Non-genetic evolution via cell-state switching drives glioblastoma (GBM) progression and therapy resistance. While transcriptional states are defined, the regulatory mechanisms that enable or constrain transitions remain unclear. Here, we mapped the epigenetic architecture of GBM plasticity and identified regulators that restrict malignant state transitions.

We analysed over one million paired single-nucleus RNA and chromatin accessibility profiles from primary IDH wildtype GBM and developed a deep learning framework (scDORI) that infers enhancer-driven gene regulatory networks at single-cell resolution. This defined state-specific programs and quantified transition potential based on activating and repressive interactions. We identified 60 key GBM regulators and performed a pooled single-cell gain-of-function screen across three patient-derived models, profiling over 250,000 perturbed cells to discover plasticity suppressors. Selected low-plasticity state regulators were mechanistically characterised through functional profiling and in vivo validation.

Multiome profiling in patients revealed a hierarchical organisation of GBM states with distinct epigenetic plasticity. Neuronal-like tumour cells occupied a low-plasticity attractor state characterised by repression of alternative lineage regulators, whereas OPC, NPC-like, and astrocytic states displayed permissive enhancer landscapes and high transition potential. A gain-of-function screen identified state master regulators and showed that enforced expression of neuronal-like factors redistributed cells towards reduced plasticity. Mechanistically, these regulators converged on the selective closure of enhancers associated with alternate-state master transcription factors. The transcription factor MYT1L emerged as a critical safeguard of GBM plasticity. MYT1L gain-of-function induced widespread chromatin compaction, with over 80% of differentially accessible regions closing, thereby repressing competing master regulators, reducing proliferation, and significantly delaying orthotopic tumour growth, with improved survival in vivo.

We provide a functional map of GBM plasticity in patients and quantify the effects of 60 state regulators on fate switching in patient-derived cells. Non-genetic GBM evolution follows a hierarchical regulatory logic governed by enhancer-driven activation and active repression. Low-plasticity attractor states exist and can be induced by plasticity repressors, which establish epigenetic barriers to state switching. Enforced suppression of plasticity programs may offer a strategy to constrain adaptive tumour evolution. Our integrated computational and experimental framework provides a blueprint for decoding and targeting plasticity regulators across heterogeneous cancers.