Citations

Abstract BACKGROUND Intraoperative electrocorticography (ECoG) may aid localization of language function and help guide eloquent brain tumor resection. However, relevance of ECoG findings to language outcomes remains unclear. METHODS Fifteen patients with brain tumors (Mdn age=60; 67% high-grade glioma) underwent awake craniotomy and intraoperative language mapping. Time-frequency ECoG gamma activity was visualized in real-time followed by direct electrical stimulation (DES) across 3 paradigms (NeuroMapper Tasks: Object, Action, Auditory). Early and late ECoG gamma modulations and positive DES sites were co-registered with postoperative MRI. A subset (N=11) completed expressive language testing (Phonemic and Semantic Fluency, Confrontation Naming) prior to and <2 months following surgery. Correlational analyses and Mann-Whitney U tests examined relationships between test results and surgical impact upon expressive task-elicited gamma activity (Object and Action Naming). RESULTS Across tasks, positive DES sites were identified in 93% of cases (Auditory, 73%; Action, 71%, Object, 57%). No positive DES sites were resected. Early and/or late gamma activations were elicited in all patients, resected in 40%, and within 1 cm of surgical impact in 90%. Four patients (36%) showed reliable decline on at least 1 outcome, which was associated with Action Naming activity <1 cm of surgical impact (phi=.67, p=.035). On individual tests, Confrontation Naming and Phonemic Fluency decline were rare (9% and 18%), though Semantic Fluency declined in 36% of patients. Change in Semantic Fluency was associated with resection of Object Naming gamma sites (Z=-1.83, p=.044) and presence of Action Naming gamma features near surgical impact (Z=-2.25, p=.014). CONCLUSIONS Despite preservation DES sites and sparing of basic confrontation naming ability, verbal fluency declined in a substantial group of patients. Decline was associated with gamma features resected or near surgical impact. While preliminary, findings suggest task-based ECoG can augment DES in informing risk of surgically-acquired language decline.

Abstract High-grade gliomas (HGGs) are aggressive, infiltrative brain tumors with poor survival outcomes and limited therapeutic options. Their spatial and molecular heterogeneity necessitates large, well-curated datasets integrating imaging, clinical, and genomic information to enable precision medicine and drive novel treatment development. Using the Medtronic StealthStation neuronavigation system, we collected over 1,000 spatially localized intraoperative biopsy samples from more than 250 patients undergoing resection for glioma at Mayo Clinic and partnering sites. Each biopsy site was recorded with 3D coordinates, labeled in real time, flash-frozen, and annotated with metadata including collection time, anatomical location, and surgical plan. Samples were processed and entered into a secure database along with the corresponding sampling coordinates, frozen pathology, whole-exome and RNA sequencing data, 5-ALA fluorescence status, patient demographics, and T1, T1Gd, T2, and T2-FLAIR MRI. This multimodal dataset supports radiogenomic research by enabling precise mapping of imaging features to molecular characteristics. A subset of this data, including 268 biopsies from 52 patients (25 male, 27 female) with reported intraoperative 5-ALA fluorescence, was used to develop a radiomics model predicting locoregional patterns of 5-ALA positivity based on preoperative MRI. Image features were extracted from the biopsy sample coordinates across multiple MRI sequences and paired with corresponding 5-ALA fluorescence status to train the model for regional fluorescence prediction. The final model achieved 83% accuracy on the training set and 86% on the validation set using a 70-15-15 data split; this included 208 samples for training and 42 for validation, stratified by patient to ensure no overlap of samples from the same individual across sets. These findings demonstrate the feasibility of predicting 5-ALA fluorescence using radiomic features from standard preoperative MRI, offering a promising noninvasive strategy for improving intraoperative decision-making and enabling more precise, personalized resections in patients with high-grade gliomas.

Abstract Computer vision now enables deep learning models to link histopathologic features to molecular features in a quantitative manner not previously achievable by human pathologists. However, human interpretability is often limited. We trained a deep neural network to predict gene expression from digital whole slide images (WSIs) using 848 central nervous system (CNS) tumors with paired WSI and RNA-sequencing data. We used inferred RNA expression levels as a surrogate for protein expression of the CNS tumor lineage markers GFAP, OLIG2, and SSTR2, as well as the proliferation marker MKI67 (Ki67). We established thresholds for categorizing the inferred expression levels as positive or negative based on levels observed in cross-validation testing in select tumor types to approximate routine clinical interpretation of immunohistochemical (IHC) staining. We tested the sensitivity and specificity of our ‘virtual’ IHC on an independent, multi-institutional cohort of over 2,000 CNS tumors with objective diagnostic labels derived from DNA methylation-based tumor classification. As a dichotomous variable, inferred GFAP expression showed a sensitivity of 74% and a specificity of 99% as a glial marker in a cohort of gliomas and meningiomas. In the same cohort, OLIG2 showed a sensitivity of 75% and a specificity of 97% as a glial marker, while SSTR2 showed a sensitivity of 83% and a specificity of 97% as a marker of meningioma. In a cohort of ependymomas and gliomas, virtual IHC for OLIG2 showed a sensitivity of 75% and a specificity of 82% in distinguishing ependymomas from gliomas. Finally, in a cohort of gliomas, inferred expression of MKI67 demonstrated a trend consistent with what would be expected by actual IHC, with increasing MKI67 expression with increasing tumor grade. Our model provides the basis for a human interpretable and clinically applicable deep neural network to aid human pathologists in the diagnosis and grading of CNS tumors.

Abstract BACKGROUND Glioblastoma (GBM) is a molecularly and phenotypically heterogeneous tumor, with regional invasion patterns driving treatment resistance and universal recurrence. In infiltrative regions, GBM cells adopt at least two predominant invasive phenotypes: Neuronal-enriched (NEU), with synaptic and neurodevelopment programs, and Glycolytic/Plurimetabolic (GPM), with heightened immune and metabolic activity. These subtypes differ in therapeutic vulnerability and microenvironmental interactions; features not captured by conventional contrast-enhanced (T1+C) MRI due to minimal blood-brain barrier (BBB) disruption. Our group sought to develop predictive models to distinguish variability of NEU/GPM subtypes within infiltrative GBM, utilizing graph convolutional networks (GCN), advanced multiparametric MRI (mpMRI), and spatially-matched image-localized biopsies with transcriptomic profiling. METHODS We analyzed 18 mpMRI features —including conventional, diffusion-weighted, and Dynamic Susceptibility Contrast (DSC) perfusion MRI— of 101 image-localized biopsies from MRI non-enhancing infiltrative margins of 64 GBM patient tumors. Features were hierarchically clustered per modality; each cluster was represented as a node in a Neo4j graph. Edges connected nodes sharing overlapping samples. Nodes were annotated with NEU or GPM subtype proportion. GCNs were trained to classify subtype-enriched nodes using graph topology and node features. RESULTS GCNs achieved high predictive performance on held-out test data (AUC = 0.95 for NEU, 0.88 for GPM; 70:30 train-test split) with similar 5-fold cross-validation performance and generalized well to an independent cohort (AUC > 0.85, n = 45). Importantly, performance was robust without relying on T1+C imaging, enabling prediction beyond BBB-disrupted regions. GCNs outperformed traditional classifiers (logistic regression, random forest). CONCLUSION Our results demonstrate that GCN-modeling of mpMRI features enables non-invasive prediction of invasive GBM subtypes at the tumor margin. Unlike convolutional neural networks (CNN), which rely on spatial continuity, GCNs leverage relational structure between heterogeneous imaging features, making them ideal for modeling sparse, non-grid data. This may support non-invasive regional vulnerability detection, surgical targeting, and improved response assessment.

Abstract The 2021 WHO classification redefined glioblastoma IDH-wildtype (GBM), integrating histopathological and tumor molecular features. Diffuse astrocytic glioma lacking histologic features of necrosis and microvascular proliferation are classified as (molecular) GBM if the tumor is IDH- and H3-wildtype and has one or more additional tumor molecular alterations: TERT promoter mutation, EGFR amplification, or +7/-10. Due to the absence of histopathological features of classic GBM and presentation without enhancement on MRI in approximately 50% of cases, molecular GBM frequently resemble low-grade IDH-mutant gliomas. We hypothesized that MRI-based deep learning models could distinguish IDH-mutant glioma from molecular GBM. We retrospectively identified 97 individuals with molecular GBM, who were matched to 97 individuals with IDH-mutant glioma by age, sex, year of diagnosis, and enhancement on MRI. Of the 97 IDH-mutant gliomas, 51% were 1p/19q codeleted, 42% grade 2, 47% grade 3, and 10% grade 4. Preoperative postcontrast T1-weighted (T1C) and T2-weighted (T2) MRI sequences were used to train a 3D-DenseNet121 model. The model was validated on an independent prospective set of 150 IDH-mutant gliomas and 71 molecular GBM. The MRI-based deep learning model achieved an area under the receiver operator curve (AUC) of 0.75 (95% CI: 0.68-0.82) in the prospective validation set. Sex-stratified AUCs were 0.72 (95% CI: 0.59-0.85) for females and 0.76 (95% CI: 0.68-0.85) for males. Age stratified AUCs were 0.71 (95% CI: 0.60-0.83) for patients under 55 years of age and 0.64 (95% CI: 0.50-0.79) for patients older than 55. These results suggest that discriminative MRI features of IDH-mutant glioma and molecular GBM exist and can be identified from deep learning models. However, radiographic ambiguity exists and underscores the need for further optimization of the deep learning architecture or by integrating additional non-invasive modalities. For example, future work entails integrating germline variants into the model, including rs55705857, which is a causal variant for IDH-mutant glioma. With further validation and integration of multi-modal data, a noninvasive tool may support more personalized decision-making.

Abstract BACKGROUND High grade astrocytoma with piloid features (HGAP) is a newly recognized tumor with a distinct DNA methylation profile. Its clinical course, treatment response, and outcomes remain poorly understood. METHODS We conducted a retrospective cohort study across three Mayo Clinic sites including adults with a HGAP diagnosis confirmed by DNA methylation profiling. RESULTS 17 patients were included (median age 38 years, Q1, Q3: 29–59; 9 male). DNA methylation-based tumor classification using the NCI/Bethesda classifier (v2) indicated a match to HGAP with high confidence (calibrated score >0.9) in 14 cases and was suggestive in 3 cases. All had MAPK pathway alterations (NF-1 mutation in 11, FGFR1 mutation in 2, KIAA1549::BRAF fusion in 2, FGFR1 transcript variant suggestive of internal tandem duplication in 1, KANK2::NTRK2 fusion in 1). All tumors also had CDKN2A/B homozygous deletion and 9 lost ATRX expression. Tumors were infratentorial (n=8), supratentorial (n=7), and spinal (n=2) at initial presentation. First line treatment included surgery followed by chemoradiation with temozolomide (n=12), surgery followed by radiation (n=4), radiation alone (n=1), and surgery alone (n=1). Gross total resection was achieved in 3/17, 8/17 had subtotal resection and 5/17 had biopsy. 14/17 received a median RT dose of 5700 cGy (range, 4005–6000) in 30 fractions. The median follow-up duration was 34.9 months. 8 patients experienced disease progression. The median progression-free survival (mPFS) was 30.3 months (95% CI: 15.4–NR). 6/8 underwent surgery; 3/8 underwent RT. Systemic therapies were administered in 5/8 (lomustine, bevacizumab, pembrolizumab, temozolomide, trametinib, tovorafenib). 3/17 patients died due to progressive disease with an OS of 54.2, 94.6 and 17.4 months, respectively. The mOS for the whole cohort was 54.2 months (95% CI: 54.2-NR). CONCLUSION While the optimal management of this newly recognized entity has yet to be established, our cohort demonstrates a clinical behavior consistent with CNS WHO grade 3, aligning with current WHO interpretations.

Abstract Molecular testing has become critical components in the diagnostic classification of the central nervous system (CNS) tumors. However, these methods require substantial resources, limiting accessibility for many patients. Recent advances in artificial intelligence (AI) and computer vision empower deep learning models to infer molecular features from histopathology images, which may often be more readily available than molecular testing, to classify CNS tumors. We trained pan-CNS tumor models to predict DNA methylation and gene expression from whole-slide images (WSIs). These molecular predictions were then used in a hierarchical machine learning framework, termed Neuropath-AI, to predict nine broad tumor families and 52 tumor types using a large diverse cohort of 5,715 histopathology samples, of which 2,988 had paired DNA methylation profiling and 848 had paired RNA-sequencing. We evaluated Neuropath-AI on another large independent multi-institutional cohort of 5,516 CNS tumors, for which Neuropath-AI predicted tumor class with an associated confidence score. Neuropath-AI designated 46% of test samples as predictable with high-confidence, which it then accuracy classified in 97% of samples. It made moderate-confidence (and above) predictions for 87% of test samples, for which it achieved a diagnostic accuracy of 80% for the top-1 prediction and 86% accuracy for a top-2 predictions. A comparison with human neuropathologists showed that Neuropath-AI is comparable at classifying tumors. A second test showed that human neuropathologists were more accurate when the Neuropath-AI classifier results were provided to them, testifying to the potential translational benefit of integrating AI classification tools into the pathologist’s workflow. Our model provides the basis for a clinically applicable deep learning assistant to improve human efficiency and accuracy of CNS tumor diagnosis. The model will be made publicly available and can be readily implemented to assist human pathologists in clinical workflows, in further expanded prospective studies.

Abstract INTRODUCTION Glioblastoma (GBM) is a brain cancer known for its complex tumor microenvironment and ability to reprogram glucose metabolism and evade host immune responses. CAR-T cell therapy has attracted considerable attention due to its potential for high specificity and efficacy across various cancer types. However, its success in GBM has been limited due to the unique and hostile metabolic landscape of the tumor microenvironment. GBM tumor cells exhibit enhanced aerobic glycolysis, resulting in elevated glucose consumption. This restricts glucose availability for T cells, impairing activation and anti-tumor response. In our study, we enhanced the tumoricidal activity of CD70 CAR T-cells by overexpressing the glucose transporters GLUT1 and GLUT3, thereby improving their metabolic fitness and enabling them to overcome tumor-induced glucose deprivation. METHODS The CD70 CAR T-cell glucose metabolic pathway was modulated via glucose transporter overexpression. The efficacy and safety of metabolically modified CD70 CAR T-cells was investigated in in vitro and in vivo platforms. RESULTS GLUTs overexpression does not negatively affect T-cell differentiation and function but improves glucose consumption and intra-tumoral T-cell trafficking. We then used CD70 CAR-T platform and overexpress GLUT1 and GLUT3 to engineer T-cells. GLUT overexpression provided metabolic advantage to CD70CAR T-cells, which showed enhanced activation and cell avidity when co-cultured with GBM cells in both glucose enriched and restricted conditions. The metabolic modifications regulate CAR T-cell efficacy by enhancing intra-tumoral trafficking and persistence, followed by improved anti-tumor response and survival in mice. The lack of significant alterations in body weight and organ morphology in mice receiving GLUT-overexpressing CD70 CAR T-cells suggests that the adoptive transfer of these metabolically enhanced T cells does not elicit toxicity, thereby demonstrating a favorable safety profile during treatment. CONCLUSION Our study demonstrates development of immunotherapeutic strategies by incorporating core principles of tumor and immune cell metabolism, thus providing evidence that targeting immune-metabolic pathways is a potential approach to treat brain tumor.

Abstract BACKGROUND The Response Assessment in Neuro-Oncology (RANO) was recently updated to RANO 2.0, which no longer requires follow-up imaging to confirm progression in recurrent glioma patients recieving immunotherapy, as previously recommended by iRANO. This study evaluates the utility of RANO 2.0 to determine whether radiographic progression on initial post-treatment MRI is sufficient to guide timely discontinuation of ineffective therapy. METHODS This retrospective study, conducted at Mayo Clinic, included patients ≥18 years receiving ≥2 cycles of off-label pembrolizumab for recurrent glioma between 1/1/2014 – 5/1/2025. Radiographic response was assessed using the first MRI performed ≥30 days after pembrolizumab initiation. Progression was defined as a ≥25% increase in the product of perpendicular diameters of the lesion, clear progression of non-measurable disease, new lesion appearance, clinical deterioration, or death. For suspected pseudo-progression, sequential MRIs were reviewed; if progression was confirmed, the original date was used. RESULTS Sixty-two patients were included (mean age: 46.1 years; 64.5% male), receiving pembrolizumab for a median duration of 2.8 (1.4 – 19.4) months. Median time to initial radiographic response assessment was 1.6 (1.1 – 3.5) months, and progression-free survival was 2.3 (0.7 – 15.2) months. Thirty-seven (59.7%) patients were identified with progression at initial evaluation. Among them, 11 were initially suspected of pseudo-progression, and continued pembrolizumab: 3 received additional radiation, 2 initiated bevacizumab, while 6 remained on their prior regimens [pembrolizumab only (3); pembrolizumab with bevacizumab (3)]. However, all were subsequently confirmed to have true progression, with a median delay of 1.4 (0.7 – 7.1) months until pembrolizumab discontinuation. No true pseudo-progression cases were observed. CONCLUSIONS Our findings support the clinical applicability of RANO 2.0 in recurrent glioma patients receiving pembrolizumab. Initial radiographic progression reliably predicted true progression, suggesting limited value in confirmatory follow-up imaging. Early recognition of treatment failure may reduce delays in initiating salvage therapies, along with associated toxicity and cost.

Abstract We describe a 46-year-old male with Neurofibromatosis type 1 (NF1) and a previously known parenchymal medullary mass identified in 2022, initially managed with surveillance. In late 2023, he developed a progressive brainstem syndrome prompting presentation to our center. MRI showed a T2/FLAIR hyperintense, heterogeneously enhancing ventral medullary mass with leptomeningeal involvement, a non-enhancing right pontine lesion, right optic nerve glioma, and nodular leptomeningeal foci at T4 and L4–L5. He received IMRT to the brainstem prior to tissue diagnosis with good local response. Due to progression at T4, biopsy revealed a hypercellular, mitotically active glioma without necrosis or microvascular proliferation. Molecular profiling demonstrated IDH1/2-wildtype, H3K27M-negative, retained ATRX, intact 1p/19q, and CDKN2A/B homozygous deletion. DNA methylation profiling classified the tumor as high-grade astrocytoma with piloid features (HGAP). Spinal lesions responded to proton therapy, but new metastatic foci emerged in the subependymal lateral ventricles, leptomeninges, and cervical spinal cord. He underwent whole-brain radiation, oral temozolomide, and intrathecal cytarabine/etoposide. In June 2024, cell-free DNA analysis showed a similar molecular profile with newly identified 1p/19q codeletion, diverging from tissue findings. The patient remains alive >3 years from initial presentation. HGAP is a novel molecular class of IDH-wildtype gliomas defined by DNA methylation profiling and MAPK pathway alterations (commonly NF1, BRAF, FGFR1), CDKN2A/B codeletion, and ATRX alterations. These tumors often resemble glioblastoma or diffuse midline gliomas on imaging and are typically infratentorial or spinal, complicating biopsy. Prognostic data are limited, with <50% 5-year overall survival and no clear prognostic value of MGMT promoter methylation. This case illustrates multifocal HGAP in a patient with NF1 with favorable survival. The acquired 1p/19q codeletion raises questions about tumor mosaicism, subclonal evolution, and prognostic implications.