Glioblastoma is characterized by its ability to adapt and resist treatment, largely due to the tumor cells’ capacity to switch between different states—a phenomenon known as plasticity. Saraswat’s team profiled over 1 million nuclei from 12 GB samples using single-cell multi-omics, combining RNA sequencing (scRNA-seq) and chromatin accessibility profiling (scATAC-seq). This massive dataset allowed them to capture the full spectrum of tumor heterogeneity, identifying distinct malignant states such as hypoxic, gliosis-like, astrocyte-like (AC-like), oligodendrocyte progenitor cell/neural progenitor cell-like (OPC/NPC-like), and neuronal-like states.
A key finding of the study is the dramatic difference in plasticity between malignant and non-malignant cells. While malignant GB cells exhibit high plasticity, enabling them to transition between states, neuronal-like tumor states are surprisingly “locked” by strong repression barriers, showing low plasticity. This discovery hints at potential therapeutic targets to restrict tumor adaptability.
To unravel the regulatory logic behind these state transitions, the team developed a novel deep learning tool called scDORI (single-cell Decomposition of Regulatory Interactions). This autoencoder model decomposes multi-omic profiles into “regulatory Topics,” each representing specific transcription factor (TF)-target gene relationships. Unlike traditional methods, scDORI does not require predefined cell types, making it scalable to millions of cells and applicable to any multi-omic dataset. The tool identified key “Topic Regulators” (TRs)—master TFs driving specific tumor states—and revealed that while only 16% of TRs are actively expressed across states, over 54% are epigenetically accessible, indicating “primed drivers” ready to activate during state transitions.
The study’s regulatory roadmap highlights an asymmetry in GB plasticity: OPC/NPC-like and AC-like states can easily transition to alternate fates, while neuronal-like states are constrained by repression barriers. Intriguingly, this regulatory framework also explains the spatial organization of GB tumors. States with easy transitions are found in close proximity within the tumor, while those separated by regulatory barriers are spatially distant. This finding aligns with a companion spatial multi-omics study by the same group, which showed that GB subclones follow conserved spatiotemporal trajectories due to shared regulatory constraints.
A major breakthrough came with the identification of MYT1L, a transcription factor predicted by scDORI as a master repressor that locks GB cells into less plastic, neuronal-like states. Experimental validation confirmed that MYT1L overexpression closed over 80% of differential chromatin regions, while its knockout reopened access to more plastic fates. MYT1L was shown to directly repress regulators of other states, with 85% of scDORI’s predicted target TFs experimentally validated. Beyond chromatin changes, MYT1L overexpression transformed GB cells into less aggressive neuronal-like cells, exhibiting reduced proliferation, decreased tumor microtube connectivity, and slower growth in vivo, leading to longer survival in preclinical models.
Published as part of a two-paper series, this work complements findings from a Science Advances study on GB heterogeneity, which also emphasized the role of region-specific regulators in tumor behavior. Saraswat’s team believes their framework can be applied to other cancers to uncover and exploit “plasticity brakes,” potentially revolutionizing cancer treatment.
The research, funded by various agencies and made possible through patient-donated samples, has been met with enthusiasm in the scientific community. “This study not only deepens our understanding of GB’s complexity but also provides a actionable roadmap for therapeutic intervention,” said a spokesperson from the Bayraktar Lab. The team is already planning further studies to explore scDORI’s applications in other cancer types, inviting collaboration from researchers worldwide.
For more details, the regulatory paper and its companion spatial multi-omics study are available online, with links shared in Saraswat’s X thread. As the fight against Glioblastoma continues, this research marks a significant step forward in harnessing the power of multi-omics and deep learning to tackle one of medicine’s toughest challenges.
#GBM #MultiOmics #CancerResearch #DeepLearning #CancerNeuroscience #GRNs #SingleCell #Cancer
