The study, published in the British Journal of Cancer, analyzed tumor specimens from 56 patients with newly diagnosed IDH-wildtype glioblastoma, the most common and most lethal molecular subtype of the disease. Rather than assuming all tumors look alike immunologically, the researchers first screened their cohort to identify cases with prominent intratumoral T-cell infiltration, a feature that only a fraction of glioblastomas display. In those T-cell-rich cases, they isolated CD3-positive, CD8-positive, CD4-negative cytotoxic T cells, the immune cells theoretically capable of killing tumor cells directly, and subjected them to single-cell RNA sequencing paired with T-cell receptor sequencing, a combination that reveals both what each cell is doing and what antigen specificity its receptor encodes.

The single-cell analysis produced a striking and consistent picture. Unsupervised clustering of the transcriptional data identified a dominant population of tumor-infiltrating lymphocytes defined by strong expression of exhaustion markers, with the checkpoint molecule LAG3 standing out as the most prominent. Exhaustion is a dysfunctional state that T cells enter after chronic antigen stimulation, in which they retain tumor recognition but lose much of their killing capacity. Critically, these exhausted cells also showed high T-cell receptor clonality, meaning the same receptor sequences appeared repeatedly across many cells. Clonal expansion of this kind is a classic hallmark of antigen-driven proliferation: it indicates that a specific stimulus, very plausibly tumor antigen, drove these cells to multiply within the tumor microenvironment.

To test whether this pattern was unique to their cohort or a general feature of glioblastoma, the team integrated their dataset with publicly available single-cell data from glioblastoma tumor-infiltrating lymphocytes and, as a comparative benchmark, with datasets from lung cancer TILs and malignant pleural effusions. The comparison revealed meaningful differences. Levels of LAG3 and of CXCL13, a chemokine that has emerged in recent years as a marker of tumor-reactive T cells in multiple cancer types, were significantly higher in the T-cell-rich glioblastoma subset from the Aichi Cancer Center cohort than in the public glioblastoma dataset. This suggests that T-cell infiltration in glioblastoma is not a uniform phenomenon but varies substantially between patients, and that the most infiltrated tumors contain a particularly pronounced exhausted, potentially tumor-specific compartment.

The most consequential step came next. The researchers applied an artificial intelligence-based prediction algorithm, developed with collaborators at NEC OncoImmunity, that estimates which T-cell receptors are likely to recognize tumor antigens based on sequence features and transcriptional context. When the algorithm was run across the datasets, it predicted that the T-cell-rich glioblastoma subset harbored a markedly higher proportion of potentially reactive T cells than either the public glioblastoma dataset or the lung cancer comparator samples. In other words, the exhausted, clonally expanded cells dominating these tumors are not merely bystanders or nonspecific inflammatory recruits; a substantial fraction of them may be genuinely tumor-reactive, held in check by their exhausted state.

These findings carry two major therapeutic implications. The first concerns LAG3, which has become one of the most closely watched immune checkpoint targets after PD-1 and CTLA-4. LAG3-blocking antibodies are already in late-stage clinical development for other cancers, and the new data provide a mechanistic rationale for testing LAG3-targeted approaches in glioblastoma, particularly in patients whose tumors show dense T-cell infiltration. If the potentially reactive T cells in these tumors are suppressed largely through LAG3-mediated inhibition, releasing that brake could reactivate a T-cell army that is already present and antigen-primed, a fundamentally different situation from trying to prime a response that does not exist.

The second implication involves antigen-specific immunotherapy, including personalized neoantigen vaccines and adoptive cell transfer. Previous work, such as the landmark neoantigen vaccine trial in glioblastoma led by researchers at Dana-Farber Cancer Institute, demonstrated that vaccine-induced T cells can traffic into these brain tumors, and other studies have shown that microbial peptides and aberrantly spliced tumor RNAs can generate antigens recognized by infiltrating lymphocytes. The new study adds a crucial piece to this puzzle by showing that in a subset of patients, the raw material for such approaches, clonally expanded, potentially tumor-reactive CD8 T cells, is already abundant. Identifying which patients fall into this T-cell-rich category could allow clinicians to select those most likely to benefit from vaccines, engineered T-cell therapies, or checkpoint blockade, moving glioblastoma immunotherapy toward a biomarker-driven strategy rather than a one-size-fits-all failure.

The study also refines the scientific understanding of T-cell exhaustion in brain tumors. Earlier single-cell analyses had described glioblastoma-infiltrating CD8 T cells as severely exhausted and had identified inhibitory receptors such as CD161 on glioma-infiltrating T cells, while other work characterized the dominant population as clonally expanded GZMK-positive effector cells. The new data reconcile these views by showing that exhaustion signatures in glioblastoma are not homogeneous: within the exhausted compartment, the clonally expanded, LAG3-high, CXCL13-positive subset is the one most likely to contain tumor-reactive cells. Distinguishing this potentially reactive exhausted state from generic dysfunction could guide the design of combination therapies that reinvigorate the right cells rather than broadly activating all T cells in the brain, where uncontrolled inflammation carries serious risks.

Important caveats remain. The prediction algorithm identifies candidate reactive T-cell receptors, but reactivity must ultimately be confirmed experimentally by identifying the antigens they recognize, and the proportion of truly tumor-specific cells among the predicted candidates is unknown. The T-cell-rich phenotype also applies to only a subset of the 56 patients studied, so the findings may not generalize to all glioblastomas. Nevertheless, the study provides a concrete foundation for two concrete therapeutic directions: antigen-specific approaches built on the T-cell receptors found in these tumors, and LAG-3-targeted immunotherapy aimed at releasing the brakes on cells that appear primed and waiting. For a disease whose median survival has barely moved in decades, the discovery that some glioblastomas already contain a recognizable, targetable immune response is a reason for cautious optimism, and a roadmap for the trials that must now follow.

Subject of Research: CD8+ tumor-infiltrating lymphocyte exhaustion signatures and potential tumor reactivity in glioblastoma

Article Title: Abundant CD8+ tumor-infiltrating lymphocytes in glioblastoma harbor distinct exhaustion signatures in potentially reactive subsets

Article References: Okamoto, T., Mizuta, R., Demachi-Okamura, A., Muraoka, D., Fukushima, Y., Ishihara, H., Sun, Y., Wang, Y., Nishida, R., Sasaki, E., Masago, K., Onoguchi, K., Tanaka, Y., Yamashita, Y., Stratford, R., Clancy, T., Muto, O., Tan, C. L., Otani, Y., … Matsushita, H. (2026). Abundant CD8+ tumor-infiltrating lymphocytes in glioblastoma harbor distinct exhaustion signatures in potentially reactive subsets. British Journal of Cancer. https://doi.org/10.1038/s41416-026-03619-3

Image Credits: AI Generated

DOI: 10.1038/s41416-026-03619-3

Keywords: glioblastoma, CD8 T cells, tumor-infiltrating lymphocytes, T-cell exhaustion, LAG3, CXCL13, single-cell RNA sequencing, T-cell receptor, immunotherapy, neoantigen, checkpoint inhibitors, brain cancer