A multimodal single-cell approach to developing personalized combinatorial treatments for glioblastoma

A major impediment to the treatment of glioblastoma (GBM) is its extensive intratumoral heterogeneity: Each GBM tumor is heterogeneous with respect to genetics, immune microenvironment, and tumor cell state. In particular, single-cell RNA-sequencing (scRNA-seq) has revealed that each GBM is a mixture of different types of tumor cells, each with different biological properties, or cell states. Consequently, it may be necessary to treat each patient with personalized drug combinations. Although scRNA-seq has enabled us to understand the complexity of GBM, its clinical potential is largely untapped. This is partly because it is difficult to physically isolate and study the cell types discovered using scRNA-seq. In previous work, we showed that cell surface markers identified through scRNA-seq can be used to purify tumor subpopulations using flow sorting. Building on this, we propose a novel approach to isolating GBM cell populations, studying their therapeutic vulnerabilities, and identifying personalized drug combinations that target multiple cell populations simultaneously. Our approach is based on CITE-seq (Cellular Indexing of Transcriptomes and Epitopes by Sequencing), an extension of scRNA-seq that will enable us to efficiently define tumor cell populations and isolate them using flow sorting. We will test the drug sensitivities of each cell population, and then identify combinations of drugs that more effectively eradicate the tumor. This approach will be developed through two Specific Aims: (1) We will isolate and characterize GBM subpopulations using CITE-seq and flow sorting. Using our preliminary scRNA-seq data, we identified a panel of candidate cell-surface markers for known GBM subpopulations. Using oligo-conjugated antibodies to these putative markers, we will perform CITE-seq in primary tumors and matched cell lines to determine which markers cleanly distinguish among tumor cell populations. Using the most promising markers, we will test and optimize our ability to separate GBM samples into distinct subpopulations using flow sorting. In doing so, we will test the hypothesis that there exist stable GBM cell states that persist after flow sorting. (2) We will perform drug-screening on flow-sorted GBM subpopulations to identify personalized drug combinations. To this end, we will isolate several subpopulations from each of five cell lines using flow sorting, then test each subpopulation (as well as the flow-through and the unsorted cell lines) for sensitivity to each of 960 FDA-approved drugs including temozolomide (current standard of care), and specific combinations thereof. The resulting drug sensitivity matrix will be analyzed using clustering and machine learning techniques to identify drugs that, when used in combination, will target multiple GBM subpopulations and achieve greater efficacy than a single drug alone. If this approach is promising in pilot experiments, we ultimately hope to expand and adapt it to the clinic, so that we can readily identify drug combinations for any individual’s tumor.