DDDR-63. Immune remodeling in CSF and tumor predicts response to CAR T therapy in glioblastoma

Abstract Glioblastoma (GBM) is the most common and lethal primary malignant brain tumor in adults. We previously reported results from a phase I clinical trial demonstrating that intracerebroventricular delivery of bivalent CAR T cells, targeting EGFR (epitope 806) and IL13Rα2, led to tumor size reductions in patients with recurrent, multifocal GBM. Despite these promising results, most patients experienced tumor relapse following initial regression. To uncover mechanisms underlying response and resistance, we performed single-cell analysis of longitudinal cerebrospinal fluid (CSF) collected at Days 0, 7, and 21 from eight patients, along with matched pre- and post-treatment tumor tissue (n = 6, three patients) and infusion products (n = 4). Immune dynamics suggest that treatment efficacy is not solely driven by CAR T cells but may equally depend on recruitment and activation of the endogenous immune system. Post-infusion, CSF-circulating CAR T cells showed increased cytotoxicity and exhaustion-related programs, suggesting target engagement. Notably, CAR T infusion drove rapid expansion and activation of endogenous immune cells, including natural killer (NK) and regulatory T cells (Tregs). NK cells increased in proportion and transitioned toward a CD56dimCD16+ cytotoxic phenotype. This NK activation was significantly associated with prolonged progression-free survival. Conversely, Tregs expanded clonally in both CSF and tumor tissue and showed increased immunosuppressive interactions, including CTLA4 and Galectin-9-based signaling, which were enriched in non-responders. Treg expansion and reduced clonal entropy negatively correlated with radiographic response. Post-treatment tumor tissues showed decreased EGFR expression and reduced immunosuppressive myeloid phenotypes, suggesting a local impact of the treatment. Rather than functioning in isolation, CAR T cells act as immune catalysts – reshaping the CSF milieu by activating cytotoxic NK cells while simultaneously triggering Treg expansion. These opposing dynamics reveal actionable points of intervention: disrupting early Treg–myeloid signaling or pre-arming with NK-supportive cues could shift the balance toward more durable tumor control.