PATH-54. A single-cell standard-of-care atlas of glioblastoma

Glioblastoma (GBM) remains the most aggressive primary brain tumor in adults with a median overall survival of less than 20 months despite standard-of-care (SOC) treatment consisting of maximal safe surgical resection, radiation therapy (RT) and temozolomide (TMZ) chemotherapy, with lomustine commonly used at recurrence. Over the past 20 years, the majority of clinical trials have failed to yield significant therapeutic advancements. In spite of this, in many clinical trials, small subsets (~5%) of patients would still exhibit exceptional clinical responses, suggesting that, with more precise selection procedures, more effective therapeutic options could still be offered. While genomic stratification seems insufficient to identify those patients, functional precision oncology (FPO) assays, whereby functional, therapeutic responses are measured in live, ex vivo tumor cells, offers a valuable alternative approach to improve patient selection. Today, the majority of FPO assays rely on cytotoxicity as a readout, limiting insights into the molecular complexity and heterogeneity of GBM. To address this, we created a single-cell therapeutic atlas of GBM by integrating cytotoxicity data from >60 patient-derived cell lines (PDCLs) exposed to the standard of care treatment options (RT, TMZ, RT+TMZ, and lomustine) with single-cell RNA-seq and clinical data. First, by comparing our PDCL single-cell transcriptomic profiles with a large public scRNA-seq dataset from GBM tumors, we found that our cell lines recapitulate the majority of molecular states observed in vivo. Furthermore, cytotoxicity responses in our cell lines were found to correlate with treatment responses observed in patients, underscoring their translational relevance. By subsequently analyzing >280 samples in control and treatment conditions, we created a cellular atlas of >1million single-cell profiles upon treatment. We identified cellular drug response metaprograms that were typically recovered in >5 samples. Additionally, we identified the gene regulatory networks driving these metaprograms and cataloged them based on their dynamics upon perturbation.