ABSTRACT
Chimeric antigen receptor (CAR) T cell therapy has shown limited success in solid tumors, mainly because of the scarcity of tumor‐specific surface antigens. To address this challenge, we developed an integrated computational and experimental pipeline for antigen discovery and applied it to melanoma. Our bioinformatics strategy introduced a novel quantitative metric, the restrictiveness ratio, to identify antigens highly expressed in tumors but minimally present in healthy tissues. Using RNA‐sequencing data from patient samples, we compared candidate targets with the established melanoma antigen chondroitin sulphate proteoglycan 4 (CSPG4) and validated their expression on melanoma cell lines. CARs were engineered against selected antigens, CAR‐T cells were generated by RNA electroporation, and expression was confirmed. Functional validation included T cell activation, cytokine secretion, and cytotoxicity assays. RNA‐seq analysis of cutaneous and uveal melanoma identified 90 and 84 candidate genes, respectively, encoding transmembrane proteins with tumor expression above 10 TPM (transcripts per million). Fifty‐six genes displayed higher restrictiveness ratios than CSPG4. Following manual curation and scFv (single‐chain variable fragment) availability assessment, five candidates were prioritized. CARs targeting ErbB3 and SLC1A5 induced strong antigen‐specific T cell activation and cytotoxicity comparable to CSPG4‐CARs. These findings identify ErbB3 and SLC1A5 as promising CAR‐T targets and establish an integrative bioinformatics‐driven framework for extending CAR‐T therapy to solid tumors.
Keywords: chimeric antigen receptor (CAR)‐T cell, Erb‐B2 receptor tyrosine kinase 3 (ErbB3), in silico/in vitro pipeline, melanoma, solid tumors, Solute Carrier Family 1 Member 5 (SLC1A5)
Integrative bioinformatics‐driven discovery of novel chimeric antigen receptor (CAR)‐T cell targets in melanoma.

1. Introduction
Chimeric antigen receptor (CAR)‐modulated T‐lymphocyte therapy (CAR‐T cell therapy), a type of targeted immunotherapy, has shown great potential as an effective treatment for cancer. CARs are fusion proteins consisting of a single‐chain variable fragment (scFv) antigen‐binding domain, a hinge region, a transmembrane domain, and at least one intracellular signaling activating domain [1]. In CAR‐T cell therapy, patients' cytotoxic T cells are genetically modified to express CARs and are subsequently returned, where they lyse antigen‐expressing tumor cells [1]. One key advantage of CARs is their human leukocyte antigen (HLA)‐independent recognition of target antigens, which makes CAR‐T cells more broadly applicable compared to other T cell therapies [2]. However, CAR‐T cells can trigger a systemic inflammatory reaction [2] or damage healthy tissues expressing the target antigen [3]. Thus, the selection of a tumor‐restricted antigen is crucial for minimizing the risk of adverse effects in CAR‐T cell therapy.
CAR‐T cell therapy has achieved high response rates in patients with various types of hematological malignancies, and consequently, has been approved for several hematological indications [4]. In contrast, the search for specific target antigens in solid tumors has been less successful [5]. A tumor antigen must fulfil various requirements to qualify as a CAR‐T cell target in solid tumors. The antigen should have no or minimal surface expression on healthy tissue but be highly expressed on tumor cells, thereby minimizing the risk of on‐target/off‐tumor toxicity [6]. The antigen should be homogeneously expressed on all tumor cells and be considered an oncogenic driver to avoid antigen escape [6].
Cutaneous melanoma (CM) and uveal melanoma (UM) are two potential solid tumors for which alternative treatment strategies, like CAR‐T cell therapy, are needed. In advanced tumor stage CM, immune checkpoint inhibitors (ICI) are the standard therapy [7]. However, a considerable fraction of patients experience severe immune‐related adverse events [8], and 40%–50% of them exhibit resistance to ICIs [9]. UM is the most common primary intraocular malignancy in adults [10]. Approximately half of all UM patients develop metastases [10], and their prognosis is particularly poor, with a mean overall survival of about one year regardless of the therapy [11]. ICI treatment of UM has shown minimal efficacy, with response rates of 15% at most [12]. Tebentafusp, a bispecific fusion protein, has shown significantly improved overall survival in metastatic UM (median 21.7 months), but is only applicable in patients with an HLA‐A*02:01 subtype [13]. Twelve CAR‐T cell antigens against CM are investigated in ongoing clinical trials (registered at ClinicalTrials.gov), three others are tested for UM (by 06.05.2025), and some candidates are preclinically tested in both tumor entities [14, 15, 16]. Although some of these antigens appear promising, gene and protein expression databases indicate that all of them are expressed in some healthy tissues and therefore carry the risk of triggering on‐target/off‐tumor toxicity.
One alternative to search for CAR‐T cell antigens is the exploitation of large public datasets of tumor sample transcriptomics data utilizing bioinformatics algorithms and (AI‐driven) molecular simulations [17]. In line with this, our study employed a combined in silico/in vitro approach to identify new tumor‐specific cell surface antigens, using CM and UM as case studies. Our computational algorithm leverages RNA sequencing data to identify potential target antigens that meet the criteria of tumor specificity and low expression in healthy tissue. The most promising antigens identified are used to generate antigen‐specific CAR‐T cells, which are tested in vitro.
2. Results
2.1. An RNA‐Seq Analysis‐Based Computational Pipeline Identifies Cell Surface Antigen Candidates for CAR‐T Cells Against CM and UM
We hypothesized that combining RNA‐Seq data from tumor and healthy samples with bioinformatics analyses would provide a systematic approach for selecting cell surface‐specific proteins with expression restricted to tumors compared to healthy tissues. These tumor‐restricted cell surface proteins could serve as target candidates for the development of novel CARs for cancer immunotherapy. We implemented a workflow to identify new tumor‐specific cell surface antigens and generate antigen‐specific CAR‐T cells (summarized in Figure 1). To test our concept, we chose UM as a case study and searched for cell surface‐specific proteins that could function as target candidates for CARs in UM (Figure 2A). We combined transcriptomics data from 80 UM samples, available from The Cancer Genome Atlas [18] and 14 UM samples collected during an in‐house clinical trial (see Section 5). We performed principal component analysis (PCA) to verify the data homogeneity across all samples (see Figure S1).
FIGURE 1.

Schematic presentation of the integrative workflow. Integrative workflow used to identify, select, and test in vitro antigen targets for CAR‐T cells in uveal melanoma (UM) and cutaneous melanoma (CM).
FIGURE 2.

Computational identification of UM CAR target antigens. (A) Schematic flowchart outlining the sequential filtering steps. Resulting numbers of gene candidates of key pipeline steps. (B) Restrictiveness (log2‐transformed ExpressionmedianTumor vs. ExpressionmaxTissue ) of all genes with a mean TPM above 10 compared to the benchmark gene CSPG4. Less restricted genes are marked in black and more restricted genes in blue. (C) Euclidean clustered heatmap of log2‐transformed expression data per sample of all genes that are more restricted than CSPG4. The expression levels of the healthy tissues with the strongest expression are depicted in the ctrl column.
We selected genes with a mean tumor expression higher than 10 transcripts per million (TPM) encoding transmembrane (TM) proteins according to the Protein Data Bank of transmembrane proteins (PDBTM) [19] and obtained a list of 84 candidates (Figure 2A). To minimize the risk of immune‐related adverse events associated with a CAR‐T therapy targeting these TM proteins and based on our previous work on the selection of tumor peptides [20], we developed a computational score named “restrictiveness ratio” (RR). High values of RR correspond to preferred cell surface proteins that are highly expressed in the tumor of interest, with comparably low expression in healthy tissues, as indicated by transcriptomic data available in the Genotype‐Tissue Expression project (GTEx) [21]. An RR of 1 means a gene's mean tumor expression equals the 75th quantile expression in the highest‐expressing healthy tissue. Since we were seeking cell surface protein candidates with superior performance to known CAR‐T antigens, we selected candidate genes with an RR value higher than that obtained for the well‐established CAR antigen CSPG4 (Figure 2B).
We computed the RR for all candidate genes and identified 56 genes with an RR value higher than the one obtained for CSPG4 (Figure 2A,B). Figure 2C displays a Euclidean clustered heatmap of log2‐normalized tumor gene expression for genes more restricted than CSPG4. We performed a similar analysis for CM (see Figure S2). We identified 90 candidate genes encoding TM proteins with mean tumor expression higher than 10 TPM. The RR‐based filter identified 56 TM candidate genes in CM with an RR value higher than the one obtained for CSPG4. An overview of the expression of all CM and UM candidate genes in healthy tissues is shown in Euclidean clustered heatmaps in Figures S3 and S4.
2.2. An Evidence‐Based Systematic Curation of Literature and Databases Identifies Feasible Target Antigens for CAR‐T Development and Experimental Testing
We manually investigated the characteristics of antigens with a higher score than CSPG4 for CM and/or UM to select suitable target antigens for further CAR‐T cell in vitro testing (Figure 3, and Tables S1 and S2).
FIGURE 3.

Evidence‐based selection of CAR target antigens for in vitro testing. The figure sketches the selection workflow followed for UM and CM. We first selected candidate antigens with a restrictiveness ratio (RR) higher than that of CSPG4, as determined by the analysis of melanoma sequencing data (Step 1; the selected gene candidates are provided for UM and CM on the outer side of the workflow scheme). Next, we used Genecard's cell localization based on COMPARTMENTS to select plasma membrane proteins (Step 2). Subsequently, we examined protein expression in cancers and healthy tissues in HPA to select antigens (Step 3). For the remaining antigens, we selected targets with a documented dominant role in cancer in general or melanoma in particular, and absence of published/patented CARs targeting the antigen (Step 4). Lastly, we checked the availability of a suitable antibody for flow cytometry‐based surface staining (Step 5). We selected all remaining antigens for in vitro analysis. In the figure, we highlighted in orange font the shared CAR‐T cell target antigens that remained after the selection procedure, and in grey an antigen that was discarded due to the lack of an available scFv sequence. See also Tables S1 and S2.
First, we investigated their predicted cellular localization according to Genecards (www.genecards.com) and selected only those antigens scored as most likely to be plasma membrane proteins. We obtained 40 candidate antigens for UM and 45 for CM. We examined protein expression in different tumors and healthy tissues using the Human Protein Atlas (HPA) and selected antigens reported to be overexpressed in cancer, in general, or in melanoma, in particular, while excluding all targets with general/ubiquitous membranous expression in healthy tissues. This combined analysis generated nine candidate antigens for UM and 11 for CM. We utilized PubMed (https://pubmed.ncbi.nlm.nih.gov) to select targets with a documented role in key cancer‐related pathways, and PubMed and Google Patents (https://patents.google.com) to exclude candidates with published or patented CARs targeting them.
For the remaining antigens, we conducted internet research to select those with commercially available antibodies that recognize their extracellular domain. This search yielded four candidate antigens for UM, namely APH1A (Aph‐1 homolog A gamma‐secretase subunit), NCSTN (Nicastrin), ERBB3/HER3 (Erb‐B2 Receptor Tyrosine Kinase 3/Human Epidermal Growth Factor Receptor 3), and ADIPOR2 (Adiponectin receptor 2). We further validated the expression of the four UM antigen candidates using a new set of transcriptomics data from 16 metastases of UM patients, included in a phase I trial at our department (NCT04335890). These data confirmed the expression of all four antigens at the RNA level in tumor samples from these patients (Figure 4 and Table S3).
FIGURE 4.

Transcriptomics data from metastatic UM (Trial NCT04335890) confirm expression of candidate genes in UM. Expression of the selected candidates, as well as currently examined targets, was compared with CSPG4 and the intracellular tumor marker PRAME, which is expressed in some, but not all UMs. RNA from biopsies from metastases of UM patients from an in‐house clinical trial was sequenced, and expression levels were calculated for the indicated genes. The log2 values of the TPM (see Table S3 for TPM counts) are shown (log2(TPM + 1)).
Furthermore, the expression levels of the candidate antigens exceeded those of the target antigens used in ongoing CAR‐T cell clinical trials (CD20 and IL‐13Rα2), and, in contrast to the well‐known intracellular tumor marker PRAME (Preferentially Expressed Antigen in Melanoma), they were present in all UM samples (Figure 4 and Table S3).
For CM, six antigens remained, from which we previously identified already four for UM. Out of the remaining two candidates, we decided to include for further in vitro testing only the one with already available scFv sequences, thereby adding SLC1A5. Taken together, the final list of candidate antigens for experimental testing included APH1A, NCSTN, ERBB3, ADIPOR2, and SLC1A5.
2.3. ErbB3 and SLC1A5 Are Expressed by CM and UM Cell Lines but Show No or Weak Expression on Healthy Human Immune Cells
We determined the surface expression of the selected tumor antigens on different CM and UM cell lines using flow cytometry. We also assessed surface expression on immune cells from healthy donors and HEK293T embryonic kidney cells as controls. Furthermore, we used the surface expression of the known CAR‐T antigen CSPG4 for comparison (Figure 5). Our control CSPG4 was strongly expressed by both CM cell lines (Mel526, A375M). However, it was absent or only weakly expressed by all tested UM cell lines and in the immune cells analyzed. ErbB3 was strongly expressed by two primary UM cell lines (92.1, Mel270), three metastatic UM cell lines (Omm1, Omm2.3, Omm2.5), a primary CM cell line (Mel526), and a metastatic CM cell line (A375M), but not in the immune cells or the HEK293T cells. APH1a and Nicastrin were expressed by two primary UM cell lines (92.1, Mel270), three metastatic UM cell lines (Omm1, Omm2.3, Omm2.5), the primary CM cell line (Mel526), and the metastatic CM cell line (A375M). In immune cells, APH1A and Nicastrin were weakly expressed on PBMCs and strongly expressed on monocyte‐derived dendritic cells (moDCs) and HEK293T. SLC1A5 was strongly expressed by the primary UM cell line (92.1), two metastatic UM cell lines (Omm1, Omm2.5), the primary CM cell line (Mel526), the metastatic CM cell line (A375M), and moDCs, but very weakly expressed on PBMCs (Figure 5).
FIGURE 5.

ErbB3, SLC1A5, APH1a, and Nicastrin are expressed by different CM and UM cell lines. The following cell lines or cells were stained (blue histograms) with antigen‐specific antibodies: (1) Mel526, (2) A375M, (3) 92.1, (4) Mel270, (5) Omm2.5, 6) Omm2.3, (7) Omm1, (8) peripheral blood mononuclear cells (PBMCs) from healthy donors, (9) monocyte‐derived dendritic cells (moDCs) from PBMCs of healthy donors, and (10) HEK293T (293T). Unstained cells (red histograms) were used as controls. For Adiponectin, we performed control stainings with only the secondary donkey‐anti‐goat antibody (depicted as orange histograms).
The staining for ADIPOR2 was inconclusive due to a very weak specific signal combined with a high background staining from the secondary antibody (Figure 5). This led us to exclude ADIPOR2 from further testing. Based on the cell surface expression results, we decided to continue examining ErbB3 and SLC1A5. We excluded APH1a and Nicastrin from further experiments because (1) their surface expression on PBMCs may interfere with antigen‐specific CARs, and (2) their strong and consistent expression on moDCs suggests that these antigens might exhibit high surface expression levels on other healthy human cells. We selected the HEK293T cell line as a negative control in all following experiments involving ErbB3‐CARs. SLC1A5 was expressed on HEK293T cells (Figure 5) and all other tested cell lines (see Figure S5); thus, we did not have access to a cell line without surface expression. However, PBMCs exhibited very low surface expression of SLC1A5 (Figure 5) and were easily accessible; therefore, we utilized primary immune cells derived from PBMCs as a negative control for further experiments with SLC1A5.
2.4. Transfection of CAR‐Encoding RNA Results in Surface Expression of ErbB3‐ and SLC1A5‐CARs on Primary Human Cytotoxic T Cells
We designed two CAR constructs against ErbB3 (CAR 22 and CAR H6) and two against SLC1A5 (CAR 1 and CAR 2). To examine their respective surface expression, we electroporated primary CD8+ T cells with mRNA encoding the antigen‐specific CAR constructs. This allowed us to rapidly generate CAR‐T cells for the functional validation of the novel target antigen in our pipeline. We used the standard CSPG4‐specific CAR construct as a positive control, and an anti‐F(ab’)2 anti‐human IgG antibody binding the Fc hinge region within all the CAR constructs for detection of CAR expression. We performed surface expression measurements 4 h and 20–24 h after electroporation (Figure 6).
FIGURE 6.

The newly developed CARs are expressed on primary human cytotoxic T cells after mRNA electroporation. (A and B) Expression levels of different CAR constructs on T cells of one representative donor (out of a total of 15 independent experiments), 4 h (A) and 20–24 h (B) after electroporation. (C) Percentage of CAR‐positive T cells 4 and 20–24 h after RNA electroporation. (D) Mean fluorescence intensity (MFI) of T cells 4 and 20–24 h after RNA electroporation. (C and D) Average values with error bars (SEM). N = 2 for CAR expression 4 h after electroporation of ErbB3‐CAR H6 and ErbB3‐CAR 22. EP = electroporated. (A‐B): Histograms were created using FlowJo 10.9. (C and D) Bar graphs were generated with GraphPad Prism10.
Compared with mock‐electroporated T cells, for all CAR constructs, 75%–90% of the transfected T cells expressed the CAR constructs on their surface both at 4 h and 20–24 h after electroporation (Figure 6C). There were no significant differences between the CARs. Except for SLC1A5‐CAR 1, which had a significantly lower percentage of CAR‐positive cells compared with CSPG4‐CAR control cells (4 and 24 h after electroporation) and SLC1A5‐CAR 2 (4 h after electroporation; Figure 6C, for significance analysis see Table S4).
We determined the mean fluorescence intensity (MFI) to quantify the density of the different CAR constructs on the surface (Figure 6D). Both SLC1A5‐specific CARs had significantly lower MFIs 20–24 h after electroporation than the CSPG4‐ and ErbB3‐specific CARs (Figure 6D and Table S4). All tested new CARs, except for ErbB3‐CAR 22, had a significantly lower MFI than the CSPG4‐specific control CAR (Figure 6D and Table S4).
Taken together, the introduction of the newly‐generated CARs specific for ErbB3 and SLC1A5 into human cytotoxic T cells by mRNA electroporation was successful.
2.5. ErbB3‐Specific and SLC1A5‐Specific CAR‐T Cells Upregulate Activation Markers and Produce Cytokines After Antigen‐specific Stimulation
To assess the activation of CAR‐transfected cytotoxic T cells upon contact with antigen‐positive or ‐negative target cells, we co‐incubated the designed CAR‐T cells overnight with different target cells at a 1:1 ratio. Next, we profiled the expression of the T cell activation markers CD25 and CD69 by flow cytometry, as well as the secretion levels of the cytokines IL‐2, TNF, and IFNγ.
In the case of the ErbB3‐specific CAR‐T cells, we used three UM cell lines (92.1, Omm1, and Omm2.5) and one CM cell line (Mel526) as target cell lines, while HEK293T cells were employed as a negative target control. We observed a significant upregulation of CD69 on the CAR‐T cells after co‐incubation with all four antigen‐positive tumor cell lines, compared with the negative target (15%–30% of CD69+ T cells, depending on the CAR and the tumor cell line; Figure 7A). For CD25, we detected upregulation after co‐incubation with all four antigen‐positive tumor cell lines for ErbB3‐CAR H6‐transfected T cells (Figure 7B).
FIGURE 7.

The newly developed CAR‐T cells upregulate the activation markers CD69 and CD25, and secrete cytokines. After overnight co‐incubation of effector T cells (CAR‐ or mock‐transfected) with several antigen‐positive or ‐negative target cells (see Methods section), the upregulation of CD69 (A, F) or CD25 (B, G) expression was determined. The bar graphs show the average percentage of activation marker expression with standard error bars (A, F + B, G). Additionally, the cytokine concentration of IL‐2 (C, H), TNF (D, I), and IFNγ (E, J) in the supernatant was determined. The bar graphs show the average cytokine concentration with standard error bars (C‐E+H‐J). The p‐values were calculated by paired or unpaired t‐test based on the difference (A, B, F, G) or the ratio (C–E + H–J) relative to control (HEK293T cells) (A–E) or CD8‐negative fraction (F–J), ns = nonsignificant, *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001. The results shown originate from six T cell activation assays with different T cell donors. (A–E) MEL526 N = 6, 92.1 and Omm1 N = 5, Omm2.5 N = 3; HEK293T was used as a control. (F–J) A375M and Omm1 N = 5; a CD8‐negative fraction of a monocyte‐depleted fraction of PBMCs was used as a control.
In contrast, in the case of ErbB3‐CAR 22‐transfected T cells, CD25 was significantly upregulated after incubation with three cell lines (Mel526, 92.1, and Omm1) when compared with the negative target (15%–35% of CD25+ T cells, depending on the CAR and the tumor cell line; Figure 7B). We did not detect significant differences in the expression of CD69 or CD25 when comparing CAR‐T cells alone with their co‐incubation with the negative target (HEK293T cells). ErbB3‐specific CAR‐T cells secreted high levels of pro‐inflammatory cytokines after co‐incubation with different CM and UM cell lines. All four examined antigen‐positive tumor cell lines induced significantly higher levels of IL‐2 and IFNγ compared to the HEK293T cells for both ErbB3‐specific CARs (Figure 7C–E). For ErbB3‐CAR H6‐transfected T cells, all four antigen‐positive cell lines also induced significantly higher levels of TNF upon co‐incubation. In contrast, for ErbB3‐CAR 22‐transfected T cells, only two UM cell lines (92.1 and Omm1) resulted in significantly higher TNF levels in comparison with the negative target (Figure 7D). We did not observe significant secretion of cytokines when comparing only CAR‐T cells with their co‐incubation with the HEK293T cells (Figure 7C‐E).
In the case of the SLC1A5‐specific CAR‐T cells, we used one CM cell line (A375M) and one UM cell line (Omm1) as target cell lines, while the CD8‐negative fraction of monocyte‐depleted PBMCs was employed as a negative target control. Compared with the negative target, we found a significant upregulation of CD69 and CD25 on SLC1A5‐specific CAR‐T cells for both tumor cell lines: 45%–65% of SLC1A5‐CAR 1‐transfected T cells and 75%–85% of SLC1A5‐CAR 2‐transfected T cells were CD69‐positive (Figure 7F), while 40% of SLC1A5‐CAR 1‐transfected T cells and 75% of SLC1A5‐CAR 2‐transfected T cells were CD25‐positive (Figure 7G). The co‐incubation of SLC1A5‐CAR 2‐transfected T cells with the negative target cells resulted in a significant upregulation of CD69 expression compared to T cells alone (65% of the CD69+ T cells), but CD69 expression was still significantly higher when induced by antigen‐positive target cells (Figure 7F). For SLC1A5‐CAR 2‐transfected T cells, we observed a relatively high background activation, with 50% of CD69‐positive cells and 40% of CD25‐positive cells in T cells alone (Figure 7F + G). T cells transfected with SLC1A5‐CAR 2 secreted comparably high levels of IL‐2 and TNF cytokines after co‐incubation with antigen‐positive tumor cell lines. However, co‐incubation with A375M resulted in higher levels of IFNγ (Figure 7H–J). In the case of SLC1A5‐CAR 1‐transfected T cells, we measured much lower cytokine concentrations and no significant difference from the negative target (Figure 7H–J). Co‐incubation of SLC1A5‐CAR 2‐transfected T cells with negative target cells resulted in significantly higher concentrations of IL‐2, TNF, and IFNγ compared to T cells alone (Figure 7H–J). In contrast, SLC1A5‐CAR 1‐transfected T cells showed a significantly higher concentration of TNF after co‐incubation with negative target cells compared to T cells alone (Figure 7I).
To investigate whether the lower cytokine production by SLC1A5‐CAR 1‐transfected T cells compared to SLC1A5‐CAR 2‐transfected T cells was caused by an upregulation of checkpoint molecules (first signs of T cell exhaustion), we co‐incubated both effector cells without and with the A375M tumor cells and measured the upregulation of PD‐1, CTLA‐4, TIM‐3, LAG‐3 at 17 h, 48 h, and 72 h after the start of the co‐incubation. Surprisingly, for the SLC1A5‐CAR 2‐transfected T cells, the expression of checkpoint molecules was already higher without co‐incubation, and increased even more after antigen‐specific stimulation with A375M (Figure S6 and Table S5).
Altogether, both new ErbB3‐specific CARs were functional when introduced into human cytotoxic T cells. In the case of SLC1A5‐CARs, SLC1A5‐CAR 1 induced expression of cell surface markers, but resulted in very low cytokine production. In contrast, SLC1A5‐CAR 2 induced cell surface marker expression and cytokine secretion, but had a much higher antigen‐unspecific background.
2.6. The Designed ErbB3‐ and SLC1A5‐Specific CAR‐T Cells Exhibit Antigen‐Specific Cytolytic Activity
To investigate the antigen‐specific cytolytic capacity of different CAR‐T cells against metastatic UM and CM cells, we performed a cytotoxicity assay based on differential carboxyfluorescein succinimidyl ester (CFSE) labeling of antigen‐negative and antigen‐positive cells. We co‐incubated overnight antigen‐positive target cells and antigen‐negative control cells, each labeled with a different concentration of CFSE, with CAR‐T cells at indicated target‐to‐effector ratios. Following dead‐cell staining, we identified living target and control cells by their respective fluorescence in flow cytometry and calculated their specific lysis percentage. We compared the designed CAR‐T cells with mock‐transfected T cells, used as control, and with the CSPG4‐specific standard CAR‐T cells.
For ErbB3‐specific CAR‐T cells, we assessed the cytolytic potential by co‐incubating them with Omm1 as antigen‐positive target cells and with HEK293T as antigen‐negative target cells. We did not observe significant differences between all three CAR‐transfected T cells. However, they differed significantly from the mock‐transfected T cells, with the CAR‐T cells exhibiting higher cytolytic potency in experiments with a target‐to‐effector cell ratio of 1:2.5 (Figure 8A and Table S6).
FIGURE 8.

ErbB3‐, SLC1A5‐, and CSPG4‐specific CAR‐T cells specifically lyse antigen‐positive uveal and cutaneous melanoma cells. After overnight co‐incubation of effector T cells (CAR‐ or mock‐transfected) with antigen‐positive and antigen‐negative target cells at different target‐to‐effector ratios, the specific lysis of target cells compared to control cells was determined. For examination of ErbB3‐specific CAR‐T cells, the metastatic UM cell line Omm1 and the ErbB3‐negative cell line HEK293T were used (A). To examine SLC1A5‐specific CAR‐T cells, the metastatic CM cell line A375M was used in combination with human immune cells (monocyte‐depleted PBMCs) as an SLC1A5‐negative control (B). Results shown originate from three (A) or four (B) independent experiments with different T cell donors.
All three CARs displayed dose‐dependent cytotoxicity, with 55%–60% lysis at a 1:1 ratio but with strong donor dependency, 80%–85% lysis at a 1:2.5 ratio, and lysis higher than 90% at ratios of 1:5 and 1:10 (Figure 8A). In contrast, mock‐electroporated T cells showed a percentage of specific lysis less than 40% at all tested target‐to‐effector ratios (Figure 8A). Cryopreserved CAR‐T cells exhibited similar results (Figure S7 and Table S7).
For SLC1A5‐specific CAR‐T cells, we assessed the cytolytic potential by co‐incubating them with A375M cells as antigen‐positive target cells and cryopreserved monocyte‐depleted PBMCs as a negative control (Figure 8B). We detected a significantly higher cytolytic potency in all CAR‐transfected T cells at all tested target‐to‐effector ratios compared to lysis caused by mock‐transfected T cells at the same ratio. However, no significant difference was observed in lysis between the different CAR‐T cells (Figure 8B and Table S8).
All three CARs displayed dose‐dependent cytotoxicity, with 50%–60% lysis at a 1:1 ratio, 70% lysis at a 1:2.5 ratio, and 75%–95% lysis at a 1:5 ratio (Figure 8B). In contrast, mock‐electroporated T cells showed a specific lysis of less than 40% at target‐to‐effector ratios of 1:1 to 1:5, and a 55% lysis at a ratio of 1:10 (Figure 8B). Cryopreserved CAR‐T cells exhibited similar results (Figure S8, Table S9).
Taken together, the antigen‐specific cytolytic activity assays indicate that our designed SLC1A5‐specific and ErbB3 CAR‐T cells efficiently lyse metastatic CM and UM cells, respectively.
3. Discussion
3.1. ErbB3 And SLC1A5 as Newly Identified CAR‐Target Antigens
We identified ErbB3 as a promising candidate for both UM and CM. ErbB3, a member of the EGFR tyrosine kinase family, signals mainly through dimerization with HER2 (HER2/ErbB2: Human Epidermal Growth Factor Receptor 2/Erb‐B2 Receptor Tyrosine Kinase 2) and activation of the PI3K/AKT/mTOR pathway [22]. It is overexpressed in multiple tumors, including melanoma, where its expression correlates with invasion, progression, and metastasis [22]. In CM, high HER3 expression is linked to poorer survival, and ErbB3 is required for metastasis formation in CM cell lines, underscoring its role in metastatic progression [23, 24].
Anti‐ErbB3 monoclonal antibodies have been investigated against various tumors, including melanoma, and have shown anti‐tumor effects in tumor cell lines and mouse models [25]. Hence, some ongoing clinical trials use anti‐HER3 antibodies (NCT05057013, NCT05919537). ErbB3 CAR‐based approaches are already being examined with a focus on breast cancer, including T cells modified with a CAR that uses the natural ligand of HER3/HER4 heterodimers as a binding domain [26], and anti‐ErbB3 CAR natural killer (NK) cells [27]. However, no study has yet investigated scFv‐based ErbB3 CAR‐T cells.
We also identified SLC1A5 as a promising candidate for CM. SLC1A5, a sodium‐dependent amino acid transporter essential for glutamine uptake, is overexpressed in various cancers and supports tumor proliferation [28, 29, 30]. Preclinical studies suggest therapeutic potential for SLC1A5 inhibitors and antibodies, although its use as a CAR target has not yet been explored [31]. While SLC1A5 plays a role in lymphocyte activation and differentiation, our experiments showed no significant adverse effects of SLC1A5 CARs on T cell function, possibly due to compensatory upregulation of alternative amino acid transporters in CD8+ lymphocytes [29, 32]. Further studies are needed to clarify the effects on T cell and CAR‐T cell activation.
3.2. Differences in Cell Surface Expression of the Newly‐Generated CARs
We observed differences in surface expression among ErbB3‐ and SLC1A5‐specific CARs, likely reflecting differences in scFv stability, which influences CAR expression and persistence [33]. Based on these findings, differences in the properties of the scFvs are a likely cause for the expression results we observed. We measured lower expression levels for both SLC1A5‐specific CARs compared to the other CARs tested, which could indicate a connection with the target antigen used, such as intracellular binding between the CAR and the target, or cross‐linking between the scFv domain and SLC1A5 on the cell surface, leading to subsequent internalization. The latter could also explain the high non‐specific background expression of CD25 and CD69. Alternatively, antigen‐independent tonic signaling of the CARs could explain this high background expression [34]. The differences in CAR expression levels could influence both in vitro and in vivo efficacy of the CAR‐T cells, with low surface expression leading to reduced T cell activation, cytokine secretion, and cytotoxicity in vitro, and high surface expression harming in vivo T cell anti‐tumor effects [35]. In our experiments, we detected differences in terms of T cell activation and cytokine secretion between the different CARs; however, we did not find a significant effect on in vitro cytotoxicity. Nonetheless, it may still have a significant influence on efficacy in vivo.
3.3. Differences in Functionality of the Newly‐Generated CARs
We observed a very low secretion of pro‐inflammatory cytokines by SLC1A5‐CAR 1‐transfected T cells compared to the SLC1A5‐CAR 2‐transfected T cells, which may be due to the lower SLC1A5‐CAR 1 surface expression. A higher expression of checkpoint molecules on SLC1A5‐CAR 1‐transfected T cells was not the cause of the lower cytokine secretion, since the SLC1A5‐CAR 2‐transfected T cells expressed more checkpoint molecules.
Since weak cytokine secretion could limit the effectiveness of the CAR‐T cells, we investigated the lytic capacity of these T cells in vitro. We found no significant difference in lytic capacity between the SLC1A5‐CAR 1‐ and SLC1A5‐CAR 2‐transfected T cells and the benchmark CSPG4 CAR‐T cells. This suggests that the lower surface expression or cytokine production does not influence the lytic activity. This may even be advantageous. A weaker cytokine secretion by the T cells could reduce the risk of a cytokine release syndrome (CRS) [36]. We also examined the effect of cryopreservation on the functionality of the newly generated CARs, as actual clinical use of CAR‐T cells generated via mRNA electroporation requires the cryopreservation of large numbers of CAR‐T cells and repetitive injections of these preserved T cells [37].
3.4. Advantages of a Combined In Silico/In Vitro Pipeline for the Identification of New Specific Tumor Antigens
Our in silico analysis generates a large pool of data‐driven candidates, which are systematically filtered and ranked until a small portion of the most promising antigens is selected for CAR design and in vitro validation. In this context, the availability of suitable transcriptomics and proteomics datasets for conducting bioinformatics analysis is one of the main challenges. RNA data from larger cohorts would increase the statistical power of the predictions, the number of target candidates, and their robustness, especially for the CM cohort.
Surface expression analysis on tumor cell lines confirmed the transferability of RNA data from patient samples to in vitro models. The exclusion of several candidates after surface staining highlights the importance of in vitro validation. Moreover, generation and testing of specific CARs showed that target antigen‐expressing tumor cells activate CARs and are lysed in vitro, supporting the clinical potential of both the targets and CARs.
Some published studies have also employed a combined in silico/in vitro method [38]. For example, Lee and co‐workers investigated the cell surface glycoproteomics of human‐derived neuroblastoma tumors in mice and the expression on normal tissue based on data from the HPA (https://www.proteinatlas.org), developed a CAR specific against one of the antigens found, and tested this CAR in vitro as well as in vivo using a mouse model [39]. A key advantage of our pipeline is that we utilize RNA samples obtained from patients to identify overexpressed antigens. This ex vivo data approach increases the likelihood that the results have actual clinical relevance compared to data obtained from cell lines or other tumor models. Another difference is the direct comparison with an established target antigen, namely CSPG4. This strategy offers the benefit of having a reference point to assess tumor restrictiveness of the different candidates, at least at the RNA level. Furthermore, since a CSPG4‐specific CAR is available and has already been examined in preclinical studies [37], we could also use CSPG4 as a benchmark in the functional testing of the new CARs. This helped us put important characteristics of the newly‐generated CARs into context.
A major advantage of this pipeline is its tumor entity‐independent design, enabling transfer to other cancer types. While the approach proved generally suitable, further optimization may be possible through refined selection criteria and in vitro experiments. For example, tumor restrictiveness could be assessed at the protein, rather than only RNA, level, although this would require reliable quantitative protein expression databases. Additionally, earlier in vitro analysis of protein expression in healthy tissues, such as by immunohistochemistry, could help exclude unsuitable targets before CAR design and functional testing.
3.5. Limitations of the Pipeline for the Identification of Novel Target Antigens
Although T cells transduced with ErbB3‐ or SLC1A5‐specific CARs were effectively activated by antigen‐positive tumor cells and demonstrated efficient antigen‐specific cytotoxicity, both targets exhibited characteristics that may limit their clinical applicability. In particular, the physiological expression and functional relevance of SLC1A5 on immune cell populations represent a potential safety concern for in vivo translation. Therefore, further investigation is required to establish whether ErbB3 and SLC1A5, and their respective CAR constructs, possess a sufficiently favorable safety profile for clinical application.
Potential on‐target/off‐tumor toxicity should initially be assessed by characterizing ErbB3 and SLC1A5 cell‐surface expression across a broad panel of healthy human tissues in vitro. Immunohistochemical analysis of clinically relevant tissue specimens could provide information on target distribution and identify tissues potentially susceptible to CAR‐mediated recognition. For SLC1A5, the effects of SLC1A5‐specific CAR‐T cells on other immune cell populations should additionally be investigated using in vitro and subsequent in vivo functional assays assessing immune cell viability and function. These studies would help determine whether physiological SLC1A5 expression results in clinically relevant off‐tumor effects.
The translational relevance of these findings should also be evaluated in patient‐derived tumor specimens from individuals with UM and CM. Immunohistochemical analyses could determine target prevalence, expression intensity, and intra‐ and intertumoral heterogeneity, thereby providing an indication of the proportion of patients who may potentially benefit from antigen‐directed CAR‐T cell therapy.
Following further validation in vitro, the anti‐tumor efficacy and safety of the different CAR constructs should be assessed in vivo using appropriate murine tumor models. Such studies could evaluate CAR‐T cell persistence, tumor control, and overall anti‐tumor activity while potentially identifying toxicities that cannot be adequately captured by in vitro systems.
Overall, the findings for ErbB3 and SLC1A5 provide encouraging evidence supporting their potential as CAR‐T cell targets. However, comprehensive preclinical validation remains necessary, particularly regarding target distribution, antigen heterogeneity, on‐target/off‐tumor toxicity, and in vivo efficacy. Given the potential severity of CAR‐T cell‐associated adverse events, these risks should be thoroughly characterized before first‐in‐human clinical trials.
Interestingly, Baker et al. independently identified ErbB3 recently as a potentially suitable melanoma CAR‐T cell target using an AI‐driven antigen discovery approach. However, as in the present study, therapeutic efficacy and safety were not evaluated in an in vivo model, leaving an important aspect of preclinical validation unresolved [40].
4. Conclusions
In summary, this study presents a novel in silico/in vitro pipeline for identifying potential CAR‐T cell targets in cancer. It demonstrates its applicability through the successful experimental validation of antigen‐specific CAR‐T cells. The in vitro data confirmed ErbB3 as a promising target, as ErbB3‐specific CAR‐T cells displayed high activity against UM cells. Furthermore, we identified SLC1A5 as a promising target for CM, which warrants further investigation. Our results align with previous evidence of the relevance of both target candidates in various tumor entities, highlighting their therapeutic potential in melanoma treatment, as well as their broader role in cancer therapy. This new approach lays the foundation for efficient and consistent target antigen discovery across various cancer entities.
5. Materials and Methods
To identify new tumor‐specific cell surface antigens and generate antigen‐specific CAR‐T cells, we implemented the following workflow (Figure 1):
Computational identification of tumor cell surface antigens: We collected and processed bulk RNA sequencing data from public repositories and in‐house samples, selected genes identified as transmembrane proteins that were sufficiently expressed in the tumor, and short‐listed antigens whose expression was negligible in healthy tissues.
Evidence‐based selection of target antigens: From the short‐listed antigens, we systematically queried the literature and databases of tissue protein expression and patents to select those plasma membrane antigens expressed in tumor but not in healthy tissue, without published CAR‐T cells and with commercially available targeting antibodies bearing a known scFv sequence.
In vitro validation of selected antigens: We confirmed that the antigens were expressed in tumor cell lines but not in relevant immune cells by flow cytometry. Next, we used RNA encoding an antigen‐specific CAR to transfect T cells, and experimentally validated the antigen‐specific immune activation and cytolytic activity of the resulting CAR‐T cells.
In the following, we report the results rendered by the approach for CM and UM. The individual steps are explained in detail below.
5.1. Collection and Processing of Transcriptomic Data From Uveal and Cutaneous Melanoma
We acquired transcriptomic data from 80 UM samples in a previous study [18] and supplemented this dataset with 16 UM samples collected during the in‐house clinical trial NCT01983748. The studies involving human subjects or material were reviewed and approved by the Ethics Committee of the Friedrich‐Alexander Universität Erlangen‐Nürnberg. The principal investigator and all other investigators ensured that the studies were conducted in full conformity with the principles set forth in the Declaration of Helsinki. In the case of CM, we obtained transcriptomic data from 12 pre‐ICI treatment biopsies of non‐responding patients ([41], GSE78220) and from 3 biopsies of non‐responding patients on ICI therapy [42]. The transcriptomics data from 80 UM samples [18] used in our analysis is accessible via CBioportal (http://www.cbioportal.org/study?id=uvm_tcga). The transcriptomics data from 12 CM samples of pre‐ICI treatment biopsies of non‐responding patients [41] used in our analysis can be accessed via Gene Expression Omnibus (GEO, GSE78220). The transcriptomics data from three biopsies of non‐responding patients on ICI therapy [42] employed in the analysis can be accessed via Gene Expression Omnibus (GEO, GSE96619). For better comparability, we transformed the fragments per kilobase million (FPKM) CM data obtained from publications to transcript per million (TPM). Regarding the 16 primary in‐house UM samples, the raw Next Generation Sequencing (NGS) data was obtained from a service provider and delivered to us along with data on technical quality control (CeGaT, Tübingen, Germany). We utilized an in‐house established NGS analysis pipeline to perform quality control, map the reads to the reference genome, and quantify gene expression [20]. Precisely, we (1) aligned the generated reads to the reference genome assembly GRCh38.p12 using the short‐read aligner STAR (Spliced Transcripts Alignment to a Reference) [43], (2) estimated the mapping rate for each sample (all samples had mapping rates above 80 %), (3) quantified transcript level gene expression using the transcript assembler StringTie [44], and (4) annotated the transcripts using the comprehensive genomic reference annotation provided by the Gencode Consortium (version 28 for the reference assembly GRch38) [45]. In the case of UM, we further used the R package limma [46] to perform batch correction to prevent the introduction of batch effects upon merging transcriptomic data from two different sources. To demonstrate homogeneity across all samples, we performed a PCA (see Figure S1).
5.2. Bioinformatics Pipeline to Identify Cell Surface‐Specific Tumor Antigens in UM and CM
To select cell surface tumor antigens in UM, we first downloaded the entire Protein Data Bank of transmembrane proteins (PDBTM) [19, 47] with their respective protein database (pdb) identifiers. To make transcriptomic and protein data comparable, we converted the pdb identifiers into Ensembl gene identifiers using the R package biomaRt [48, 49], and intersected all gene identifiers from the PDBTM with genes detected in UM at any TPM. Next, we removed all UM cell surface‐expressed genes with a mean expression lower than 10 TPM. To ensure that the selected cell surface antigens are as constrained to the tumor as possible, we acquired data on 53 healthy tissues from the whole‐body map sequencing project GTEx [21], and calculated summary statistics for all the 33,731 genes found in the database (mean, median, 25th, 50th, and 75th percentile). To select tumor‐restricted cell surface antigens, we defined the RR of a gene as follows:
| (1) |
Here, stands for the expression of in the set of considered tumor samples, and is the 75th quantile expression in the healthy tissue showing the highest expression.
Gene expression data was visualized after log2‐transformation using the R package ComplexHeatmap [50, 51].
We calculated the RR score for all cell surface‐specific tumor antigens and compared it to the RR score for the well‐established CAR antigen chondroitin sulphate proteoglycan 4 (CSPG4), which is used here as a threshold to select superior antigens for further investigation. For the tumor antigen candidates scoring better than CSPG4, we performed a manual selection based on predicted and measured cellular localization, combined protein expression and literature research.
We followed almost the same procedure for the CM cell surface‐specific tumor antigen selection. Due to the low number of on‐treatment samples available (n = 3), we performed the first part of the analysis for both the pre‐treatment and the on‐treatment datasets. After filtering for genes expressed at least 10 TPM in the pre‐treatment dataset, we intersected the data with the on‐treatment results to keep only genes present in both groups. This procedure ensures a more robust analysis because it uses data from more samples while still selecting genes expressed in samples undergoing ICI therapy.
The scripts of this pipeline are stored and freely accessible at https://zenodo.org/records/17242857.
5.3. Manual Selection of Target Antigens
After the computational preselection of candidates, manual research was performed for all remaining targets.
5.3.1. Investigation of Predicted Localization
The predicted membrane expression was investigated based on data from the “COMPARTMENTS subcellular localization database” [52] found in Genecards (www.genecards.com). COMPARTMENTS localization data is integrated from literature manual curation, high‐throughput microscopy‐based screens, predictions from primary sequence, and automatic text mining. This database predicts the subcellular expression of an antigen including a confidence score out of 5. Unified confidence scores of the localization evidence are assigned based on evidence type and source. Since plasma membrane expression is a prerequisite for a candidate protein to be recognized by a CAR, we only considered antigens with the highest score (i.e., 5) as relevant, and therefore, all antigens with a predicted plasma membrane expression with a confidence score of 4 or less were excluded from further examination.
5.3.2. Combined Protein Expression Score
To assess the protein expression of the different antigens in healthy tissues, data from The HPA (www.proteinatlas.org) was used [53, 54]. Additionally, the reported protein expression on different tumors was investigated using data from The HPA as well. Based on these two factors, a combined score was developed with the following criteria: (1) no general/ubiquitous protein expression on healthy tissues/cells and (2) at least moderate protein expression in different tumor entities or in melanoma in particular. Antigens that fulfilled both of these criteria were chosen for further investigation.
5.3.3. Literature Research
A literature search in PUBMED (https://pubmed.ncbi.nlm.nih.gov) was performed to investigate whether there are any publications that suggest a relation of the antigens to cancer and possibly a role as an oncogenic driver. For this, the different synonyms of the antigens were used as keywords, in combination with keywords such as “tumor,” “cancer,” “tumorigenesis,” “oncogene” and other related terms.
Additionally, a literature search was performed using PUBMED and Google Patents (https://patents.google.com/) to determine whether there were any CARs directed against the antigens published and/or patented or any CAR‐based therapies published at the time of the antigen selection process. In addition to the synonyms for the antigens, keywords that were used included “CAR,” “chimeric antigen receptor,” “CAR‐T cell,” and “CAR‐T cell therapy” and synonyms of these terms.
An internet search for the availability of commercial antibodies recognizing the extracellular part of the protein was performed using Labome (www.labome.com) and Abcam (www.abcam.com) to identify suitable antibodies for flow cytometric measurement of antigen surface expression.
To identify published and/or patented sequences of single chain variable fragments (scFvs) specific for the different candidate antigens, an internet search was performed using PUBMED and Google Patents. For this, synonyms of the different antigens were used, combined with keywords such as “scFv,” “single chain variable fragment,” and others.
5.4. Human Immune Effector Cells and Tumor Cell Lines
Primary immune effector cells were obtained by isolating peripheral blood mononuclear cells (PBMCs) using density gradient centrifugation with Lymphoprep reagent (#1114547, Axis‐Shield, Oslo, Norway). Fresh whole blood of healthy donors was used after written informed consent and approval by the institutional review board (Ethics Committee of the Friedrich‐Alexander‐Universität Erlangen‐Nürnberg, Erlangen, Germany: Ref. no.16‐251_4‐B; date of approval July 21, 2023).
For some experiments, PBMCs were separated into a non‐adherent fraction (NAF) and monocytes by plastic adherence, as described previously in detail [55]. Alternatively, CD14‐positive cells were isolated from PBMCs using magnetic anti‐CD14‐microbeads (#130‐050‐201, Miltenyi Biotec, Bergisch Gladbach, Germany) according to the manufacturer's instructions.
PBMCs, NAF, or CD14‐negative fractions were used to isolate CD8+ cytotoxic T cells using magnetic anti‐CD8‐microbeads (#130‐045‐201, Miltenyi Biotec) according to the manufacturer's instructions. Afterwards, isolated CD8+ T cells were incubated overnight at a concentration of 2–3 × 106 cells/mL in R10 medium (consisting of RPMI 1640 (#P04‐18047, PAN‐Biotech, Aidenbach, Germany), supplemented with 10% foetal calf serum (#P30‐3033, heat‐inactivated; PAN‐Biotech), 1% L‐glutamine (#17‐605E BE17‐605E, 200 mM, Lonza, Verviers, Belgium), 0.2% HEPES buffer (1 M, PAA, GE healthcare, Piscataway, NY, USA), 0.04% β‐mercaptoethanol (#P07‐05020, 50 mM, PAN‐Biotech), and 0.04% gentamycin (#P06‐13100, 50 mg/mL, PAN‐Biotech)), supplemented with 1000 IU/mL interleukin 7 (#130‐095‐367, IL‐7; Miltenyi Biotec), and then used in further experiments.
To generate large numbers of T cells, a T cell expansion was performed as described in [37].
Cell lines included HEK293T, a derivate of human embryonic kidney cells, the CM cell lines Mel526 ([56]; kindly provided by Prof. Dr. H. Abken [primary CM]) and A375M ([57]; kindly provided by Dr. C. Aarnoudse [metastasis of CM]), and the UM cell lines 92.1 ([58]; from M.J.J. [primary UM]), Mel270 (primary UM), Omm2.5, Omm2.3 ([59]; from Dr. B. R. Ksander, Schepens Eye Research Institute, Boston, USA; Omm‐1,3, Omm‐1,5 from liver metastases from the same patient as Mel270. Also known as Omm2.3 and Omm2.5), and Omm1 (subcutaneous metastasis of UM) ([60]; from Prof. Dr. G. P. M. Luyten, Rottterdam University Hospital, Rotterdam, The Netherlands). All human immune effector cells and cell lines were cultured in R10 medium. ErbB3 positive: MEL526, 92.1, Omm1, Omm2.5; ErbB3 negative: HEK293T. SLC1A5 positive: A375M, Omm1; SLC1A5 negative: CD8‐negative fraction of a monocyte‐depleted fraction of PBMCs.
5.5. Generation of CARs and Transfection Into T Cells
Five different second‐generation CARs were used for transfer into T cells: (1) a CAR directed against CSPG4 (CSPG4HL‐CD28/CD3ζ‐CAR), which has been described previously [33], (2) an ErbB3‐specific CAR with a scFv based on an anti‐ErbB3 antibody [61] with the complete scFv sequence deposited in Genebank (https://www.ncbi.nlm.nih.gov/nuccore/AF048774.1) (referred to as ErbB3‐CAR H6), (3) an ErbB3‐specific CAR with a scFv described by Zhang et al. [62] (referred to as ErbB3‐CAR 22), (4) an SLC1A5‐specific CAR with a scFv based on a patented antibody sequence (patent number EP2311946B1 or US8268592B2; inventors: Norihiko Shiraishi et al.)(referred to as SLC1A5‐CAR 1), and (5) a SLC1A5‐specific CAR with a scFv based on another patented antibody sequence (patent number US20190367605A1 or EP3538150A4; Inventors: Nabendu Pore et al.) (referred to as SLC1A5‐CAR 2). All new CARs were designed with the same backbone as the CSPG4HL‐CD28/CD3ζ‐CAR [33].
All CAR constructs were cloned into the pGEM4Z[‐64A] vector backbone. The encoding template plasmids were linearized and in vitro transcription (IVT) was performed as described previously [33] using a mMESSAGE mMACHINE T7 Ultra kit (#AM1345‐5, Ambion, ThermoFisher Scientific, Waltham, MA, USA) according to the manufacturer's instructions. Polyadenylation was performed to enhance the stability of the mRNA according to the manufacturer's instructions (#AM1350, Invitrogen (by ThermoFisher Scientific), Eugene, OR, USA). Purification of mRNA was performed with an RNeasy Mini‐kit (#74106, QIAGEN GmbH, Hilden, Germany) following the manufacturer's instructions. The mRNAs were then aliquoted in the desired quantities and stored at ‐20°C for later experimental use.
Transfection of isolated CD8‐positive T cells or expanded T cells with the different CAR constructs was performed using RNA electroporation, as described previously [63]. The electroporated cells rested in the incubator at 37°C for 3 to 4 h before performing further experiments.
5.6. Flow Cytometry
Antigen expression by different cell lines and ex vivo human immune effector cells was determined by flow cytometry, using either a phycoerythrin (PE)‐labeled (1) anti‐human CSPG4 antibody (mouse anti‐chondroitin sulphate [clone 9.2.27]; BioLegend, San Diego, California), (2) anti‐human erbB3/HER‐3 mouse monoclonal antibody (#324706, clone 1B4C3; BioLegend), (3) anti‐SLC1A5 antibody (#ABIN6988141, Anti‐Solute Carrier Family 1 Member 5 [SLC1A5] [AA465‐515]; Antibodies‐online, Limerick, PA, USA), (4) anti‐APH1a polyclonal rabbit antibody (#BSS‐BS‐4259R‐PE, Biozol, Hamburg, Germany), (5) anti‐Nicastrin rabbit monoclonal antibody (#ab275621, Abcam, Cambridge, UK), or (6) primary unlabeled anti‐ADIPOR2 polyclonal goat antibody (#ab77612, Abcam) combined with a secondary PE‐labeled donkey anti‐goat antibody (Beckman Coulter Inc., Brea, CA).
Surface expression of the transfected CAR constructs on T cells was analyzed by flow cytometry 4 and approximately 20–24 h after electroporation, as well as 4 and 24 h after thawing cryopreserved cells using a goat‐F(ab’)2 anti‐human IgG antibody (#2042‐02, Southern Biotech, Birmingham, AL, USA; Fluorescein Isothiocyanate (FITC)‐conjugated. A summary of the gating strategy that has been used for analysis can be found in Figure S9.
All samples were measured using the FACS Canto II (BD, Franklin Lakes, NJ, USA). Data analysis was performed using FlowJo, version 10.9 (BD Biosciences, San José, CA, USA).
5.7. Antigen‐Specific CAR‐T Cell Activation
To evaluate whether CAR‐T cells were activated upon antigen‐specific contact, the upregulation of the expression of the surface activation markers CD25 and CD69 on the T cells was analyzed. Additionally, anti‐CD3 or anti‐CD8 antibodies were added to specifically gate for (cytotoxic) T cells. CD8+ T cells were electroporated as described and co‐cultured overnight to allow for protein production, expression, or secretion at a 1:1 ratio with target cells in a 96‐well round‐bottom plate at 37°C. After approximately 17 h, the supernatant was harvested for later cytokine analysis. The co‐cultured cells were stained using an anti‐CD25‐PE (#555432, BD Biosciences, Franklin Lakes, NJ, USA) and anti‐CD3‐FITC (#349201, BD Biosciences) combined with a staining of anti‐CD69‐PE (#654975, BD Biosciences) and anti‐CD3‐FITC (BD Biosciences). Alternatively, a staining mix consisting of anti‐CD8‐PerCP (#980916, BioLegend, San Diego, California), anti‐CD25‐FITC (#347643, BD Biosciences), and anti‐CD69‐PE (BD Biosciences) was used. A summary of gating strategies that were used for analysis can be found in Figures S10 and S11.
To assess the cytokine secretion by CAR‐T cells upon antigen‐specific target cell contact, the supernatants of the activation assay described above were analyzed with a BD Cytometric Bead Array (CBA) Human Th1/Th2 Cytokine Kit II (#551809, BD Biosciences, San José, CA, USA) following the manufacturer's instructions. Measurements were performed using the FACS Canto II (BD, Franklin Lakes, NJ, USA) and the data were analyzed using FCS Express software version 5 (DeNovo Software, Pasadena, CA, USA). For a better comparison, a ratio of the cytokine levels to the concentrations in the T cell‐only condition or the negative control condition (containing the different CAR‐T cells together with HEK293T cells or CD8‐negative fraction cells) was calculated before performing statistical tests.
5.8. Cytotoxicity Assay
To assess the cytotoxicity of the ErbB3‐ and SLC1A5‐specific CAR‐T cells, they were co‐cultured with antigen‐negative target cells and with antigen‐positive target cells. Each antigen‐specific CAR was compared with CSPG4‐specific CAR‐T cells and mock‐transfected CAR‐T cells. To examine ErbB3‐specific CAR‐T cells, the ErbB3‐negative cell line HEK293T was used as a negative target while the metastatic UM cell line Omm1 was used as an ErbB3‐positive target. To assess the cytotoxic capacity of SLC1A5‐specific CAR‐T cells, a frozen monocyte‐depleted fraction of PBMCs (NAF) was thawed and used as a negative control while the melanoma cell line A375M was used as an SLC1A5‐positive target.
5.8.1. CFSE Labeling of Cells
To label the target cells, CFSE (#65085084; 5(6)‐CFDA, SE [5(6)‐carboxyfluorescein diacetate, succinimidyl ester] [500 µg]; Invitrogen) was used. HEK293T cells were labeled with 0.25 µM CFSE, while Omm1 cells were labeled with 3.0 µM CFSE. NAF cells were labeled with 1.0 µM CFSE, and A375M cells were labeled with 6.0 µM CFSE. For CFSE labeling, 10–20 × 106 cells per staining condition were resuspended in 1–2 mL of the corresponding staining solution and incubated in the dark for 10 min at room temperature before stopping the staining with 1–2 mL foetal calf serum (FCS) and filling the Falcon tube with phosphate‐buffered saline‐ethylenediaminetetraacetic acid (PBS‐EDTA) (4°C) to reach a total volume of 15 mL. The staining was stopped by keeping the Falcon tube on ice in the dark for 5 min. The cell suspension was then centrifuged at 200 g for 10 min. The cells were resuspended in PBS‐EDTA (4°C) and counted. The cells were centrifuged again at 200 g for 10 min and then resuspended in the calculated volume of R10 medium to adjust to a concentration of 2 × 106 cells/mL.
5.8.2. FACS‐Based Cytotoxicity Measurement
HEK293T cells and Omm1 cells were mixed at a 3:7 ratio, while NAF cells and A375M cells were mixed at a 1:1 ratio. The target cells were then distributed onto a 48‐well‐plate. Electroporated CAR‐T cells and mock‐electroporated T cells were used as effector cells by co‐incubating them with the different target cells for 18 to 22 h at 1:1, 1:2.5, 1:5, and 1:10 target‐to‐effector cell ratios in R10 medium in a 48‐well plate in duplicates. After overnight co‐incubation of target cells and CAR‐T cells, the cells were harvested and stained in 100 µl of ZombieNIR (Zombie Near‐Infrared viability dye) staining solution to label dead cells. The solution consisted of ZombieNIR Dye (#423106, BioLegend, San Diego, California, USA; prepared according to manufacturer's instructions) diluted in PBS (4°C) at a 1:58 ratio. The samples were measured using a FACS Canto II (BD) and analyzed with FlowJo, version 10.9 (BD Biosciences). The percentage of living control cells (ZombieNIR negative, CFSE low) and living target cells (ZombieNIR negative, CFSE high) was determined by analyzing the fluorescence intensities. A summary of gating strategies that have been used for analysis can be found in Figures S12 and S13. The percentage of lysis was calculated. For this, the ratios of living target cells and living control cells in conditions with the different target‐to‐effector ratios (T cell‐mediated lysis) were compared to the same ratio in a condition without any specific effector cells present (background apoptosis) according to the following formula:
| (2) |
5.9. Assessment of Cryopreserved CAR‐T Cells
To assess whether cryopreserved CAR‐T cells show a similar cytotoxic potency, the T cells were frozen 3–5 h after electroporation and stored at −80°C for up to 10 months. After thawing, the T cells were rested for about an hour. They were then co‐incubated with target and control cells.
5.10. Statistical Analysis and Data Visualization
GraphPad Prism, version 10 (GraphPad Software, San Diego, CA, USA) was used to perform statistical analysis and to create graphs. Depending on the experiment, a paired t‐test or an unpaired t‐test with Welch correction was performed to calculate p‐values. Details of the different statistical methods used in the individual experiments can be found in the respective figure legends.
For p > 0.05, differences were considered not significant (displayed as “ns” in figures). For p ≤ 0.05, differences between groups were considered significant (displayed in figures using the following symbols: * indicates p ≤ 0.05, ** indicates p ≤ 0.01, and *** indicates p ≤ 0.001).
Author Contributions
Kilian F. Juraschek: methodology, data analysis, writing – original draft preparation, writing – review and editing. Esther Güse: methodology, data analysis, writing – original draft preparation, writing – review and editing. Maartje M. C. Hof: methodology, data analysis, writing – review and editing. Christopher Lischer: methodology, data analysis, writing – review and editing. Niklas Feuchter: methodology, data analysis, writing – review and editing. Beatrice Schuler‐Thurner: data analysis, writing – review and editing. Elias A. T. Koch: data analysis, writing – review and editing. Martine J. Jager: writing – review and editing. Markus Biburger: writing – review and editing, supervision. Carola Berking: writing – review and editing. Jan Dörrie: writing – review and editing, supervision. Julio Vera: conceptualization, writing – original draft preparation, writing – review and editing, supervision, funding acquisition. Niels Schaft: conceptualization, writing – original draft preparation, writing – review and editing, supervision, funding acquisition. All authors have read and agreed to the published version of the manuscript.
Funding
This work was supported by a grant of the Hiege‐Stiftung/Die Deutsche Hautkrebs‐Stiftung to J.V. and N.S.; and by a grant of the Forschungsstiftung Medizin am Universitätsklinikum Erlangen to N.S. and J.V.
Ethics Statement
The blood of healthy donors was obtained after written informed consent and approval by the institutional review board (Ethics Committee of the Friedrich‐Alexander‐Universität Erlangen‐Nürnberg, Erlangen, Germany: Ref. no. 16–251_4‐B; date of approval July 21, 2023).
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Supporting File 1: mco271154‐sup‐0001‐SuppMat.pdf
Acknowledgments
We express our gratitude to all voluntary blood donors and to the medical staff of the Uniklinikum Erlangen from the Department of Dermatology for their support with blood donations. In addition, we would like to thank Annett Hamann, Nadine Hellmuth, and Manuel Wiesinger for valuable advice and technical assistance. We thank Hinrich Abken (Leibnitz Institute, Regensburg, Germany) for providing us with the cutaneous melanoma cell line Mel526 and the CAR backbone. We also thank Corlien Aarnoudse (Universiteit Leiden, The Netherlands) for the cutaneous melanoma cell line A375M. The present work was performed in (partial) fulfilment of the requirements forobtaining the degree ‘Dr. med.’ from the Friedrich‐Alexander‐Universität Erlangen‐Nürnberg (FAU) for Kilian Juraschek.
Open access funding enabled and organized by Projekt DEAL.
Juraschek K. F., Güse E., Hof M. M. C., et al. “Integrative Bioinformatics‐Driven Discovery of the Novel Chimeric Antigen Receptor (CAR)‐T Cell Targets Erb‐B2 Receptor Tyrosine Kinase 3 (ErbB3) and Solute Carrier Family 1 Member 5 (SLC1A5) in Melanoma.” MedComm 7, no. 12 (2026): e71154. 10.1002/mco2.71154
Julio Vera and Niels Schaft share senior authorship.
Data Availability Statement
The dataset(s) supporting the conclusions of this article are available in repositories or are included within the article and its additional files as listed: The transcriptomics data from 80 UM samples used in our analysis are accessible via CBioportal (http://www.cbioportal.org/study?id = uvm_tcga). The transcriptomics data from 12 CM samples of pre‐ICI treatment biopsies of non‐responding patients used in our analysis can be accessed via Gene Expression Omnibus (GEO, GSE78220). The transcriptomics data from three biopsies of non‐responding patients on ICI therapy employed in the analysis can be accessed via Gene Expression Omnibus (GEO, GSE96619). The transcriptomics data from metastatic UM patient material (Trial NCT04335890) of the selected candidate target antigens, as well as currently examined targets, CSPG4, and PRAME (depicted in TPM) are summarized in Table S3. The scripts of the pipeline are stored and freely accessible at https://zenodo.org/records/17242857. Other datasets used and/or analyzed during the current study not listed above are available from the corresponding author on reasonable request.
References
- 1. Dabas P. and Danda A., “Revolutionizing Cancer Treatment: A Comprehensive Review of CAR‐T Cell Therapy,” Medical Oncology 40, no. 9 (2023): 275. [DOI] [PubMed] [Google Scholar]
- 2. June C. H. and Sadelain M., “Chimeric Antigen Receptor Therapy,” New England Journal of Medicine 379, no. 1 (2018): 64–73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Casucci M., Hawkins R. E., Dotti G., and Bondanza A., “Overcoming the Toxicity Hurdles of Genetically Targeted T Cells,” Cancer Immunology, Immunotherapy 64, no. 1 (2015): 123–130. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Brownlee E., Turley M., and Nations H., “Chimeric Antigen Receptor Therapy in Hematologic Malignancies,” JAAPA 38, no. 2 (2025): 17–24. [DOI] [PubMed] [Google Scholar]
- 5. Amorós‐Pérez B., Rivas‐Pardo B., Gómez Del Moral M., Subiza J. L., and Martínez‐Naves E., “State of the Art in CAR‐T Cell Therapy for Solid Tumors: Is There a Sweeter Future?” Cells 13, no. 9 (2024): 725. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Schaft N., “The Landscape of CAR‐T Cell Clinical Trials Against Solid Tumors—A Comprehensive Overview,” Cancers 12, no. 9 (2020): 2567. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Swetter S. M., Johnson D., Albertini M. R., et al., “NCCN Guidelines® Insights: Melanoma: Cutaneous, Version 2.2024: Featured Updates to the NCCN Guidelines,” Journal of the National Comprehensive Cancer Network 22, no. 5 (2024): 290–298. [DOI] [PubMed] [Google Scholar]
- 8. Cao J., Ding X., Ji J., Zhang L., and Luo C., “Efficacy and Safety of Immune Checkpoint Inhibitors Rechallenge in Advanced Solid Tumors: A Systematic Review and Meta‐Analysis,” Frontiers in Oncology 14 (2024): 1475502. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Seidel J. A., Otsuka A., and Kabashima K., “Anti‐PD‐1 and Anti‐CTLA‐4 Therapies in Cancer: Mechanisms of Action, Efficacy, and Limitations,” Frontiers in Oncology 8 (2018): 86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Jager M. J., Shields C. L., Cebulla C. M., et al., “Uveal Melanoma,” Nature Reviews Disease Primers 6, no. 1 (2020): 24. [DOI] [PubMed] [Google Scholar]
- 11. Rantala E. S., Hernberg M., and Kivelä T. T., “Overall Survival After Treatment for Metastatic Uveal Melanoma: A Systematic Review and Meta‐Analysis,” Melanoma Research 29, no. 6 (2019): 561–568. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Koch E. A. T., Heppt M. V., and Berking C., “The Current State of Systemic Therapy of Metastatic Uveal Melanoma,” American Journal of Clinical Dermatology 25, no. 5 (2024): 691–700. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Piulats J. M., Watkins C., Costa‐García M., et al., “Overall Survival From Tebentafusp Versus Nivolumab Plus Ipilimumab in First‐Line Metastatic Uveal Melanoma: A Propensity Score‐Weighted Analysis,” Annals of Oncology 35, no. 3 (2024): 317–326. [DOI] [PubMed] [Google Scholar]
- 14. Forsberg E. M. V., Lindberg M. F., Jespersen H., et al., “HER2 CAR‐T Cells Eradicate Uveal Melanoma and T‐Cell Therapy‐Resistant Human Melanoma in IL2 Transgenic NOD/SCID IL2 Receptor Knockout Mice,” Cancer Research 79, no. 5 (2019): 899–904. [DOI] [PubMed] [Google Scholar]
- 15. Harrer D. C., Dörrie J., and Schaft N., “CSPG4 as Target for CAR‐T‐Cell Therapy of Various Tumor Entities‐Merits and Challenges,” International Journal of Molecular Sciences 20, no. 23 (2019): 5942. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Ventin M., Cattaneo G., Arya S., et al., “Chimeric Antigen Receptor T Cell With an Inducible Caspase‐9 Suicide Gene Eradicates Uveal Melanoma Liver Metastases via B7‐H3 Targeting,” Clinical Cancer Research 30, no. 15 (2024): 3243–3258. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Mohanty R., Manoswini M., Dhal A. K., and Ganguly N., “In Silico Analysis of a Novel Protein in CAR T‐Cell Therapy for the Treatment of Hematologic Cancer Through Molecular Modelling, Docking, and Dynamics Approach,” Computers in Biology and Medicine 151, no. Pt A (2022): 106285. [DOI] [PubMed] [Google Scholar]
- 18. Robertson A. G., Shih J., Yau C., et al., “Integrative Analysis Identifies Four Molecular and Clinical Subsets in Uveal Melanoma,” Cancer Cell 32, no. 2 (2017): 204–220.e15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Kozma D., Simon I., and Tusnády G. E., “PDBTM: Protein Data Bank of Transmembrane Proteins After 8 Years,” Nucleic Acids Research 41 (2013): D524–D529. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Lischer C., Eberhardt M., Flamann C., et al., “Gene Network‐Based and Ensemble Modeling‐Based Selection of Tumor‐Associated Antigens With a Predicted Low Risk of Tissue Damage for Targeted Immunotherapy,” Journal for ImmunoTherapy of Cancer 12, no. 5 (2024): e008104. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. The_GTEx_Consortium . “The Genotype‐Tissue Expression (GTEx) Project,” Nature Genetics 45, no. 6 (2013): 580–585. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Majumder A., “HER3: Toward the Prognostic Significance, Therapeutic Potential, Current Challenges, and Future Therapeutics in Different Types of Cancer,” Cells 12, no. 21 (2023): 2517, 10.3390/cells12212517. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Reschke M., Mihic‐Probst D., van der Horst E. H., et al., “HER3 is a Determinant for Poor Prognosis in Melanoma,” Clinical Cancer Research 14, no. 16 (2008): 5188–5197. [DOI] [PubMed] [Google Scholar]
- 24. Tiwary S., Preziosi M., Rothberg P. G., Zeitouni N., Corson N., and Xu L., “ERBB3 is Required for Metastasis Formation of Melanoma Cells,” Oncogenesis 3, no. 7 (2014): e110. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Li W. J., Xie C. Y., Zhu X., Tang J., Wang L., and Lou L. G., “SIBP‐03, a Novel Anti‐HER3 Antibody, Exerts Antitumor Effects and Synergizes With EGFR‐ and HER2‐Targeted Drugs,” Acta Pharmacologica Sinica 45, no. 4 (2024): 857–866. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Zuo B. L., Yan B., Zheng G. X., et al., “Targeting and Suppression of HER3‐Positive Breast Cancer by T Lymphocytes Expressing a Heregulin Chimeric Antigen Receptor,” Cancer Immunology, Immunotherapy 67, no. 3 (2018): 393–401. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Lee J., Song J., Yoo W., et al., “Therapeutic Potential of Anti‐ErbB3 Chimeric Antigen Receptor Natural Killer Cells Against Breast Cancer,” Cancer Immunology, Immunotherapy 74, no. 2 (2025): 73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Scalise M., Pochini L., Console L., Losso M. A., and Indiveri C., “The Human SLC1A5 (ASCT2) Amino Acid Transporter: From Function to Structure and Role in Cell Biology,” Frontiers in Cell and Developmental Biology 6 (2018): 96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Ren W., Liu G., Yin J., et al., “Amino‐Acid Transporters in T‐Cell Activation and Differentiation,” Cell Death & Disease 8, no. 3 (2017): e2655. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. van Geldermalsen M., Wang Q., Nagarajah R., et al., “ASCT2/SLC1A5 Controls Glutamine Uptake and Tumour Growth in Triple‐Negative Basal‐Like Breast Cancer,” Oncogene 35, no. 24 (2016): 3201–3208. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Wang W., Pan H., Ren F., Chen H., and Ren P., “Targeting ASCT2‐Mediated Glutamine Metabolism Inhibits Proliferation and Promotes Apoptosis of Pancreatic Cancer Cells,” Bioscience Reports 42, no. 3 (2022): BSR20212171. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Edwards D. N., Ngwa V. M., Raybuck A. L., et al., “Selective Glutamine Metabolism Inhibition in Tumor Cells Improves Antitumor T Lymphocyte Activity in Triple‐Negative Breast Cancer,” Journal of Clinical Investigation 131, no. 4 (2021): e140100. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Krug C., Birkholz K., Paulus A., et al., “Stability and Activity of MCSP‐Specific Chimeric Antigen Receptors (CARs) Depend on the scFv Antigen‐Binding Domain and the Protein Backbone,” Cancer Immunology, Immunotherapy 64, no. 12 (2015): 1623–1635. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Chen J., Qiu S., Li W., et al., “Tuning Charge Density of Chimeric Antigen Receptor Optimizes Tonic Signaling and CAR‐T Cell Fitness,” Cell Research 33, no. 5 (2023): 341–354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Rodriguez‐Marquez P., Calleja‐Cervantes M. E., Serrano G., et al., “CAR Density Influences Antitumoral Efficacy of BCMA CAR T Cells and Correlates With Clinical Outcome,” Science Advances 8, no. 39 (2022): eabo0514. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Shimabukuro‐Vornhagen A., Gödel P., Subklewe M., et al., “Cytokine Release Syndrome,” Journal for ImmunoTherapy of Cancer 6, no. 1 (2018): 56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Wiesinger M., März J., Kummer M., et al., “Clinical‐Scale Production of CAR‐T Cells for the Treatment of Melanoma Patients by mRNA Transfection of a CSPG4‐Specific CAR Under Full GMP Compliance,” Cancers 11, no. 8 (2019): 1198. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Schäfer D., Tomiuk S., Küster L. N., et al., “Identification of CD318, TSPAN8 and CD66c as Target Candidates for CAR T Cell Based Immunotherapy of Pancreatic Adenocarcinoma,” Nature Communications 12, no. 1 (2021): 1453. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Lee J. Y., Jonus H. C., Sadanand A., et al., “Identification and Targeting of Protein Tyrosine Kinase 7 (PTK7) as an Immunotherapy Candidate for Neuroblastoma,” Cell Reports Medicine 4, no. 6 (2023): 101091. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Baker D. J., Frommer L. M., Uslu U., et al., “AI‐Driven Discovery of GPNMB CAR T Cells as a Multi‐Cancer Therapy,” Cell 189, no. 13 (2026): 3871–3882.e12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Hugo W., Zaretsky J. M., Sun L., et al., “Genomic and Transcriptomic Features of Response to Anti‐PD‐1 Therapy in Metastatic Melanoma,” Cell 165, no. 1 (2016): 35–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Garcia‐Diaz A., Shin D. S., Moreno B. H., et al., “Interferon Receptor Signaling Pathways Regulating PD‐L1 and PD‐L2 Expression,” Cell Reports 19, no. 6 (2017): 1189–1201. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Dobin A., Davis C. A., Schlesinger F., et al., “STAR: Ultrafast Universal RNA‐Seq Aligner,” Bioinformatics 29, no. 1 (2013): 15–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Pertea M., Pertea G. M., Antonescu C. M., Chang T. C., Mendell J. T., and Salzberg S. L., “StringTie Enables Improved Reconstruction of a Transcriptome From RNA‐Seq Reads,” Nature Biotechnology 33, no. 3 (2015): 290–295. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Frankish A., Diekhans M., Ferreira A. M., et al., “GENCODE Reference Annotation for the Human and Mouse Genomes,” Nucleic Acids Research 47, no. D1 (2019): D766–D773. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. Ritchie M. E., Phipson B., Wu D., et al., “limma Powers Differential Expression Analyses for RNA‐Sequencing and Microarray Studies,” Nucleic Acids Research 43, no. 7 (2015): e47. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47. Tusnády G. E., Dosztányi Z., and Simon I., “Transmembrane Proteins in the Protein Data Bank: Identification and Classification,” Bioinformatics 20, no. 17 (2004): 2964–2972. [DOI] [PubMed] [Google Scholar]
- 48. Durinck S., Spellman P. T., Birney E., and Huber W., “Mapping Identifiers for the Integration of Genomic Datasets With the R/Bioconductor Package biomaRt,” Nature Protocols 4, no. 8 (2009): 1184–1191. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Durinck S., Moreau Y., Kasprzyk A., et al., “BioMart and Bioconductor: A Powerful Link Between Biological Databases and Microarray Data Analysis,” Bioinformatics 21, no. 16 (2005): 3439–3440. [DOI] [PubMed] [Google Scholar]
- 50. Gu Z., Eils R., and Schlesner M., “Complex Heatmaps Reveal Patterns and Correlations in Multidimensional Genomic Data,” Bioinformatics 32, no. 18 (2016): 2847–2849. [DOI] [PubMed] [Google Scholar]
- 51. Gu Z., “Complex Heatmap Visualization,” iMeta 1, no. 3 (2022): e43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52. Binder J. X., Pletscher‐Frankild S., Tsafou K., et al., “COMPARTMENTS: Unification and Visualization of Protein Subcellular Localization Evidence,” Database 2014 (2014): bau012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53. Uhlén M., Fagerberg L., Hallström B. M., et al., “Proteomics. Tissue‐Based Map of the Human Proteome,” Science 347, no. 6220 (2015): 1260419. [DOI] [PubMed] [Google Scholar]
- 54. Uhlén M., Björling E., Agaton C., et al., “A Human Protein Atlas for Normal and Cancer Tissues Based on Antibody Proteomics,” Molecular & Cellular Proteomics 4, no. 12 (2005): 1920–1932. [DOI] [PubMed] [Google Scholar]
- 55. Berger T. G., Feuerstein B., Strasser E., et al., “Large‐Scale Generation of Mature Monocyte‐Derived Dendritic Cells for Clinical Application in Cell Factories,” Journal of Immunological Methods 268, no. 2 (2002): 131–140. [DOI] [PubMed] [Google Scholar]
- 56. Marincola F. M., Shamamian P., Alexander R. B., et al., “Loss of HLA Haplotype and B Locus Down‐Regulation in Melanoma Cell Lines,” Journal of Immunology 153, no. 3 (1994): 1225–1237. [PubMed] [Google Scholar]
- 57. Kozlowski J. M., Hart I. R., Fidler I. J., and Hanna N., “A Human Melanoma Line Heterogeneous With Respect to Metastatic Capacity in Athymic Nude Mice,” JNCI 72, no. 4 (1984): 913–917. [PubMed] [Google Scholar]
- 58. De Waard‐Siebinga I., Blom D. J., Griffioen M., et al., “Establishment and Characterization of an Uveal‐Melanoma Cell Line,” International Journal of Cancer 62, no. 2 (1995): 155–161. [DOI] [PubMed] [Google Scholar]
- 59. Chen P. W., Murray T. G., Uno T., Salgaller M. L., Reddy R., and Ksander B. R., “Expression of MAGE Genes in Ocular Melanoma During Progression From Primary to Metastatic Disease,” Clinical & Experimental Metastasis 15, no. 5 (1997): 509–518. [DOI] [PubMed] [Google Scholar]
- 60. Luyten G. P., Naus N. C., Mooy C. M., et al., “Establishment and Characterization of Primary and Metastatic Uveal Melanoma Cell Lines,” International Journal of Cancer 66, no. 3 (1996): 380–387. [DOI] [PubMed] [Google Scholar]
- 61. Merchant A. M., Zhu Z., Yuan J. Q., et al., “An Efficient Route to Human Bispecific IgG,” Nature Biotechnology 16, no. 7 (1998): 677–681. [DOI] [PubMed] [Google Scholar]
- 62. Zhang J., Zhang X., Liu Q., et al., “Mammalian Cell Display for Rapid Screening scFv Antibody Therapy,” Acta Biochim Biophys Sin 46, no. 10 (2014): 859–866. [DOI] [PubMed] [Google Scholar]
- 63. Schaft N., Dörrie J., Müller I., et al., “A New Way to Generate Cytolytic Tumor‐Specific T Cells: Electroporation of RNA Coding for a T Cell Receptor Into T Lymphocytes,” Cancer Immunology, Immunotherapy 55, no. 9 (2006): 1132–1141. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting File 1: mco271154‐sup‐0001‐SuppMat.pdf
Data Availability Statement
The dataset(s) supporting the conclusions of this article are available in repositories or are included within the article and its additional files as listed: The transcriptomics data from 80 UM samples used in our analysis are accessible via CBioportal (http://www.cbioportal.org/study?id = uvm_tcga). The transcriptomics data from 12 CM samples of pre‐ICI treatment biopsies of non‐responding patients used in our analysis can be accessed via Gene Expression Omnibus (GEO, GSE78220). The transcriptomics data from three biopsies of non‐responding patients on ICI therapy employed in the analysis can be accessed via Gene Expression Omnibus (GEO, GSE96619). The transcriptomics data from metastatic UM patient material (Trial NCT04335890) of the selected candidate target antigens, as well as currently examined targets, CSPG4, and PRAME (depicted in TPM) are summarized in Table S3. The scripts of the pipeline are stored and freely accessible at https://zenodo.org/records/17242857. Other datasets used and/or analyzed during the current study not listed above are available from the corresponding author on reasonable request.
