Artificial intelligence(AI)-based feature extraction from whole slide images (WSIs) shows significant promise for glioma prognosis. However, while foundation pathology models have potential, their features typically lack biological interpretability and clear links to disease mechanisms, limiting their clinical translational value. We propose a clustering Vision Transformer (ViT)–based framework for pathological image analysis. Tens of thousands of image patches extracted from each WSI were first clustered into 100 groups. The foundation-model features of the 100 cluster centroids, together with the proportional representation of each cluster across the whole slide, were then integrated and input into the ViT model for prognostic prediction. The image-derived risk scores were further integrated with matched RNA-seq data to identify risk-associated genes and biological pathways through differential expression and pathway enrichment analyses. Prognostic prediction was developed and evaluated in a total of 2552 cases, including 559 cases from Huashan Hospital for training and internal testing, 309 cases from Shanghai International Hospital as external validation cohort 1, and 1684 cases from TCGA as external validation cohort 2. Pathogenomic analyses were performed in 567 TCGA cases with matched WSI and RNA-seq data and further validated in 657 CGGA cases. Risk scores produced by the prognostic model showed a significant correlation with patient survival (log-rank P < 0.001). When integrated into existing risk stratification frameworks, these risk scores enhanced survival predictions via a nomogram based on clinical information and risk score compared to a traditional clinical-molecular nomogram, achieving a C-index of 0.90, superior calibration and classification accuracy (net reclassification improvement = 0.99, P < 0.001). Specific genes and pathways demonstrated significant associations with risk scores, and the mean expression levels of these key genes exhibited strong prognostic value across both the pathogenomics cohort and the CGGA dataset (log-rank P < 0.001). These results demonstrate that risk scores derived from pathological imaging can effectively identify key genes and biological pathways critical for stratifying glioma patients into high- and low-risk subgroups. Consequently, therapeutic strategies that modulate the tumor immune microenvironment may be particularly beneficial for high-risk patients identified by this approach.
Immunomodulatory therapies have emerged as promising treatments for cancer, where poly(I:C), a synthetic double-stranded RNA, is currently under clinical investigation for its ability to control myeloid cell function and T cell infiltration. While poly(I:C) mimics viral infection and activates Toll-like receptor 3 (TLR3), its exact mechanism of action in cancer is unknown. Our survey of multiple human tumors demonstrated that TLR3 is enriched in gliomas, particularly low-grade gliomas (LGGs; pilocytic astrocytoma, PA) in individuals with the neurofibromatosis type 1 (NF1) cancer predisposition syndrome. Using two independently generated murine models of NF1-LGG (optic pathway glioma, OPG), we evaluated the efficacy of poly(I:C) and examined tumor-associated myeloid cell (TAM) function, chemokine expression, and signaling pathway activation in vivo and in vitro. We found that poly(I:C) durably suppressed tumor growth in these two Nf1-OPG strains, with sustained glioma suppression and normalized retinal nerve fiber thickness 16 weeks following treatment cessation in vivo. Importantly, poly(I:C) induced TAM Ccl3 production through both TLR3 and MDA5-dependent pathways to control Ccl3-mediated intratumoral CD8+ T cell trafficking. Since we found that Ccl3 is enriched in microglia from both human PA and mouse Nf1-OPG TAM, we used fresh murine Nf1+/− microglia to demonstrate that these independent pathways activated distinct intracellular cascades in vitro, including TLR3/MEK and MDA5/NFκB signaling, which both converge on ERK activation. Collectively, these findings identify the innate immune mechanisms by which poly(I:C) regulates TAM-mediated T cell infiltration and LGG growth, highlighting the critical role of innate immune sensors in regulating non-neoplastic cells within the brain tumor microenvironment. The immunomodulatory therapy, poly(I:C), suppresses brain tumor growth by reducing tumor-associated monocyte T cell chemoattraction through two distinct pathways that converge on MEK/ERK signaling.
Rhabdoid tumors (RT) are highly aggressive malignancies that typically present in very young children. The most common anatomic sites are the central nervous system, (atypical teratoid/rhabdoid tumor (AT/RT)), kidney (malignant rhabdoid tumor or rhabdoid tumor of the kidney (RTK)) and extra-renal sites. The tumors are characterized predominantly by homozygous inactivation of the SMARCB1 tumor suppressor gene. Comprehensive epigenomic profiling using bisulfite conversion and array hybridization previously demonstrated distinct methylation subgroups, especially within AT/RTs, which have helped to define the origin and transcriptional programs in rhabdoid tumors from different sites. Integrating genome-wide methylation and high-resolution copy number analysis from samples with low DNA input remains challenging, which will ultimately be required in prospective studies with targeted treatment protocols that increasingly utilize cell-free DNA (cfDNA) from cerebrospinal fluid liquid biopsies to diagnose patients and monitor response. In this study, we applied enzymatic conversion followed by whole genome sequencing (EM-Seq) to a cohort of 24 central nervous system (CNS), 14 renal and 13 extrarenal RTs with and without germline SMARCB1 predisposing mutations or copy number alterations, as well as five cerebrospinal fluid (CSF) cfDNA samples from patients with AT/RT. Global methylation patterns confirmed the presence of five previously described subgroups (ATRT-TYR, ATRT-SHH, ATRT-MYC, RTK-like, and extra-renal-like). We leveraged EM-Seq to characterize genome-wide copy number alterations and copy-neutral loss of heterozygosity as well as exon-level copy number gains and losses in SMARCB1. We further demonstrate the feasibility of extending this approach to CSF cfDNA, highlighting the clinical potential for minimally invasive tumor classification and molecularly informed surveillance that could inform precision medicine for this aggressive entity. These findings position EM-Seq as a unified platform for integrated methylation-based classification and copy number profiling of RT, with applicability to both tissue and liquid biopsy samples.
Mitochondria are dynamic organelles essential for neuronal survival and synaptic function, and their dysfunction is a key consequence of excitotoxicity following traumatic brain injury (TBI). While intercellular mitochondrial transfer and exogenous mitochondrial transplantation have emerged as mechanisms to restore cellular bioenergetics, its in vivo relevance in the central nervous system remains incompletely understood. Here, we used astrocyte and neuron-specific mitochondrial reporters (GFP or Dendra2) in mice to assess cell-type-specific mitochondrial morphology, bioenergetics, and transfer 24 h after TBI. Neurons exhibited marked mitochondrial dysfunction, including altered morphology and reduced bioenergetic capacity across somatic, synaptic, and non-neuronal fractions. In contrast, astrocytic mitochondria showed morphological changes but preserved bioenergetic function. Concomitantly, astrocyte-to-neuron mitochondrial transfer was significantly increased following injury, although transfer to synapses remained limited. Single-cell RNA sequencing of astrocytes revealed upregulation of genes involved in extracellular vesicle (EV) biogenesis and mitochondrial translation following injury compared to controls. In vitro co-culture studies confirmed that astrocytes transfer mitochondria to neurons via EVs containing mitochondria (EV-mito). Isolated EV-mito from astrocyte-conditioned media improves neuronal mitochondrial function under NMDA (N-methyl-D-aspartate) induced excitotoxic conditions. Together, these findings demonstrate that TBI drives astrocyte-mediated mitochondrial transfer as an adaptive response, occurring alongside neuronal mitochondrial dysfunction. This process preserves neuronal bioenergetics in the soma and neurites but not at synapses, highlighting both its therapeutic potential and spatial limitations.
This study aimed to (1) develop a quantitative approach to automate detection of cortical CAA-positive vessels separate from β-amyloid plaques, (2) validate digital pathology classifier for CAA against manual scoring, and (3) characterize the spatial distribution of perivascular β-amyloid plaques in relation to CAA-positive vessels. Digital pathology classifiers were trained on the HALO AI platform to distinguish cortical CAA-positive vessels from β-amyloid plaques in the calcarine cortex. Classifiers were applied to digitized histopathologic slides (n=100), previously rated by expert neuropathologists. Slide level digital quantitative measures for CAA (CAA classified area and vascular β-amyloid percent positivity) and for β-amyloid plaques (β-amyloid plaque classified area) were compared with manual CAA scores and Alzheimer’s disease neuropathologic change (ADNC). Perivascular plaques, defined as β-amyloid plaques within a 0- to 150-micron concentric distance from the center of CAA-positive vessels, were identified. Participants had a mean age of 91.1 (SD=5.6) years at death, with 76
