bims-malgli Biomed News
on Biology of malignant gliomas
Issue of 2026–11–22
two papers selected by
Oltea Sampetrean, Keio University



  1. Commun Biol. 2026 Jul 27. pii: 1530. [Epub ahead of print]9(1):
      Metabolic rewiring fuels tumorigenicity of cancer stem cells contributing to universal glioblastoma (GBM) recurrence. Invasive glioblastoma stem cells (GSCs), already unresectable, acquire tumorigenic properties that confer therapy resistance. Using samples from both recurrent and newly diagnosed GBM patients, we identified fatty acid oxidation (FAO) as a key pathway supporting GSC tumorigenicity. Transcriptomic profiling, compared against non-malignant human astrocytes from epilepsy patients, highlighted an FAO-enriched signature in GSCs. In vivo studies showed that tumor formation by these GSCs could be modulated by altering the availability of metabolizable lipids in the brain, with beneficial effects observed under a low carbohydrate high fat diet (LCHFD). The lipid metabolism contributing to GSC mediated tumor formation was confirmed using metabolomic and lipidomic analyses. Gut microbiome analyses revealed that LCHFD shapes the microbiome and modulates the circadian network. By targeting FAO in the postoperative period before chemoradiation, LCHFD offers a non-pharmacologic strategy to delay GBM recurrence. This study suggests a mechanistic framework for dietary modulation as a therapeutic approach to improve GBM outcomes.
    DOI:  https://doi.org/10.1038/s42003-026-10516-1
  2. Neuro Oncol. 2026 Nov 14. pii: noag286. [Epub ahead of print]
      Artificial intelligence (AI) has become central to neuro-oncological imaging, impacting clinical and scientific workflows. In clinical practice, AI-based image reconstruction and noise reduction expedite the generation of advanced multiparametric MRI and PET scans. Automated segmentation and volumetric analysis improve reproducibility, speed, and standardization of tumor quantification and response assessment. In research, large language and multimodal models extract structured information from radiology reports and medical records. Collaborative initiatives, such as BraTS and FeTS, have established standardized, privacy-compliant frameworks for large-scale quantitative analyses. Tumor-specific applications include noninvasive molecular profiling in gliomas, grading and recurrence prediction in meningiomas, detection and response assessment in brain metastases, and pediatric-specific modeling aligned with RAPNO. However, AI models remain sensitive to technical and biological variations across institutions, limiting their clinical utility unless appropriate training and validation is performed. Nevertheless, progress is evident in prospective, biopsy-validated infiltration mapping, assessment of resection extent aligned with RANO-Resect, and standardization through AI-RANO. Emerging directions include foundation models, uncertainty estimation, federated learning, AI-enhanced ultra-low-field MRI, and dual-tracer PET. When used as a decision-support tool rather than a replacement for the imaging team, AI can increasingly improve the utility, reproducibility, and efficiency of quantitative, multiparametric imaging in patients with brain tumors.
    Keywords:  AI-RANO; PET/MRI; clinical translation; federated learning; foundation models
    DOI:  https://doi.org/10.1093/neuonc/noag286