N Nature Cancer · Dec 04, 2025 Retargeted oncolytic viruses engineered to remodel the tumor microenvironment for glioblastoma immunotherapy Glioblastoma (GBM) is an aggressive, immunotherapy-resistant brain tumor. Here, we engineered an oncolytic virus platform based on herpes simplex virus 1 for GBM viroimmunotherapy. We mutated the highly cytopathic MacIntyre strain to increase spread and oncolytic activity, limit genetic drift, prevent neuron infection and enable PET tracing. We incorporated microRNA target cassettes to attenuate replication in healthy brain cells. Moreover, we engineered the gD envelope protein to specifically target GBM using EGFR-specific or integrin-specific binders. Lastly, we incorporated five immunomodulators to remodel the tumor microenvironment (TME) by locally expressing IL-12, anti-PD1, a bispecific T cell engager, 15-hydroxyprostaglandin dehydrogenase and anti-TREM2 to target T cells and myeloid cells in the GBM TME. A single intratumoral injection increased survival in GBM preclinical models, while promoting tumor-specific T cell, natural killer cell and myeloid cell responses in the TME. In summary, we engineered a retargeted, safe and traceable oncolytic virus with strong cytotoxic and immunostimulatory activities for GBM immunotherapy. Cancer Cancer therapy Tumour immunology biology
N Nature Cancer · Nov 19, 2025 SMMILe enables accurate spatial quantification in digital pathology using multiple-instance learning Spatial quantification is a critical step in most computational pathology tasks, from guiding pathologists to areas of clinical interest to discovering tissue phenotypes behind novel biomarkers. To circumvent the need for manual annotations, modern computational pathology methods have favored multiple-instance learning approaches that can accurately predict whole-slide image labels, albeit at the expense of losing their spatial awareness. Here we prove mathematically that a model using instance-level aggregation could achieve superior spatial quantification without compromising on whole-slide image prediction performance. We then introduce a superpatch-based measurable multiple-instance learning method, SMMILe, and evaluate it across 6 cancer types, 3 highly diverse classification tasks and 8 datasets involving 3,850 whole-slide images. We benchmark SMMILe against nine existing methods using two different encoders—an ImageNet pretrained and a pathology-specific foundation model—and show that in all cases SMMILe matches or exceeds state-of-the-art whole-slide image classification performance while simultaneously achieving outstanding spatial quantification. Cancer Cancer imaging Computational biology and bioinformatics Tumour biomarkers Tumour heterogeneity Cancer Machine Learning Digital Pathology Multiple-Instance Learning
N Nature Cancer · Nov 14, 2025 Characterization of the tumor microbiome of brain metastases and glioblastoma reveals tumor-type-specific and location-specific microbial signatures Brain tumors, including glioblastoma multiforme (GBM) and brain metastases, present a notable clinical challenge. Recent research highlights the presence of intratumor bacteria across many tumor types, yet the microbiome of brain tumors remains largely underexplored. Here we show that the microbiome of 322 brain tumors differs markedly by tumor type and location. Using multiple approaches to visualize, culture and sequence bacterial communities, we found that brain metastases harbor higher bacterial richness and diversity than GBM, with distinct microbial compositions. Moreover, metastases in posterior brain regions exhibited greater diversity than those in anterior regions. Pathway analysis revealed enrichment of bacterial metabolic pathways associated with tumor spread and metastasis in brain metastases while GBM was enriched with pathways supporting alternative phosphorus use. These findings provide valuable insights into the microbial landscape of brain tumors, highlighting tumor-type-specific and location-specific variation and suggesting potential roles for bacteria in brain tumor biology. Cancer Cancer microenvironment CNS cancer Microbiome biology
N Nature Cancer · Nov 06, 2025 Spatiotemporal control of SMARCA5 by a MAPK–RUNX1 axis distinguishes mutant KRAS-driven pancreatic malignancy from tissue regeneration Acute pancreatitis-induced acinar-to-ductal metaplasia involves global chromatin remodeling and contributes to normal tissue regeneration. Oncogenic KRAS hijacks this process to promote PDAC formation. Here we show that regeneration and KRASG12D-driven oncogenesis can be decoupled from tissue regeneration through a chromatin remodeler, SMARCA5. We show that SMARCA5 maintains KRASG12D-dependent chromatin accessibility at regions specifically required for malignancy, without affecting chromatin opening that occurs during normal regeneration. Without SMARCA5, regeneration can be restored in the presence of KRASG12D. Mechanistically, regeneration-related or malignancy-related chromatin remodeling activities occur in a time-sensitive manner. The activity of SMARCA5 is controlled spatiotemporally by transcription factor RUNX1, which only accumulates at sufficient levels with sustained MAPK signals. We further show that inhibition of the SMARCA5-containing NoRC complex specifically inhibits the growth of PDAC organoid but not that of normal tissue derived from patients. Cancer Chromatin remodelling Pancreatic cancer biology mouse experiments
