Computation of protein and cellular architecture from cryomicroscopy images

Electron cryomicroscopy reveals biological structure in a native, frozen– hydrated state. Images are recorded under low dose conditions in order to reduce radiation damage. They are therefore low in contrast and require computational analysis to extract structural information. This thesis describes the development of computational methods for structure determination by single particle analysis and tomography and their application to several biologically important specimens. The reliable determination of reconstruction parameters in single particle analysis is the first subject addressed. To determine a structure from projection images, relative orientations must be assigned to each view of the specimen as well as microscope parameters describing magnification and contrast transfer. A new method of magnification determination is described that provides an automatic estimate of magnification using apoferritin as a reference. Modelbased refinement of particle orientation parameters is susceptible to “model bias” in the case of noisy experimental images and can result in incorrect orientation determination. A new implementation of tilt pair cross-validation was created in order to validate orientation parameters. The software has been implemented as a fully automatic web server toward the goal of providing a standard cross-validation tool. The latter part of the thesis is devoted to problems in the computation and interpretation of low contrast, sparsely sampled tomographic reconstructions. The first model of an intact Weibel-Palade Body (WPB) was built based on tomograms computed from tilt series of frozen-hydrated endothelial cells. A template matching procedure was introduced in order to facilitate the analysis of von Willebrand Factor packaging in WPBs leading to an understanding of the morphology and physical properties of WPBs. New analytical tools were also required to facilitate the interpretation of the spatial organization of glycoproteins on the surface of influenza A virus that may have important consequences for influenza biology. A three dimensional model of the influenza haemagglutinin positions on the virion surface was constructed based on analysis of cylindrical projections