Diffuse midline glioma is the kind of diagnosis that ends conversations. It grows into the brainstem, the piece of tissue that runs breathing, heartbeat and swallowing, and it is one of the few brain cancers that mostly strikes children. Surgeons cannot cut it out. Most children die within a year. Engineered immune cells, the technology that transformed blood cancers over the past decade, have repeatedly failed to dislodge it.
On October 28, UC San Francisco announced a result that may change the arithmetic. A team there built CAR-T cells that stay switched off until they arrive in the brain, then turn on and attack. In mice carrying human DMG tumors, the therapy cleared the cancer in every animal tested across two tumor models, and the cells were still patrolling, and still working, more than 70 days later. The paper appeared the same day in Science Translational Medicine, and a clinical trial at UCSF Health could begin as early as April 2027.
Why CAR-T keeps losing to brain tumors
A CAR-T cell is a patient’s own T cell, re-engineered in a lab to carry a chimeric antigen receptor, a synthetic sensor that recognizes a protein on tumor cells and tells the T cell to kill. Against certain leukemias and lymphomas the approach produces durable remissions. Against solid tumors, and brain tumors in particular, it has mostly disappointed.
Part of the problem is targeting. The proteins that mark a brain tumor also tend to appear, at lower levels, on healthy tissue elsewhere in the body. Build a receptor aggressive enough to clear the tumor and it can attack organs on the way there. Build one cautious enough to be safe and the tumor survives. As UCSF neurological surgery professor Hideho Okada, MD, PhD, a co-corresponding author on the paper, put it in the university’s announcement, “It’s been a struggle to make CAR-T cells that overcome DMG without causing unacceptable side-effects.”
The other problem is stamina. T cells that are switched on for too long become what immunologists call exhausted: still present, no longer effective. A cell that starts firing in the bloodstream may have little left by the time it reaches a tumor.
A two-step switch, and a better trigger
Okada and UCSF cellular and molecular pharmacology professor Wendell Lim, PhD, have spent years on a workaround: do not give the T cell one instruction, give it two. Their earlier design, E-SYNC, used a cellular sensor called synNotch to detect EGFRvIII, a mutant protein found on some glioblastomas. Detecting it did not trigger an attack directly. It triggered the cell to build a CAR aimed at EphA2 and IL13Rα2, two proteins common across tumor cells. Location first, kill second.
That worked better than a conventional single-step CAR, but it had a narrow door. Only about 20% of patients have tumors carrying EGFRvIII, and according to UCSF’s account, even those tumors can shed the protein as the therapy eliminates the cells that carry it. The trigger disappears mid-treatment.
So the team went looking for something more reliable, and chose a strange target: brevican, a protein found throughout the brain rather than specifically on cancer. The resulting design, B-SYNC, uses brevican as a geographic signal. The T cell asks, in effect, “am I in the brain?” If yes, it builds the tumor-killing CAR and starts hunting cells carrying EphA2 or IL13Rα2, whether or not those cells ever had EGFRvIII. Postdoctoral researcher Senthilnath Lakshmanachetty, PhD, in Okada’s lab, tested whether the glioblastoma-oriented design would also suit DMG, since both cancers display the same two target proteins.
What happened in the mice
Lakshmanachetty first gave B-SYNC cells intravenously to mice implanted with human DMG. Some crossed into the brain, produced the CAR there and attacked the tumor. Critically, B-SYNC cells recovered from the blood and spleen did not express the CAR at all, which is the whole point of the design and, in the researcher’s words, happened “just as we designed it.”
The team then delivered B-SYNC directly into the brain by intracerebroventricular injection. According to UCSF, that route cleared tumors in every mouse tested across two human DMG models, while conventional CAR-T cells again failed to control the cancer.
The durability test came next. Nearly seven weeks after treatment, the researchers implanted a second round of DMG cells in mice that had already been treated. Every B-SYNC-treated mouse rejected the new tumors. “We didn’t expect B-SYNC CAR-T cells to necessarily persist,” Lakshmanachetty said. “But they are still patrolling for cancer over 70 days out, and they’re still very functional.”
The researchers’ working explanation is that holding the cells inactive until they arrive may spare them from exhausting themselves en route. If that holds up, it is a general lesson about cell therapy design, not a quirk of this one tumor.
From bench to a trial, unusually fast
UCSF says the first trial will be led by Marie Jaeger-Krause, MD, and Sabine Mueller, MD, PhD, director of the UCSF Pediatric Brain Tumor Center, and will test both intracerebroventricular and intravenous delivery in different combinations. For the ICV arm, a shunt would connect the brain ventricles to a reservoir on the scalp, allowing repeat doses without repeat surgery. Each patient’s own T cells would be converted into the B-SYNC therapy.
“Mice with DMG are being cured, and we are incredibly excited to bring this therapy to the children and young adults we care for,” Mueller said in the announcement. The work was funded by the National Institutes of Health, ARPA-H, the Botha-Chan Project Fund, the Valhalla Foundation, the UCSF Cell Design Institute and the German Research Foundation.
What it means
Read this carefully. Mice are not children, and the history of oncology is thick with therapies that cured rodents and did nothing for patients. Nobody has yet given B-SYNC to a person. A trial starting in 2027 produces readable safety data years before it produces anything resembling a survival answer.
What is genuinely notable is the engineering idea. Most of the cell therapy field has pursued better targets: find the protein that appears on the tumor and nowhere else. B-SYNC accepts that such a protein may not exist and solves the problem with logic instead, gating an aggressive, broadly-reactive weapon behind a location check. That approach is portable. Any tumor sitting in a tissue with a distinctive resident protein is, in principle, a candidate.
For anyone tracking the cell therapy sector, the signal is that programmable, multi-input cell circuits are moving out of synthetic biology papers and toward the clinic, with the manufacturing and regulatory complexity that implies: two genetic modules per cell instead of one, and a trial design testing two delivery routes at once. For families facing a DMG diagnosis, the honest summary is that there is a trial being prepared, not a treatment available.