Understanding glioblastoma : cell identity in tissue space

<p dir="ltr"><b>Abstract</b></p><p dir="ltr">Glioblastoma is the most prevalent form of brain cancer among adults. Inherently malignant and aggressive, glioblastoma cells persist through treatment - tumors grow back and uniformly prove lethal, oftentimes within months of diagnosis. Therefore, there exists an urgent need to understand which tumor components confer these highly malignant properties. Substantial advances on this front have been made in recent years using single-cell transcriptomics technologies: the general landscape of glioblastoma cell identities has been defined, with neurodevelopmental and a mesenchymal-like cell states comprising the highly heterogeneous tumor compartment. However, the full range of neurodevelopmental tumor cell states is still under debate, as new identities are occasionally reported, whereas the mesenchymal designation has been linked to hypoxia or injury responses, although the nature of this association remains contested. This indicates that the molecular diversity of glioblastoma has not yet been mapped exhaustively, nor are the defined cell identities fully understood.</p><p dir="ltr">the full range of neurodevelopmental tumor cell states is still under debate, as new identities are occasionally reported, whereas the mesenchymal designation has been linked to hypoxia or injury responses, although the nature of this association remains contested. This indicates that the molecular diversity of glioblastoma has not yet been mapped exhaustively, nor are the defined cell identities fully understood.</p><p dir="ltr">Cell identity, however, is inherently linked to spatiotemporal self-organization of tissues, as seen in highly dynamic contexts such as embryonic development or wound healing. Given that glioblastoma has been proposed to partly recapitulate both, we hypothesized that glioblastoma cell identity is, too, best understood from a spatial tissue organization perspective. We therefore set out to create novel tools that would allow spatial transcriptomics profiling of human glioblastoma at scale and to deploy these tools to refine the classification and understanding of glioblastoma cell identities.</p><p dir="ltr">We firstly present EEL-FISH, a single-molecule RNA fluorescence in situ hybridization method based on electrophoretic transfer of RNA onto a glass slide. The electrophoretic transfer combined with subsequent removal of other tissue components solved several key limitations of preceding single-molecule spatial transcriptomics methods: the need to image z-stacks spanning an entire tissue section and the need to overcome tissue autofluorescence through signal amplification or background quenching. We furthermore implemented multi-color combinatorial signal barcoding to increase gene-target multiplexing. Together, these improvements constitute one of the first truly high-throughput (888 gene panel on square centimeter scale tissue sections) single-molecule spatial transcriptomics methods.</p><p dir="ltr">We next deployed this method to profile 25 human glioblastoma and 2 oligodendroglioma tumors, thus providing one of the largest integrated and spatially resolved single-cell glioblastoma datasets to date. We mapped the diversity of tumor cell states captured within our data to single-cell RNA sequencing references of glioblastoma and early neurodevelopment, thus creating a detailed classification that links tumor cells to precise developmental counterparts, a fibroblast-like state, or a range of wound response meta-states. We argue that the last two collectively account for the mesenchymal-like identity. Characterizing large tissue areas and samples from anatomically distinct regions of same-patient tumors, we furthermore found the wound response states to be a phenomenon of the tumor core, internally organized by successive activation of distinct wound response modules nested around blood vessels and associated with increasingly pro-tumor myeloid infiltration. We complemented these findings with in vitro perturbation experiments on patient-derived glioblastoma organoids to demonstrate that wound response states may indeed be hypoxia-inducible and transient.</p><p dir="ltr">Lastly, we aimed to expand the toolkit of cell identity investigation by taking dynamical cell behavior into account. To that end, we integrated live imaging of organotypic human glioblastoma slice cultures with post hoc spatial transcriptomics readout, thus allowing to link measures of cell motility to transcriptional identity. We provide a detailed protocol and showcase the feasibility of using this approach to study otherwise-inaccessible dimensions of cell identity. Taken together, the work presented in this thesis opens new avenues of spatial transcriptomics profiling and supports the notion that cell identity in glioblastoma might be best understood in respect to tissue space.</p>