Investigating the causes and consequences of altered subcellular spatial composition in the immune system and beyond

Schmacke, Niklas Arndt (2023): Investigating the causes and consequences of altered subcellular spatial composition in the immune system and beyond. Dissertation, LMU München: Fakultät für Chemie und Pharmazie

Abstract

NLRP3 priming by translocation Working on the role of NEK7 in NLRP3 activation, we had discovered that, in con- trast to the role of NEK7 in mouse cells, human cells activate NLRP3 independently of NEK7. “Transplanting” mouse NLRP3 into a model of human monocytes rescued the activity of mouse NLRP3 in the absence of NEK7. From this result we concluded that rather than a difference between the two NLRP3 orthologues, a difference be- tween cellular signalling must be responsible for the differential requirement of NEK7 for NLRP3 activation. Coupled with the finding that TLR4 stimulation via LPS can bypass the requirement for NEK7 in mouse cells, we concluded that a pathway activated downstream of TLR4 can bypass NEK7 by priming NLRP3. Tracing the signalling cascade of TLR4 by genetically knocking out its components, we arrived at the kinase IKKβ. Indeed, experiments with knockouts of IKKβ in mouse and human cells explained both phenotypes: LPS could no longer bypass NEK7 in mouse cells and NLRP3 signalling in human cells was blunted. Why human cells are incapable of using NEK7 to prime NLRP3 in the absence of IKKβ remains unclear. Using human induced pluripotent stem cell-derived macrophages that we could genetically engineer to lack NEK7 as a model system, we confirmed that human cells in contrast to mouse cells do not require NEK7, but instead fully rely on IKKβ to prime NLRP3. Elucidating the mechanism by which IKKβ primes NLRP3 for NEK7-independent inflammasome activation, we found that IKKβ activity recruits a fluorescently tagged NLRP3 variant to the trans-Golgi network, a finding we corroborated by mass spec- trometry analysis of subcellular fractions. Our results define recruitment of NLRP3 to a specific organelle as a new priming modality of the NLRP3 inflammasome. CRISPR screening for subcellular spatial phenotypes at genome scale The development of charge coupled device (CCD) chips has enabled the acquisition of digital images at high resolution (Boyle and Smith 1970). In combination with modern microscopes the latest development of such chips has enabled the collection of large digital datasets representing the spatial composition of cells. A technology that can profile this composition and connect it to the genetic identity of individual cells at scale could generate insights into complex cellular biology. Here we developed a new genetic screening technology for image-based phenotypes. We first generated a library of 40 million human U2OS cells with one genetic knockout in each cell using CRISPR/Cas9. The cells in this library had been genetically engineered to express the fluorescently labelled autophagosome marker LC3 (mCherry-LC3). We stimulated these cells with the mTOR inhibitor Torin-1 to induce autophagy, during which LC3 gets redistributed onto autophagosomes. We then acquired microscopy images of this library and segmented these images into single cells using a nuclear stain to identify individual cells and a membrane stain to associate a the cytosol of a cell with its nucleus. This resulted in a dataset of single cell images across three channels: Nucleus and membrane that were used for segmentation and an image corresponding to the distribution of LC3 in each cell. Given that each cell in this library harboured a different genetic knockout, we expected some cells to have been unable to redistribute LC3 onto autophagosomes following Torin-1 stimulation owing to the lack of a gene that is essential for this process. We then sought to identify these cells based on their LC3 images. Since these data are inherently large and complex, we made use of the recent breakthrough in image analysis by machine learning models (LeCun et al. 2015). Using a knockout of ATG5, an essential autophagy gene, as a positive control, we trained a binary classifier based on a convolutional neural network to differentiate between images of cell undergoing autophagy and images of cells that had a blunted response to Torin-1 or were left unstimulated, and therefore not undergoing autophagy. With this classifier we were able to identify individual cells in our library that were incapable of forming autophagosomes in response to Torin-1. We then used fully automated laser microdissection to isolate the nuclei of these cells and subsequently sequenced their genetic perturbations. Here we found almost all genes known to be essential for autophagy to be defective in this pool of selected cells. This experiment demonstrates that our technology is capable of associating image-based phenotypes with the genotype of individual cells at genome scale.

• Dokumententyp: Dissertationen (Dissertation, LMU München)
• Themengebiete: 500 Naturwissenschaften und Mathematik ; 500 Naturwissenschaften und Mathematik > 540 Chemie
• Fakultäten: Fakultät für Chemie und Pharmazie
• Sprache der Hochschulschrift: Englisch
• Datum der mündlichen Prüfung: 21. September 2023
• 1. Bericht­erstatter:in: Hornung, Veit