One of our premises has always been that when we understand these rules, we can not only observe what the cells are doing, but we can also change what they are doing,
Moritz Mall said of new findings published in Nature. Glioblastoma kills over 100,000 people worldwide each year, driven by tumor cells that constantly shift between distinct cellular states to evade anti-cancer treatments. Two companion studies from researchers at EMBL, the German Cancer Research Center (DKFZ), and the Wellcome Sanger Institute show that this state-shifting follows predictable, hierarchical patterns governed by specific gene regulatory networks rather than occurring at random.
Glioblastoma remains the deadliest and most common malignant brain tumor in adults. Individual cells inside these tumors vary wildly, with some resembling nerve cells, others looking like stem cells, and some trapped in an aggressive, injury-like state. Traditionally, researchers viewed this cellular plasticity as an unpredictable obstacle to therapy.
Building a Molecular Atlas from Patient Samples
To challenge that assumption, Omer Bayraktar’s team at the Wellcome Sanger Institute examined DNA, epigenetic marks, and RNA in tumors from 12 patients, generating one of the largest glioblastoma molecular atlases to date. Funded by a Wellcome Leap grant, the team used single-cell spatial technologies to map cellular diversity. To process the complex dataset, they partnered with Oliver Stegle’s team at EMBL Heidelberg and DKFZ.
We asked – can we go back to the drawing board?
Bayraktar said, describing the quantitative approach. By analyzing over 1 million patient-derived glioblastoma cells, Stegle’s team developed scDORI, a deep learning method that infers gene regulatory networks at single-cell resolution. The tool extracts rules from natural variations across millions of cells, far outstripping what researchers could generate in a laboratory setting.
Steering Tumor Cells with Safeguard Regulators
The analysis revealed a structured regulatory program governing cellular plasticity along predictable paths. Researchers also discovered specific ‘safeguard’ regulators that arrest cells in low-plasticity states. Testing this hypothesis in patient-derived brain tumor cells, Mall’s team found that overexpressing MYT1L, one such safeguard, reduced cellular plasticity and tumor malignancy.
When transplanted into mice, these modified tumors resulted in better survival rates than control tumors. Our approach predicts the regulators for all of the major tumour cell states, and we tested those predictions experimentally,
Mall explained. We were able to steer tumour cells into specific states by manipulating just a handful of regulators.
A second machine-learning method mapped the spatial architecture of these cell states inside the tumor, showing that different states occupy distinct spatial niches.
Common Questions About Glioblastoma Plasticity
Why did previous researchers view glioblastoma state-shifting as random?
High tumor heterogeneity meant that individual cells carried vastly different combinations of genetic and epigenetic changes, making patterns difficult to detect without large-scale single-cell sequencing and advanced artificial intelligence tools.
How does the scDORI deep learning method work?
Developed by Oliver Stegle’s team, scDORI compares how regulators and their target genes change together across more than a million cells to pinpoint which regulators steer cells toward specific states.
What are the next steps for this research?
Researchers are currently applying this spatial and AI-driven framework to other cancers and brain disorders to identify common molecular patterns and potential therapeutic targets.