Publicly available datasets [The Cancer Genome Atlas (TCGA) (https://www.cancer.gov/tcga); Genotype Tissue Expression (GTEx) (25); Human Protein Atlas (HPA) (26)] were accessed through the online resource Recount2 (https://jhubiostatistics.shinyapps.io/recount/) (27). Read alignment against the hg38 reference genome and mRNA quantification were part of the Recount2 pre-processing pipeline. Raw count tables were obtained and combined into one comprehensive dataset. For each distinct primary cancer tissue from the TCGA 30 samples were randomly chosen. Random sampling was also applied for the GTEx dataset, with maximum number of 20 samples, if available. Regarding the HPA dataset, all samples were included (3-5 samples per tissue). The compiled dataset consisted of a total of 2202 samples and was normalized utilizing the EdgeR package and its Relative Log Expression (RLE) method (28, 29) in R (v3.4.3). Finally, the dataset was filtered to retain only those genes showing evidence of expression in ovarian cancer, as defined by a minimum mean of 100 read counts (16855 genes in total). Differential gene expression analysis was performed using the EdgeR package after fitting a quasi-likelihood negative binomial generalized log-linear model to the count data. Genes were defined to be DE in ovarian cancer when they exhibited an absolute minimum fold change (FC) of ≥ 20 and FDR adjusted p-value of ≤ 0.05. Mean expression in ovarian cancer was compared against most of the healthy tissues present in the dataset, only tissues from reproductive organs and tumors were excluded.

Seven solid primary OVCA patient samples derived from different patients (2 – 20 gram) were collected and dissociated using the gentleMACS (Miltenyi Biotec) procedure ( Supplemental Methods ). Also one ascites OVCA patient sample (6*10⁹ cells) and three primary acute myeloid leukemia (AML) samples (65 – 500*10⁹ cells) were collected. Furthermore, various cell lines were expanded up to at least 2*10⁹ cells ( Supplementary Table 3 ). Cell lines transduced with HLA alleles, CLDN6 and/or CTCFL were first enriched for marker gene expression via magnetic-activated cell sorting (MACS) or fluorescence-activated cell sorting (FACS). HLA typing of all samples/cell lines was performed and gene expression was quantified by Quantitative Polymerase Chain Reaction (qPCR) ( Supplemental Methods ).

Cell pellets were lysed and subjected to an immunoaffinity column to collect bound peptide-HLA complexes. Peptides were subsequently separated, fractionated and analyzed by data-dependent MS/MS ( Supplemental Methods ). Proteome Discoverer V.2.1 (Thermo Fisher Scientific) was used for peptide and protein identification, using the mascot search node for identification (mascot V.2.2.04) and the UniProt Homo Sapiens database (UP000005640; Jan 2015; 67,911 entries). Peptides were in-house synthesized using standard Fmoc chemistry and PE-conjugated pMHC-multimers were generated with minor modifications ( Supplemental Methods ).

T cells were cultured in T-cell medium (TCM) and (re)stimulated every 10-14 days with PHA and irradiated autologous feeders ( Supplemental Methods ). OVCA cell lines COV-318/-362.4/-413b/-434/-504/-641 were established at the department of Medical Oncology (LUMC, NL) (30). OVCA cell lines OVCAR-3 and SK-OV-3 were obtained from the ATCC and A2780 from the ECACC. Primary patient-derived OVCA cells were either isolated from bulk tumor tissue using gentle MACS and immediately frozen (OVCA-L11) or isolated from the ascites fluid by centrifugation (>70% EpCAM positive cells and >95% CD45 negative cells) and immediately frozen (OVCA-L23). Both OVCA-L11 and OVCA-L23 were derived from an HLA-A*02:01 positive OVCA patient. The primary patient-derived OVCA cells (p0) were thawed three days before being used as target cells in screening experiments. Additionally, primary patient-derived OVCA-L23 cells expanded in vitro which allowed retroviral introduction of HLA-A*24:02 or B*07:01, followed by MACS-enrichment. OVCA-L23 cells transduced with HLA-A*24:02 or B*07:01 (passage 10) were included as target cells in screening experiments. Tumor cell lines and primary patient-derived OVCA cells were cultured in different media ( Supplemental Methods ). CD14-derived mature and immature dendritic cells (mDCs and imDCs), and activated CD19 cells were isolated from peripheral blood mononuclear cells (PBMCs) of different healthy donors and generated as previously described (24). Purity of the generated cells was assessed using flow cytometry ( Supplemental Methods ). Fibroblasts and keratinocytes, both cultured from skin biopsies, were cultured as previously described (24). PTECs derived from kidney tubules were isolated and cultured as previously described (31).

Buffy coats of healthy donors were collected after informed consent (Sanquin). PBMCs were isolated using Ficoll gradient separation and incubated with the selection of pMHC-multimers for 1 hour at 4°C or 15 minutes at 37°C. pMHC-multimers were only included if the healthy donor was negative for the restricted HLA allele. pMHC-multimer bound cells were MACS enriched using anti-PE MicroBeads (Miltenyi Biotec/130-048-801). The positive fraction was stained with CD8 (AF700) and CD4, CD14 and CD19 (FITC). pMHC-multimer and CD8 positive cells were single-cell sorted using an Aria III cell sorter (BD Biosciences) in a 96 well round bottom plate containing 5x10⁴ irradiated PBMCs (35Gy) and 5x10³ EBV-JY cells (55Gy) in 100 μL TCM with 0.8 µg/mL phytohemagglutinin (PHA). T-cell recognition was assessed 10 – 14 days after stimulation, followed by restimulation or storage of the selected T-cell clones.

T-cell recognition was measured by an IFN-γ ELISA (Sanquin or Diaclone). 5,000 T cells were cocultured overnight with target cells in various effector-to-target (E:T) ratios in 60 μL TCM in 384-well flat-bottom plates (Greiner Bio-One). To upregulate HLA expression, all adherent target cells were treated with 100 IU/mL IFN-γ (Boehringer Ingelheim) for 48 hours before coculture. All T cells and target cells were washed thoroughly before coculture to remove expansion-related cytokines. Supernatants were transferred during the ELISA procedure using the Hamilton Microlab STAR Liquid Handling System (Hamilton company) and diluted 1:5, 1:25 and/or 1:125 to quantify IFN-γ production levels within the linear range of the standard curve. T-cell mediated cytotoxicity was measured in a 6-hour ⁵¹chromium release assay ( Supplemental Methods ).

TCR α and β chains of the selected T-cell clones were identified by sequencing with minor modifications ( Supplemental Methods ). The TCR α (VJ) and β (VDJ) regions were codon optimized, synthesized, and cloned in MP71-TCR-flex retroviral vectors by Baseclear. The MP71-TCR-flex vector already contains codon-optimized and cysteine-modified murine TCR α and β constant domains to optimize TCR expression and increase preferential pairing (32). Apart from the OVCA-specific TCRs, a murinized CMV-specific TCR (NLVPMVATV peptide presented in HLA-A*02:01) was included as a negative control. CD8+ T cells were isolated from PBMCs of different donors by MACS and TCRs were introduced via retroviral transduction two days after stimulation with PHA and irradiated autologous feeders. Seven days after stimulation, CD8+ T cells were MACS enriched for murine TCR. Ten days after stimulation, purity of TCR-T cells was checked by flow cytometry and used in functional assays (more details in Supplemental Methods ).

NOD-scid-IL2Rgamma^null (NSG) mice (The Jackson Laboratory) were intravenously (i.v.) injected with 2*10⁶ U266 multiple myeloma (MM)_cells. U266 cells were transduced with and enriched for Luciferase-tdTomato Red and HLA-A24 (NGFR) when indicated. On day 14, mice were treated i.v. with 5*10⁶ purified PRAME TCR-T cells (n = 6) or CMV TCR-T cells (n = 4). TCR-T cells were used seven days after second stimulation with PHA and irradiated autologous feeder cells. Tumor outgrowth (average radiance) was measured at regular intervals after intraperitoneal injection of 150 mL 7.5 mM D-luciferine (Cayman Chemical) using a CCD camera (IVIS Spectrum, PerkinElmer). All mice were sacrificed when control mice reached an average luminescence of 1*10⁷ p/s/cm²/sr. This study was approved by the national Ethical Committee for Animal Research (AVD116002017891) and performed in accordance with Dutch laws for animal experiments.

DAC (5-aza-2′-deoxycytidine) (A3656, Sigma-Aldrich) was solved in dimethyl sulfoxide (DMSO). Target cells were at 50% confluency at start of treatment and were treated with 1 µM DAC on day 1 and 4. DMSO treated cells served as negative control. On day 7, cells were harvested for T-cell reactivity assays and RNA isolation to determine gene expression by qPCR.

Statistical analysis was performed using GraphPad Prism software (Version 9.0.1.). Statistical tests used are indicated in the figure legends, P < 0.05 was considered significant. Significance levels are indicated as p <.05 *, p <.01 **, p <.001 ***, and p <.0001 ****.

Samples of healthy donors and AML patients were used from the LUMC Biobank for Hematological Diseases, after approval by the Institutional Review Board of the LUMC (approval number 3.4205/010/FB/jr) and the METC-LDD (approval number HEM 008/SH/sh). The OVCA patient samples were obtained according to the Code of Conduct for Responsible Use of human tissues or in the context of study L18.012 that was approved by the Institutional Review Board of the LUMC (approval number L18.012) and Central Committee on Research Involving Human Subjects (approval number NL63434.000.17). Studies were conducted in accordance with the Declaration of Helsinki and after obtaining informed consent.

To identify genes with immuno-therapeutic potential in ovarian cancer, we obtained mRNA-seq data of 2202 samples from three independent sources (TCGA, GTEx, and HPA) representing 120 different healthy or tumor tissues. We combined these tissues into one comprehensive dataset to perform an elaborate differential gene expression analysis ( Figure 1A and Supplementary Table 1 ). Genes were defined to be differentially expressed (DE) in ovarian cancer when they exhibited an absolute FC of ≥ 20 compared to the different healthy tissues present in the dataset, using the mean expression values. Tissues from reproductive organs were excluded from this comparison, as expression in the reproductive compartment does not form an unacceptable toxicity risk for ovarian cancer patients. The FC values for all 16,855 genes with ≥ 100 read counts in ovarian cancer are listed in Supplementary Table 2 , of which 9 genes were DE with a FC ≥ 20 in ovarian cancer. We plotted for all genes the minimum FC against the adjusted p-value to visualize the minimal extent of differential expression in ovarian cancer ( Figure 1B ).

Six of the nine DE genes are not expressed on protein level and were therefore not considered target candidates for T-cell therapy. SLC25A3P1, small nuclear RNU1-27P and small nuclear RNU1-28P are pseudogenes which are assumed not to be translated (33). Furthermore, microRNA MIR3687-1, antisense RNA ELFN1-AS1 and an uncharacterized long non-coding RNA gene are classified as non-protein coding RNAs, although they do exhibit several gene regulating functions of other genes (34). The final three genes, PRAME, CTCFL and CLDN6, were considered interesting target candidates. These genes were at least 20 times higher expressed in ovarian cancer compared with healthy tissues, except for some reproductive organs ( Supplementary Figure 1 , summarized in Figure 1C ). In line with their classification as CTA, PRAME and CTCFL were highly expressed in testis (35). PRAME was also found to be expressed in healthy endometrium and ovary, and CLDN6 in placenta. According to the TCGA data, in particular PRAME is expressed in various other tumor types as well ( Supplementary Figure 2 ).

To confirm expression of the three selected genes in ovarian cancer, we quantified gene expression by qPCR in primary solid tumor patient samples and malignant ascites patient samples, and in OVCA cell lines ( Figure 2A ). We quantified relative gene expression compared with three housekeeping genes. PRAME and CLDN6 expression was demonstrated in most primary patient samples and OVCA cell lines. Expression of CTCFL was high (>30% relative expression) in 10/12 solid tumor patient samples, but limited expression was observed in ascites patient samples and cell lines.

The number of previously identified peptides derived from PRAME, CTCFL and CLDN6 binding in different common HLA class I molecules is limited, as well as solid evidence of processing and presentation in the context of HLA class I on ovarian tumors. The PRAME TCRs currently investigated in clinical trials all target the SLLQHLIGL or VLDGLDVLL peptide presented in HLA-A*02:01. To establish a dataset of peptides that can be targeted by TCRs, we determined the HLA class I ligandome of eight primary OVCA patient samples and two OVCA cell lines ( Supplementary Tables 3A, B ). In order to enlarge the dataset, various tumor cell lines and primary AML patient samples expressing the selected genes were additionally included ( Supplementary Table 3C ), some of these cell lines were transduced with CTCFL, CLDN6 and/or HLA class I molecules ( Supplementary Table 3D ). All best scoring peptides for each gene with preferably a minimal Best Mascot Ion score of 20 and a mass accuracy of 10 ppm were considered in the first round of selection. As CLDN6 and CTCFL share homology with ubiquitously expressed family members, only those peptides that were unique for the target genes and did not demonstrate major sequence overlap with Claudin-family members (n=47) or paralog CTCF (n=6) were selected. In addition, we only continued with peptides binding to common HLA molecules according to netMHC peptide binding algorithm that matched with the HLA typing of the material from which the peptides originated ( Supplementary Tables 3A–D ) (36). Identified peptides were validated by comparing mass spectra of eluted peptides and synthetic peptides ( Figure 2B and Supplementary Figure 3 ). HLA binding was confirmed by stable pMHC-monomer refolding. In total 23 PRAME peptides, 8 CTCFL peptides and 3 CLDN6 peptides were validated ( Supplementary Table 4 ). As a result of alternative splicing, at least 15 protein variants derived from CTCFL isoforms are known (37). 7/8 CTCFL peptides are present in all 15 CTCFL variants, 1/8 CTCFL peptides, KLHGILVEA in HLA-A*02:01, is only located in the unique region of CTCFL variant 13 ( Supplementary Figures 4A, B ) (37). Since no substantial differences in gene expression were observed between variant 13 and the other CTCFL variants we also continued with this peptide ( Supplementary Figure 4C ).

To isolate high-avidity T cells reactive against PRAME-, CTCFL- and CLDN6-derived peptides, peptide MHC-multimers (pMHC-multimers) were generated for a selection of 17 peptides binding in different common HLA class I alleles ( Table 1 ). Of these peptides 16 were identified in our mass spectrometry analysis and 1 peptide was previously identified (38). These pMHC-multimers were incubated with PBMCs of 25 healthy HLA typed donors, pMHC-multimer+ cells were enriched by MACS, and pMHC-multimer+ CD8+ cells were subsequently single-cell sorted ( Figure 2C ). pMHC-multimers were only included if the donor was negative for the HLA allele, to ensure identification of T cells from the allogeneic T-cell repertoire, and thereby circumventing self-tolerance. On average 618*10⁶ PBMCs were used per donor and between 21 and 368 pMHC-multimer+ CD8+ T-cell clones could be expanded after single-cell sorting. To test for functional peptide-specificity, T-cell clones were cocultured with Raji cells loaded with a pool of all target peptides. T-cell clones specifically recognizing the peptide pool were subsequently tested for recognition of target cells transduced with OVCA genes, to select T-cell clones potent enough to recognize endogenously processed and presented peptide. T-cell clones that were only reactive against peptide-loaded cells, nonreactive, reactive against one specific HLA allele independent of added peptides, or reactive against all target cells were excluded ( Figure 2D ). In addition to our search in healthy donors, we searched within the allogeneic T-cell repertoire of an AML patient after HLA-mismatched stem cell transplantation that was published previously (39).

In total, 56 T-cell clones specific for 6/9 PRAME and 3/5 CTCFL peptides that recognized cells transduced with the respective OVCA gene were selected of which 28 clones are shown in Supplementary figure 5A, B . For CLDN6, T-cell clones were isolated that recognized peptide-loaded target cells ( Supplementary figure 5C ), however, CLDN6 transduced cells were not recognized and therefore these CLDN6-specific T-cell clones were not of sufficient avidity and excluded from further screenings.

To select TCR candidates for clinical development, 3 additional screenings were performed. First, tumor recognition was assessed using a panel of naturally expressing PRAME or CTCFL positive tumor cell lines, all expressing the target HLA allele. OVCA cell lines were included to screen the PRAME T-cell clones and for the CTCFL T-cell clones K562 and Ca Ski cell lines were included since OVCA cell lines did not express CTCFL ( Figure 2A ). Second, cross-reactivity with other peptides presented in the target HLA allele was assessed using a panel of PRAME or CTCFL negative tumor cell lines and healthy cell subsets. Third, HLA cross-reactivity was assessed using a panel of Epstein-Barr virus transformed lymphoblastoid cell lines (EBV-LCL) expressing all HLA alleles with an allele frequency ≥ 1% present in the Caucasian population. In total, four T-cell clones were selected as TCR candidates for clinical development. Three T-cell clones target a PRAME-derived peptide: clone DSK3 specific for QLLALLPSL in HLA-A*02:01 (PRAME/QLL/A2), clone 16.3C1 specific for LYVDSLFFL in HLA-A*24:02 (PRAME/LYV/A24) and clone 8.10C4 specific for SPSVSQLSVL in B*07:02 (PRAME/SPS/B7). One T-cell clone targets a CTCFL-derived peptide: clone 39.2E12 specific for KLHGILVEA in HLA-A*02:01 (CTCFL/KLH/A2). These T-cell clones effectively recognized all PRAME or CTCFL positive OVCA/tumor cell lines ( Figure 3A ). Of the PRAME and CTCFL negative cells, only clone 8.10C4^PRAME/SPS/B7 showed low recognition of PRAME negative healthy imDCs ( Figure 3B ). To prevent unwanted toxicity, this recognition should be investigated further using TCR-T cells. Furthermore, clone 16.3C1^{PRAME/LYV/A24} showed cross-reactivity against HLA-B*37:01 and HLA-B*38:01 positive EBV-LCLs ( Figure 3C ). The global frequencies of these HLA alleles are low (HLA-B*37:01: 3.23% and HLA-B*38:01: 1.72%) (40). The excluded T-cell clones exhibited either limited recognition of PRAME or CTCFL positive OVCA/tumor cell lines (25/56), or were cross-reactive against peptides in commonly expressed HLA alleles (27/56).

To investigate the clinical potential of the selected PRAME T-cell clones for TCR gene therapeutic strategies, the TCR α and β chains were sequenced and transferred using retroviral vectors into CD8+ T cells of at least four different donors. TCR-T cells were enriched based on murine TCR (mTCR) expression and functionally tested. In Figure 4A we demonstrated, by pMHC-multimer staining, that PRAME TCR-T cells efficiently expressed the three newly identified TCRs at the cell surface. As a reference, the previously identified HSS3 TCR^PRAME/SLL/A2 (patent: WO2016142783A2) that will be clinically tested in the near future was included (39). Most TCR-T cells exhibited high peptide sensitivity in peptide titration experiments, only TCR 8.10C4^PRAME/SPS/B7 demonstrated limited peptide sensitivity ( Figure 4B ). Additionally, ovarian cancer reactivity of the different PRAME TCR-T cells was studied against various OVCA tumor cell lines and primary patient-derived ovarian cancer cells (OVCA-L23) ( Figure 4C ). The OVCA-L23 cells positive for HLA-A*02:01 expanded in vitro which allowed additional retroviral introduction of HLA-A*24:02 or B*07:01. Uncultured OVCA-L23 (p0) cells were therefore included as target for TCR DSK3^PRAME/QLL/A2 and HLA-A*24:02 or B*07:01 transduced cells (p10) were included as targets for all the PRAME TCR-T cells. All PRAME TCR-T cells recognized the primary patient-derived OVCA-L23 cells as well as all seven PRAME positive OVCA tumor cell lines. In addition, the specificity of the PRAME TCR-T cells was tested against various healthy cell subsets. By qPCR relative PRAME expression was observed in mDCs (3.2%), PTECs (1.3%) and stimulated CD19 cells (0.3%) ( Supplementary Figure 6 ). mDCs were slightly recognized by the PRAME TCRs, as was previously observed for the HSS3 TCR^PRAME/SLL/A2 (39), but no other reactivity was observed ( Figure 4D ). Although clone 8.10C4^PRAME/SPS/B7 had exhibited some reactivity against imDCs ( Figure 3B ), the TCR-T cells did not show any signs of recognition in repeated experiments ( Figure 4D ).

Anti-OVCA cytotoxic reactivity was further investigated in a six-hour ⁵¹chromium release assay. Transfer of the different PRAME TCRs to CD8+ T cells of four different donors resulted in efficient killing of OVCA tumor cell lines and the primary patient-derived OVCA cells (OVCA-L23 p0 or p10) ( Figure 5A ). Comparable killing percentages were observed by positive control TCR HSS3^PRAME/SLL/A2 ( Figure 5A ), and peptide-loaded targets were similarly lysed ( Supplementary Figure 7 ). No off-target killing of Raji cells (0% PRAME), imDCs (0% PRAME), and target HLA negative COV362.4 cells was observed ( Figure 5B ). In vivo killing potential of the PRAME TCRs was tested in an established model for multiple myeloma (MM) (23), since PRAME is also expressed in MM. Despite low PRAME expression (4%), all three newly identified PRAME TCR-T cells and positive control TCR HSS3^PRAME/SLL/A2 reduced tumor burden for at least 6 days after infusion ( Figure 5C ). TCR 16.3C1^{PRAME/LYV/A24} and the positive control demonstrated the strongest effect. In conclusion, the three PRAME TCRs (DSK3^PRAME/QLL/A2, 16.3C1^{PRAME/LYV/A24} and 8.10C4^PRAME/SPS/B7) demonstrated potent antitumor reactivity in vitro and in vivo without harming healthy cell subsets in vitro and are considered promising TCRs for TCR gene therapy.

Next, the CTCFL-specific TCR 39.2E12^CTCFL/KLH/A2 was tested for anti-ovarian cancer reactivity and specificity. Generated CTCFL TCR-T cells efficiently expressed the TCR at the cell surface ( Figure 6A ) and demonstrated high peptide sensitivity in a peptide titration ( Figure 6B ). CTCFL TCR-T cells generated from three different donors recognized primary patient-derived OVCA-L11 cells harvested from an HLA-A*02:01 positive patient ( Figure 6C ). Furthermore, in line with the lack of CTCFL expression in any of the included healthy cell subsets ( Supplementary Figure 6 ), healthy cell subsets were not recognized by CTCFL TCR-T cells ( Figure 6D ).

Despite high CTCFL expression in primary OVCA patient samples, OVCA tumor cell lines did not express CTCFL ( Figure 2A ). In contrast, cervical cancer cell line Ca Ski is positive for CTCFL and this correlates with expression in part of primary cervical carcinoma samples ( Supplementary Figure 2B ). As demonstrated in Figure 6E , the Ca Ski cells were efficiently recognized by the CTCFL TCR-T cells. Since CTCFL expression is epigenetically regulated and treatment with demethylating agent DAC has previously been shown to upregulate expression of CTCFL in OVCA tumor cell lines (42), we investigated whether DAC can make tumor cell-lines more susceptible to CTCFL-mediated killing. Seven days of DAC treatment clearly resulted in CTCFL upregulation in OVCA cell lines, compared to not treated cells ( Figure 6E ). In line with the upregulation, recognition of DAC-treated OVCA tumor cell lines COV413b and A2780 was significantly increased for CTCFL TCR-T cells ( Figure 6E ). CTCFL upregulation by DAC was restricted to tumor cells, as DAC treatment of healthy fibroblasts did not upregulate CTCFL expression and did not induce recognition by CTCFL TCR-T cells ( Figure 6F ). In line with increased cytokine production, cytotoxic capacity of CTCFL TCR-T cells towards DAC-treated COV413b was significantly increased ( Figure 6G ). DAC treatment did not increase killing by allo-HLA-A*02:01 T cells ( Figure 6G ) neither did it influence killing of peptide-loaded target cells ( Supplementary Figure 8 ), suggesting DAC treatment does not generally increase susceptibility of these target cells to T-cell mediated killing. In OVCA tumor cell lines we also observed increased PRAME expression after DAC treatment, which slightly increased recognition and killing potential by HLA-A*02:01-restricted PRAME TCR-T cells ( Supplementary Figure 9 ). In conclusion, CTCFL-specific TCR 39.2E12^CTCFL/KLH/A2 demonstrate anti-OVCA reactivity against (DAC-treated) CTCFL positive tumor cells without harming healthy cell subsets and is considered a promising TCR for TCR gene therapy of ovarian cancer.