Scientists decode the hidden rules that steer brain cancer cell states

Glioblastoma, the deadliest and most common malignant brain tumour in adults, has long frustrated researchers because its cells refuse to behave as a single, uniform target. Individual tumour cells differ dramatically from one another, shifting dynamically between distinct molecular states, some resembling nerve cells, others looking like stem cells and some trapped in an aggressive, injury-like condition. This plasticity allows the cancer to dodge anti-cancer treatments, since a therapy that eliminates one cell state may leave others untouched. The prevailing assumption in the field has been that this state-shifting is essentially random and impossible to predict. Two new studies from researchers at the European Molecular Biology Laboratory, the German Cancer Research Center and the Wellcome Sanger Institute, published in the journal Nature, now challenge that assumption directly.

The first study set out to rebuild the picture of glioblastoma from first principles. Omer Bayraktar, Group Leader at the Wellcome Sanger Institute and a newly appointed professor at the UCL Cancer Institute, described the motivation bluntly: can researchers assume they know nothing about the disease and reconstruct how a cancer cell behaves through quantitative approaches? His team, funded by a Wellcome Leap grant, examined DNA, epigenetic marks and RNA in tumours from 12 patients, generating what the researchers describe as the largest glioblastoma molecular atlas to date. Epigenetic marks, the molecular annotations that regulate gene expression, were mapped alongside the genetic sequence and the RNA readouts, giving an unusually complete view of what each cell was doing.

To make sense of this highly complex dataset, Bayraktar’s team joined forces with Oliver Stegle’s group at EMBL Heidelberg and DKFZ, specialists in applying advanced computational methods to fundamental life science problems. Stegle framed the central question: is there a specific rule-based system that determines how cells transition between these states, and if so, what drives those rules? The answer the collaboration found was emphatic. Contrary to the long-standing view that state-shifting is unpredictable, glioblastoma cells followed certain stereotypical, predictable paths between cell states. Moreover, different cell states occupied different spatial niches inside the tumour, a pattern the researchers say was present in every single glioblastoma tumour from every patient studied.

That spatial finding depended on integrating single-cell data with spatial transcriptomics, a technique that records where molecular activity occurs within intact tissue. Bayraktar explained that the combined data revealed a meticulous and very specific spatial architecture of cancer cell states. The team then tested whether this architecture held outside the patient samples. They grew patient-derived tumour cells as organoids, three-dimensional cultures on a scaffold that closely mirror conditions inside the brain, and found that similar cell state trajectories emerged in these organoids as those deduced from the atlas data. The convergence between patient tissue and laboratory-grown tumour material strengthened the case that the trajectories were a genuine feature of the disease rather than an artefact of sampling.

With predictable trajectories established, the second study asked how the transitions are regulated. Led by Moritz Mall’s team at DKFZ, the researchers examined gene regulation signatures in more than one million patient-derived glioblastoma cells and used this information to work out how tumour cells move between states. Gene regulatory networks, the rules and molecular regulators that determine which genes are expressed in which cells and when, turned out to govern the process. The team identified a structured regulatory program controlling cellular plasticity that was both predictable and hierarchical, meaning some regulators sit upstream and shape the behaviour of others downstream.

Among the most consequential discoveries were a set of regulators the researchers call safeguards. These molecules arrest cells in low-plasticity states, preventing them from transitioning into other states. That property makes them attractive candidates for future therapies, because they could potentially be targeted genetically or pharmacologically. Mall noted that one of the team’s premises has always been that understanding these rules would allow researchers not only to observe what cells are doing but to change it, perhaps reprogramming them into a state that is less malignant and less dangerous.

The team tested that idea in patient-derived brain tumour cells. Overexpressing MYT1L, one of the safeguard regulators identified in the study, reduced cellular plasticity and tumour malignancy. When these modified tumours were transplanted into mice, they resulted in better survival than control tumours. Mall stressed that the MYT1L results, while interesting, are only part of the story: the approach predicts the regulators for all of the major tumour cell states, and the team tested those predictions experimentally, steering tumour cells into specific states by manipulating just a handful of regulators. In other words, the model’s predictions can be used to engineer the identity of a tumour cell.

The computational engine behind these predictions is scDORI, a deep learning method developed by Stegle’s team that infers gene regulatory networks at single-cell resolution. The diversity that makes glioblastoma so hard to treat became the raw material for the analysis. Each tumour, and each cell within it, carries its own combination of genetic and epigenetic changes, and across a million cells this produces a vast number of natural variations in gene regulation, far more than researchers could ever generate in the laboratory. By comparing how regulators and their target genes change together across more than a million cells, scDORI pinpoints which regulators steer cells toward one state or another. Those predictions were the starting point for the reprogramming experiments carried out by Mall’s team. Stegle put the philosophy succinctly: a tumour generates an almost endless number of variations on its own regulatory programme, and instead of seeing this heterogeneity only as an obstacle, researchers can use AI to learn from it and extract the rules that determine how cells choose their fate.

A second machine-learning method complemented scDORI by mapping where each cell state sits within the tumour, revealing the spatial architecture that the atlas analysis had described. The two studies therefore converge on a single picture: glioblastoma cells occupy defined states, follow predictable routes between them, occupy specific spatial niches, and are governed by identifiable regulatory programs that can be computationally inferred and experimentally manipulated. The researchers are now applying the approach to other cancers and brain disorders. While clinical applications may still be some way off, the framework is already helping researchers understand tumour cells at a deep molecular level and spot common patterns across diseases.

The immediate significance lies in what the work reframes. Glioblastoma kills more than 100,000 people each year worldwide, and its resistance to treatment has often been attributed to an untameable diversity of cells. These studies suggest that diversity is not noise but structure, governed by rules that can be read out with single-cell and spatial technologies and rewritten, at least in experimental systems, by manipulating a small number of regulators. Mall noted that with the help of the atlas and the new AI tool, the team identified new regulatory sequences that enable them to potentially drive specific cell states in a tumour-specific setting, adding that with the framework in hand it can now be applied to many other questions and cancers. For a disease with almost no curative options, the demonstration that tumour cell identity can be predicted and experimentally steered offers a new direction for future therapies, even as the path from mouse models to patients remains long.

Subject of Research: How gene regulatory networks govern cellular plasticity in glioblastoma

Article Title: Understanding the rules of brain cancer

Article References: Understanding the rules of brain cancer. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: glioblastoma, single-cell omics, gene regulatory networks, cellular plasticity, MYT1L, spatial transcriptomics, deep learning, scDORI, organoids, tumour heterogeneity, Nature, reprogramming