Modeling of Image Formation in Cryo-Electron Microscopy

Abstract

Knowledge of the structure of biological specimens is crucial for understanding life. Cryo-electron microscopy (cryo-EM) permits structural studies of biological specimen at their near-native state. The research performed in this thesis represents one of two subprojects of the FOM industrial partnership program with FEI Company. The common aim is to obtain higher resolution in cryo-EM of biological specimens. Currently, the resolution is limited by: i) the noise and blurring of the detector; ii) the oscillatory and dampening character of the contrast transfer function (CTF) originating from defocusing which is employed to produce contrast; and iii) the radiation damage which limits the integrated electron flux that can be used, resulting in images with poor signal-to-noise ratio. Simulation of image formation (forward modeling) provides possibility to easily and cost-effectively investigate the influence of a certain physical parameter on the final image. The main goal of this thesis is to improve our understanding of the relevant physical processes that govern image formation and to develop a quantitative forward model. Such a model is essential for optimizing the acquisition strategy, assisting the regularization (introduction of prior information) in the 3D reconstruction, improving image interpretation, and achieving a resolution beyond the limits imposed by the oscillatory CTF. This thesis addresses the following challenges: i) construction of the electron-specimen interaction potential based on elastic and inelastic electron scattering properties and adequate description of the electron propagation through the specimen; ii) accurate estimation of the CTF parameters, in particular defocus and astigmatism and their uncertainties, iii) characterization of the detector including all relevant statistics; iv) better understanding of certain aspects of radiation damage such as specimen heating, dose-rate effects, and beam-induced movements. The validation of forward model is based on a systematic comparison between simulated and experimental images under various experimental conditions. All parameters are based on physical principles. Defocus and astigmatism as well as detector parameters are accurately estimated from independent measurements using the methods developed in this thesis. Software tools for image simulations, accurate defocus and astigmatism estimation and detector characterization have been developed and are freely available for non-commercial use. The theory and methods presented in this thesis form the essence of an expert system that would optimize the data collection strategy. Furthermore, the influence of new hardware components could be inexpensively and efficiently investigated.