Lillian Bussema

The progression of isocitrate dehydrogenase-mutant glioma (IDH-G) from slow-growing tumor to fatal disease is associated with transcriptional and DNA methylation changes that remain poorly understood. Here, we profiled a longitudinal cohort of 36 IDH-G samples from 19 patients by joint-capture multi-omic single-nucleus DNA methylation, single-nucleus RNA sequencing and bulk exome sequencing. We show that IDH-G progression is associated with an increase in malignant stem-like states, decreased differentiation and methylation loss, which marks tumors with worse clinical outcome. Methylation loss was uniformly observed across malignant cells within individual tumors, suggesting that it may underlie rather than result from the increase in stem-like states. Analysis of cell-state heritability and plasticity using high-resolution phylogenetic trees links DNA methylation loss to alterations in glioma cell-state encoding and heritability. Our study offers insights into how DNA methylation loss reshapes cellular transitions and how it may mark clinically more aggressive tumors across IDH-G subsets.

Gliomas with mutant isocitrate dehydrogenase (IDH) are malignant brain tumours that typically arise in early to mid-adulthood and nearly always recur following treatment ^1,2 . However, the genetic and cellular-state changes that drive IDH-mutant glioma progression under treatment remain incompletely understood. Here we integrated single-nucleus transcriptomic profiles, chromatin accessibility profiles and bulk DNA and RNA sequencing from 75 temporally separated gliomas across 35 patients comprising both the oligodendroglioma and astrocytoma IDH-mutant glioma tumour types. We show that malignant cell states transcriptionally resemble stages of normal glial–neuronal lineage development or a reactive mesenchymal-like state, mirroring states previously described in IDH wild-type glioblastoma ^3,4 . Malignant cell states displayed distinct chromatin accessibility profiles that were comparable between both IDH-mutant glioma types. The abundance of less differentiated malignant cells increased with grade and with genetic alterations such as PDGFRA amplification. Longitudinal analysis highlighted two major malignant cell-state transition patterns. First, reduced lineage differentiation and increased proliferative malignant cells at recurrence were enriched in gliomas that acquired recurrence-associated genetic events. These included treatment-associated hypermutation, increased copy number changes and cell cycle alterations. Second, increased mesenchymal-like-state abundance occurred independently of acquired genetic alterations and instead coincided with elevated macrophage expression. Overall, our findings provide an integrative model that traces the cell intrinsic and extrinsic factors that shape cellular states during IDH-mutant glioma disease progression.

Glioblastoma invasion into brain parenchyma presents significant challenges for treatment but remains poorly understood. In this study, we combine single-cell RNA sequencing, spatial transcriptomics, and multiplexed imaging of orthotopic xenograft models to investigate glioblastoma invasion. We first screen 20 patient-derived gliomasphere models for their distal (i.e., extending to the contralateral hemisphere) and local invasive potential in mice. We show that models with distal invasion potential are enriched with oligodendrocyte progenitor-like cells, while models with only local invasion potential are enriched with mesenchymal-like cells. These patterns reflect predominantly peri-axonal vs peri-vascular invasion routes, respectively. Next, we analyze the transcriptomes of invading cells within models (compared to tumor core) and identify programs associated with distal and local invasion. Thus, we decouple transcriptional features associated with invasion potential from those associated with the process of invasion. We validate our findings by spatial transcriptomics and multiplexed imaging, further describing the spatial niche of invasive cells. Taken together, our results provide a blueprint for the invasive potential of glioblastoma cell states and of the programs associated with invasion across different scales.

Adenosine deaminase acting on RNA (ADAR) converts adenosine to inosine within double-stranded RNA (dsRNA) and can be co-opted for therapeutic RNA editing by introducing dsRNA substrates in trans using programmable guide RNAs (gRNAs). However, ADAR's natural promiscuity necessitates sophisticated gRNA designs to achieve efficient and specific editing. Here we present Helix, a predictive model that achieves highly accurate, zero-shot per-adenosine editing predictions for any target sequence. Helix's performance arises from two architectural choices: a transformer framework that scales effectively with increasing and imbalanced training data; and a structure-aware attention mechanism that incorporates predicted RNA secondary structure, a key determinant of ADAR activity. Helix's predictive accuracy enables seamless integration with DeepREAD, our previously reported generative model, in a noisy-student distillation framework termed DeepHelix. This workflow supports both zero-shot gRNA design and the generation of complex, constraint-based designs. We demonstrate DeepHelix's utility by designing gRNAs that efficiently edit a therapeutically relevant adenosine and by leveraging its flexible design space to engineer species cross-reactive gRNAs to accelerate pre-clinical development.

The evolution of isocitrate dehydrogenase (IDH)-wildtype glioblastoma (GBM) after standard-of-care therapy remains poorly understood. Here we analyzed matched primary and recurrent GBMs from 59 patients using single-nucleus RNA sequencing and bulk DNA sequencing, assessing the longitudinal evolution of the GBM ecosystem across layers of cellular and molecular heterogeneity. The most consistent change was a lower malignant cell fraction at recurrence and a reciprocal increase in glial and neuronal cell types in the tumor microenvironment (TME). The predominant malignant cell state differed between most matched pairs, but no states were exclusive or highly enriched in either time point, nor was there a consistent longitudinal trajectory across the cohort. Nevertheless, specific trajectories were enriched in subsets of patients. Changes in malignant state abundances mirrored changes in TME composition and baseline profiles, reflecting the co-evolution of the GBM ecosystem. Our study provides a blueprint of GBM’s diverse longitudinal trajectories and highlights the treatment and TME modifiers that shape them.

In isocitrate dehydrogenase wildtype glioblastoma (GBM), cellular heterogeneity across and within tumors may drive therapeutic resistance. Here we analyzed 121 primary and recurrent GBM samples from 59 patients using single-nucleus RNA sequencing and bulk tumor DNA sequencing to characterize GBM transcriptional heterogeneity. First, GBMs can be classified by their broad cellular composition, encompassing malignant and nonmalignant cell types. Second, in each cell type we describe the diversity of cellular states and their pathway activation, particularly an expanded set of malignant cell states, including glial progenitor cell-like, neuronal-like and cilia-like. Third, the remaining variation between GBMs highlights three baseline gene expression programs. These three layers of heterogeneity are interrelated and partially associated with specific genetic aberrations, thereby defining three stereotypic GBM ecosystems. This work provides an unparalleled view of the multilayered transcriptional architecture of GBM. How this architecture evolves during disease progression is addressed in the companion manuscript by Spitzer et al.