Here’s a neat paper that’s shown up on arXiv.org on protein structures. The authors, from Yale and Edinburgh, are specifically comparing X-ray crystallographic structures with NMR-determined ones in solution. It’s widely known that when you look at the same (or nearly the same) proteins by both those methods that you see small but real differences: the question is why those exist. Crystal-packing in the solid state is certainly a different situation than floating around in buffer solution, so that’s a first-order answer, but what specifically is going on? And can you be sure that what you’re seeing is a real difference in structure, or is it just an artifact of the two experimental techniques?
The team identified 16 cases where proteins have been solved with high-quality data by both techniques (although, as they note, there isn’t a universally agreed-on framework for quality of NMR protein structures, so they used the ones where all residues had the appropriate number of restraints applied). Edit: originally misconstrued! Separately, they also created lists of high-quality protein structures solved by either method without the other one available, to try to exclude effects caused by ease of crystallization, solubility, and other factors peculiar to one system or the other. The differences that people had seen in the past definitely showed up this time – specifically, root-mean-square deviation (RMSD) of core alpha-carbon positions (higher in the NMR structures), higher “packing fraction” in the cores of NMR structures, changes in backbone and/or side chain dihedral angles, etc.
What’s interesting is that they go on to suggest a common physical basis for all these effects. They find that computational simulations of packing that don’t have an explicit temperature component (athermal) generate protein cores that look like the X-ray data, while adding some thermal annealing generates cores that look much more like the NMR structures (in fact, they can end up even more tightly packed). That second link above is an earlier paper that suggested that the differences seen have to do with the greater dynamics of the solution structures (as opposed to the limited mobility of the crystals), and this work ties in well with that.
So the “thermalized” systems can pack more tightly in their cores than the athermal ones, and that might well be just a function of the greater mobility in the NMR situation. But there might still be room for a “just an artifact of the technique” explanation when you consider that NMR structures are not only in solution, they’re done around room temperature. And X-ray structures are not only from solid crystals, they’re usually done at very low temperature (to slow the atomic motions and get tighter data). I suppose that one way to deal with that would be experimentally, but there’s a limit to how cold you can get buffer solutions (and the decreased motion might well broaden the NMR resonances?) And you could take RT protein crystal X-ray measurements, but the increased motion will almost certainly decrease the quality of the data for those as well. Never the twain shall meet, or not?