Computational methods for high-resolution structure determination of macromolecular complexes imaged in situ using cryo-electron tomography

Abstract

Cryo-electron tomography (CET) combined with sub-volume averaging (SVA) is a powerful imaging technique for determining macromolecular structures in situ. To resolve structures at high-resolution, large numbers of volumes containing copies of the protein of interest are aligned and averaged in three dimensions. Using this strategy, the structures of highly ordered virus capsid proteins and large ribosomes have been resolved at near-atomic resolution. However, CET studies of proteins of lower molecular weight (<1000 kDa) or targets present in their crowded native context have been limited to sub-nanometer resolutions. This is due to limitations in the accuracy of image alignment resulting from the low image contrast generated by the smaller scattering masses and the presence of overlapping objects in the cellular environment. While recent advances in high-throughput tomography that use beam image-shift accelerated data acquisition (BISECT) allow producing enough data for SVA, demanding storage and processing requirements associated with analyzing large numbers of particles often make structure determination impractical. The overarching goal of this thesis is to build a computational framework for CET/SVA structural determination that streamlines and extends the applicability of the technique to a wider class of biomedically-relevant targets while improving the resolution of structures to near-atomic resolution. To achieve this, this thesis focuses on: (1) filling gaps in the current CET data analysis workflow by designing a comprehensive end-to-end platform for SVA, (2) improving the resolution of structures by developing methods for improved alignment of protein images and better extraction of high-resolution information, and (3) validating our workflows by determining low-molecular weight structures and native membrane-bound proteins at near-atomic resolution. To routinely convert raw tilt-series into high-resolution structures, we developed high-throughput data collection approach, implemented robust strategies for tilt-series alignment and particle picking, and designed a scalable platform for distributed image analysis that makes analysis of large datasets feasible. To improve resolution, we used a constrained image alignment approach that uses parameters from the tilt geometry to overcome the low contrast and crowdedness of tomographic data. In addition, we efficiently recovered high-resolution signal contained in the raw data using per-tilt CTF correction and data-driven exposure weighting. These advances allowed the structure determination of low-molecular weight complexes such as dGTPase (300-kDa) and of immature human endogenous retrovirus K (HERV-K) Gag and immature human immunodeficiency virus 1 (HIV-1) Gag at near-atomic resolution. Our methods for CET/SVA allowed routine determination of structures of biomedically important targets both in-vitro and in situ at high enough resolution to elucidate mechanistic details governing virus assembly and infection. These advances will represent an important step towards closing the resolution gap between high-resolution strategies used to study molecular assemblies reconstituted in-vitro and techniques for in situ structure determination.