The central obstacle in solid tumor CAR T therapy has always been target selection. The therapy works by collecting a patient’s own T cells, genetically engineering them to recognize a specific protein on the surface of cancer cells, and reinfusing them so they hunt down and destroy anything bearing that molecular signature. In blood cancers, this is relatively straightforward because malignant B cells display molecules such as CD19 that are dispensable elsewhere in the body. Solid tumors are different. Most of their defining abnormalities hide inside the cell, driven by mutated or fused genes operating in the nucleus, while the proteins displayed on the cell surface tend to be shared with healthy tissues. Attacking them risks catastrophic off-tumor toxicity, and tumors that do express a target often do so unevenly, allowing antigen-negative cells to survive and seed relapse.

The research team approached this problem from a different angle: instead of searching for surface proteins common to broad cancer types, they looked for surface proteins that are directly commanded into existence by the specific gene fusions that drive certain rare cancers. Alveolar soft-part sarcoma, or ASPS, is a striking example. This aggressive sarcoma, which disproportionately strikes adolescents and young adults, is caused by a chromosome rearrangement that fuses the TFE3 transcription factor gene to the ASPSCR1 gene. Translocation renal cell carcinoma, similarly, arises from fusions involving MiT/TFE-family transcription factors such as TFE3 or TFEB. Because these fusion proteins are aberrant transcription factors, they rewire the cell’s gene expression program wholesale, and the investigators reasoned that this rewiring might force cancer cells to display unique combinations of surface molecules.

Using gene expression profiling across primary and relapsed tumor samples, the team identified glycoprotein NMB, or GPNMB, as a molecule that fits the bill. GPNMB is a transmembrane glycoprotein that in these fusion-driven cancers is expressed at high levels, with striking uniformity across tumor cells and remarkable stability over time. In ASPS and translocation renal cell carcinoma samples, including tumors that had relapsed after prior therapies, essentially every malignant cell carried GPNMB on its surface. This homogeneity is exactly what a CAR T target needs, because it denies tumor cells the escape route of simply switching the target off. Equally important, GPNMB expression in normal tissues is low and restricted, raising the prospect of a workable therapeutic window.

With the target validated, the researchers engineered a CAR T cell product they named GCAR1. The construct couples an antibody-derived recognition domain that binds GPNMB to intracellular signaling modules that activate the T cell upon contact, triggering killing of the target cell and proliferation of the engineered population. In the laboratory, GCAR1 cells showed potent, selective cytotoxicity against patient-derived tumor cells, lysing GPNMB-positive cancer cells while sparing matched normal cells that lacked the protein. The activity extended beyond flat cultures of cells into three-dimensional patient-derived organoids, which better recapitulate the architecture and drug resistance of real tumors, and into xenograft models in which human tumors were implanted in immunodeficient mice. In those animal models, GCAR1 infusions produced marked tumor control, establishing preclinical proof that the approach could work in living tissue.

The pivotal step came with the launch of a first-in-human, open-label, individual-participant clinical trial, registered as NCT07104682, designed to test GCAR1 in patients with relapsed or refractory fusion-driven solid tumors. The study reported here includes an interim analysis of a participant with metastatic ASPS whose disease had progressed despite standard treatment. Following lymphodepleting chemotherapy to clear space for the engineered cells, the patient received a single infusion of GCAR1. The clinical response, while not a complete remission, was notable: imaging showed stable disease sustained for up to three months, and, strikingly, many of the patient’s non-target lesions, smaller metastatic deposits not formally measured as primary endpoints, resolved entirely on follow-up scans. In a cancer as relentless as ASPS, with few effective systemic options and a median survival historically measured in a few years from diagnosis, even disease stabilization with lesion regression represents a meaningful clinical signal.

Just as important as the efficacy signal was the safety profile. GCAR1 was well tolerated, without the severe cytokine release syndrome, neurotoxicity, or on-target off-tumor organ damage that has plagued some solid tumor CAR programs. The engineered cells were detectable in the patient’s peripheral blood for about a month after infusion, expanding as a polyclonal population, meaning that multiple distinct T cell clones carrying the receptor expanded in parallel rather than a single clone dominating. Polyclonal persistence is generally viewed favorably, as it suggests a robust, diverse immune response less vulnerable to outgrowth of tumor variants that could evade any single clone. The one-month persistence window is also consistent with a controlled, self-limited therapy, which may explain the clean toxicity profile even against a target like GPNMB that has low-level normal tissue expression.

The study did not stop at the celebration of a response, however. One lesion in the patient proved treatment-resistant, and the team subjected it to an advanced molecular interrogation known as spatial transcriptomics, a technique that maps which genes are active at precise locations within intact tissue. The analysis revealed that the resistant lesion harbored immunosuppressive niches, microanatomical pockets enriched for pathways and cell types that suppress T cell function, effectively creating local sanctuaries where the CAR T cells could not operate even when they reached the tumor. This finding transforms an apparent failure into a roadmap: resistance, in this case, was not about loss of the GPNMB target but about the tumor microenvironment building walls around the attacking cells.

That mechanistic insight pointed directly at a rational combination strategy. Immune checkpoint blockade, the class of drugs that includes antibodies against molecules such as PD-1 and its ligand PD-L1, works by releasing molecular brakes that tumors place on T cells. The researchers tested whether combining checkpoint blockade with GCAR1 would overcome the immunosuppressive niches, and in a xenograft model the combination produced synergy, with the two modalities together controlling tumors more effectively than either alone. For a translational program, this is a crucial result, because checkpoint inhibitors are already approved, widely available, and clinically familiar. A future trial testing GCAR1 alongside checkpoint blockade is an obvious and achievable next step, and the preclinical synergy data provide the justification.

The broader conceptual contribution of the study may ultimately matter more than any single clinical result. The MiT/TFE-family fusion proteins that drive ASPS and translocation renal cell carcinoma are master regulators, and the demonstration that their activity can be exploited through a surface readout like GPNMB establishes a general paradigm: oncogenic gene fusions, though intracellular and classically considered undruggable, can be converted into actionable surface targets by mapping the transcriptional programs they impose. The same strategy could in principle be extended to other fusion-driven cancers, a category that includes many pediatric sarcomas, leukemias, and carcinomas for which targeted drugs remain elusive. Rather than trying to inhibit an undruggable fusion protein directly, clinicians could train a patient’s immune system to recognize the distinctive surface signature that the fusion creates.

Challenges remain before GCAR1 or its successors become standard care. The clinical experience so far involves a single participant in an interim analysis, and larger cohorts will be needed to confirm response rates, define the optimal dosing, and fully characterize toxicities. The three-month duration of disease control, while encouraging, will need to extend into durable remissions, likely through combinations with checkpoint inhibitors or other microenvironment-modulating agents suggested by the spatial transcriptomics findings. Questions about whether resistance can emerge through GPNMB loss in other patients, and whether GPNMB expression levels in normal tissues vary enough between individuals to cause occasional toxicity, will require larger datasets. Nevertheless, the trajectory from target discovery through organoid and xenograft validation to a controlled, tolerated, clinically active infusion in a patient with one of oncology’s most feared sarcomas has been completed in a single study, a bench-to-bedside arc that few experimental therapies achieve so cleanly. For patients with ASPS, translocation renal cell carcinoma, and other fusion-driven solid tumors, the message is that the wall separating CAR T success in blood cancers from success in solid tumors is no longer impenetrable, and the first engineered cells are already through it.