single crystals to ~4.4 >> Angstrom (at a BM-line, so no translation), where you can clearly see >> that P212121 is most likely the right spacegroup (ISa for P212121 > >> 30 whereas ISa for P4 related spacegroups is never greater than 2, >> also determined using pointless as well). In addition, these crystals >> are composed of a multi-protein complex, where 75% of the components >> are structurally known; therefore, molecular replacement in p212121 >> gives a clear solution with no symm-related clashes, and initial >> rigid/bgroup refinement gives Rwork/Rfree values of 0.29/0.34. Maps >> also show a good fit for the known parts of the model, with clear >> electron density in the difference map for the novel components. >> >> As for the radiation damage, I wonder why this would cause such a >> problem, since we are only losing the high res reflections? So while >> we have datasets to 4.5 Angstrom, at that resolution there is no way >> to trace the novel components (se-mets are sparse, and generally >> reside in predicted disordered loop regions which would not be much >> help for tracing). So translating after a few degrees is the only way >> to maintain the high intensity reflections we need at the higher >> resolution (~3.5 Angstrom, which already is pretty weak). However, >> reflections less than 4.2 stay consistent (at least visually), so why >> would radiation damage between the last frame from one wedge and >> first frame of the next wedge cause such a dramatic drop even for the >> low res data, and is there any way I can correct for this? >> >> Thanks, >> Christopher