Brain-infused CAR-T cells showed early promise against recurrent glioblastoma

A small early-phase trial infused B7-H3-targeting CAR-T cells, called TX103, directly into the brains of people with recurrent glioblastoma and reported that the treatment was tolerated, with a median overall survival of 19.1 months and no dose-limiting toxicities [s1]. That is an encouraging signal, but a preliminary one: it comes from 15 patients in a single-arm phase 1 study with no control group, so it establishes safety and a hint of activity rather than proven benefit [s1] [s2].

Glioblastoma is the most common malignant brain tumour in adults, and when it recurs there are few effective options [s1]. CAR-T cells — a patient's own T cells re-engineered to recognise a tumour target — have transformed some blood cancers, but solid tumours, and brain tumours in particular, have been far harder ground. The results, published in Nature Medicine, are worth reading precisely because they are narrow.

What the trial did

The study was an open-label, 3-plus-3 dose-escalation trial — the classic design for finding a tolerable dose — in adults aged 18 to 75 whose recurrent glioblastoma expressed the target antigen B7-H3 on at least 30% of tumour cells [s1]. Rather than infusing the cells into a vein, clinicians delivered TX103 directly into the brain at three dose levels: 2 x 10^7, 6 x 10^7 and 1.5 x 10^8 cells per infusion [s1].

Fifteen patients received a total of 72 intracranial infusions, and 13 of them underwent repeated dosing [s1]. The trial is registered as NCT05241392 and was run by Beijing Tiantan Hospital [s2].

What happened on safety

The primary endpoints were safety, the maximum tolerated dose and the recommended phase 2 dose [s1]. On the first of those, the trial's central claim is that no dose-limiting toxicities occurred and no maximum tolerated dose was reached — meaning the cells did not force the escalation to stop [s1].

Adverse events attributed to the treatment were common but mostly low-grade. Low-grade cytokine release syndrome — the immune-activation reaction familiar from other CAR-T therapies — occurred in 86.7% of patients [s1]. Sinus tachycardia, vomiting and hypertension each affected 53.3%, and raised intracranial pressure affected 46.7% [s1]. Three grade 3 treatment-related events were logged as serious: raised intracranial pressure, epilepsy and depressed consciousness, two of them at the highest dose level [s1].

Those last events are the ones to sit with. Infusing an inflammatory cell therapy into a confined space — the skull — carries a specific hazard that a vein does not, and pressure and neurological events are exactly the signals that matter. The investigators report that repeated infusions produced no cumulative toxicity, and that immune activity was concentrated where it was wanted: cerebrospinal fluid showed a marked rise in CAR gene copies and cytokines, with minimal activity in the blood [s1].

What happened on survival

Survival and response were secondary endpoints, which is the correct weight to give them in a dose-finding study [s1]. The 12-month overall survival rate was 66.7%, and median overall survival was 19.1 months from the first infusion, with a 95% confidence interval running from 8.93 months to a value not yet reached [s1].

Disease control — stable disease or better — was achieved in 8 of 14 patients with measurable disease, including one complete response that was still holding at the latest follow-up [s1]. A single durable complete response in recurrent glioblastoma is a genuinely uncommon thing to be able to write down.

The number to hold in mind alongside all of this is 15. A median survival estimate from 15 people, in a study with no comparison group, is a description of what happened to those patients, not an effect size that can be attributed to the drug. Recurrent glioblastoma outcomes vary widely with age, molecular subtype and how much tumour was resected, and none of that can be balanced without randomisation.

What it does and does not establish

The authors' conclusion is appropriately bounded: intracranial TX103 showed acceptable safety and encouraging efficacy, supporting a future phase 2 trial at the recommended dose — dose level 2, or 6 x 10^7 cells per infusion [s1]. That is a green light to test the therapy properly, not a verdict that it works.

What the study establishes is that the cells can be delivered into the brain repeatedly without runaway toxicity, that they act locally, and that a signal worth chasing exists. What it cannot establish, by design, is whether TX103 lengthens survival compared with standard care, which patients benefit, or whether the effect seen here would survive contact with a randomised trial and a larger, more varied population.

What to watch

The next question is the phase 2 evaluation at the recommended dose, and whether it is designed with a comparator that can turn a 19.1-month figure into an actual comparison [s1]. B7-H3 is drawing wide interest as a solid-tumour target — it is also the antigen behind antibody-drug conjugates now in lung cancer, so read-across between programmes will accumulate. And the broader lesson for immunotherapy in hard-to-treat cancers holds here too: an early signal is a reason to run the real trial, not a substitute for it.

This article describes trial results. It is not advice about any treatment, and nothing here should be used to start, stop or change care.

Sources

• Intracranial delivery of B7-H3-targeting CAR-T cells for recurrent glioblastoma, Nature Medicine, 6 August 2026
• Safety and Efficacy Study of Anti-B7-H3 CAR-T Cell Therapy for Recurrent Glioblastoma (NCT05241392), ClinicalTrials.gov