Abstract High-grade gliomas are a leading cause of cancer-related death in adults and children. They are highly heterogeneous diseases in which both inter- and intra-tumoral heterogeneity contribute to disease progression and therapeutic failure. In HGG, defined cellular states with key phenotypic characteristics are selected for during tumorigenesis, drive tumor evolution, and underlie resistance to therapy and invasion. In particular, HGG are thought to be driven by glioma stem cells (GSC), subpopulations of cells recapitulating aspects of neural development that have the preferential capacity to self-renew and to generate differentiated cancer cells. Traditional methodologies to identify GSC rely on functional assays with important caveats and thus do not allow a comprehensive characterization of cellular states in human patients. Additionally, while models of GSC and HGG are extensively used for research, very little is known about their capacity to comprehensively mirror the spectrum of cellular states present in patient samples; due to these limitations, vulnerabilities identified in models frequently do not translate to clinical settings. Accordingly, we propose that the range of cellular states that drive HGG should first be defined directly from patient samples, at single cell resolution, and subsequently be functionally tested in animal and cell-based models. More specifically, we will leverage single-cell RNA-sequencing and a comprehensive systems biology approach in order to (I) identify tumor subpopulations unbiasedly across different genetic clones in human HGG and in matched models of disease, at single-cell resolution; (II) functionally test the capacity of these subpopulations to initiate tumors and to re-generate the diversity of states present in patients; (III) identify faithful cell models that can recapitulate defined cellular states observed in patients and utilize them to experimentally identify regulators with potential utility in clinical settings. Successful completion of the research will fill a fundamental and large gap of knowledge in understanding brain cancer in patients and in models and will provide novel opportunities to target key cellular states that are driving these incurable malignancies. Furthermore, the proposed approach could be extended to other malignancies, and will provide a proof-of-concept for uncovering subpopulations that drive tumor growth directly from patient samples, and subsequently identifying regulators with potential clinical relevance.