Abstract
2D electron crystallography can be used to study small membrane proteins in their native environment. Obtaining highly ordered 2D crystals is difficult and time-consuming. However, 2D crystals diffracting to only 10-12 Angstrom can be prepared relatively conveniently in most cases. We have developed image-processing algorithms allowing to generate a high resolution 3D structure from cryo-electron crystallography images of badly ordered crystals. These include movie-mode unbending, that corrects for crystal imperfections and locally-varying beam-induced sample deformations, while preventing overfitting and computationally optimizing the electron dose used for each resolution range individually; refinement over sub-tiles of the images in order to locally refine the sample tilt geometry within different tile locations on the images; and implementation of different CTF correction schemes. Additionally, the sample tilt angle achievable while collecting 2D crystal data, is typically limited to 60 degrees. This leads to a missing conical region in 3D Fourier space known as the 'missing cone'. In real space, the missing cone makes the densities look elongated in the vertical direction. In addition, data can also be missing in other regions, depending on the tilt sampling. Here we report an iterative method that applies known constraints in the real and reciprocal space to approximate amplitudes and phases in missing regions. We applied the new procedure to a three-dimensional (3D) dataset consisting of ~400 movies of the voltage-gated ion-channel MloK1 crystallized in 2D. The 3D density map and its quality measurements improved significantly. The refined map shows more distinct densities than the 3D reconstruction obtained by classical crystal unbending and is easier to interpret.
- Posted March 3, 2017.
