Multiplatform molecular profiling and functional genomic screens identify prognostic signatures and mechanisms underlying MEK inhibitor response in somatic NF1 mutant glioblastoma

Abstract

NF1 is recurrently mutated in glioblastoma yet the molecular landscape and efficacy of targeted therapies remain unclear. Here, we combine bulk and single cell genomics of human somatic NF1 mutant, IDH-wildtype glioblastomas with functional genomic analysis of cell lines and mouse intracranial tumor models to identify molecular subgroups within NF1 mutant glioblastomas and mechanisms underlying MEK inhibitor response. Targeted DNA sequencing showed homozygous deletion of the cell cycle regulator CDKN2A/B is a poor prognostic marker in somatic NF1 mutant, but not NF1 wildtype, tumors. DNA methylation array profiling revealed three epigenetic groups highlighted by distinct clinical features, co-mutation patterns, and reference methylation classifier identities. Genome-wide CRISPRi screens in glioblastoma cells revealed cell cycle regulators are conserved mediators of cell growth while response to the MEK inhibitor selumetinib converges on Ras/RAF/MEK pathway activation. Repression of the RAF regulator SHOC2 sensitizes glioblastomas to selumetinib in vitro and in vivo in mouse intracranial glioblastoma models. Single cell RNA-sequencing of mouse intracranial glioblastomas treated with the MEK inhibitor selumetinib reveals distinct responses between mesenchymal-like (MES-like) and non MES-like subpopulations suggesting non-MES like cells are intrinsically resistant to MEK inhibition. Finally, single nuclear RNA-sequencing (snRNA-seq) of human NF1 mutant, CDKN2A/B deleted glioblastomas reveals MES-like tumor cells are associated with selumetinib sensitivity signatures while non MES-like cells exhibit increased cell cycle progression and lack selumetinib sensitivity, further supporting the notion that MEK inhibition specifically targets MES-like tumor cell subpopulations. Taken together, our data underscores the heterogeneity between and within somatic NF1 mutant glioblastomas and delineates mechanisms of MEK inhibitor response across distinct tumor subpopulations, guiding the development of future therapeutic strategies that may synergize with MEK inhibition for NF1 mutant tumors. The tumor suppressor NF1 is mutated in 15% of glioblastomas,1-3 the most common malignant brain tumor with poor outcomes and few effective treatments.4 NF1 is a GTPase activating protein (GAP) that negatively regulates Ras, and thus, NF1 loss leads to induction of Ras/RAF/MEK/ERK signaling, driving tumorigenesis and comprising a targetable molecular cascade.5,6 Genomic analysis of glioblastoma demonstrates NF1 mutation is associated with a mesenchymal-like (MES-like) transcriptomic tumor cell subpopulation and altered tumor microenvironment.7,8 More broadly, DNA methylation analysis reveals multiple epigenetic subgroups with overlapping relationships to transcriptomic subtype and DNA alterations, underscoring the complex relationship between genetic drivers and molecular signatures.9 While the updated 2021 Central Nervous System WHO tumor classification incorporates an ever increasing amount of molecular criteria for diffuse astrocytic tumors,10 the existence and clinical significance of molecular subgroups within somatic NF1 mutant, IDH-wildtype glioblastomas based on genetic co-mutations, epigenetic profile, or transcriptomic signatures remain unclear. Preclinical data support the utility of MEK inhibition in NF1 mutant gliomas,11,12 and the MEK inhibitor selumetinib is FDA approved for tumors arising in patients with syndromic neurofibromatosis type 1 (NF-1) harboring a germline NF1 mutation.13,14 In NF-1 associated gliomas, MEK inhibition demonstrates efficacy in a limited case series,15 and combined BRAF/MEK inhibition shows efficacy in BRAF p.V600E mutant gliomas,16 further supporting the translational potential of Ras/Raf/MEK/ERK blockade within genetically defined glioma subtypes. Nevertheless, treatment resistance to molecular monotherapy remains a challenge,17-20 and the mechanisms underlying MEK inhibitor resistance in NF1 mutant glioma are unknown. Here, we integrate targeted DNA sequencing, DNA methylation profiling, and single nuclear RNA-sequencing (snRNA-seq) of human patient somatic NF1 mutant, IDH-wildtype glioblastomas with single cell RNA-sequencing (scRNA-seq), genome-wide clustered regularly interspaced short palindromic repeats interference (CRISPRi) screens, and pharmacologic studies in cell lines and mouse intracranial glioblastoma models to define molecular subgroups and functional mediators of MEK inhibitor response. Targeted DNA sequencing of NF1 mutant, IDH-wildtype glioblastomas (n=186 tumors) revealed CDKN2A/B deletion was associated with poor outcomes in NF1 mutant, but not NF1 wildtype, glioblastomas. DNA methylation profiling (n=129 tumors) demonstrated three epigenetic subgroups with distinct clinical features, co- mutation patterns across cell cycle genes, and reference methylation classifier identities. Genome-wide CRISPRi screens in mouse SB28 and human GBM43 glioblastoma cells identified a conserved cell cycle gene network mediating cell growth, consistent with the clinical importance of additional hits affecting the cell cycle in human somatic NF1 mutant glioblastomas. Moreover, genome-wide mediators of selumetinib response converged upon two Ras pathway effectors mediating selumetinib sensitivity: BRAF and SHOC2. SHOC2 repression in glioblastoma cells significantly improved selumetinib response both in vitro and in intracranial allografts in vivo. Single cell RNA-sequencing (scRNA-seq) of mouse intracranial glioblastomas treated with the MEK inhibitor selumetinib revealed MES-like tumor cells correlated with CDKN2A retention and the CRISPRi screen selumetinib sensitivity signature, with selumetinib resistant cells displaying Ras pathway induction. In contrast, non-MES like tumor cells were CDKN2A deficient and lacked expression of the CRISPRi screen selumetinib sensitivity signature, with selumetinib resistant cells inducing a glial de-differentiation program. Finally, snRNA-seq of NF1 mutant, CDKN2A/B deleted, IDH-wildtype glioblastomas (n=9) showed non MES-like tumor cells exhibit increased cell cycle progression and were not associated with the CRISPRi screen selumetinib sensitivity signature. MES-like tumor cells within newly diagnosed, but not recurrent, tumors retained expression of the CRISPRi screen selumetinib sensitivity signature, suggesting resistance can arise both between and within specific transcriptomic glioblastoma cell tumor cell subpopulations. Taken together, our data identifies clinically important subgroups of NF1 mutant, IDH-wildtype glioblastomas and supports a model in which heterogeneity between tumors and within tumor cell subpopulations underlies MEK inhibitor response, supporting the need for additional synergistic therapeutic approaches beyond maximal Ras pathway blockade for NF1 mutant glioblastomas.