Intravesical mesothelin-based CAR T cells targeting MUC16 effectively control bladder cancer in preclinical models

In 2023, about 600,000 new cases of bladder cancer (BCa) were diagnosed worldwide and caused nearly 200,000 deaths. In the United States, urothelial cell carcinoma, which arises from the bladder epithelial surface, accounts for >90% of 80,000 reported new cases annually (Siegel et al., 2022). Most BCa patients present with nonmuscle invasive urothelial cell carcinoma (NMIBC), for which the standard of care is transurethral resection followed by intravesical therapy, which includes chemotherapy—typically, mitomycin C or gemcitabine—and bacillus Calmette–Guerin (BCG) immunotherapy (Chang et al., 2016). While these therapies induce remission in certain subsets of patients, they are associated with high recurrence and progression rates, adverse events, and global BCG supply shortages. Newer agents, including adenoviral-based therapies (Boorjian et al., 2021), immune checkpoint blockade (anti-PD-1, pembrolizumab) (Balar et al., 2021; Meghani et al., 2022), and cytokine-based therapies (e.g., IL-15 superagonist) (Chamie et al., 2022), have shown promising, yet modest, efficacy, often leading many patients to undergo radical cystectomy (RC) as a salvage therapy. RC with urinary diversion is recommended for patients with high-risk NMIBC, including those with pathologies consistent with variant histological subtypes, and with muscle-invasive BCa (MIBC), but is a life-changing surgery associated with major comorbidities and complications approaching 60% within 30 days of surgery (Chang et al., 2017). Given these challenges, there is significant unmet clinical need, leading to renewed interests in bladder-sparing therapies for patients with high-risk NMIBC and organ-confined MIBC who are unfit or unwilling to undergo RC (Tyson et al., 2023).

Chimeric antigen receptor (CAR) T cell therapy, in which T cells are reprogrammed to express an artificial receptor to a known target, is an immunotherapeutic approach that is clinically effective in hematologic malignancies (June and Sadelain, 2018). Generally, this method involves targeting a tumor-associated antigen by genetically engineering a receptor, composed of a single-chain variable length fragment (scFv) derived from a monoclonal antibody, fused to the cytoplasmic signaling domains of a costimulatory receptor and the CD3ζ chain of the T cell receptor (van der Stegen et al., 2015). Often described as “living drugs,” CAR T cell therapies offer more specific and durable responses than systemic treatments and have the potential to induce long-lasting remissions (Park et al., 2018). However, CAR T cell therapies have had limited success with solid tumors. The underlying reasons are complex and multifactorial, including failure of CAR T cells to traffic and persist in the tumor microenvironment (TME) (Albelda, 2024). Additionally, major systemic toxicities, including on-target off-tumor effects, cytokine release syndromes, and neurotoxicities, further limit the efficacy of CAR T cell therapies in solid tumors (Neelapu et al., 2018). While these toxicities can be safely managed (Brudno and Kochenderfer, 2024), several studies have reported that locoregional cavitary delivery of CAR T cells—including intratumoral (Tchou et al., 2017), intrathecal (Theruvath et al., 2020), intracerebroventricular (Donovan et al., 2020), and intrapleural (Adusumilli et al., 2014) routes—can enhance their trafficking into tumors and minimize systemic side effects. The barrier function of the bladder normally limits systemic delivery of drugs to the bladder surface (Kong et al., 2022) but also limits extravasation of intravesical contents systemically (Benedict et al., 2004); a feature that is exploited in urologic practice when delivering high doses of toxic therapeutics directly into the bladder to enhance efficacy (Tyagi et al., 2006). In addition to limiting systemic exposure, a unique advantage of intravesical administration is the ability to evacuate bladder contents before administration, which can be used to promote agent-specific antitumoral activity under optimized conditions.

To date, no CAR T cell therapy has been specifically designed to target BCa antigens while avoiding cross-reactivity with healthy tissues. Existing CAR T cell approaches targeting MUC1 (Yu et al., 2021), EGFR (Grunewald et al., 2021), CD44v6 (Grunewald et al., 2021), and PD-L1 (Parriott et al., 2020) recognize antigens that are also expressed in other tissues, raising safety concerns. Furthermore, none of these CAR T cell therapies have demonstrated efficacy in preclinical orthotopic BCa models. Instead, CAR T cells targeting MUC1, EGFR, and CD44v6 have only been tested in vitro using cell lines (Grunewald et al., 2021; Yu et al., 2021), with PD-L1–targeting CAR T cells evaluated in a subcutaneous flank BCa model (Parriott et al., 2020). In this study, we employed an antigen discovery pipeline and identified MUC16 as a relevant targetable antigen for CAR T cell therapy in BCa. MUC16 is a large mucinous glycoprotein composed of several domains including a tandem-repeat domain of >60 repeats of 156 amino acids, which can extend upward of 500 nm from the plasma membrane (Gipson et al., 2014). MUC16 displays pleiotropic functions, including barrier functions as part of the glycocalyx matrix, lubrication, cell adhesion, and signaling functions, which can promote metastasis, chemotherapeutic resistance, and tumorigenesis (Lakshmanan et al., 2017). In BCa, MUC16 is commonly mutated in smokers (Fantini et al., 2019), and its expression correlates with tumor stage and grade, with reported detection of 30% in high-grade T1 tumors and up to 40% in MIBC, which increases by pathologic T stage (Cotton et al., 2017; Yamashita et al., 2023). In muscle-invasive disease, MUC16 expression correlates with worse overall and recurrence-free survival (Cotton et al., 2017), as well as resistance to both gemcitabine and cisplatin chemotherapy (Cotton et al., 2017; Yamashita et al., 2023). Interestingly, MUC16 is also cleaved into a soluble fragment, CA-125, which serves as a biomarker in ovarian adenocarcinoma due to its restricted and polarized expression to the surface of female reproductive organs and the aerodigestive tract (Bottoni and Scatena, 2015). Serum CA-125 is also a clinically relevant prognostic biomarker in BCa, correlating with invasiveness and chemotherapeutic resistance (Ahmadi et al., 2014; Laukhtina et al., 2021; van de Merbel et al., 2021). However, to date, there are no preclinical or clinical studies that exploit MUC16 as a therapeutic target for BCa. We developed MUC16-targeted CAR T cells and demonstrated their therapeutic efficacy in BCa. Intravesical delivery of these CAR T cells reduced the growth of BCa xenografts and prolonged survival in xenograft-bearing mice, showing superior efficacy compared with typical systemic CAR T cell administration. Our findings not only establish MUC16 as a clinically relevant target for anti-BCa CAR T cell therapy, but also suggest that intravesical delivery, a commonly used administration route in urological practice, represents a viable, easy-to-implement, and more effective strategy of antitumoral adoptive CAR T cell transfer.

To identify relevant targets for CAR T cell therapy in BCa, we developed an antigen-identification pipeline using large-scale transcriptomic data from The Cancer Genome Atlas (TCGA) and other BCa datasets covering NMIBC, MIBC, and BCG recurrent high-risk NMIBC tumors. We utilized paired normal bladder tissues where possible, allowing for a direct comparison. We first identified differentially expressed gene transcripts in tumor versus normal bladder. Next, we sorted for genes that encode membrane-bound proteins. Finally, we used the Genotype–Tissue Expression (GTEx) project dataset to filter out target transcripts that were significantly expressed on normal tissues to minimize the risk of off-target toxicities. The top four tumor antigens most significantly expressed on bladder tumor but not normal bladder, with minimal pan-tissue expression, were MUC16, ADAM2, TPTE, and ZPLD1 (Fig. 1 A). Among them, MUC16 and ZPLD1 exhibited the lowest expression across all healthy tissues, including testis, within the GTEx dataset (Fig. 1 B), highlighting the selectivity for these antigens in BCa over other tissues. We next examined the mRNA expression of these four candidates in established BCa cell lines from the Cancer Cell Line Encyclopedia (CCLE), representing an array of molecular subtypes. Notably, MUC16 expression was detectable in around 60% of the evaluated cell lines (Fig. 1 C). We then validated the expression of these candidate antigens across a spectrum of bladder tumors from multiple well-studied BCa cohorts, representing NMIBC (n = 535, Lindskrog et al.’s dataset; Lindskrog et al., 2021), high-grade T1 (n = 73, Robertson et al.’s dataset; Robertson et al., 2020), and MIBC (n = 404, subsets T2: n = 130, T3: n = 140, T4: n = 134, TCGA; Cancer Genome Atlas Research Network et al., 2013). MUC16 was the most broadly expressed candidate, detected in ∼30–60% of cases depending on tumor grade and stage (Fig. 1 D). MUC16 expression was especially enriched in subsets of primary BCa tumors with divergent differentiation or histological subtypes, such as micropapillary, plasmacytoid, and poorly differentiated (n = 280, Fig. 1 E and Fig. S1 A), which are known to have poor oncological outcomes in response to existing intravesical therapies. Moreover, MUC16 expression remained stable in both high-risk NMIBC tumors that recur after BCG intravesical therapy (de Jong et al., 2023) (Fig. 1 F and Fig. S1 B) and in matched pre- and posttreatment MIBC tumors not responsive to neoadjuvant pembrolizumab (Necchi et al., 2018) (Fig. 1 G and Fig. S1 C). To place MUC16 in context among other BCa-associated antigens, we assessed several previously reported preclinical and clinical targets of precision-guided BCa therapies alongside MUC16. Most of these antigens were excluded by our transcriptomics-based pipeline due to high RNA expression in healthy bladder and other normal tissues (Fig. S1 D). This observation was further supported by protein expression analysis using Human Protein Atlas immunohistochemistry data, which revealed membranous staining on normal urothelium (Fig. S1 E), and by GTEx gene expression data demonstrating elevated expression across multiple normal tissues (Fig. S1 F). Given its favorable expression profile, absence in normal bladder, and high expression across a broad spectrum of bladder tumors analyzed collectively spanning a total of 1,292 patients, including those recalcitrant to existing therapies, MUC16 was selected as the lead candidate for BCa-specific CAR T cell therapy development.

To select BCa cell lines for preclinical testing of MUC16-targeting CAR T cell therapy, we first assessed MUC16 expression in multiple BCa cell lines identified from the CCLE database, which exhibited varying levels of MUC16 mRNA, using quantitative RT-PCR (qRT-PCR) (Fig. 2 A). We then confirmed cell-surface MUC16 protein expression in these cell lines using two separate MUC16-specific monoclonal antibodies (OC125 and FAB5609R), which corroborated our qRT-PCR findings with detectable expression found in TCCSUP, UM-UC-7, 5637, and HT-1376 cell lines (Fig. 2 B). MSLN, a differentiation antigen with restricted expression to the mesothelial lining of the peritoneum, has a well-characterized high-affinity interaction with tandem-repeat domains found throughout MUC16 (Gubbels et al., 2006; Kaneko et al., 2009). Using a human MSLN–Fc fusion protein (hMSLN-Fc), we found that hMSLN-Fc binds to MUC16⁺ but not MUC16⁻ BCa cell lines (Fig. 2 B). This interaction is highly specific, as soluble CA-125, the cleaved form of MUC16, reduced surface binding of soluble hMSLN-Fc to MUC16⁺ HT-1376 cells in competitive binding assays (Fig. 2 C). Furthermore, hMSLN-Fc binding intensities directly correlated with those of a MUC16-specific monoclonal antibody (OC125), as well as with normalized MUC16 transcripts across multiple MUC16⁺ but not MUC16⁻ BCa cell lines, confirming that the hMSLN:MUC16 interaction is dependent on MUC16 expression (Fig. 2 C).

After confirming that MUC16 is a targetable antigen on the surface of BCa cells, we verified that MUC16 is not significantly expressed in resting T cells or in T cells following activation (Fig. S2 A). We next designed second-generation CD28-based MUC16-targeting CAR T cells for evaluation of a potential CAR T cell immunotherapy (Fig. 2 D). The CD28 costimulatory endodomain designs were utilized due to their rapid and potent effector function and enhanced activity in low-antigen contexts (Hamieh et al., 2019; Heitzeneder et al., 2022; Majzner et al., 2020; Priceman et al., 2018). We compared two different MUC16-specific CARs derived from: (1) scFv from 3a5 (3a5-28z), a humanized mAb that binds to the tandem-repeat domains of MUC16 (Chen et al., 2007); and (2) a fragment corresponding to the full length of the MSLN protein (MSLN-28z), which similarly binds to tandem repeats throughout MUC16 (Kaneko et al., 2009). Both the MUC16 scFv (3a5) and MSLN CARs were fused to a backbone composed of a Myc-identification tag and CD28-CD3ζ signaling endodomains. CAR expression on primary T cells after gammaretroviral transduction was confirmed by detection of the Myc-tag by flow cytometry, which yielded similar transduction efficiencies, as well as CD8⁺/CD4⁺ and memory T cell subset ratios for both constructs (Fig. 2 E and Fig. S2, B and C). When cocultured with BCa cell lines representing a range of MUC16 protein levels, both 3a5-28z and MSLN-28z CAR T cells effectively lysed MUC16⁺ but not MUC16⁻ cells compared with untransduced (UTD) control T cells (Fig. 2 E). Notably, MSLN-28z CAR T cells demonstrated higher cytotoxicity and cytokine production compared with 3a5-28z CAR T cells (Fig. 2, E and F), with this difference being most apparent against BCa cells with lower MUC16 expression and across a broad range of effector-to-target (E:T; T cell:tumor cell) ratios (Fig. S2 E).

To further evaluate MUC16-specific CAR T cells as a clinically relevant BCa-directed therapy, we assessed their functionality against three-dimensional patient-derived BCa tumor organoids, an in vitro precision-guided medicine platform that more accurately captures the heterogeneity, morphological and functional traits of cancer–T cell interactions than two-dimensional cell lines (van de Merbel et al., 2021). We derived three organoids from surgically excised primary bladder tumors, which expressed varying levels of MUC16 (none, positive-low, and positive-high) and different pathologies (T2 UCC and T1 UCC with neuroendocrine differentiation) (Fig. 3 A). These organoids were serially passaged while maintaining a stable phenotype including canonical uroplakins and MUC16 expression levels (Fig. 3 A). Both 3a5-28z and MSLN-28z CAR T cells exhibited potent cytolytic capacity against patient-derived tumor organoids (PDTOs), which correlated with the level of MUC16 expression in the organoids (Fig. 3 B). Consistent with the observations from the two-dimensional cell lines, we observed similar cytotoxicity between 3a5-28z and MSLN-28z against the positive-high PDTO but greater MSLN-28z CAR T cell–mediated cytotoxicity against the positive-low PDTO and minimal cytotoxicity against the MUC16-negative PDTO across three independent donors (Fig. 3 B). Further, the production of IL-2, IFN-γ, and TNF-α was also most significantly elevated in the MSLN-28z CAR T cell coculture supernatants, with concentrations of these cytokines also proportionally corresponding to tumor MUC16 expression levels (Fig. 3 C). We also evaluated the 4H11-28z CAR, a well-characterized construct targeting the MUC16-retained ectodomain (Chekmasova et al., 2010). However, both the 4H11 antibody and 4H11-28z CAR T cells demonstrated minimal recognition (Fig. S3 A) and activity across our MUC16⁺ 2D (Fig. S3 B) and 3D PDTO (Fig. S3 C) BCa culture models, suggesting a lack of accessible epitopes in BCa. Based on these findings, MSLN-28z was selected as the lead MUC16-specific candidate CAR for further characterization and in vivo validation.

We next assessed the specificity of MSLN-28z CAR T cells by evaluating their antitumor function dependence on CAR binding to MUC16. To do this, we generated MUC16-knockout HT-1376 cells (HT-1376-MUC16^KO), lacking surface MUC16, using CRISPR/Cas9-mediated gene ablation. Through this loss-of-function approach, we demonstrated that MUC16 on HT-1376 cells is necessary for MSLN-28z CAR T cell cytotoxicity, as MSLN-28z CAR T cells, like UTD T cells, failed to effectively lyse HT-1376-MUC16^KO cells (Fig. 4 A). As an orthogonal gain-of-function approach, we overexpressed a truncated form of MUC16 (MUC16^TR6) composed of six tandem repeats in a MUC16-negative cell line, UM-UC-3. The overexpression of MUC16^TR6 was sufficient to induce MSLN-28z CAR T cell cytotoxicity, leading to the killing of UM-UC-3-MUC16^TR6 cells (Fig. 4 B). As expected, the cytolytic function of MSLN-28z CAR T cells required both engagement of MUC16 on the target cell and T cell signaling through the intracellular CD28-CD3ζ signaling domains, as ablation of the intracellular signaling domains (MSLN-Del) resulted in loss-of-function of these CAR T cells (Fig. 4 C). Additionally, surface MUC16 and intracellular CD28-CD3ζ signaling were required for MSLN-28z CAR T cells to produce the proinflammatory cytokines IL-2, IFN-γ, and TNF-α (Fig. 4 D). These cytokines were produced by both polyfunctional CD8⁺ and CD4⁺ subsets of MSLN-28z CAR T cells (Fig. 4 E). Given that the TME in locally advanced BCa can be rich in CA-125 (Yamashita et al., 2023), which could potentially compete with MUC16 binding and impair MSLN-28z CAR T cell function, we quantified CA-125 concentrations in urine from patients with bladder tumors and age- and gender-matched healthy volunteers. CA-125 was detectable and elevated in BCa samples compared with healthy controls (P = 7.8e−5), including NMIBC (Ta + Tis, T1) and MIBC (T2-4) subgroups, with a trend toward higher levels by the pathologic T stage (Fig. 5 A). To assess whether soluble CA-125 could affect CAR T cell cytotoxicity, we performed cross-blocking assays using graded concentrations of soluble CA-125. Even at 100 nM, over 1,000-fold higher than the maximum urinary level detected (9.4 U/ml, ∼59 pM), soluble CA-125 did not reduce cytotoxicity against antigen-positive HT-1376 or UM-UC-3-MUC16^TR6 cells, nor did it increase killing of antigen-negative HT-1376 MUC16^KO or UM-UC-3 cells (Fig. 5 B). These findings suggest that MSLN-28z CAR T cell function is not significantly impaired by soluble CA-125, allowing them to effectively target MUC16⁺ tumors even in environments with far excess soluble CA-125 found in the urine of BCa patients. MSLN, like CA-125, can also be both shed by tumor cells and physiologically excreted in the urine, raising a possibility that soluble MSLN could compete with the MSLN-based binding moiety found on the MSLN-28z CAR. We found that soluble MSLN was detectable at low levels in healthy controls and elevated in the urine of patients with bladder tumors (Fig. 5 C). In contrast to CA-125, however, soluble MSLN-mediated blockade of MUC16 in cocultures with MUC16⁺ tumor cells could reduce MSLN-28z CAR T cell cytotoxicity at concentrations >6 nM in vitro, which was approximately twofold higher than the highest value detected in urine specimens (Fig. 5 D). This result reinforces our loss- and gain-of-function studies, confirming that MSLN-28z CAR T cells require access to MSLN-binding sites within the tandem-repeat domains of MUC16⁺ BCa tumor cells for their effector function. Taken together, these findings demonstrate that MSLN-28z CAR T cells require tumor cell-surface MUC16 expression, engagement, and intracellular signaling domains to exert cytotoxic functions, which are preserved in the presence of soluble CA-125 but can be competitively inhibited in the presence of excess of the soluble MSLN protein.

To evaluate the preclinical therapeutic potential of MSLN-28z CAR T cells in vivo, we used a xenograft model in which we engrafted human BCa cells orthotopically into the bladders of NOD-SCID gamma (NSG) mice (Fig. 6 A). Despite the challenges associated with orthotopic BCa models, including their time-consuming nature, technical complexity, and variable engraftment consistency between lines, we chose this model because it more accurately recapitulates organ-confined BCa (Zuiverloon et al., 2018). In our xenograft model, uncontrolled HT-1376 tumor growth resulted in obstructive uropathy and subsequent demise typically within 30–45 days. A single dose of 5 × 10⁶ MSLN-28z CAR T cells, delivered intravesically, significantly increased survival compared with UTD T cells (Fig. 6 B). In contrast, intravenous (IV) adoptive transfer of 5 × 10⁶ MSLN-28z CAR or UTD T cells failed to control tumor outgrowth (Fig. 6 B). Endpoint analysis revealed that mice treated with intravesical MSLN-28z CAR T cells exhibited smaller tumor burdens than mice treated with intravesical UTD T cells (Fig. 6 C). Immunohistochemical analysis of treated tumors at the endpoint showed higher infiltration of human CD45⁺ lymphocytes in the group receiving intravesical adoptive transfer compared with the IV group (Fig. 6 C). Notably, neither MSLN-28z CAR T nor UTD T cells administered intravesically were detectable in the peripheral blood of mice within 7 days of transfer (Fig. 6 D). Consistent with the observed lack of systemic engraftment following intravesical adoptive transfer, inflammatory human cytokines IFN-γ and TNF-α were detectable only in the urine (Fig. 6 E) of intravesically treated mice. These observations suggest that cytokine secretion by intravesical CAR T cells is localized to the local bladder environment, potentially minimizing systemic toxicities. To validate these findings in an independent orthotopic BCa model with a different MUC16 antigen expression level, we utilized the human UM-UC-3 cell line engineered to overexpress the MUC16 tandem-repeat domain (UM-UC-3-MUC16^TR6) (Fig. 6 F). Consistent with the HT-1376 model, we observed enhanced survival (Fig. 6 G), significantly reduced tumor burden (Fig. 6 H), minimal detectable systemic engraftment within 7 days following adoptive transfer (Fig. 6 I), and elevated levels of urinary IFN-γ and TNF-α cytokines (Fig. 6 J) after intravesical adoptive transfer of MSLN-28z CAR T cells. Taken together, these results demonstrate that intravesical delivery of MSLN-28z CAR T cells effectively controls orthotopic BCa outgrowth while restricting systemic exposure, highlighting a promising localized therapeutic strategy.

Given the efficacy observed with intravesical adoptive transfer in the orthotopic BCa models, we next investigated the trafficking kinetics of MSLN-28z CAR T cells delivered by each route of administration. To enable sensitive, longitudinal monitoring, we used a bicistronic vector composed of MSLN-28z, a p2a element, and a membrane-anchored Gaussia luciferase protein (MSLN-ExtGLuc), which enables in vivo bioluminescence imaging (BLI) of CAR T cell localization (Smith et al., 2018). We employed the UM-UC-3-MUC16^TR6 cell line, which engrafts robustly in both bladder and subcutaneous flank models, to directly compare CAR T cell trafficking following intravesical or IV adoptive transfer. In the flank tumor model, IV adoptive transfer produced detectable BLI signal in the tumor as early as day 5, which persisted through day 21 (Fig. 6 K and Fig. S4 A). In contrast, intravesical adoptive transfer did not yield significantly detectable signal in flank tumors over the same period, which goes in line with the lack of detectable engraftment seen previously. When the same BCa line (UM-UC-3-MUC16^TR6) was implanted in the bladder, IV adoptive transfer resulted in minimal signal within the bladder and aligning with the limited efficacy observed in the prior therapeutic studies. By comparison, intravesical adoptive transfer resulted in a strong BLI signal localized to the bladder on day 0, which declined to below the limit of detection by day 5 (Fig. 6 L and Fig. S4 B), corresponding to an estimated intravesical residence half-life (t_1/2) of ∼0.8 days. Importantly, there was no evidence of BLI signal outside the bladder after intravesical adoptive transfer—including the peritoneal cavity or kidneys—through day 21, supporting the confinement of CAR T cells to the bladder lumen. To evaluate whether these observations were specific to MSLN-28z CAR T cells or could extend to other CARs and/or targets, we tested a second-generation CAR directed against B7-H3 (CD276), another cancer-associated antigen reported to be expressed on bladder tumors. UM-UC-3 cells express B7-H3 (Fig. S4 C) and can be targeted in vitro by a CAR composed of the MGA271 scFv (Loo et al., 2012) fused to the CD28-CD3ζ signaling domains (Fig. S4 D). Like MSLN-28z CAR T cells, 5 × 10⁶ B7H3-28z CAR T cells after intravesical adoptive transfer enhanced survival (Fig. S4, E and F), reduced endpoint tumor burdens (Fig. S4 G), were minimally detectable in the peripheral blood (Fig. S4 H), and resulted in elevation of urinary IFN-γ and TNF-α cytokines (Fig. S4 I). Together, these data demonstrate that intravesical delivery of CAR T cells, targeting either MUC16 or B7-H3, provides potent and localized antitumor activity in orthotopic BCa models while minimizing systemic dissemination, establishing a strong preclinical rationale for BCa-directed CAR T cell therapies.

We next evaluated the potential activation of MSLN-28z CAR T cells by normal murine cells when 5 × 10⁶ MSLN-28z CAR or UTD T cells were injected IV or intravesically. No acute infusion–related systemic toxicities were observed via either mode of adoptive transfer, including changes in body weight, relevant hematologic, liver, and renal clinical chemistries within 14 days of transfer (Fig. 7 A). A subset of mice was then subjected to complete necropsy, revealing hyperproliferation of human CD45⁺ lymphocytes in the lung of one mouse that received IV UTD (control) T cells, consistent with physiologic hyperproliferation or early graft-versus-host disease. However, no significant pathological changes were observed in major organs (lung, liver, spleen, brain, or heart) with either mode of transfer of 5 × 10⁶ MSLN-28z CAR T cells (Fig. 7 B), suggesting that both IV and intravesical delivery of MSLN-28z CAR T cells are well tolerated in NSG mice, consistent with prior preclinical studies with MUC16-specific antibody and CAR T cell–based therapies (Casey et al., 2024; Chen et al., 2007).

While MSLN-28z CAR T cells were well tolerated in NSG mice in our study and the sequence homology of MSLN is conserved in mice and humans (Kaneko et al., 2009), MUC16 differs significantly (Wang et al., 2008). Likewise, the B7H3-28z CAR is derived from a humanized antibody that recognizes a tumor-restricted epitope of human B7-H3 (Loo et al., 2012; Majzner et al., 2019). To evaluate potential on-target off-tumor toxicities, even in the setting of low-level systemic escape after intravesical adoptive transfer, we used two orthogonal models with sensitive and well-defined hematologic and tissue toxicities as readout biosensors for toxicity. Tyrosinase-related protein 1 (TRP-1/gp75) is a melanoma differentiation antigen and target of interest for immunotherapy in metastatic melanoma (Turk et al., 2002). Therapeutic preclinical murine models and clinical studies targeting TRP-1 commonly manifest in fur and skin changes, respectively, owing to destruction of shared TRP-1 expression in normal melanocytes (Hara et al., 1995; Jilani et al., 2024; Spreafico et al., 2024). In the TRP-1 transgenic TCR model, IV adoptive transfer of as few as 2 × 10³ naïve TRP-1 CD4⁺ T cells is sufficient to induce fur depigmentation in RAG1^−/− C57BL/6 strains (Fig. S5 A) (Muranski et al., 2008), which correlates with antitumor responses (Kreft et al., 2010). After adoptive transfer of 5 × 10⁴ TRP-1 T cells IV, representing 25-fold escalation of the minimum dose required for detectable fur depigmentation, all mice developed fur depigmentation within 30 days of transfer, whereas no mouse administered the equivalent dose intravesically developed this autoimmune toxicity (Fig. S5 A), suggesting minimal to no systemic engraftment with intravesical adoptive transfer. As a second model, we tested systemic escape using murine-specific CAR T cells that target CD19 (clone 1D3–based m19m28z) (Davila et al., 2013), an antigen found on normal murine and human B cells. B cell aplasia is both a toxicity and pharmacodynamic marker of CD19-targeted CAR T cell therapy in B-ALL (Davila et al., 2013). IV administration of 5 × 10⁶ m19m28z CAR T cells resulted in a significant decrease in peripheral B cell counts, whereas intravesical adoptive transfer did not cause any measurable B cell aplasia (Fig. S5 B), supporting the lack of systemic escape observed in the autoimmune TRP-1 model in a CAR T cell context. To further validate the absence of systemic toxicity with intravesical adoptive transfer, we used the syngeneic MB49 BCa model as a proof-of-principle. We generated MB49 cells overexpressing murine CD19 (MB49-CD19), which were effectively killed by m19m28z CAR T cells in vitro (Fig. S5 C). Like the NSG models, survival of orthotopic MB49-CD19–bearing mice was significantly extended with intravesical adoptive transfer of m19m28z CAR T cells (Fig. S5 D) without significant B cell aplasia, compared with IV administration (Fig. S5 D). Taken together, these results suggest that intravesical adoptive transfer of both CAR T and non-CAR T cell therapeutic lymphocytes results in inefficient systemic engraftment and, consequently, reduced systemic on-target off-tumor toxicity.

In this study, we leveraged a computational antigen-identification pipeline, which prioritized high tumor specificity and minimal pan-tissue expression to rationally identify MUC16 as a potential target for BCa-directed CAR T cell therapy. MUC16 and its soluble form, CA-125, have previously been identified as prognostic biomarkers for BCa; our study extends these observations to a therapeutic target relevant to a range of pathologic T stages (Cancer Genome Atlas Research Network et al., 2013; Lindskrog et al., 2021; Robertson et al., 2020), including high-risk NMIBC recurrent after intravesical BCG (de Jong et al., 2023), MIBC unresponsive to neoadjuvant pembrolizumab (Necchi et al., 2018), and tumors with divergent differentiation and histological subtypes. These findings suggest that MUC16 may be a candidate for CAR T cell therapy for both initial treatment and treatment-refractory subsets of both NMIBC and MIBC tumors, providing a therapeutic option for patients who may have limited therapeutic alternatives besides early or salvage RC.

MUC16 has been explored as a CAR T cell therapy target in several malignancies, most notably ovarian cancer. Early studies targeting MUC16 primarily focused on the retained ectodomain (MUC-CD), based on the assumption that the cleaved fragment, CA-125, might compete and impair CAR T cell function. A recent phase I clinical trial using the 4H11-28z CAR targeting MUC-CD reported that the therapy was well tolerated but achieved only stable disease responses (Koneru et al., 2015; O’Cearbhaill et al., 2020). Although several factors may explain the lack of clinical remissions and why we saw limited activity of the 4H11-28z CAR in our 2D and 3D PDTO BCa cell culture models, studies suggest that CA-125 can remain associated with the retained MUC16 ectodomain after cleavage, potentially limiting antigen accessibility (Aithal et al., 2018). In contrast, CARs recognizing shared epitopes between soluble CA-125 and membrane-bound MUC16 are not hindered by this process and likely require cross-linking by membrane-tethered antigens (Casey et al., 2024). We compared two MUC16-specific CARs: a scFv-based CAR (3a5-28z) and a natural-ligand MSLN-based CAR (MSLN-28z). Across multiple two-dimensional BCa cell lines and three-dimensional PDTOs spanning a range of MUC16 densities, MSLN-28z CAR T cells exhibited superior cytotoxicity and proinflammatory cytokine secretion compared with 3a5-28z CAR or control UTD T cells, particularly at lower MUC16 levels. These key effector functions required both MUC16 engagement and CD28-CD3ζ signaling, confirming antigen-dependent activity. Importantly, soluble CA-125, which was detectable in urine derived from BCa patients, did not impede the cytotoxicity or specificity of MSLN-28z CAR T cells, similar to prior studies evaluating scFv-based MUC16-specific CAR T cells targeting tandem-repeat domains of MUC16 (Casey et al., 2024). In contrast, MSLN, reported to be rarely expressed in BCa (Weidemann et al., 2021) but detectable in urine from BCa patients, reduced in vitro cytotoxicity only at supraphysiologic concentrations. Although this inhibition occurred at MSLN levels exceeding the highest urinary outlier in our dataset, it could limit MSLN-28z CAR or other MUC16-targeted T cell therapy scenarios of concomitantly elevated soluble MSLN (e.g., systemic or intraperitoneal administration for mesothelioma or ovarian cancers [Weidemann et al., 2021]). This concern is less relevant for intravesical adoptive transfer, as tumor-shed proteins in the bladder are cleared by voiding, and the bladder can be evacuated and washed prior to adoptive transfer. Overall, our findings underscore the promise of MUC16-targeted CAR therapies while highlighting potential confounding factors for non-BCa applications.

In vivo, MSLN-28z CAR T cells controlled orthotopic BCa xenografts following intravesical but not IV adoptive transfer. To our knowledge, this is the first demonstration of CAR T cell efficacy against BCa in an orthotopic local disease model. Intravesical delivery of CAR T cells was effective at controlling bladder tumor growth, outperforming systemic adoptive transfers at equivalent doses across multiple orthotopic tumor models evaluated in our study. Importantly, intravesical adoptive transfer resulted in inefficient systemic engraftment and was accompanied by inflammatory T cell–derived cytokines, which were detected in the urine of mice treated with intravesically administered CAR T cells. BLI revealed that only intravesical adoptive transfer produced high CAR T cell signals within the bladder, which was both immediate and transient. We hypothesize that this rapid decline in T cell signal is driven by a combination of mechanical elimination via micturition and the potentially unfavorable urinary microenvironment. Interestingly, despite this lack of engraftment and long-term persistence—which is typically required for systemic CAR T cell efficacy in hematologic malignancies—intravesical delivery achieved durable orthotopic bladder tumor control. These data suggest that high local concentrations of CAR T cells facilitate tumor eradication prior to clearance without evidence of extravesical dissemination, in a “hit-and-run”–type mechanism. To further confirm the absence of systemic toxicities in the event of escape below the limit of detection, we used adoptive transfers of T cells targeting well-characterized antigens found in normal murine tissues and demonstrated a lack of B cell aplasia with m19m28z CAR T cells and an absence of fur depigmentation with transgenic TRP-1 CD4⁺ T cells after intravesical adoptive transfers. These findings are in agreement with prior observations evaluating intravesical adoptive transfer of tumor-infiltrating lymphocytes in the MB49 preclinical BCa model, which similarly controlled tumor growth, and extend them to a CAR T cell therapy–based context (Bunch et al., 2020). Taken together, localized intravesical administration not only enhances the antitumor efficacy of CAR T cells against bladder tumors but also minimizes systemic escape, a common limitation of manufactured T cell therapy products for other solid tumors (Lamers et al., 2013; Morgan et al., 2010). In contrast to prior locoregional CAR T cell adoptive transfer studies (Katz et al., 2016; Theruvath et al., 2020; Wang et al., 2021), which demonstrate similar dose-matched locoregional superiority phenomena, we observed limited efficacy with IV adoptive transfer in the orthotopic bladder tumor models. This may be in part because the bladder, while not considered a classically immune-privileged site, is a barrier and excretory organ that functions to store and empty urinary waste with limited resorptive capacity, and in part because our preclinical models are organ-confined and therefore more reliant on the more limited blood supply of the bladder (Zuiverloon et al., 2018), unlike subcutaneous flank models. Together, these reasons likely explain why intravesical CAR T cells traffic inefficiently out of bladder tumors and why IV CAR T cells traffic inefficiently into bladder tumors. This inefficiency may be overcome with higher IV dosing, as was reported targeting the antigen B7-H3 with 10-fold higher IV dosing than intrathecal dosing to the brain (Theruvath et al., 2020); however, in our experience IV adoptive transfers of >5 × 10⁶ human lymphocytes into NSG mice often result in graft-versus-host disease, which limit interpretations of efficacy. These findings support an immunotherapy model initially proposed for optimal BCG therapy by Zbar and Rapp (1974), requiring close locoregional contact, appropriate therapeutic dosage, and adequate tumor sizes for effective responses. Our study underscores the benefit of intravesical treatment approaches in organ-confined BCa, which can maximize therapeutic impact while minimizing systemic exposure, which extends to multiple therapies including BCG (Morales et al., 1976), gemcitabine (FJ et al., 2024), mitomycin C (Thrasher and Crawford, 1992), IL-15 superagonists (Furuya et al., 2019), Fc-enhanced CD40 antibody agonists (Garris et al., 2021), and adenoviral-based therapies (Boorjian et al., 2021; Tao et al., 2006); all of which demonstrate similar transient but durable antitumoral effects and localized adverse events such as hematuria, dysuria, and pollakiuria, with negligible systemic toxicities when administered intravesically.

Because intravesical administration results in delivery of CAR T cells into a confined anatomic compartment with inefficient systemic engraftment, this route is suboptimal for treating extravesical and metastatic disease and will have transient antitumoral effects, as seen in our study. BCa recurrences in the prostatic urethra or upper tract urothelium (ureter/renal pelvis) can occur after intravesical therapy with different agents (Chamie et al., 2022; Herr, 1998), this is a generalized limitation of this route of administration (von Rundstedt and Lerner, 2014). This limitation, however, presents additional avenues for research. Intraurethral (Varol et al., 2004) or upper tract instillation via retrograde or percutaneous approaches (Studer et al., 1989) have been described and successfully used to overcome this anatomic limitation with respect to BCG, and we surmise they can similarly be exploited for T cell–based therapies. Because the effects of intravesical therapy are localized to the bladder, similar to intratumoral delivery (Haydar et al., 2021), preconditioning lymphodepleting chemotherapy with comorbid agents like cyclophosphamide/fludarabine or irradiation, which is common with systemic CAR T cell therapies to promote T cell engraftment (Klebanoff et al., 2005), may not be required for intravesical CAR T cell therapy. Furthermore, this route of adoptive transfer is also amenable to repeat administration, with reduced concern for accumulated systemic toxicities and hypofunctionality associated with IV CAR T cell dosing strategies (Albelda, 2024).

The degree of antigen heterogeneity in solid tumors presents an additional challenge for CAR T cell therapy. While MUC16 was identified as a target based on high differential expression in BCa tumors and minimal pan-tissue expression, prior studies demonstrate a wide range of MUC16 expression within positive BCa cases (Cotton et al., 2017). This variability could impact clinical responses, particularly in patients with low or heterogeneous MUC16-expressing tumors. To address this, targeting second or third antigens with an independent expression pattern, such as TPTE, ADAM2, and ZPLD1, which were also identified in our antigen-identification pipeline, or more broadly expressed targets such as B7-H3 (Boorjian et al., 2008), such as shown here, or NECTIN-4 (Challita-Eid et al., 2016) could be employed in a similar fashion to other affinity-tuned and dual-targeted CAR T cells designed to overcome antigen escape (Fernandez de Larrea et al., 2020; Ruella et al., 2016). Given the inefficient systemic engraftment seen with intravesical adoptive transfer in our model, further target antigens may also be identified through less stringent criteria with respect to pan-tissue expression. Another active area of investigation in CAR T cell biology is armored augmentation (i.e., cytokines, antibodies) (Jaspers et al., 2023; Rafiq et al., 2018) to enhance antitumor activity through overcoming immunological checkpoints, and enhance bystander killing or altering the TME, which can carry additional risks of systemic toxicities, as seen in certain IL-15 (Chen et al., 2019) and IL-18 models (Ma et al., 2020). More recently, the glycocalyx matrix itself has been identified as a barrier to immune activity, which can be addressed by engineering CAR T cells with mucinase activity (Park et al., 2024), though this approach requires validation for when the CAR targets themselves are also mucinous antigens. Investigating such strategies as an intravesical CAR T cell therapy could potentially enhance the depth and duration of antitumor responses for more comprehensive tumor coverage with attenuated systemic side effects, owing to reduced systemic engraftment and escape of intravesical cytokines.

Our study is the first to identify MUC16 as a targetable antigen in BCa and demonstrate that MSLN-based CAR T cells could serve as a potential intravesical therapy for patients with organ-confined MUC16⁺ BCa tumors, including those resistant to existing treatments. Furthermore, our findings lay the groundwork for refining CAR T cell therapies targeting other antigens for BCa.

To identify candidate antigens that are unique and specific to BCa, we used RNA-seq data from tumors and paired tumor-adjacent normal bladder tissue from TCGA (Cancer Genome Atlas Research Network et al., 2013). Starting with 20,502 genes, DESeq2 version 1.42.0 (Love et al., 2014) was used to run differential expression analysis and extract significantly expressed genes in primary tumors versus normal tissues; genes that had log fold change <2.5 or a P value >0.01 were filtered, resulting in 707 genes. The list of genes was further reduced to only include cell-surface genes using a list of genes queried from UniProt (UniProt Consortium, 2023). To further filter for genes that are not expressed in critical tissues, a last filtration step was performed using RNA-seq data from normal tissues obtained from the GTEx project (GTEx Consortium, 2020). All the genes that had log2(TPM) expression >1 in any tissue were excluded, resulting in 4 lead candidate genes. Data accessed from studies (access codes) were as follows: (Robertson et al., 2020) (GSE154261), Lindskrog et al. (2021) (EGAS00001004693), (Necchi et al., 2018) (EGAC00001002276), and de Jong et al. (2023) (EGAS00001006879 and Zenodo: 10.5281/zenodo.7883518). For analysis of variant histological subtypes: following written and informed consent in accordance with Institutional Review Board (IRB) approval at Memorial Sloan Kettering Cancer Center (MSKCC) (MSK IRBs #06-107 and #12-245), bladder tumors were collected from patients after RC or transurethral resection. Specimens were centrally reviewed by a board-certified genitourinary pathologist and histomorphologically classified according to the World Health Organization classification of tumors, fifth edition. Dual DNA/RNA extraction was performed from formalin-fixed, paraffin-embedded (FFPE) tissue following standard protocols. Bulk RNA whole-transcriptome sequencing was then performed using the Illumina NovaSeq 6000 or X according to the manufacturer’s instructions. RNA-seq reads were aligned against human genome assembly hg19 by STAR 2-pass alignment.

5637 (HTB-9; ATCC), T24 (HTB-4; ATCC), HT-1376 (CRL-1472; ATCC), TCCSUP (HTB-5; ATCC), SV-HUC-1 (CRL-9520; ATCC), UM-UC-3 (CRL-1479; ATCC), UM-UC-7 (Sigma-Aldrich), OVCAR3 (HTB-161; ATCC), and MB49 (from M. Glickman, MSKCC, New York, NY, USA) were stably transfected with a retroviral vector expressing GFP and luciferase and were maintained in DMEM (Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS), L-glutamine, and penicillin/streptomycin. Gpg29 fibroblasts (BioVec H29) and GALV-pseudotyped 293 (BioVec GALV9) packaging cells were maintained in DMEM supplemented with 10% FBS, L-glutamine, and penicillin/streptomycin. MUC16 knockout HT-1376 cells were generated by nucleofection of HT-1376 cells with Cas9 and MUC16 sgRNA (Synthego) and FACS-isolated as a MUC16⁻ population. MUC16-TR6–overexpressing UM-UC-3 cells were generated by spinoculation of retroviral supernatant containing overexpression vector corresponding to the proximal six tandem-repeat domains of MUC16 and FACS-isolated as a MUC16⁺ population.

Plasmids encoding CAR constructs in the SFG y-retroviral vector were used to transfect H29 cells with ProFection Mammalian Transfection System (Promega) according to the manufacturer’s instructions. The retroviral supernatants were used to transduce GALV9 cells to generate stable retroviral particle–producing cell lines. All vectors were generated by restriction enzyme digest and Gibson assembly (New England BioLabs). For the 3a5-28z CAR, we used the scFv for the 3a5 monoclonal antibody, Myc-tag, human CD28 transmembrane and intracellular domain, and human CD3ζ intracellular domain. For the MSLN-28z CAR, we used a cDNA fragment corresponding to AA296-598 of the human MSLN protein, which was similarly fused to the Myc-tag (EQKLISEEDL), human CD28 transmembrane and intracellular domain, and human CD3ζ intracellular domain. For the MSLN-Del CAR, a construct lacking the CD28 intracellular and CD3ζ domains was cloned. For the B7H3-28z CAR, we used the scFv for the MGA271 monoclonal antibody, Myc-tag, human CD28 transmembrane and intracellular domain, and human CD3ζ intracellular domain. For the MSLN-ExtGLuc CAR, a construct composed of MSLN-28z, p2a element, and ExtGLuc transgene was cloned. The MUC16-TR6 overexpression vector was created corresponding to the SFG backbone.

Human T cells were isolated, activated, and transduced as previously described. Briefly, peripheral blood mononuclear cells were isolated from buffy coats (New York Blood Center) from which T cells were isolated using the EasySep Human T Cell Isolation Kit (StemCell Technologies). T cells were activated with anti-CD3/CD28 beads (bead/cell ratio 1:1) supplemented with 5 ng/ml IL-7 and 5 ng/ml IL-15 (PeproTech) and cultured in RPMI 1640 supplemented with 10% FBS, L-glutamine, and 1% penicillin/streptomycin. After 48 h of activation, the T cells were then debeaded and transduced by spinoculation on 5 μg/ml RetroNectin-coated plates (Takara) with retroviral supernatant from viral packaging cells supplemented with 5 ng/ml of IL-7 and IL-15.

Mice were euthanized, and their spleens were harvested. Following tissue dissociation and RBC lysis, CD3⁺ T cells were enriched via negative selection using EasySep Mouse T Cell Isolation Kit (StemCell). Cells were then expanded in vitro by culturing in RPMI 1640 supplemented with 10% FBS, nonessential amino acids, 1 mM sodium pyruvate, 10 mM HEPES, 2 mM L-glutamine, 1% penicillin/streptomycin, 11 mM glucose, 2 μM 2-mercaptoethanol, 100 IU of recombinant human IL-2 (PeproTech), and anti-CD3/28 Dynabeads (Life Technologies) at a bead:cell ratio of 1:2. At 24 and 4 h after initial expansion, T cells were spinoculated with viral supernatant collected from Phoenix-ECO cells. After the second spinoculation, cells were rested for 1 day and then used in adoptively transfer studies. For TRP-1 T cell isolation, TRP-1 CD4⁺ TCR transgenic mice were obtained from the N. Restifo laboratory (National Institutes of Health, Bethesda, MD, USA), and crossed to Rag1^−/− Trp1^−/− CD45.1 background. All mice were bred at MSKCC. These mice were euthanized and T cells isolated as described and immediately used for adoptive transfer.

The collection and use of PDTOs followed guidelines established by IRB (NewYork-Presbyterian/Weill Cornell Medical Center, IRB # 1305013903). Primary tumor specimens (>2 cm) were isolated from three BCa patients. A small portion of each tumor was preserved in formalin for histopathological review by a board-certified pathologist. PDTOs were developed following a modified protocol as previously described. Briefly, fresh tissue was mechanically dissected into 2-mm pieces. Media containing loose cells or cell clumps were separated as a predigest fraction and cultured without enzymatic digestion. Enzymatic digestion was performed using collagenase IV (Thermo Fisher Scientific) media, incubated on a shaker at 200 rpm and 37°C for 30 min until cloudy. Both the suspension and predigest fraction were centrifuged at 300 g for 3 min, and cell pellets were washed. Cells were resuspended separately in PDTO culture media (DMEM/F12 [Gibco], 2 mM GlutaMAX [Thermo Fisher Scientific], 2 mM HEPES [Gibco], 100 U/ml P/S [Gibco], B27 supplement [Gibco], 10 mM nicotinamide [Sigma-Aldrich], 1.2 mM N-acetylcysteine [Sigma-Aldrich], 1 ng/ml FGF-b [PeproTech], 20 ng/ml FGF-10 [Peprotech], 500 nM A-83-01 [Tocris], and FGF-basic [PeproTech]). Up to 10 100-μl drops of Matrigel/cell suspension were plated onto a 6-well culture plate and polymerized for 30 min, and then, 3 ml PDTO culture media were added per well. Media were replaced every 3 days. When PDTOs reached 300–500 μm, they were passaged using TrypLE Express for 10 min at 37°C. Single cells and small clusters were replated using the same procedure. Monthly Mycoplasma screening was performed using the Abm Mycoplasma PCR Detection Kit. PDTOs were cryopreserved in Recovery Cell Culture Freezing Medium (Thermo Fisher Scientific) and stored in liquid nitrogen. The histopathology of PDTOs was verified by comparing FFPE sections of passage 5 PDTO blocks with parent tumor sections using cytology and histology. PDTOs were suspended in a fibrinogen/thrombin gel pellet, fixed with 4% paraformaldehyde in PBS, and embedded in paraffin. H&E-stained sections of these FFPE blocks were verified as tumor cells and compared with H&E-stained parent tumor sections for matching morphology by a Weill Cornell Medicine pathologist.

Human CAR T cells were cocultured at varying E:T ratios with 1 × 10³ target tumor cells in 96-well plates in 200 μl media for 20 h. 75 ng of D-luciferin (Gold Biotechnologies) in 10 μl of PBS was added to each well and luminescence measured with a microplate reader. Wells with target cells alone were used to determine the maximum signal, and the percentage of lysis was calculated as 1 − (sample/maximum) × 100. For PDTOs:CAR T cell cocultures, a previously established protocol (Dijkstra et al., 2018) was adapted for use. Briefly, 3 days prior to coculture experiments, PDTOs were stained with CellTrace Far Red (red) and embedded in 100 μl of 66% Matrigel droplets. After 72 h, the stained PDTOs were harvested with cell recovery solution and seeded into 96-well plates with 100 μl of RPMI/10% FCS containing 1 µM NucView 488 caspase-3 substrate (green). T cells were added at a 2:1 E:T ratio in 100 μl of media. The plates were imaged every hour for 36 h using the Incucyte S3 (Sartorius), recording four fields per well. A minimum threshold area of 50 µm² was set for quantifying red events (PDTOs). Apoptotic PDTOs were identified as double-positive cells (green and red signals). The percentage of apoptotic PDTOs was calculated based on the total number of red events per field.

For analysis of in vitro cytokine secretion, cell-free supernatant cytokines were quantified using Luminex-based bead multiplex immunoassays per the manufacturer’s protocol (EMD Millipore). For in vivo cytokine analysis, serum and urine were collected as described. For urine samples, the pH was neutralized in 1:1 with assay buffer. Samples were analyzed using Luminex xMAP INTELLIFLEX System (Luminex Corp). For intracellular cytokine staining, rested CAR or UTD T cells were cocultured with or without tumor cells at 1:1 ratios in RPMI/10% FCS containing BD GolgiStop and GolgiPlug Protein Transport Inhibitors (BD) for 6 h per the manufacturer’s protocol.

Antibodies and stains were titrated for optimal staining using flow cytometric buffer (PBS/0.5% BSA/2 mm ethylenediaminetetraacetic acid [EDTA]) and analyzed on CyTek Aurora (CyTek) or Attune NxT Flow Cytometer (Thermo Fisher Scientific). Analysis was performed with FlowJo Software (FlowJo, LLC). The following antibodies/stains were used: Myc-tag-Alexa 647 (9B11, 2276S; Cell Signaling), CA125/MUC16-Alexa 647 (986808, FAB5609R; R&D), CA125 (OC125; Sigma-Aldrich), CD4-BV650 (SK3, 563875; BD), CD8-B548 (SK1, R7-20097; CyTek), CD45RA-BUV737 (H100, 612846; BD), CD62L-BUV563 (SK11, 749211; BD), CD8-cFluor B548 (SK1, R7-20097; BD), CD3-APC/Fire810 (SK7, 344858; BioLegend), Live/Dead Fixable Aqua (L34957; Thermo Fisher Scientific), IL-2-PE (MQ1-17H12, 12-7029-42; Thermo Fisher Scientific), IFN-γ-PE/Dazzle-594 (B27, 506530; BioLegend), TNFa-PE-Cy7 (mAb11, 557647; BD), murine B220-PE (104, 12-04542-82; Thermo Fisher Scientific), murine CD19-BV605 (6D5, 115539; BioLegend), hMSLN-Fc (ab215637; Abcam), human MSLN-Alexa 647 (AFR3265-020; R&D), and goat anti-human IgG-Fc Alexa 647 (A55749; Thermo Fisher Scientific). For intracellular staining, cells were stained for cell-surface markers and subsequently fixed, permeabilized, and stained for intracellular markers using BD CytoFix/CytoPerm (554723; BD) per the manufacturer’s protocol.

RNA from cultured cells was isolated using RNeasy Mini Kit (Qiagen) and used to make cDNA with High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific) according to the manufacturer’s protocols. qPCRs were assembled using TaqMan 2× Gene Expression Master Mix and probes specific for MUC16 (Hs01065175_m1, Thermo Fisher Scientific) and GAPDH (Hs02786624_g1, all from Thermo Fisher Scientific), and subsequently analyzed on QuantStudio Real-Time PCR System (Thermo Fisher Scientific). Gene expression levels were normalized to GAPDH.

Urine samples from patients diagnosed with BCa (n = 50) and healthy volunteers (n = 12) were sourced from commercial biobanks (BioIVT and Medix Biochemica, respectively). There were no significant differences in age (mean 69.5 [BCa] versus 69.8 [HC], P = 0.93) or sex (26 versus 33% female, P = 0.88) between the two cohorts. Urine samples were processed by centrifugation at 500 × g for 10 min, and supernatants were aliquoted and frozen for each assay until use. ELISAs were used to quantify urinary biomarker concentrations in duplicate against a standard curve (4PL fit) according to the manufacturer’s instructions. For MSLN (DMSLN0; R&D Systems), a dilution of 1:50 was used (sensitivity at this dilution: 1.1 ng/ml [27.5pM]). For CA125 (ab274402; Abcam), a dilution of 1:3 was used (sensitivity at this dilution: 0.06 U/ml). Concentration estimates below the manufacturer-reported sensitivity were set to zero.

Mice were housed under specific pathogen–free conditions in the MSKCC Research Animal Resource Center, and all experiments were performed in accordance with Institutional Animal Care and Use Committee–approved protocols (2022-0023). 6- to 8-wk-old female NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice were purchased from The Jackson Laboratory (The Jackson Laboratory) and used for all xenograft experiments. Mice were first anesthetized with isoflurane and catheterized per urethra with 24-g angiocatheter (Braun Introcan), intravesical positioning was confirmed through urine evacuation, and the bladder was then washed with PBS, after which 50 μl of 0.01% poly-L-lysine (Sigma-Aldrich) was instilled for 30 min and subsequently evacuated. For tumor engraftment, approximately 1 × 10⁶ HT-1376 cells, 0.5 × 10⁶ UM-UC-3-MUC16^TR6 cells, or 0.5 × 10⁶ UM-UC-3 in 100 μl PBS were instilled into the bladder for 1 h, after which the bladder contents were evacuated. Tumor engraftment was confirmed at day 7 by bioluminescence, and ∼50% of mice successfully engraft with HT-1376, 90% with UM-UC-3, and 90% with UM-UC-3-MUC16^TR6. For engrafting UM-UC-3-MUC16^TR6 heterotopically into the flank, ∼0.5 × 10⁶ cells were suspended in 100 μl of Matrigel (Corning) per the manufacturer’s protocol, and injected subcutaneously into the flank with 25-g needle, tumor engraftment was confirmed at day 7 by bioluminescence and physical palpation, and ∼100% of mice successfully engraft in the flank. The tumor-bearing mice were randomized before CAR T cell transfer. For IV adoptive transfer experiments, 5 × 10⁶ T cells resuspended in 200 μl PBS were injected via tail vein. For xenograft intravesical adoptive transfer experiments, mice were anesthetized with isoflurane and catheterized per urethra with a 24-g angiocatheter after which 5 × 10⁶ T cells in 100 μl PBS were instilled in the bladder for 2 h. For the CD19 intravesical model, C57BL/6J (6–8 wk female) mice were obtained from The Jackson Laboratory after which 5 × 10⁶ CAR T cells suspended in 200 μl PBS were injected via tail vein or 5 × 10⁶ CAR T cells suspended 100 μl PBS were instilled in the bladder for 3 h. For the MB49-CD19 intravesical model, 0.5 × 10⁶ MB49-CD19 cells were instilled with 24-g angiocatheter per urethra in C57BL/6J (6–8 wk female) for 1 h. Approximately 90% of mice successfully engraft with MB49-CD19 at day 7. After randomization of tumor-bearing mice, approximately 5 × 10⁶ CAR T cells suspended in 200 μl PBS were injected via tail vein or 5 × 10⁶ CAR T cells suspended 100 μl PBS were instilled in the bladder for 3 h. For the TRP-1 model, C57BL/6J RAG1^−/− (6–8 wk female) mice were obtained from The Jackson Laboratory. 5 × 10⁴ TRP-1 T cells suspended in 200 μl were injected via tail vein, or 5 × 10⁴ TRP-1 T cells suspended in 100 μl were instilled intravesically for 2 h. Blood was sampled via retro-orbital access using microhematocrit capillary tubes (Thermo Fisher Scientific). Blood for cellular analysis was collected into EDTA collection tubes, and red blood cells were lysed using ACK lysis buffer (Thermo Fisher Scientific). Serum was isolated from blood collected into 1.5-ml centrifuge tubes, allowed to clot for 30 min at room temperature, and centrifuged to separate serum. Urine samples were obtained via 24-g angiocatheter and collected into 1.5-ml microcentrifuge tubes. For ExtGLuc imaging, mice were anesthetized using isoflurane and retro-orbitally administered 100 μg of water-soluble coelenterazine (NanoLight) in 100 μl of PBS, and images were immediately captured using a Spectrum CT IVIS machine (Xenogen). Images were analyzed using Living Image software (PerkinElmer). The estimated t_1/2 of in vivo functional residence time was determined using first-order exponential decay kinetics using the following formula: t_1/2 = [t × ln(2)]/[ln(N₀/N_t)], where t is the elapsed time interval over which the signal decayed (5 days), and N₀ and N_t are the mean bioluminescence flux at days 0 and 5, respectively.

For necropsy and histopathology studies, mice were euthanized with CO₂, following gross examination, and all organs were fixed in 10% neutral buffered formalin, followed by decalcification of bone in a formic acid solution (Surgipath Decalcifier I, Leica Biosystems). Tissues were then processed in ethanol and xylene and embedded in paraffin in a Leica ASP6025 tissue processor. Paraffin blocks were sectioned at 5 μm, stained with H&E, and examined by a board-certified veterinary pathologist. The following tissues were processed and examined: heart, thymus, lungs, liver, gallbladder, kidneys, pancreas, stomach, duodenum, jejunum, ileum, cecum, colon, lymph nodes (submandibular, mesenteric), salivary glands, skin (trunk and head), urinary bladder, uterus, cervix, vagina, ovaries, oviducts, adrenal glands, spleen, thyroid gland, esophagus, trachea, spinal cord, vertebrae, sternum, femur, tibia, stifle joint, skeletal muscle, nerves, skull, nasal cavity, oral cavity, teeth, ears, eyes, pituitary gland, and brain. In some instances, BCa tumors were stained with human CD45 and quantified using QuPath (Bankhead et al., 2017). For hematologic studies, blood was collected via cardiac puncture into tubes containing EDTA. Automated analysis was performed on an IDEXX ProCyte DX hematology analyzer (IDEXX), and the following parameters were determined: white blood cell count, red blood cell count, hemoglobin concentration, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, red blood cell distribution width standard deviation and coefficient of variance, reticulocyte relative and absolute counts, platelet count, platelet distribution width, mean platelet volume, and relative and absolute counts of neutrophils, lymphocytes, monocytes, eosinophils, and basophils. For serum chemistry, blood was collected via cardiac puncture into tubes containing a serum separator. Serum chemistry was then performed on a Beckman Coulter AU680 analyzer (Beckman Coulter), and the concentration of the following analytes was determined: alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, creatine kinase, gamma-glutamyl transpeptidase, albumin, total protein, globulin, total bilirubin, blood urea nitrogen, creatinine, cholesterol, triglycerides, glucose, calcium, phosphorus, chloride, potassium, and sodium.

Statistical analysis was performed using GraphPad Prism (GraphPad Software). All statistical tests are two-tailed. Unless otherwise indicated, log-rank Mantel–Cox tests were used for survival curves and an unpaired t test was used for comparison of experimental groups with controls.

Fig. S1 provides an antigen landscape analysis of potential therapeutic targets in BCa detailing ZPLD1, TPTE, ADAM2, and known antibody-based targets across various histological subtypes, treatment-resistant tumors, and normal bladder and other human tissues. Fig. S2 characterizes phenotypic markers and in vitro cytotoxicity of 3a5-28z and MSLN-28z CAR T cells. Fig. S3 demonstrates limited recognition and cytotoxicity of 4H11-28z CAR T cells against BCa cell lines and PDTOs. Fig. S4 contains localization studies of MSLN-ExtGLuc in the UM-UC-3 heterotopic and orthotopic BCa models and confirms intravesical adoptive transfer of B7H3-28z CAR T cells results in effective tumor control. Fig. S5 evaluates the safety of intravesical adoptive transfer in TRP-1 autoimmune and CD19-directed B cell aplasia models demonstrating reduced systemic toxicity compared with IV adoptive transfer. Data S1 lists supplementary amino acid sequences.

All data generated in this study are presented in the manuscript and/or supplementary material. Any further information required for replicating experimental procedures will be made available by the corresponding authors upon reasonable request.

We thank L. Morgado-Palacin for her helpful input and assistance with manuscript editing and figure refinement.

This work was supported in part by the New York Academy of Medicine Ferdinand C. Valentine Fellowship Award to P. Abrahimi, Swim Across America, the Ludwig Institute for Cancer Research, and the Parker Institute for Cancer Immunotherapy (C-04306).

Author contributions: Parwiz Abrahimi: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, validation, visualization, and writing—original draft, review, and editing. Jonathan F. Khan: conceptualization, data curation, formal analysis, investigation, validation, and writing—review and editing. Alyssa Duren-Lubanski: formal analysis, investigation, methodology, and validation. Winson Cai: investigation and methodology. Yacine Marouf: formal analysis, software, visualization, and writing—review and editing. Nan Chen: investigation. Daniel Hirschhorn: data curation and investigation. Renata Mammone: data curation, formal analysis, methodology, visualization, and writing—review and editing. Ileana C. Miranda: investigation, resources, visualization, and writing—review and editing. Jacob E. Tallman: data curation, investigation, resources, and writing—review and editing. Alejandra Vela-Moreno: data curation and investigation. Mohamad Hamieh: resources. Bishoy M. Faltas: conceptualization and writing—original draft, review, and editing. Thomas M. Carroll: data curation, formal analysis, project administration, resources, visualization, and writing—review and editing. Micaela L. Everitt: investigation and methodology. Hari K. K. Subramanian: investigation. Hikmat A. Al-Ahmadie: data curation, investigation, and writing—review and editing. Olivier Elemento: funding acquisition, investigation, supervision, and writing—review and editing. Benjamin D. Hopkins: methodology and resources. Douglas S. Scherr: conceptualization, data curation, formal analysis, investigation, methodology, resources, supervision, validation, visualization, and writing—review and editing. Renier J. Brentjens: conceptualization, funding acquisition, investigation, methodology, project administration, resources, supervision, and writing—original draft, review, and editing. Jedd D. Wolchok: conceptualization, funding acquisition, investigation, methodology, project administration, resources, supervision, visualization, and writing—original draft, review, and editing. Taha Merghoub: conceptualization, funding acquisition, investigation, methodology, project administration, resources, supervision, and writing—original draft, review, and editing.