Purpose: The success of immunotherapy for the treatment of metastatic cancer is contingent on the identification of appropriate target antigens. Potential targets must be expressed on tumors but show restricted expression on normal tissues. To maximize patient eligibility, ideal target antigens should be expressed on a high percentage of tumors within a histology and, potentially, in multiple different malignancies.
Design: A Nanostring probeset was designed containing 97 genes, 72 of which are considered potential candidate genes for immunotherapy. Five established melanoma cell lines, 59 resected metastatic melanoma tumors, and 31 normal tissue samples were profiled and analyzed using Nanostring technology.
Results: Of the 72 potential target genes, 33 were overexpressed in more than 20% of studied melanoma tumor samples. Twenty of those genes were identified as differentially expressed between normal tissues and tumor samples by ANOVA analysis. Analysis of normal tissue gene expression identified seven genes with limited normal tissue expression that warrant further consideration as potential immunotherapy target antigens: CSAG2, MAGEA3, MAGEC2, IL13RA2, PRAME, CSPG4, and SOX10. These genes were highly overexpressed on a large percentage of the studied tumor samples, with expression in a limited number of normal tissue samples at much lower levels.
Conclusion: The application of Nanostring RNA counting technology was used to directly quantitate the gene expression levels of multiple potential tumor antigens. Analysis of cell lines, 59 tumors, and normal tissues identified seven potential immunotherapy targets for the treatment of melanoma that could increase the number of patients potentially eligible for adoptive immunotherapy. Clin Cancer Res; 19(18); 4941–50. ©2013 AACR.
Accurate quantitation of RNA levels is an essential step in the identification of potential tumor antigens. Nanostring is a solution-based gene expression profiling technology that accurately counts individual RNA molecules in the small amounts of total RNA (250 ng or less) that are often obtained from biopsy samples. We used this technology to study potential tumor antigen gene expression in 59 metastatic melanoma samples and identified seven genes as potential targets for adoptive immunotherapy.
The development of successful immunotherapy for metastatic cancer requires the identification of appropriate target antigens. As immunotherapeutic strategies become increasingly sophisticated and powerful, finding antigens that are overexpressed in malignancies but have restricted expression in normal tissue becomes challenging. To date, the most successful immunotherapeutic approach is the adoptive cell transfer (ACT) of tumor-infiltrating lymphocytes (TIL), with objective response rates of more than 70% and complete response rates of approximately 40% reported in trials treating patients with metastatic melanoma (1). This strategy necessitates the acquisition of tumor specimens for generation of TIL and has primarily shown success in treating melanoma. An alternative approach is the infusion of lymphocytes that have been harvested from the patient and genetically engineered to recognize tumor-associated antigens (1, 2).
Tumor antigen-reactive T-cell receptor (TCR) gene therapies have been used with success in multiple histologies; however, the number of patients that can be treated is somewhat limited as they must express a specific human leukocyte antigen (HLA; e.g. HLA-A*0201; refs. 2, 3). The restricted expression of certain target antigens in tumor cohorts can also limit a therapy's potential use. For example, effective TCR therapies have been reported targeting the cancer-testis antigen (CTA) NY-ESO-1; however, only around 20% to 30% of melanomas express this antigen (4, 5). Clinical trials using non–MHC-restricted chimeric antigen receptor (CAR) therapy can potentially expand the number of patients eligible for ACT if the target antigen is expressed on the cell surface (2). Recently, CAR therapies have shown success in treating non-melanoma and non-solid organ cancers, specifically the ability of the anti-CD19 CAR to effect regression of advanced B-cell malignancies (6, 7).
The most effective way to maximize the number of patients potentially eligible for a therapy would be to target an antigen expressed on a high percentage of tumors within a given histology. This study uses Nanostring technology to achieve gene expression profiling of melanoma cell lines, metastatic melanoma tumors, and normal human tissue samples to identify potential target antigens for immunotherapy. This robust technology uses unique digital color-coded barcodes that hybridize directly to specific nucleic acid targets and allow for detection and quantitation of hundreds of transcripts in a single reaction. Unlike microarray approaches, it facilitates the direct measurement of mRNA expression levels for subsequent gene expression analysis, and has been shown to be highly reproducible and as sensitive as real-time PCR assays while still allowing for the measurement of multiple genes at one time (8).
Materials and Methods
Sample collection, RNA isolation, and cell lines
Patients seen at the Surgery Branch, National Cancer Institute (NCI; Bethesda, MD) for treatment of metastatic melanoma underwent surgical excision of metastatic lesions for harvest of TILs in protocols approved by the Institutional Review Board and Food and Drug Administration. Viable-appearing fragments of these tumors were freed from surrounding normal tissue, collected, and either stored in RNAlater (Ambion) or flash-frozen and stored at −80° Celsius, until RNA isolation was conducted using a RNEasy Mini Kit (Qiagen). Analyzed tumor samples were collected between September 2007 and December 2012. RNA isolation was conducted in the same fashion for established human melanoma cell lines initiated at Memorial Sloan-Kettering (SKmel23) or the NCI Surgery Branch (all others). The lines were grown under standard conditions in RPMI-1640 with 10% FBS medium at 37°C, 5%CO2. For normal tissues, commercially available RNA samples were used (Agilent, Ambion, Biochain, Clontech).
All tumors samples were confirmed to be metastatic melanoma at the time of harvest by pathological evaluation including immunohistochemistry. Immunohistochemical staining was carried out for expression of the antigen NY-ESO-1 (encoded by gene CTAG1B) with the specific anti-NY-ESO-1 monoclonal antibody E978 (Invitrogen; ref. 5). Immunohistochemical scores were assigned for intensity of staining and percentage of tumors cells that stained positive.
Using the Nanostring nCounter Analysis System (Nanostring Technologies), gene expression analysis was conducted for each sample as previously described using a custom-designed codeset containing 97 genes (8). Each reaction contained 250 ng of total RNA in a 5 μL aliquot, plus reporter and capture probes, and 6 pairs of positive control and 8 pairs of negative control probes. Analysis and normalization of the raw Nanostring data was conducted using nSolver Analysis Software v1.1 (Nanostring Technologies). Raw counts were normalized to internal levels of 7 reference genes: CNOT2, GAPDH, HPRT1, PHGDH, SUMO2, SYS1, and WDR45L. A background count level was estimated using the average count of the 8 negative control probes in every reaction plus 2 SDs.
Gene expression analysis
Principal component analysis (PCA) and ANOVA analysis were conducted using the Partek Genomic Suite (Partek Incorporated). PCA was used to characterize samples on the basis of their gene expression profiles. ANOVA analysis was used to identify differentially expressed genes (significant P value < 0.05) and samples were clustered by hierarchic clustering.
Flow cytometry and quantitative-PCR
Flow cytometry (FACS) was conducted using conjugated monoclonal antibody (mAb) specific for human chondroitin sulfate proteoglycan 4 (CSPG4) according to the manufacturers' instructions (R&D Systems). Reverse transcription (RT) was conducted using High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Quantitative PCR was carried out with the TaqMan Fast Universal PCR Master Mix (Applied Biosystems) by the use of a 7500 Fast Real-Time PCR System (Applied Biosystems). Copy numbers were generated using a standard curve from CSPG4 plasmid and results were normalized against β-actin (ACTB).
Samples and Nanostring probeset
Gene expression profiling using Nanostring technology was conducted on RNA from five melanoma lines (mel1300, mel526, mel624.38, mel888, and SKmel23), 59 resected metastatic melanoma tumor deposits, and 31 normal tissue samples. A total of 97 genes were included in the probeset (Table 1). Immune-related genes were included to investigate the immune characteristics of melanoma samples; however, these would not be considered as potential immunotherapy target antigens. After elimination of the control genes (n = 7) and the immune genes (n = 18), 72 candidate genes remained. The candidate genes were grouped as follows for organizational purposes: melanoma-related, cancer testis antigen (CTA), glioblastoma-related, and other tumor-related genes. The sites of resection for the tumor deposits were primarily subcutaneous tissue (46%), lymph node (24%), lung (15%), and liver (8%). One tumor was resected from each of the following sites: adrenal, pelvic mass, retroperitoneum, and small bowel. Before detailed analysis of the data, we used PCA to determine whether anatomic location of the tumor influenced gene expression profiling. PCA did not show differentiation of gene expression profiles based on the site of tumor harvest (Supplementary Fig. S1). All 59 patients who underwent tumor resection had stages III or IV melanoma. They ranged in age from 19 to 66 years (average 47 ± 13 years) and 71% of patients were male.
Validation of Nanostring data
To compare the sensitivity of Nanostring to other methods of gene expression analysis, we used quantitative real-time PCR (RT-PCR) and fluorescence antibody staining. Nanostring counts for the high-molecular weight melanoma-associated antigen (HMW-MAA) gene CSPG4 for the 5 melanoma lines showed high correlation with CSPG4 copy numbers generated by RT-PCR (R2 = 0.99435; Fig. 1A) and mean fluorescence intensity (MFI) values generated by flow cytometric analysis (R2 = 0.97418; Fig. 1B). Of the 59 tumors samples, 41 had immunohistochemical staining conducted for NY-ESO-1 at the time of resection. Despite the fact that immunohistochemical analysis was conducted by multiple individuals over several years, results of both the intensity of staining and the percentage of tumor cells that stained were shown to compare well with Nanostring RNA counts for NY-ESO-1 (Fig. 1C and 1D).
Selection of potential target genes
On the basis of the counts obtained with the 8 pairs of negative control probes (average, 9.0; STD, 5.8), 20 was chosen as the background level for gene detection. To set a value that would be likely for potential immune recognition (PIR), we used our previously published data on the ability of the engineered T cells to recognize 3 tumor antigens: NY-ESO-1 (encoded by gene CTAG1B), MART-1 (encoded by gene MLANA), and CSPG4 (formerly known as HMW-MAA, encoded by gene CSPG4; refs. 9–11). These reports used the same melanoma lines analyzed in this study as targets and we observed that tumor cell lines with Nanostring counts greater than 100 for a given target gene consistently showed effector cytokine release when they were cocultured with genetically modified peripheral blood lymphocytes (PBL), whereas lines with counts lower than 100 did not reproducibly demonstrate such reactivity (Supplementary Tables S1 and S2).
For each gene, the percentage of tumors showing PIR positivity (PIR+, defined as a Nanostring count > 100) was determined as was the average Nanostring count of the PIR+ tumors (Table 2; Supplementary Table S3). Of the 72 potential target genes, 5 were not expressed in any of the tumor samples (Fig. 2; Supplementary Table S3). Of the remaining 67 genes, 33 were PIR+ in more than 20% of tumor samples. Twenty percent was chosen as a cut-off value based on our previous experience targeting NY-ESO-1 (5). RNA from 31 normal tissues was then subject to Nanostring Analysis with the same codeset (Supplementary Table S4). Analysis of these data by PCA showed clear differences in gene expression profiles between normal tissues and tumor samples (Fig. 3A). The positioning of the normal tissues testis and spleen were likely attributable to the expression of the cancer testis genes and immune genes, respectively. We then conducted hierarchic clustering of the combined tumor and normal tissue datasets and observed that 27 of the initial 72 candidate genes were differentially expressed between tumors and normal tissues, with at least a two-fold higher expression in tumors (Supplementary Fig. S2). Of these 27 genes, 20 were PIR+ in more than 20% of tumors samples (Table 2, Fig. 3B). Of those 20, MAGEA12 was eliminated from further consideration because of severe reported toxicities in a previous clinical trial (12). The melanocyte differentiation antigens MLANA, PMEL, and TYR were eliminated because they were targeted in previous trials and, although clinical responses were achieved, patients suffered associated skin, eye, and ear toxicities (13, 14). Sixteen genes remained as potential target antigens for consideration (Table 3).
MITF was expressed in tumors at an average level that is 8.5 times higher than the average expression in normal tissues (Table 3); however, high levels of expression (counts >1,000) in diaphragm, muscle, and uterus (Supplementary Table S4) eliminated it from consideration. ST3GAL5 was eliminated secondary to expression at very high levels in multiple tissues including the brain, adrenal, thyroid, spleen, and artery (Supplementary Table S4). Likewise, PTPRZ1 was eliminated because of high levels of expression in the brain, brainstem, and artery samples (Supplementary Table S4). AURKB, IGF2BP3, and KIF20A all showed low levels of expression on limited normal tissues; however, the average level of expression detected on tumors was sufficiently low (ratio < 2.0) to render them non-ideal targets (Table 3). Of the remaining 10 genes, 8 encode CTAs and exhibit little-to-no normal tissue expression outside of testis. Of these 8 genes, the highest levels of tumor RNA expression (all >1,000) were seen for CSAG2, IL13RA2, MAGEA3, MAGEC2, and PRAME, all of which warrant further consideration as possible targets (Table 3). The remaining non-CTA candidate genes were CSPG4 and SOX10. Both exhibited low levels of expression in a number of normal tissues (skin, trachea, vein, heart, lung, diaphragm, muscle, adipose, uterus, prostate, thymus, spleen, bone marrow, and gastrointestinal organs for CSPG4 and brain, brainstem, trachea, spleen, artery, and breast for SOX10); however, the average levels of expression in tumors for both were substantially higher than in normal tissues (6.6 times higher for CSPG4 and 4.6 times higher for SOX10). Therefore, seven genes are identified as potential immunotherapy targets: CSAG2, MAGEA3, MAGEC2, IL13RA2, PRAME, CSPG4, and SOX10.
The identification of target antigens for immunotherapy is a complex process that involves assessing the expression of antigens on tumors and normal tissues. CTAs are of particular interest as immunotherapy targets because they are expressed in multiple cancers of diverse histologic origin, including breast cancer, prostate cancer, non-small cell lung cancer, gastrointestinal cancers such as colon and esophageal cancers, bladder cancer, and melanoma. Aside from expression in male germ cells, CTA expression in normal human tissues is relatively restricted (15). Previous studies have examined gene expression profiles of this antigen group in melanoma and other cancers using microarray technology and RT-PCR (16, 17). Herein, we identified five genes encoding CTAs, which are expressed at high levels on a large percentage of melanoma tumor samples studied: CSAG2, MAGEA3, MAGEC2, IL13RA2, and PRAME.
CSAG2, also known as taxol-resistance-associated gene-3 (TRAG3), was overexpressed on 71% of studied melanoma tumors with no significant expression on any normal tissues. It is overexpressed in multiple other histologies including carcinoma of the bladder, cervix, breast, esophagus, bile duct, stomach, colon, and lung, and its expression has been correlated with poor prognosis in both ovarian cancer and osteosarcoma (18–21). Preclinical studies have identified a potential target CSAG2-directed T-cell epitope capable of inducing cytotoxic lymphocytes (CTL; ref. 22). MAGEA3 is a very appealing target given its overexpression on a high percentage of metastatic melanoma tumors (75% in this study) and its potential as a target for ACT in other histologies such as colorectal cancer, lung cancer, breast cancer, esophageal cancer, and glioblastoma (23, 24). A particular epitope, MAGEA3 112–120, was targeted in TCR gene therapy trials in MAGEA3+ patients with observed cancer regression but severe associated neurologic toxicity and mortality, likely secondary to TCR cross-recognition of this epitope in MAGEA12, which was found to be expressed in the brain (12). The generation of a completely MAGEA3-specific TCR, however, could allow for its targeting in future studies. MAGEC2, formerly known as hepatocellular carcinoma-associated antigen 587 (HCA587), is also overexpressed in a variety of cancers aside from melanoma, including hepatocellular carcinoma, gallbladder carcinoma, medulloblastoma, multiple myeloma, and squamous cell cancer of the head and neck (25–28). It was overexpressed on 39% of tumors in this study, with no expression on any normal tissues aside from testis. Several preclinical studies have identified potential target epitopes and have generated CTL that show functionality against target cells (29, 30).
IL13RA2 was overexpressed on 32% of studied tumors and showed low levels of expression on the liver and adrenal samples. It has been associated with adrenocortical carcinoma, glioblastoma multiforme, and systemic sclerosis, and IL13RA2-specific CARs have shown efficacy against glioma cell targets in preclinical studies (31–33). PRAME was overexpressed in 86% of melanoma samples, and its average count on overexpressing tumors (2899) was the highest among any of the potential CTA targets. It was absent on any normal tissues in this cohort although other studies have reported some expression on normal tissues such as endometrium, ovaries, and adrenals (34, 35). Its expression has been widely reported in other cancers including carcinoma of the lung and kidney, squamous cell carcinoma of the head and neck, sarcomas, mammary cancer, multiple myeloma, and acute leukemias (34). It is one of the few CTAs that is commonly expressed in leukemic malignancies. Preclinical studies have generated PRAME-specific T cells that showed activity across multiple histologies encompassing both solid organ and hematologic malignancies (36, 37).
CSPG4 and SOX10 are both melanoma-associated genes that were expressed in 92% and 90% of tumors in this study, respectively. Unlike CTA, they do exhibit expression on multiple normal tissues, although at much lower levels than on tumors. This does raise concerns regarding application in the clinical setting and argues for safety measures such as dose-escalation trials or the implementation of a suicide gene (38). CSPG4 is a highly immunogenic cell surface proteoglycan which was identified on melanoma cells in the 1970s, and it has been found on glioblastoma, triple-negative breast cancer, head and neck squamous cell cancer, mesothelioma, renal cell cancer, and sarcoma as well (39–43). It has been targeted with vaccines in clinical trials with no reported toxicity, and antitumor effects have been reported in preclinical immunotherapy models with melanoma and head and neck squamous cell cancers targets (9, 40, 44). SOX10 is a transcription factor that is expressed on neural crest cells and melanocytes, and has been shown to be crucial for the maintenance of neoplastic cells (45, 46). In addition to being widely expressed on melanomas and other lesions such as giant congenital nevi, it has also been identified on cancers of the breast and prostate (45, 47–49). Naturally occurring anti-SOX10 CTLs were identified in a patient with a dramatic response to immunotherapy (50).
This study identified genes that have potential as immunotherapy targets based on their expression in a high percentage of studied melanoma tumors with limited expression in normal tissues. One possible study limitation is the correlation between gene expression and antigen expression. We have shown that, for CSPG4 and NY-ESO-1, there is a strong association between the level of gene expression as assessed by Nanostring and the degree of antigen expression, as determined by both FACS and immunohistochemistry; however, this may not be the case for every gene. Importantly, these data do not directly address issues involving the potential safety of a given target gene.
To have the potential for clinical application, each new target gene would require the generation of reagents (TCR or CAR) that can mediate specific antigen recognition and, the target antigen must not be expressed on a vital tissue. In the case of CTAs, while it is widely reported that these genes are cancer specific, this is not universally true. As we recently reported, the MAGEA12 gene is strongly expressed in rare neurons in the human brain and expression in these isolated cells was likely sufficient to initiate a destructive immune response leading to death in some patients (12). On the other hand, we have used the identical strategy of TCR gene therapy to target NY-ESO-1, and in more than 40 patients treated, have not observed any target-related toxicities. Clearly, more detailed studies (e.g., multiple tissue immunohistochemistry) would be needed before any of these new antigens can be targeted in clinical trials; however, Nanostring provides a reliable way to test multiple candidate genes at once and select attractive potential targets for further investigation.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: R.E. Beard, D. Abate-Daga, R.A. Morgan
Development of methodology: R.E. Beard, D. Abate-Daga, Z. Zheng
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.E. Beard, S.F. Rosati, J.R. Wunderlich
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.E. Beard, D. Abate-Daga, S.F. Rosati, S.A. Rosenberg, R.A. Morgan
Writing, review, and/or revision of the manuscript: R.E. Beard, D. Abate-Daga, S.F. Rosati, S.A. Rosenberg, R.A. Morgan
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.E. Beard, S.A. Rosenberg, R.A. Morgan
Study supervision: R.A. Morgan
This work is supported by the Intramural Research Program of the Center for Cancer Research, National Cancer Institute, NIH.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The authors thank Arnold Mixon and Shawn Farid for technical support with FACS analysis and the Laboratory of Pathology Department at the National Cancer Institute for their role in immunohistochemical staining.
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
- Received May 7, 2013.
- Revision received June 20, 2013.
- Accepted July 9, 2013.
- ©2013 American Association for Cancer Research.