Delivering CAR-T Into the Brain: Phase 1 Trial in Recurrent Glioblastoma Shows Early Efficacy Signal

Once glioblastoma recurs, treatment options are limited. Although CAR-T cells have transformed the treatment of some blood cancers, the blood–brain barrier, variability in tumor antigens, and the immune environment within the brain have made it difficult to reproduce the same results in solid brain tumors. A phase 1 trial published in *Nature Medicine* instead used local infusion, delivering CAR-T cells called TX103 directly into the brain and obtaining preliminary signals of safety and disease control in 15 patients with recurrent disease.

This open-label, dose-escalation study enrolled adults aged 18 to 75 whose tumors expressed B7-H3 in at least 30% of cells. B7-H3, also known as CD276, is an immunoregulatory protein found on the surface of multiple types of tumors. Researchers collected and engineered the patients’ own T cells to recognize B7-H3, then injected the cells into the surgical resection cavity or a ventricle through an Ommaya reservoir, thereby bypassing the limited efficiency with which systemic infusions reach the brain.

The 15 patients received a total of 72 infusions, with 13 receiving repeat treatment. The per-infusion doses were 20 million, 60 million, and 150 million cells. The trial observed no dose-limiting toxicities and did not identify a maximum tolerated dose; however, “tolerable” does not mean risk-free. Low-grade cytokine release syndrome occurred in 86.7% of patients, while tachycardia, vomiting, and hypertension each occurred in 53.3%, and increased intracranial pressure occurred in 46.7%.

The study recorded three grade 3 serious adverse events considered treatment-related: increased intracranial pressure, seizure, and decreased consciousness. Two occurred in the highest-dose group. The team ultimately selected the intermediate dose of 60 million cells per infusion as the recommended phase 2 dose, indicating that increasing the cell count may not necessarily produce a more favorable risk–benefit balance.

Regarding efficacy, the overall survival rate 12 months after the first infusion was 66.7%, and median overall survival was 19.1 months; the lower bound of its 95% confidence interval was 8.93 months, while the upper bound had not been reached at the time of analysis. Among 14 patients with measurable lesions, eight achieved stable disease or a better response, including one whose complete response was maintained through the most recent follow-up. These figures provide a basis for subsequent trials but cannot prove that TX103 prolonged survival because the study had no randomization or comparator treatment.

Cerebrospinal fluid analysis added a layer of biological evidence for the local-delivery strategy: after infusion, CAR transgene copy numbers and multiple cytokines increased markedly in the cerebrospinal fluid, while activity in peripheral blood was relatively limited; no cumulative toxicity was observed after repeat infusions. This indicates that the cells were indeed active in the intended region, but biological activity and stable, reproducible clinical benefit are still two different things.

The next key step is to determine in a larger phase 2 study with an appropriate comparison benchmark whether the survival and tumor responses can be reproduced, while clarifying which patients are most likely to benefit. At this stage, the results remain limited by patient selection, the small sample size, and the complexity of imaging interpretation. The study was also supported by Tcelltech, a company involved in the development of TX103, and some authors were its employees. The funder also participated in study design, data analysis, and manuscript preparation, making independent validation particularly important.