Mesothelin-Specific CAR T Cells Target Ovarian Cancer

Ovarian cancer is the most lethal gynecologic malignancy and the fifth leading cause of cancer-related death among women. The majority of patients are diagnosed at advanced disease stages (stage III–IV) and the five-year survival for these patients is 29% (1). Current standard line of treatment for ovarian cancer consists of surgical tumor debulking complemented by taxane- and platinum-based chemotherapy (2), in certain cases in combination with Avastin (bevacizumab). Maintenance with PARP-inhibitors has proven an effective treatment option in advanced or recurrent disease settings, particularly for approximately 20% of the patients with a tumor-associated Brca mutation (3). Despite radical treatment regimens, the majority of patients with advanced ovarian cancer unfortunately suffer from disease recurrence leaving them in need of a therapeutic alternative. While significant clinical advances have been achieved within the field of cancer immunotherapy in recent years, only Avastin is FDA approved to treat ovarian cancer (4, 5).

Targeted immunotherapies using adoptive cell transfer of engineered autologous chimeric antigen receptor (CAR) T cells have shown remarkable clinical results in hematologic malignancies (6, 7). Advances made in CAR design by the introduction of an intracellular costimulatory domain have resulted in remarkable responses in patients with refractory CD19⁺ malignancies. CD19-directed CAR T-cell therapy with costimulatory domain CD28 (axicabtagene ciloleucel) and 4–1BB (tisagenlecleucel) are FDA and European Medicines Agency approved to treat acute lymphoblastic leukemia and non-Hodgkin lymphoma (7–10).

In contrast, CAR T-cell approaches against solid tumors have so far only led to marginal responses (11). Hurdles for efficient CAR T-cell therapy against solid tumors include impaired CAR T-cell proliferation and persistence in vivo, the immunosuppressive tumor microenvironment, limited CAR T-cell trafficking to the tumor, and long-term deterioration of CAR T-cell functionality (11).

A major bottleneck for CAR T-cell therapy in solid tumors is the limited availability of suitable tumor-associated antigens. Ideally such an antigen is homogeneously overexpressed on malignant cells with minimal or no expression levels on healthy cells, avoiding off-tumor on-target toxicity. Antigens such as mesothelin (MSLN, a glycoprotein), MUC16 (a transmembrane mucin glycosylated protein), folate receptor-α (a carbonyl phosphatidylinositolanchoring protein), and HER2 are overexpressed in ovarian cancer and therefore represent attractive CAR T-cell targets (12, 13). CAR T-cell clinical trials have been initiated against all these targets, with MSLN being the most commonly targeted antigen. MSLN is overexpressed on several solid tumors other than ovarian cancer, including malignant pleural mesothelioma (MPM) and pancreatic ductal adenocarcinoma, with low expression levels on normal mesothelial tissues (pleura, pericardium, peritoneal mesothelial cells; ref. 13).

The application of MSLN-CD28z–directed CAR T cells in an orthotopic preclinical model of MPM has shown promising efficacy (14). We established an orthotopic model of ovarian cancer to compare the in vivo efficacy of fully human MSLN-directed CAR T cells with either a CD28 (M28z) or a 4–1BB (MBBz) costimulatory domain. CD28 and 4–1BB costimulatory domains are the most commonly used costimulatory domains in CAR T-cell design and CAR T cells using either domain have shown great potency in expansion, persistence, and activity in vivo and in vitro in several reports (15). The trafficking and phenotype of CAR T cells was analyzed at different time points. In addition, we identified different tumor immune escape mechanisms such as tumor antigen loss and upregulation of coinhibitory ligands. We found that treatment of SKOV3 tumor–bearing mice with M28z CAR T cells resulted in significantly prolonged survival, while MBBz CAR T cells induced persistent remissions in some mice. This study further highlights the importance of CAR T-cell design to optimize persistence and limit functional exhaustion of CAR T cells.

MSLN-directed CAR constructs, containing a human MSLN-specific scFv (m912; ref. 14) followed by either the CD28-CD3ζ or 4–1BB-CD3ζ signaling domain and linked to a truncated EGFR (EGFRt) sequence by the 2A self-cleaving peptide [kindly provided by Prof. M. Sadelain, Memorial Sloan Kettering Cancer Center (MSKCC), New York, NY]. T cells from healthy donors were transduced with γ-retroviral vectors encoding M28z or MBBz CAR constructs (Fig. 1A). The anti-human CD19 CAR with a CD28 costimulatory domain (Supplementary Fig. S1A), was used as control (generously donated by Prof. S. Rosenberg, NCI, Bethesda, MD). CAR transduction efficiency was determined by flow cytometric analysis of EGFRt cell surface expression, while CAR expression was detected by anti-scFv staining (Fig. 1B). Retroviral supernatant production was performed according to previous published reports (14, 16, 17). Peripheral blood mononuclear cells were isolated from healthy donor buffy coats (Karolinska University Hospital, Huddinge, Sweden), activation and viral transduction of T cells was performed as described previously (16).

The Swedish Ethical Review Agency (Etikprövningsmyndigheten) approved the collection of samples from patients with metastatic high-grade serous carcinoma (HGSC; 2016/1197–31/1, 2016/2060–32), and the patient information is provided in Moyano-Galceran and colleagues report (18). SKOV3 and OVCAR3 tumor cells (human ovarian adenocarcinoma, ATCC, HTB-77 and HTB-161, respectively) were retrovirally transduced with human MSLN variant 1 (SFG vector, gifted by Prof. M. Sadelain, MSKCC, New York, NY) to ensure stable MSLN expression. These cells were subsequently retrovirally transduced with the GFP/firefly luciferase fusion protein (SFG vector, gifted by Prof. M. Sadelain, MSKCC, New York, NY) to generate MSLN⁺GFP⁺ffLuc⁺ ovarian tumor cell lines. SKOV3 MSLN⁺GFP⁺ffLuc⁺ ovarian tumor cells were polyclonal pools sorted by FACS (BD FACSAria, Becton Dickinson) based on expression of GFP and MSLN. SKOV3 tumor cells were cultured in McCoy 5A culture medium (Sigma-Aldrich) supplemented with 10% FBS (GE Healthcare) and OVCAR3 tumor cells were cultured in RPMI1640 (Gibco) supplemented with 20% FBS and 0.01 mg/mL insulin (Sigma-Aldrich). OVCAR4 cells transduced with the pMx-Renilla luciferase–GFP as described previously (18), were kindly provided by Prof. Kaisa Lehti (Karolinska Institutet, Stockholm, Sweden). GFP⁺ OVCAR4 were retrovirally transduced with human MSLN variant 1 as described in the Methods section. OVCAR4 MSLN⁺GFP⁺Luc⁺ were sorted using the SONY MA900 cell sorter (Sony Biotechnology Inc.) to ensure a pure population of OVCAR4 MSLN⁺GFP⁺Luc⁺. OVCAR4 MSLN⁺GFP⁺Luc⁺ tumor cells were cultured in RPMI1640 supplemented with 10% FBS. Transduced SKOV3 and OVCAR4 cells were tested for microbiology and Mycoplasma the day of microsurgery (MycoAlert Mycoplasma Detection Kit, Lonza).

All animal experiments were approved by the Swedish Board of Agriculture (Jordbruksverket, Dnr. 5.2.18–7079/16 and 19354–2019). Six- to 8-week-old female NOD scid gamma (NSG) mice were purchased from Charles River (more information is provided in Supplementary Materials). To establish orthotopic ovarian tumors, microsurgery was performed and 0.5E6 SKOV3 MSLN⁺GFP⁺ffLuc⁺ cells (passage 36 and 37) were injected into the bursa of the left ovary as described previously (19). Mice were randomly grouped and received 1E6 CAR T cells (2E6 total T cells) through tail vein injections at 26–28 days of tumor engraftment. CD19 CAR T cells were administered at the same dose (control for antigen specificity and xeno-reactivity). Tumor burden, general status and weight (Supplementary Fig. S1B) were assessed weekly through clinical examinations and bioluminescence imaging (BLI). BLI data were acquired and analyzed using the Living Image 4.3.1 software (PerkinElmer). Mice were intraperitoneally injected with 150 mg/kg d-luciferin (PerkinElmer) in PBS to quantify the GFP⁺ffLuc⁺ bioluminescence signal using the In Vivo Imaging System (IVIS, PerkinElmer). To longitudinally analyze the phenotype of tumors and CAR T cells, some mice from each group were sacrificed at intermediate time points (10–11 and 21–25 days post-CAR treatment, regardless of tumor burden). The remaining mice were kept for survival analysis and sacrificed upon reaching humane endpoints due to high tumor burden; M28z (n = 11), MBBz (n = 10), and CD19z (n = 8). Two CD19z mice were sacrificed for reasons (weight variation) not related to high tumor burden or treatment and were excluded from survival analysis. In both mice, weight changes were associated with changes in the posture, general condition (mice activity level), and body shape, indicating that the mice have reached the humane endpoint according to our ethical permit. Mice surviving ≥7 days than control CD19z mice were classified as responders. The tumor burden, monitored by BLI signal 23–27 days post tumor inoculation (1–3 days prior to CAR administration), was used as the threshold to determine treatment response; a BLI signal below threshold was classified as tumor regression and mice in which regression was maintained for ≥14 days (i.e., for at least 2 consecutive weekly BLI measurements) were deemed to be in remission. Mice were sacrificed when tumor was deemed to be 1.5 cm by clinical examination and/or development of ascites (except for mice that were sacrificed at scheduled intermediate time points). In one of the last surviving mice in the control group (CD19z), small masses were found in the peritoneal cavity, liver, and intestine, indicating transcelomic metastasis. In addition, tumor infiltrates were found in the lungs of several mice from all groups, indicative of hematogenous metastasis. These findings were confirmed by IVIS, hematoxylin and eosin staining, FACS analysis, and/or physical examination upon sacrifice.

Every 3 weeks, blood samples were collected from the tail veins. More information on organ collection at sacrifice and processing are found in Supplementary Materials.

0.5E6 OVCAR4 MSLN⁺GFP⁺Luc⁺ were injected by microsurgery as described above for SKOV3 cells. Following randomization of the three groups, 13 mice received 1E6 CAR T cells (M28z, MBBz, and CD19 CAR) 16 days post microsurgery. Tumor burden, general status, and weight were assessed through regular clinical examinations and weekly BLI.

For CAR staining; cells were stained with a primary biotinylated anti-human Fab (for MSLN-CARs) or anti-mouse Fab (CD19-CARs) antibody (Jackson ImmunoResearch Lab) then blocked with mouse gamma globulin (Jackson ImmunoResearch Lab). Alternatively, to assess CAR transduction efficiency, MSLN CAR T cells were stained with a primary biotinylated anti-EGFRt/cetuximab antibody (R&D Systems). Cells were washed twice prior to staining with extracellular antibodies. The antibody panels used for surface staining of tumor cells can be found in Supplementary Material. Intracellular cytokine staining was performed as described previously (16). Cells were acquired on a CytoFlex (Beckman Coulter) or FACSCanto (Becton Dickinson) flow cytometer and data were analyzed using FlowJo software (FlowJo).

Details on cytotoxicity assays, IHC quantification of tumor microarrays (TMA), measurement of soluble MSLN-related peptides (SMRP), and cytokines can be found in Supplementary Material.

Data analysis was performed with GraphPad Prism software. Data are all represented as median. When comparing two groups of mice, the Mann–Whitney test was used and for comparing multiple groups, nonparametric Kruskal–Wallis test followed by Dunn post-test was used. Wilcoxon matched pairs signed rank test was performed to compare two groups of paired samples. Correlation was assessed by linear regression (r² value) after log transformation of the data (0 values were ascribed a nominal value of 0.01). Kaplan-Meier survival was analyzed using the log-rank test for survival. The significance threshold was set at 0.05. Online CIMminer was used to construct a two-dimensional clustered image map using the Euclidian distance method and the average linkage cluster algorithm to cluster both human plasma level of soluble factors and individual mice. Soluble factors were log transformed before hierarchical cluster analysis of mice and plasma levels of soluble factors.

Transduced T cells displayed comparable CAR surface expression (scFv, 48.2% and 52.8%) and transduction rates (EGFRt, 61,9% and 66.7%). Despite donor heterogeneity, both M28z- and MBBz-transduced T cells mediated superior antigen-specific killing of OVCAR3 and SKOV3 cells transduced with MSLN compared with untransduced T cells (Fig. 1C). M28z- and MBBz-transduced CD4⁺ and CD8⁺ T cells specifically produced IFNγ and TNF, while IL2 production was minimal when cocultured with MSLN⁺ SKOV3 cells (Fig. 1D).

Following 26–28 days of tumor engraftment of MSLN⁺GFP⁺ffLuc⁺ SKOV3 cells (Supplementary Fig. S2A), mice received M28z or MBBz CAR T cells intravenously (Fig. 2A). CD19 CAR T cells were administered at the same dose (control for antigen specificity, and xeno-reactivity).

Treatment with M28z CAR T cells significantly augmented survival relative to the control group (P = 0.02; Fig. 2B and C). While the majority (64%) of mice responded to M28z CAR T-cell therapy, the treatment effect was transient and merely resulted in delayed tumor growth. Only 2 of 7 M28z responder mice showed transient tumor regression and the tumor burden increased within 7 days after the initial response (Fig. 2B and C). The difference in relative tumor growth at an earlier (+13–17 days) and later (+36–38 days) time point post-CAR treatment revealed that M28z CAR T cells mediated an initial quicker treatment effect compared with MBBz CAR T cells, but this advantage was lost at later time points (Supplementary Fig. S2B).

Although the proportion of responding MBBz mice was lower (50%) compared with M28z mice, the treatment was extremely effective in the responding mice; with 4 of 10 mice reaching remission. The BLI signal decreased below threshold (BLI before CAR injection) at some time points in these mice, highlighting that when antitumor immunity was effective it led to clearance of the tumor (Fig. 2A–C; Supplementary Fig. S2C). On the basis of the BLI signal, however, all 4 mice eventually relapsed. Two of the MBBz responder mice were deemed to be under long-term remission, because tumor regression was maintained up to 63 days post-CAR treatment. Of note, no palpable tumor was found in the two mice under long-term remission and no SKOV3 tumor cells were detected by flow cytometry in the ovary, spleen, and lung upon sacrifice of these mice, indicating disease control upon termination of the experiment (91 days post-CAR infusion).

Interestingly, in both the control and MSLN-CAR–treated groups, the serum levels of the human SMRP correlated with tumor burden (as detected by BLI, Fig. 2D; Supplementary Fig. S2D).

To assess the in vivo efficacy of the MSLN CAR T cells in a model of HGSC, OVCAR4 MSLN⁺GFP⁺Luciferase⁺ ovarian tumors were established in NSG mice (same protocol followed as for SKOV3 tumor inoculation, and with CAR T-cell injection 16 days after microsurgery of MSLN⁺ OVCAR4 cells, see Supplementary Material). The OVCAR4 model was more aggressive with a faster tumor growth than the SKOV3 model (Fig. 2E). In this HGSC model, mice treated with MSLN-targeted CAR T cells showed a significant prolonged survival (P < 0.01) with a median survival of 48 days (MBBz group) and 49 days (M28z group) as compared with the control group (CD19z group) that showed a median survival of 35 days (Fig. 2F).

We compared the expression level of MSLN protein in OVCAR4 and SKOV3 cells transduced with MSLN and tumors from ascites. MSLN protein level in MSLN+ SKOV3 cells (13.6) was within the range of the MSLN protein levels (2.8 to 18.0) measured in tumor cells isolated from the ascites of patients with HGSC, while OVCAR4 cells showed lower MSLN protein level (1.1; Fig. 3A).

To assess MSLN expression on tumor cells in vivo, SKOV3 tumor cells (identified as CX3CR1⁺ and GFP⁺ cells) were isolated from ovarian tissues in 55 of 59 mice and assessed by flow cytometry for MSLN surface expression and phenotyping. No SKOV3 tumor cells were detected in the ovaries of two mice sacrificed at control time points.

Upon CAR treatment, MSLN surface expression was slightly decreased (83.5%) compared with expression at the time of tumor inoculation (95%–97.3%; Fig. 3B). MSLN surface expression progressively decreased throughout the course of the study in all groups (Fig. 3C). Already 10–15 days post-CAR treatment, MSLN surface expression decreased to 62.7% in M28z mice while expression remained high in MBBz mice (median 92.4%, P = 0.009). However, IHC of TMAs for MSLN (which detects extra and intracellular expression) showed a high frequency of MSLN⁺ cells among nucleated cells ranging from 59.8% to 96.5% at time of sacrifice (regardless of the time or group). A low frequency of MSLN⁺ cells (28.4%) was found in sections of a MBBz mouse that achieved long-term remission (Fig. 3D). Furthermore, when SKOV3 cells isolated from primary tumors were placed in ex vivo culture, we observed that >80% of SKOV3 cells reexpressed cell surface MSLN (Supplementary Fig. S3A). Surface MSLN expression was not homogenously lost on the metastatic tumor sites found in the peripheral organs (spleen, lung, and/or ascites) of four M28z mice and three relapsed (BLI signal below threshold prior to tumor recurrence) MBBz mice (Fig. 3E).

We observed a gradual PD-L1, HLA-DR, and galectin-9 upregulation in MSLN⁺ tumor cells in M28z and MBBz mice (Fig. 3F; Supplementary Fig. S3B). In M28z mice, PD-L1 and HLA-DR expression was significantly higher on MSLN⁺ compared with MSLN- SKOV3 cells 10–25 days post-CAR infusion and this upregulation on MSLN⁺ SKOV3 cells became more pronounced ≥31 days. In MBBz mice, PD-L1 and HLA-DR expression was only significantly higher on MSLN+ SKOV3 cells ≥31 days post-CAR T-cell infusion. The expression levels of HLA-DR and PD-L1 were negligible on MSLN+ SKOV3 tumor cells in the CD19z CAR group (Supplementary Fig. S3C). A significant increase in galectin-9 expression was only observed on the MSLN⁺ tumor cells from M28z mice sacrificed ≥31 days after CAR treatment.

Next, we determined the expression of immune checkpoint ligands on MSLN⁺ SKOV3 cells over time. The frequency of MSLN⁺ tumor cells negative for the expression of inhibitory ligands decreased substantially from the first intermittent time point to >45 days post-CAR treatment (Fig. 3G). After ≥31 days, MSLN+ tumor cells from M28z and MBBz mice presented similar (co-)expression profiles of PD-L1, HLA-DR, and galectin-9. There was a sharp increase in the (co-)expression of PD-L1, HLA-DR, and galectin-9 on the tumors from mice sacrificed >45 days after CAR treatment compared with tumors collected between 31 and 40 days.

On MSLN⁺ metastases, PD-L1 expression tended to decrease. The opposite was found for HLA-DR, with comparable or elevated expression levels on metastatic MSLN⁺ tumor cells. Galectin-9 expression was almost absent on lung and spleen metastases (Supplementary Fig. S3B).

Plasma levels of human soluble factors secreted by T cells (IFNγ, GrzA, TNF, FasL, IL3, IL2, TNF and GrzB) and/or SKOV3 cells (MMP-2, IL8, Fas, VEGF, and SMRP) were evaluated. In mice sacrificed due to high tumor burden, higher levels of IFNγ, TNF, Fas (P < 0.05) and FasL and GrzA (P = 0.08) were found in MSLN-CAR–treated mice compared with controls (CD19z, Supplementary Fig. S4A). Unbiased hierarchical clustering for soluble factors revealed three distinct clusters in M28z- and MBBz-treated mice: cluster A, B, and C (Fig. 4A). Cluster A (n = 12) was characterized by overall low levels of both T cell and SKOV3 cell–derived soluble factors and mainly included mice that were sacrificed at the first intermediate time point with lower tumor burden (median of 11 days post CAR treatment). Of note, mouse 36, one of the MBBz mice under long-term remission (no detectable tumor upon sacrifice), clustered in cluster A. Mice in cluster B (n = 15, median time sacrifice of 24 days) were characterized by increased production of tumor secreted factors (MMP-2, IL8, SMRP, and Fas) and factors of T-cell origin. Cluster C (n = 10, median time sacrifice of 54 days) was composed of mice with a similar tumor-associated soluble factor profile as group B (increased plasma levels of MMP-2, Fas, IL8, SMRP) combined with significantly higher levels of IFNγ, GrzA, Fas. CD19 CAR treated mice clustered in cluster A (n = 7) and cluster B (n = 8) with a median sacrificing time of 11 and 28 days post CAR treatment, respectively (Supplementary Fig. S4B). Taken together, these data show that levels of plasma soluble factors were dependent on the time of sacrifice, which reflected tumor burden (clusters A and B) and increased T-cell responses (cluster C).

This significant difference in cytokines produced by T cells between cluster B and C was remarkable among the mice kept for survival analysis: higher levels of IFNγ, GrzA, GrzB, TNF, FasL (P < 0.001, Fig. 4B), and of Fas, IL10, TGFα (P < 0.05, Supplementary Fig. S4C) were observed in cluster C. Mice from cluster C, characterized by high T-cell produced soluble factors, showed a prolonged survival (P = 0.003, difference median of 23 days) as compared with those from cluster B (Fig. 4C).

Mice from cluster B had a higher frequency of triple-negative coinhibitory ligands tumor cells compared with mice from cluster C (P < 0.001). Reversely, MSLN⁺ tumor cells in the mice from cluster C had a higher expression of HLA-DR alone or in combination with PD-L1 as compared with mice from cluster B (P < 0.05, Fig. 4D). A similar profile was found in MSLN- tumor cells (Supplementary Fig. S4D).

M28z-treated mice had the highest proportion of infiltrating T cells to tumor cells ≥10–25 days post-CAR treatment (Fig. 5A). M28z T cells also had higher CX3CR1 expression than MBBz T cells ≥10–25 days after CAR administration (P = 0.004). Notably, that profile changed drastically overtime with a higher CX3CR1 expression in MBBz T cells than M28z T cells ≥31 days post-CAR treatment (27.75% vs. 16.15% respectively, Supplementary Fig. S5A).

The proportion of transduced M28z and MBBz T cells (EGFRt+) increased rapidly within 10–25 days following treatment (Fig. 5B). As tumor burden increased (≥31 days), the fraction of transduced T cells in tumors decreased to a median of 77.45% for M28z and 82.3% for MBBz (P = 0.011, Supplementary Fig. S5B). Of note, CAR surface expression (scFv+) was lower relative to the fraction of EGFRt+ T cells for both M28z (P = 0.004) and MBBz (P = 0.016 and P = 0.039), suggesting CAR internalization.

The CD4⁺/CD8⁺ ratio in M28z- and MBBz-transduced T cells was 0.95 and 1.48, respectively, upon CAR infusion. The ratio (0.39) notably decreased, in tumors from mice sacrificed ≥10–25 days post-CAR treatment, indicating a shift toward CD8⁺ T cells in MBBz-transduced T cells as compared with the treatment start and M28z-transduced T cells (0.697, P < 0.001; Fig. 5C). At later time points (≥31 days), the CD4⁺/CD8⁺ ratio in MBBz CAR T cells increased back to 0.639 (P = 0.003).

IHC analysis showed a higher infiltration of CD8⁺ T cells in the primary ovarian tumor in mice from cluster C compared with those in cluster B (P = 0.004, Fig. 5D). There was no difference in the frequency of infiltrating CD4⁺ CAR T cells. The highest CD8⁺ T-cell infiltration was observed in the two MBBz mice that achieved long-term remission (M36 and M38).

The CAR T-cell memory phenotype composition changed rapidly in vivo as the naïve (CD45RA⁺CCR7⁺) and central memory (Tcm, CD45RA⁻CCR7⁺) subsets diminished after 10–25 days in both M28z and MBBz mice (Supplementary Fig. S5C). The majority of T cells were effector memory cells (Tem, CD45RA-CCR7-) at time of sacrifice.

Both M28z and MBBz T cells showed evidence of trogocytosis of the MSLN antigen from SKOV3 cells as detected by MSLN staining on T-cell surface (Supplementary Fig. S5D). MBBz T cells mediated significantly more trogocytosis 10–25 days post-CAR treatment compared with M28z (P = 0.004) and a similar trend was observed ≥31 days. There was more trogocytozed MSLN antigen on transduced (EGFRt+) M28z and MBBz T cells relative to untransduced (EGFRt-; Fig. 5E). There was significantly more trogocytosis by MBBz CD4⁺ than CD8⁺ T cells (P = 0.016).

Prior to infusion, M28z- and MBBz-transduced T cells displayed similar cell surface frequencies of LAG3 and TIM3, while PD-1 expression was the highest on the MBBz CAR T cells (Supplementary Fig. S6A). Expression of all coinhibitory receptors (except for PD-1 expression on MBBz T cells) was significantly increased on transduced (EGFRt+) compared with untransduced (EGFRt-) T cells in MSLN-CAR–treated mice (Fig. 6A). More M28z CAR T cells expressed PD-1 and TIM3 relative to MBBz CAR T cells. For both M28z and MBBz, PD-1 levels were elevated on CD4⁺ compared with CD8⁺ CAR T cells (Fig. 6B). The opposite was found for LAG3 and TIM3, with higher expression levels on CD8⁺ CAR T cells (Fig. 6B).

We analyzed the frequency of single, double, or triple positive PD-1, LAG3, and TIM3 cells (Fig. 6C). Already 10–15 days post-CAR treatment, PD-1, TIM3, and LAG3 triple positive CD4⁺ and CD8⁺ MBBz CAR T cells (median of 26.6% and 12.2%, respectively) were found, suggesting the onset of exhaustion (for technical reasons we could not analyze M28z CAR T cells at this time point). The frequency of triple coinhibitory receptor positive MBBz CAR T cells swiftly increased (∼ 2-fold) after 21–25 days and similar frequencies were observed within M28z CAR T cells 21–25 days posttreatment. Minor fluctuations were observed in the frequency of PD-1⁺LAG3⁺TIM3⁺ CD8⁺ T cells on later time points (≥31 days), as between 44% and 50% of both M28z and MBBz CAR T cells remained triple positive. In the longest surviving M28z mice (≥45 days), the median proportion of triple coinhibitory receptor–positive CD4⁺ T cells declined to 39% while the fraction of double positive PD-1⁺TIM3⁺ T cells increased to 36.6%. CAR T cells were detected in the ovary of the MBBz-treated mice that achieved long-term remission. Notably, as shown in Fig. 6C for one long-term remission mouse these MBBz CAR T cells had the least exhausted phenotype from all analyzed samples as 39.6% of CD4⁺ and 34.9% of CD8⁺ T cells did not express any coinhibitory receptors while only 4.4% (CD4⁺) and 27.8% (CD8⁺) were triple positive.

Among the mice kept for survival analysis, grouping mice treated with MSLN-CAR T cells according to the hierarchical clusters defined using soluble factors revealed that CD4⁺ CAR T cells of cluster B mice had the highest frequency of triple coinhibitory receptor positive (P = 0.006). CD4⁺ CAR T cells of cluster B were characterized by a higher frequency of LAG3 alone or in combination with TIM3 compared with cluster C mice (Fig. 6D). Interestingly, CD4⁺ CAR T cells found in the tumor of cluster C mice presented a higher expression of PD-1 alone or coexpressed with TIM3 compared with cluster B mice. CD8⁺ CAR T cells did not present significant differences (Supplementary Fig. S6B).

Interestingly, the T cells found in the periphery (ascites, lung, spleen, and blood) displayed a distinct phenotype compared with primary tumor (SKOV3 model) infiltrating T cells. The proportion of M28z transduced T cells (EGFRt+) declined from (median) 88% in the primary tumor to 45% in the lung and spleen and 18% in the blood (P = 0.002; Fig. 7A). On the other hand, the proportion of MBBz-transduced cells was similar for all sites. In metastases, the T-cell to tumor-cell ratio revealed that the proportion of infiltrating M28z T cells was higher as compared with MBBz (Fig. 7B). No significant differences were observed in PD-1 expression on peripheral M28z and MBBz CAR T cells relative to tumor-infiltrating CARs. There was a trend toward lower PD-1 expression on lung- and spleen-derived M28z CAR T cells compared with the primary tumor (P = 0.067; Fig. 7C). Frequency of LAG3⁺ M28z CAR T cells in the lung and spleen was significantly decreased (P < 0.001) relative to tumor-infiltrating CAR T cells. Furthermore, the frequency of TIM3⁺ M28z CAR T cells was considerably lower in lung, spleen, and blood as compared with primary tumor. For MBBz mice, the difference was not as pronounced as for the peripheral M28z CAR T cells but a similar trend of decreased LAG3 and TIM3 expression on lung- and spleen-derived CAR T cells was observed relative to tumor-infiltrating CAR T cells.

To differentiate between phenotypical and functional exhaustion, CAR T cells were isolated from tumor (SKOV3 model), lung and/or spleen tissue for functional assays. Following overnight resting, PD-1, LAG3, and TIM3 expression decreased significantly on M28z CAR T cells with a similar trend in MBBz CAR T cells (Fig. 7D). Following 1 to 4 days of rest ex vivo, CAR T cells isolated from different tissues were cocultured with SKOV3 MSLN⁺ cells. Despite low sample numbers (n = 4–14), MSLN⁺ SKOV3 lysis increased at higher ratios (P = 0.013, Fig. 7E). Interestingly, at the 2:1 E:T ratio MBBz CAR T cells (median lysis 35.2%) derived from the primary tumor showed a trend toward increased SKOV3 lysis compared with M28z CAR T cells (median lysis 3.6%). Killing was similar for spleen- and/or lung-derived M28z and MBBz-transduced T cells. The supernatants of the tumor-derived CAR T cells cocultured with SKOV3 (2:1 ratio) were used to detect soluble Fas, FasL, IFNγ, GrzA, GrzB, and TNF. The median Fas, FasL, IFNγ, TNF, and GrzB production was higher by MBBz than M28z CAR T cells, with IFNγ and TNF being significantly elevated (P < 0.05, Fig. 7F; Supplementary Fig. S6C). In addition, there was a positive correlation between GrzB and FasL levels in the supernatant (r² = 0.7099, Supplementary Fig S6D).

There is an urgent need for novel treatment modalities to improve the outcome for patients with advanced ovarian cancer, particularly for the patients who do not respond to conventional therapies. Earlier reports showing antitumor activity and improved survival in subcutaneous and intraperitoneal preclinical models raised CAR T-cell therapy to a promising treatment modality for those patients (20, 21). Furthermore, clinical trials using MSLN-directed CAR T cells have so far demonstrated no on-tumor/off-target toxicity (22–24). Yet, phase I studies using MSLN-targeted CAR T cells with murine-derived scFv in patients with solid tumors showed potential but limited effect (25, 26).

In this study, we show that fully human anti-mesothelin CD28 (M28z) and 4–1BB (MBBz) CAR T cells were able to control ovarian cancer in vivo using two different cell lines (SKOV3 and OVCAR4) in an orthotopic preclinical model of ovarian cancer. M28z CAR T cells provided prolonged survival of SKOV3 tumor bearing mice but eventually disease progressed in all mice. On the other hand, while MBBz CAR T-cell treatment nonsignificantly increased the median survival of SKOV3 tumor–bearing mice, 40% of the MBBz-treated mice reached remission with two mice maintaining tumor control until termination of the experiment. Despite lower response rates, MBBz CAR T cells thus can induce long-term remission, which was not observed in mice treated with M28z CAR T cells. These results are in concordance with previous findings in preclinical models where CD19-CD28z CAR T cells more rapidly eliminated tumors compared with CD19–4-1BBz CAR T cells, while CD19–4-1BBz CAR T cells had longer persistence (27). Also, in solid tumor settings such as MPM and prostate cancer, 4–1BB-containing MSLN- and PSCA-directed CAR T cells, showed more durable antitumor responses compared with CD28-containing CAR T cells (10, 14).

Tumors employ several immune escape mechanisms such as antigen loss and expression of inhibitory ligands. Longitudinal analysis of SKOV3 cells (from the primary tumor) revealed a progressive loss overtime of MSLN surface expression that occurred in all groups (including control mice), but more rapidly in M28z CAR T-cell–treated mice. This suggests that M28z CAR T cells exerted an increased and/or earlier pressure on tumor cells. Loss of MSLN cell surface expression is most likely not a result of increased shedding because SMRP (which can be generated either by alternative splicing or cleavage of MSLN variant 1; ref. 28) levels correlated with tumor burden progression in all groups and did not appear to accumulate overtime as a result of increased shedding. Furthermore, IHC detection showed that most of the nucleated cells in the tumors analyzed were positive for MSLN, suggesting a cell surface downregulation and intracellular accumulation. In addition, the ex vivo reexpression of cell surface MSLN indicates that MSLN⁺ cells were not selected against but rather that internalization/recycling might have been responsible for loss of cell surface MSLN.

MSLN surface expression by SKOV3 cells varied in different metastasized tissues; high levels of MSLN⁺ SKOV3 cells were detected in the ascites and lungs, suggesting an organ-specific modulation of MSLN surface expression. Our data revealed that MBBz and M28z CAR T cells isolated from primary tumors had acquired MSLN surface expression by trogocytosis, which in turn could partially explain the earlier loss of MSLN surface expression by tumor cells in M28z CAR T-cell–treated mice. Trogocytosis has previously been described to promote tumor antigen escape in response to CD19-CAR T-cell treatment but also fratricide killing and exhaustion of T cells (29). The loss of cell surface MSLN represents a limitation to MSLN-targeted therapy. In addition, due to the heterogeneity of MSLN expression (not all tumor cells express MSLN and within MSLN⁺ tumors the expression levels may vary), not all tumor cells will be targeted by MSLN CAR T cells, but activation of MSLN CAR T cells can in turn mediate killing of MSLN⁻ tumor cells via by-stander killing and epitope spreading (30, 31). Interestingly, absence of detection of MSLN cell surface expression by FACS analysis does not necessarily imply complete absence of cell surface expression (due to the limit of detection by FACS) nor precludes specific recognition of very low levels of target antigens by CAR T cells. This is supported by presence of MSLN protein in parental (untransduced) SKOV3 cells while no MSLN was detectable on the cell surface by FACS analysis. In our study, M28z and MBBz CAR T cells were able in vitro to lyse parental SKOV3 cells, albeit to a lower degree than MSLN⁺ SKOV3 cells. These data demonstrate the capacity of M28z and MBBz CAR T cells to recognize and lyse ovarian cancer cells with undetectable MSLN surface expression by FACS. Furthermore, Nerreter and colleagues have shown that CD19 negative myeloma by FACS analysis did express cell surface CD19 at low levels (e.g., <100 molecules/cell), which could be measured by direct stochastic optical reconstruction microscopy, and that CD19 CAR T cells were capable of lysing specifically these very low CD19-expressing myeloma cells (32).

Our data showed that high levels of soluble IFNγ, FasL, TNF, GrzA and B (cluster C) identified MSLN CAR–treated mice with longer survival, as well as higher frequencies of MSLN⁺ tumor cells expressing HLA-DR (alone or in combination with PD-L1) and higher tumor infiltration of CD8⁺ T cells, as compared with mice with lower levels of soluble IFNγ, GrzA and FasL (cluster B). IFNγ and TNF have been shown to induce HLA-DR expression and increased HLA-DR expression may reflect an increased immune infiltration (33, 34). Furthermore, the frequency of tumor-infiltrating CD4⁺ T cells expressing the ligand for HLA-DR and LAG3 was decreased in cluster C mice, suggesting that HLA-DR expression did not induce nor increased T-cell suppression mediated via LAG3. Tumor expression of HLA-DR is correlated with better prognosis in some cancers, and a report recently showed that MHCII⁺ tumors correlated with T cell infiltration and higher expression of immune checkpoint receptors (35). Altogether, this suggests that in our preclinical model of ovarian cancer, HLA-DR expression by tumor cells correlates with greater immune response.

Our study showed in vivo antigen-dependent combined upregulation of PD-1 and LAG3 on tumor-infiltrating MSLN CAR T cells, especially those belonging to cluster B. The expression of PD-1, LAG3, and TIM3 was reversible in the absence of antigen stimulation ex vivo, indicating that expression of immune checkpoint and cytotoxic potential of MSLN CAR T cells can be rescued. MSLN CAR T cells, especially MBBz CAR T cells, isolated from primary tumors showed cytotoxic capacity toward SKOV3 tumor cells ex vivo. We suggest that the higher expression levels of immune checkpoints on M28z CAR T cells contributed to their lower cytotoxic potential upon ex vivo antigen stimulation, implying a critical role of the functional exhaustion of MSLN-directed CAR T cells. The crucial role of exhaustion in MSLN CAR T-cell efficacy was further confirmed by the MBBz-treated mice under long-term remission, as its CAR T cells displayed the least exhausted phenotype. Several strategies exist to overcome CAR T-cell inhibition via immune checkpoints. Recently, a phase I/II study combining MSLN CAR T-cell treatment in MPM with pembroluzimab is ongoing (NCT02414269; ref. 36). Furthermore, combined use of checkpoint blockades targeting PD-1 and LAG3 better restored the effector function of tumor-specific CD8⁺ T cells than blockade of LAG3 or PD-1 alone in human and mouse models (37, 38). It would, therefore, be of utmost interest to combine PD-1/LAG3 checkpoint blockade with MSLN-directed CAR T cells for the treatment of ovarian cancer.

Our orthotopic mouse model of ovarian cancer allowed us to describe many clinically relevant features, as we observed two mechanisms of immune escape by the primary tumor cells, namely the loss of MSLN surface expression and overexpression of immune checkpoint ligands. In parallel, tumor-infiltrating MSLN-directed CAR T cells upregulated PD-1 and LAG3 checkpoint receptor expression in vivo, retained high TIM3 expression levels and demonstrated trogocytic activity. In addition, we identified that HLA-DR expression by tumor cells was increased in mice (cluster C) with longer survival and a higher frequency of tumor infiltrating CD8⁺ T cells.

In conclusion, we show in a preclinical model, that MSLN CD28- and 4–1BB-containing CAR T cells were able to control SKOV3 ovarian cancer with different kinetics, as M28z increased survival significantly, but only MBBz provided long-term remission. In SKOV3 tumor-bearing mice under long-term remission, MBBz CAR T cells showed less signs of exhaustion further emphasizing the beneficial potential of combining checkpoint blockade with MSLN CAR T-cell therapy. In a HGSC model using OVCAR4 cells, both M28z and MBBz CAR T cells significantly prolonged the mice survival.

No disclosures were reported.

E. Schoutrop: Conceptualization, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. I. El-Serafi: Conceptualization, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. T. Poiret: Conceptualization, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. Y. Zhao: Conceptualization, investigation. O. Gultekin: Formal analysis, visualization. R. He: Investigation. L. Moyano-Galceran: Investigation, visualization. J.W. Carlson: Investigation, methodology. K. Lehti: Investigation, visualization, methodology. M. Hassan: Conceptualization, investigation, visualization, methodology. I. Magalhaes: Conceptualization, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Mattsson: Conceptualization, resources, funding acquisition, methodology, writing–original draft.

The authors would like to express their gratitude to Prof. Michel Sadelain (Memorial Sloan Kettering Cancer Center (MSKCC), New York, NY) for providing the vectors encoding the MSLN CAR constructs, MSLN-1 and GFP/firefly luciferase, and helpful discussion, and to Prof. Steven Rosenberg (National Cancer Institute, Bethesda, MD) for providing the vector encoding for the CD19 CAR construct. They are also grateful to Dr. Johan Lundberg and Dr. Gregory Winn for their excellent technical assistance with TMA preparation. The authors thank Doctors Ulrika Joneborg, Emelie Wallin and Elisabet Hjerpe to provide access and information on patient samples. I. El-Serafi was supported by Karolinska Institutet Foundation (KI funds 2018–02344). K. Lehti was funded by Cancerfonden (2018/858) and Vetenskapsrådet (2019–01541). M. Hassan was funded by the Swedish Childhood Cancer Foundation (PR2017–0083) and Radiumhemmets research funding (161082). I. Magalhaes was funded by Cancerfonden (19 0002 FE) and Clas Groschinskys Minnesfond (M18224). J. Mattsson was supported by grants from Cancerfonden (19 0359 Pj 01 H9), Radiumhemmets research funding (181201), and the Mix private donation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.