Protein structures solved by X-ray crystallography might not be as accurate as originally thought, according to a recent article in Structure. The technique, which has been invaluable for characterizing protein interactions and for aiding structure-based drug design, relies on building models on the basis of an observed diffraction pattern that is created by the scattering of X-rays through a protein in a crystallized form. However, Mark A. DePristo and colleagues explain that if only a single model is used to fit the experimental data, the heterogeneous structure and anisotropic motion of proteins intrinsic to their function is unaccounted for, and the structure is therefore likely to be inaccurate.

In a crystalline form, proteins are ordered by packing within the crystal lattice, but the high solvent content in most crystals means that heterogeneity remains. Modelling of these features is currently only possible for proteins that diffract to high resolution, whereas most proteins diffract to worse than 1.6 Å resolution and are therefore solved as a single conformation with isotropic motion — a kind of 'one size fits all' approach. The accuracy of macromolecular models determined by X-ray crystallography is still a subject of debate, but now it seems that the limitations of the models used for theoretical calculations are partly to blame.

The authors generated an ensemble of alternative solutions for three previously solved protein structures using diffraction data obtained from the Protein Data Bank (PDB). These alternative solutions were refined to fit the experimental data at least as well as the original PDB structure, but remained different in detail from both the PDB structure and each other. This enabled the authors to study differences between the final models and estimate their accuracy. Because the same protocol was used to determine each structure, differences observed in the final models must result from intrinsic structural heterogeneity, because they cannot result from experimental variation or subjective human observations.

One example of heterogeneity was described for human interleukin-1β (h-IL1-β), which showed extensive main- and side-chain variability between the authors' models. Variation was most pronounced in the main chain, which differed by as much a 1 Å between models, and was consistent with previously reported regions of disparity. This demonstrates that it is not possible to accurately model the crystal structure of h-IL1-β using an isotropic protocol.

The authors suggest that rather than concentrating on a single model, future X-ray crystallography models should actively incorporate heterogeneity, in a manner similar to the approach taken with nuclear magnetic resonance spectroscopy. Until then, features that depend on the exact positions of atoms, as well as small structural differences resulting from site-directed mutagenesis or ligand binding, should be treated with caution.