Engineering CAR T-Cells to Function Where Others Fail
The first patient has been dosed with GB-5267, an investigational armored CAR T-cell therapy for ovarian cancer designed in collaboration with Roswell Park Comprehensive Cancer Center. This milestone marks the first clinical test of a CAR T-cell therapy we designed by intentionally programming how T-cells behave in solid tumors—how they kill cancer cells, expand into a larger immune response, and persist long enough to finish the job.
Until now, CAR T development has been constrained by what could be tested. Most CAR binders were selected from a small set of affinity-matured antibodies, evaluated empirically, with limited ability to tune T-cell behavior itself. Our generative biology platform removed that constraint. By designing and evaluating CAR architectures at scale, we were able to optimize GB-5267 directly for T-cell function—engineering potency, proliferation, and persistence together, which along with armoring allow CAR T-cells to operate inside the hostile tumor microenvironment. That design shift is why we believe a function-first CAR T has a meaningfully better chance of succeeding in solid tumors, where prior approaches have fallen short. That ability to program T-cell function, rather than hope it emerges, is what makes this moment fundamentally different.
To discuss how this function-first, AI-enabled approach shaped GB-5267, and what it could mean for CAR T therapy in solid tumors, we spoke with Dinesh De Alwis, Senior Vice President and Head of Clinical Drug Development, Jason Yi, Director of Experimental Therapeutic Biology, and Victoria Szenes, Senior Director of Clinical Development.
Q: What makes CAR-T so much harder in solid tumors?
A: Dinesh de Alwis: There are several layers to the challenge. First, solid tumors create an environment that actively suppresses immune activity. Between inhibitory signaling and immune-suppressive cells within the tumor, CAR T-cells are often shut down before they can do meaningful work.
Second, there’s a physical barrier. Abnormal blood vessels and dense tumor tissue make it difficult for T-cells to even get into the tumor and reach cancer cells.
Third, finding the right target is surprisingly difficult. You need something that’s highly expressed on tumor cells but largely absent from normal tissue, and that also triggers a productive T-cell response when engaged. We have that in the membrane-bound form of MUC16 on tumor cells—the same protein clinically measured as CA125 but anchored on the cancer cell surface rather than circulating in the bloodstream. This form is highly expressed in roughly 80% or more of ovarian cancer patients and is a well-established biomarker in the disease.
And finally, even when CAR T-cells do infiltrate, they often don’t last. You may see initial expansion, but without sustained persistence, the response fades as T-cells become exhausted. These are the core obstacles we’ve been focused on overcoming with GB-5267.
Q: How did AI shape your approach to designing CAR T-cells, especially for solid tumors?
A: Jason Yi: Throughout my career, I’ve watched the CAR T field operate empirically, essentially throwing spaghetti at the wall to see what sticks. With CAR T, you couldn’t screen hundreds of thousands of constructs the way you can with other drug modalities. You were typically limited to a very small pool of candidates, given how few “plug-and-play” clinically validated antibodies could be converted into working CAR binders, and how uncertain it was that newly generated antibody clones would function once built into a CAR. You were selecting from that narrow set, hoping one would work in the clinic.
That’s where AI becomes transformative. It lets us move beyond a handful of limited constructs and scale our design space. We can test and program functions into each CAR T construct in ways that didn’t exist before. That programmability is the real innovation.
Q: In solid tumors, what combination of functional traits actually determines whether a CAR T-cell succeeds?
A: Jason Yi: In solid tumors, success really comes down to a combination of behaviors, not a single attribute. You need potency, the ability to kill tumor cells, but that alone isn’t sufficient. Those cells also need to proliferate, to build a large enough T-cell population, and they need to persist, or maintain immune activity long enough to keep working in an environment that’s actively trying to shut them down. Those three behaviors—potency, proliferation, and persistence—are inseparable in solid tumors. If one of them falls short, the therapy doesn’t hold up.
Q: Can you tell us more about the shift from designing for function rather than affinity?
A: Jason Yi: For a long time, CAR T design relied on antibody affinity as a proxy for performance. The assumption was that if a CAR bound its target tightly enough, effective T-cell activity would follow. However, what the field has learned — especially in solid tumors — is that binding affinity does not reliably predict CAR performance. You can engineer a CAR with very high affinity and still see poor outcomes because the T cells don’t expand sufficiently, or exhaust too quickly once they encounter the tumor environment.
What we did differently with GB-5267 was stop treating function as something you discover after the fact and start designing for it directly. Using our generative biology platform, we were able to explore CAR designs at a scale that wasn’t previously possible and evaluate them based on how T-cells actually behave, how long they stay active, how well they proliferate, and how consistently they kill in a hostile environment.
The result is a CAR T product that isn’t just potent initially but remains active and continues expanding in response to antigen over time. In our preclinical work, we’ve reached the limits of what we can detect in vitro and in animal models in terms of both potency and persistence. That’s why we feel confident that designing for function rather than affinity gives GB-5267 a meaningfully better chance of succeeding where prior solid-tumor CAR Ts have fallen short.
Q: What makes ovarian cancer the right place to start with solid tumor CAR-T when it’s struggled in other solid tumors?
A: Dinesh de Alwis: Ovarian cancer is not the right place to start because it is an easy solid tumor. It is a rational place to start because it gives us one of the clearest opportunities to test whether a next-generation CAR‑T can overcome the problems that limited earlier solid-tumor programs.
First, platinum-resistant ovarian cancer continues to carry a major unmet need. That creates a risk–benefit setting in which a carefully controlled CART study can be justified. In addition, ovarian cancer is commonly concentrated within the peritoneal cavity. That gives us a defined disease compartment in which both intravenous and potentially intraperitoneal administration can be evaluated. Tumor and ascites can often be sampled to determine whether the CAR‑T cells are reaching the disease, expanding, persisting, engaging MUC16 and remaining functional.
So, ovarian cancer does not avoid the challenges of solid-tumor CAR‑T. It is attractive because the unmet need, MUC16 target biology as mentioned earlier, peritoneal disease compartment and ability to interrogate trafficking and persistence all align—giving us a credible setting in which to determine whether the new AI optimized armored CAR‑T design has solved those challenges.
Q: What does patient safety look like for someone receiving first-in-human CAR T-cell therapy, and what are clinicians watching most closely after treatment?
A: Victoria Szenes: For patients, safety means knowing what to expect and having an experienced care team ready to recognize changes and respond quickly. Because this is a first-in-human study, patients are monitored very closely after treatment. Before treatment can escalate to a higher dose level, the Data and Safety Monitoring Committee reviews the available safety data. If a safety issue occurs, the team evaluates its severity, the treatment required, and the patient’s recovery.
Among the side effects the clinicians monitor most closely are cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS). CRS may include fever, low blood pressure, and low oxygen levels, while ICANS may cause confusion, difficulty speaking, reduced alertness, or other neurologic changes. These complications require prompt recognition and can often be treated with medications and supportive care. The care team checks vital signs, neurologic status, laboratory results, organ function, and any new symptoms or side effects. Since this is a first-in-human therapy, the team also remains alert for unexpected effects. Patients and caregivers are asked to report new symptoms promptly rather than waiting to see whether they resolve.
The goal is to identify and address safety issues early, determine whether each dose is tolerable, and learn from every patient’s experience. That close partnership between patients and the clinical team is central to conducting a first-in-human study.
Q: What are you hoping to get out of the Phase 1 study of GB-5267?
A: Dinesh de Alwis: At its core, the Phase 1 study is about making sure we are doing right by patients. Our priority is to establish the safety profile of GB-5267 and to identify a dose that can be given reliably and assess any anti-tumor activity. As with any first-in-human CAR T-cell therapy, careful dose escalation is essential, not only to manage immune activation appropriately, but to ensure patients are treated in settings with deep experience in CAR T delivery, manufacturing, and the management of cytokine-related effects.
Beyond safety, this study is also about learning whether GB-5267 behaves in patients the way it was designed to. We will be closely tracking whether our CAR T-cells demonstrate intended potency, proliferation, and persistence. For patients with ovarian cancer who have limited options, these early insights matter. Phase 1 is therefore not just a procedural first step—it is a meaningful opportunity to evaluate whether a function-first CAR T-cell approach can begin to address the longstanding challenges of solid tumors, and to guide how we move forward with intention.
Q: Can you tell us more about the MUC16 target of GB-5267 and why it is exciting for ovarian cancer?
A: Dinesh de Alwis: MUC16 is a strong starting point for ovarian cancer because it is highly prevalent and consistently expressed on tumor cells in the majority of patients, while remaining limited on normal tissue. Clinically, it is a well-established biomarker, which means the disease biology is well understood and the unmet need is clear.
What makes MUC16 particularly compelling for GB-5267 is the membrane-bound form of the protein on tumor cells. By designing the CAR to recognize that membrane-associated form, we can focus T-cell activity where it matters—on the tumor itself—rather than being diverted by circulating antigen. Importantly, MUC16 has shown signs of immune responsiveness in prior approaches, but those therapies lacked the durability and functional strength needed for meaningful impact. GB-5267 pairs a validated tumor target with a CAR T-cell engineered to function effectively in solid tumors, which is where we believe the real opportunity lies.