The central aim of this thesis was to visualize changes in cellular architecture induced by environmental factors such as drug treatment or viral infection. High-resolution microscopic analysis of such alterations is crucial to elucidate the mechanisms of disease, as well as to develop potential therapies. The importance thereof was clearly underscored by the COVID 19 pandemic, whose causative agent, the novel severe acute respiratory syndrome coronavirus 2 (SARS CoV 2), infected and killed millions of people across the globe. By using high-resolution 3D imaging by cryo soft X ray tomography (cryo SXT), rapid changes in the cellular ultrastructure upon infection with SARS CoV 2 or feline infectious peritonitis virus (FIPV) could be visualized. The changes induced by viral infection were affected by treatment with FDA-approved drugs, which had previously emerged as candidates from drug-repurposing screens to combat SARS CoV 2. Cryo SXT analysis revealed formation of lysosome-associated dark-rimmed vesicles (DRVs), which were demonstrated to be multilamellar lipid deposits by transmission electron microscopy. The observed striking interplay of drug- and virus-induced alterations on the level of size and number of lysosome-associated DRVs suggests involvement of lysosomal function in the inhibition of the viruses by the drugs. It is likely that this inhibition is either due to impairment of lysosomal escape or due to reduced lipid availability for membrane remodeling essential for virus replication. To comprehensively test these models, complementary screening tools are required, which allow ultrastructural analysis at a higher throughput than cryo SXT. For that, the recently developed ultrastructure expansion microscopy (U ExM) is a promising technique. Comparing the structural preservation of cell organelles after the U ExM sample preparation to that of cells acquired under near-native conditions in cryo SXT, it emerged that the morphology of large organelles such as nuclei and plasma membrane can be visualized efficiently by U ExM, while very small or lipid-dense structures were more difficult to preserve accurately. These results highlight the benefits of cryo SXT, but also the potential of U ExM to complement cryo SXT with its higher throughput of ultrastructure visualization, even if its resolution is lower than that of cryo SXT. Therefore, U ExM can be used for experimental characterization for cryo SXT, as one step towards more efficiency of cryo SXT. Another step is to automate the segmentation of cryo SXT tomograms. To that end, a deep learning platform was trained and validated on a large pool of cryo SXT data acquired in this work. This convoluted neural network performs full annotation of cellular features in cryo SXT tomograms within less than half an hour on consumer-grade GPUs, thereby significantly reducing the time required for data analysis. Taken together, these results illustrate how cryo SXT, U ExM and deep learning can be used complementarily to address highly relevant biological questions.
