Electron Crystallography

In the electron crystallographic approach to structure determination, phases are extracted from images of two-dimensional (2D) crystals and combined with amplitudes measured from electron diffraction patterns. Diffraction data can however be obtained far more quickly, and often to higher resolution, than images. The success of refinement procedures with bacteriorhodopsin has shown that diffraction amplitudes are adequate to determine and refine a structure (although phase residuals can be used to guide the refinement as well). We therefore assumed that diffraction data can be recorded with sufficient accuracy that it should be possible to use starting phases from the model of a homolog (or from partial image data) and to carry out other computational procedures as in X-ray crystallography. Molecular replacement – the use of one structure to determine phases for diffraction data from crystals of a related structure – is an especially powerful method in X-ray crystallographic structure analysis, because the evolution of proteins, including membrane proteins, has generated large families of polypeptides with related folds. Molecular replacement therefore enormously facilitates structure determination in the increasingly frequent cases where a model for a structurally related protein is available. Hence, in collaboration with Stephen Harrison (Harvard Medical School), we began to develop the application of related methods to 2D crystals of membrane proteins. We are also exploring ways to exploit non-crystallographic symmetry, multiple crystal forms, and fine sampling along lattice lines to refine and extend phases. A long-range goal is to develop experience in identifying the correct combination of improved diffraction data and limited phase information from images to implement a general strategy for accelerated structure determination. This strategy will involve essentially the same kinds of phase improvement calculations that routinely yield interpretable maps in X-ray crystallography.

Since the crystal structure of the homologous bovine AQP1 was available, we determined the structure of the AQP0 membrane junction by molecular replacement, thus avoiding the cumbersome and time-consuming process of collecting high-resolution images of tilted specimens. The structure of AQP0 (bottom panel) was determined by molecular replacement in MOLREP using as search model the AQP1 structure (top panel) lacking residues 36-47 and 122-134. Crowther Fast Cross-Rotation Function calculations identified an orientation of a single subunit in the asymmetric unit as the top solution with an RF signal twice the value of the second peak. The initially determined unit cell axis a = b of 68 Å was refined over ± 6 Å in 0.2 Å increments by repeating the search and optimizing the signal. The top solution was identified as 65 Å. The translation function with the monomer in the refined unit cell gave a top solution with an R-factor of 52.5% and a correlation coefficient of 42.7% (average values for the top 50 translation peaks were 57.8% and 25.3%, respectively). In the initial rounds of refinement we evaluated both electron and X-ray scattering factors. As expected for 3 Å resolution data, there was little difference in the resulting refinement statistics. In all subsequent refinement rounds we used X-ray scattering factors. At this stage the residues in AQP1 that differed from those in the AQP0 sequence were replaced by Ala (or Gly if the substitution was to Gly). The model was refined using CNS version 1.1. In a first step the rigid body refinement was repeated, varying the unit cell dimension in 0.2 Å increments. The R-free minimum showed that the cell axis had to be adjusted to 65.5 Å, and this value was used in further refinements. Subsequent model building was performed with O using 2Fo-Fc density modified by solvent flipping, and simulated annealing composite omit-maps. The model was refined by simulated annealing with a number of refinement parameters (starting temperature, cooling rate, stereochemical weight term, energy constant of dihedral restraints for α-helical regions, range around restrained angles for helical regions) optimized by running several refinement cycles with a wide range of starting values and ultimately selecting the ones that gave the best R-free.