AQP3 NOT gating mitigates lethal on-target off-tumor pulmonary toxicity of MSLN-directed CAR-T cells in ovarian cancer

Background: Mesothelin (MSLN) is overexpressed across multiple solid tumors and represents a promising therapeutic target. However, MSLN-directed chimeric antigen receptor (CAR)-T cells have been associated with lethal pulmonary on-target off-tumor toxicity. Here, we implemented a “NOT” logic-gating strategy to engineer dual-input MSLN/AQP3 lung-Specific Attenuated Functionally Engineered (lung-SAFE) CAR-T cells that retained potent antitumor activity against ovarian cancer while preventing fatal pulmonary toxicity. Methods: A phase I trial (NCT05141253) was ongoing to evaluate the safety of MSLN-directed CAR-T cells in patients with solid tumors. Human MSLN (hMSLN) knock-in NCG mice bearing SKOV3 xenografts received intravenous MSLN-CAR-T cells and were longitudinally monitored for signs of toxicity. Major organs and peripheral blood were collected at moribund stages. MSLN expression, CAR-T cell infiltration, and lung injury were analyzed. Integrated analyses of in-house single-cell ovarian cancer datasets and publicly available single-cell lung datasets were performed to identify candidate inhibitory targets, which were subsequently validated in vitro and in vivo. CAR-T cell activation and cytotoxicity were assessed using flow cytometry, cell lysis assays, and cytokine quantification in co-culture supernatants. Results: In the phase I trial, two patients with ovarian cancer experienced fatal pulmonary toxicity following MSLN-CAR-T cell infusion. To elucidate the underlying mechanism, we generated hMSLN knock-in NCG mice, which faithfully recapitulated the lethal pulmonary injury observed in patients. This toxicity was mediated by MSLN expression on alveolar cells and CAR-T cell-driven immunopathology. To overcome this limitation, we integrated single-cell transcriptomic analyses of ovarian cancer and healthy lung tissues to identify molecules selectively enriched in alveolar cells but minimally expressed in ovarian cancer cells. Cross-validation in cell lines, organoids, and human tissue specimens identified AQP3 as the most selective target. We then engineered a dual-input MSLN/AQP3 lung-SAFE CAR-T cells, wherein AQP3 delivered an inhibitory signal to suppress MSLN-CAR-T cell activation in the lung. The resulting lung-SAFE CAR-T cells maintained potent cytotoxicity against MSLN+ ovarian cancer cells while sparing AQP3+ alveolar cells in vitro. Conclusions: This study established hMSLN knock-in NCG mice as a robust preclinical platform to evaluate the safety of MSLN-targeted CAR-T cells, identified AQP3 as a key molecular switch to mitigate their pulmonary toxicity in ovarian cancer, and introduced a generalizable framework that integrated single-cell transcriptomics with “NOT” logic-gating design for precision control of organ-specific CAR-T cell toxicity, advancing the safer translation of CAR-T cell therapies for solid tumors.