A Genetic Polymorphism in CTLA-4 Is Associated with Overall Survival in Sunitinib-Treated Patients with Clear Cell Metastatic Renal Cell Carcinoma Free

Renal cell carcinoma (RCC) is the most common type of kidney cancer in adults. Almost 70% of RCC is of clear cell histology, and 20% to 25% of patients have metastatic spread by the time they are diagnosed with RCC (1, 2). Even though new cases of RCC remain relatively few compared with other more frequent malignancies, the global incidence and mortality are steadily increasing at a rate of 2% to 3% per decade (2).

For decades, the outcome for metastatic RCC (mRCC) treatment with IL2 and IFNα was disappointing. With the discovery of von Hippel–Lindau (VHL) gene mutations in the majority of clear cell RCC, therapies targeting angiogenesis were introduced in early 2003 including tyrosine kinase inhibitors (TKI), mTOR inhibitors, and the anti-VEGF mAb bevacizumab. With an improved understanding of how tumor cells evade antitumor response and how T cells can be redirected through activating or inhibiting receptors, the immune checkpoint pathway broadens the knowledge of the biological behavior of RCC. Checkpoint inhibitors, such as programmed death-1 (PD-1) inhibitors (pembrolizumab and nivolumab), PD-1 ligand (PD-L1) inhibitor (atezolizumab), and cytotoxic T-lymphocyte antigen 4 (CTLA-4) inhibitor (ipilimumab), have emerged as promising RCC treatments (3). Given the fact that angiogenesis pathway inhibitors interact with immune responses (4, 5), the potential synergy between checkpoint inhibitors and angiogenesis inhibitors has led to several combination trials in mRCC, such as sunitinib or pazopanib combined with nivolumab (6).

Upon activation, some T-cell receptors (such as CD28) positively regulate T cells, leading to tumor destruction, whereas others (such as PD-1 and CTLA-4) negatively regulate T cells, resulting in prolonged tumor survival. When PD-1 binds to PD-L1, the activation of T cells is inhibited, resulting in the suppression of T-cell attack and tumor immune escape (7). After binding to B7.1 (CD80) and B7.2 (CD86), CTLA-4 reduces the activated T-cell response to tumor cells, thereby failing to halt tumor progression (8). Generally, there is a balance among the receptors which makes the immune response of T cells maintain a proper intensity in order to protect normal cells from collateral damage (7). However, overexpression of PD-L1 and CTLA-4 was observed in patients with many tumor types, including clear cell mRCC (cc-mRCC; refs. 8–11).

Recently, the prognostic and/or predictive value of aforementioned immune checkpoints has been evaluated in several tumor types. It was reported in RCC patients that higher PD-L1 expression was associated with a larger tumor size and a higher risk of death (10, 12, 13). Subsequently, the association of higher PD-L1 tumor expression with shorter progression-free survival (PFS) and/or overall survival (OS) in mRCC patients treated by TKIs was reported by Choueiri and colleagues (14) and Fukuda and colleagues (15). The possible association between CTLA-4 expression and OS has been investigated in different tumor types with controversial results. A meta-analysis was conducted to pool all data, but consistent results were lacking possibly due to heterogeneity of the tumor types, differences in experimental methods (immunohistochemistry, ELISA, or PCR), and varying sample sources (blood or tumor) used in the studies (16).

Thus far, there are no studies investigating the role of single-nucleotide polymorphisms (SNP) in genes related to the immune checkpoints in cc-mRCC patients. In this study, we investigated whether genetic variants within CD274 (PD-L1), PDCD1 (PD-1), and CTLA-4 could be useful as predictive biomarkers for sunitinib treatment outcome in cc-mRCC patients.

The discovery cohort was composed of patients who participated in the EuroTARGET project (17). In brief, the EuroTARGET is a multicenter observational study aiming to identify biomarkers for prediction of response to TKI treatment in mRCC, within which a hypothesis-free genome-wide association study was conducted on sunitinib efficacy in cc-mRCC patients. Thus, clinical information and germline DNA variant chip data (Illumina Human OmniExpress BeadChip) were available (17). After quality control checks, 679,324 SNPs met the quality criteria applied (Supplementary Document S1). A total of 6,540,327 SNPs were available after imputation by impute2 using the 1000 Genomes (phase III integrated data set of 2,504 individuals) as reference panel. In the present study, patients with cc-mRCC who were treated by sunitinib as first TKI were included in the discovery cohort.

In order to capture the genetic variation in the target genes, genomic sequences of CD274, PDCD1, and CTLA-4 genes were retrieved from the 1000 Genomes Project (GRCh37.p13). For our study, we selected tagging SNPs using the SNP Tagger approach of the Haploview software package (version 4.2). Pairwise tagging with r² threshold 0.8 was used. Based on the 1000 Genomes Project, only SNPs with minor allele frequency (MAF) of 0.05 or higher were included. This approach allows to cover the common genetic variations in candidate genes (18), while reducing the number of tested SNPs.

Genotyping results of the selected tagging SNPs were extracted from the EuroTARGET project. To assess the quality of imputation, an estimated imputation “info” score was used, which was computed by impute 2 and took values between 0 and 1. An “info” score near 1 indicates that an SNP has been imputed with high certainty. SNPs with an “info” score lower than 0.9 were excluded from statistical analysis.

Positive hits were tested in an independent validation cohort consisting of cc-mRCC patients treated with sunitinib as first TKI. Details have been described previously (19). In brief, patients were enrolled between 2004 and 2010 in five medical centers in the Netherlands and in the Cleveland Clinic Foundation Taussig Cancer Institute in the United States. Germline DNA was isolated from whole blood, serum, plasma, or peripheral blood mononuclear cell samples and genotyped. A flowchart of the study design is shown in Fig. 1. Genotyping methods are described in Supplementary Document S2.

Study endpoints consisted of PFS and OS. PFS was defined as the time in months between the first day of sunitinib treatment and the date of progressive disease (according to RECIST version 1.1) or death due to cancer. If no progression was observed during sunitinib treatment, PFS was censored at the time of the last follow-up or death due to other reasons (whichever occurred first). OS was defined as the time in months between the first day of sunitinib treatment and the date of death or the date at which patients were last known to be alive. In the discovery cohort, associations between tagging SNPs and endpoints were univariately tested using an additive genetic model. Individual SNPs with P values lower than 0.1 were tested in a multivariable Cox-regression model together with well-established covariates age, gender, and Heng prognostic risk group (20), which have been repeatedly associated with PFS and OS. Subsequently, individual SNPs with a P value < 0.05 in the multivariable analysis were tested in the validation cohort using the same Cox-regression model and endpoint definitions. Finally, data from the discovery and validation cohort were pooled using a fixed effect meta-analysis. Patient characteristics between discovery and validation cohorts were compared by t test, χ² test, Mann–Whitney U test, or Kaplan–Meier survival analysis with the log-rank test depending on the type of data. All tests were two-sided and carried out using SPSS Statistical Package for Windows (version 23.0, IBM Corp). R (version 2.3.2) and package MICE (Multivariate Imputation by Chained Equations; ref. 21) were used to impute missing values for variables included in Heng prognostic risk group (World Health Organization performance status, hemoglobin, neutrophil count, thrombocytes, calcium, and time from diagnosis until start of sunitinib).

We used several bioinformatic tools to assess the possible functional relevance for genetic variants showing significant associations with endpoints in the final pooled analysis. The Genotype-Tissue Expression (GTEx, http://www.gtexportal.org/home/) portal was employed to identify potential associations between genetic variants and gene expression levels (eQTL) in all available tissues (22). The OncoLnc (http://www.oncolnc.org/; ref. 23) was used to link mRNA expression levels to the survival data from The Cancer Genome Atlas Kidney Renal Clear Cell Carcinoma (TCGA-KIRC) collection (24), which included 537 patients with different stages of cc-RCC.

A total of 550 and 138 sunitinib-treated cc-mRCC patients were included in the discovery and validation cohort, respectively. Patient characteristics for the discovery and validation cohorts are presented in Table 1. In brief, patients included in the discovery cohort were older (median age 63) than those in the validation cohort (median age 59). Nearly 93% of patients in the discovery cohort were classified in the intermediate or poor Heng prognostic risk group, whereas in the validation cohort, only 72% belonged to these prognostic groups. Consequently, patients in the validation cohort had longer PFS and OS compared with those in the discovery cohort. Only 86 (16%) patients in the discovery cohort and 14 (10%) patients in the validation cohort had one or more values missing as variables required for inclusion in a Heng prognostic risk group.

In the discovery analysis, we selected a total of 34 tagging SNPs in CD274, PDCD1, and CTLA-4 according to our predefined tagging SNP approach. Genotype results for 27 of 34 tagging SNPs were imputed based on 1000 genomes project. Seven SNPs with “info” score lower than 0.9 were excluded from statistical analysis (Supplementary Table S1).

In the univariate analysis, CTLA-4 rs231775 and CD274 rs7866740 were associated with PFS and OS with P values lower than 0.1 (Supplementary Table S1). After correction for age, gender, and Heng prognostic risk group, CTLA-4 rs231775 and CD274 rs7866740 remained significantly associated with OS [HR, 0.84; 95% confidence interval (CI), 0.72–0.98; P = 0.028 and HR, 0.73; 95% CI, 0.54–0.99; P = 0.047, respectively). The minor allele carriers of both SNPs had a better OS compared with that of major allele carriers. The association of CTLA-4 rs231775 and CD274 rs7866740 with PFS did not reach the significance threshold in the multivariate analysis (shown in Table 2).

In the validation cohort, the genotype call rate for CTLA-4 rs231775 and CD274 rs7866740 was 99.3% and 98.6%, respectively. Both SNPs were in Hardy–Weinberg equilibrium (P > 0.05). The MAF of both SNPs was similar in both the discovery cohort and the National Center for Biotechnology Information (NCBI) dbSNP database (CEU population, which represents Utah residents with Northern and Western European ancestry). Only 4 patients with the CD274 rs7866740 GG genotype were detected. As a consequence, a dominant model was used for the genetic association analysis.

In the validation study, none of the SNPs was significantly associated with PFS or OS either in the univariate analysis or after correction for age, gender, and Heng prognostic risk group. Nevertheless, CTLA-4 rs231775 showed the same direction of effect and comparable effect size for OS (HR, 0.74; 95% CI, 0.55–1.01; P = 0.057; Table 2) than that calculated in the discovery cohort (HR, 0.84; 95% CI, 0.72–0.98; P = 0.028).

In the pooled cohort, CTLA-4 rs231775 showed a significant association with OS (HR, 0.83; 95% CI, 0.72–0.95; P = 0.008). Patients with the GG genotype had longer OS (35.1 months) compared with the patients with an AG or AA genotype (30.3 and 24.3 months, respectively). The Kaplan–Meier plot for OS is presented in Supplementary Fig. S1. No significant associations were found for PFS (shown in Table 2).

We utilized data from GTEx and OncoLnc to provide possible explanations for our interesting finding that CTLA-4 rs231775 was associated with OS. In the vast majority of tissues in GTEx (Release V6p), GG genotype of CTLA-4 rs231775 was associated with a decreased mRNA expression of the CTLA-4 gene compared with the AA and AG genotypes (Supplementary Fig. S2A). Due to the strict significance threshold, the above association between CTLA-4 rs231775 and mRNA expression was only observed in testis tissue (P = 1.0 × 10⁻⁷; Supplementary Fig. S2B). By OncoLnc, OS was compared between cc-RCC patients with lower and higher CTLA-4 mRNA expression. It was revealed that patients with lower CTLA-4 mRNA expression (lower 50 percentile) had a longer OS (P = 0.00255, Supplementary Fig. S3).

In the present study, we explored genetic variants in the checkpoint-related genes CD274 (PD-L1), PDCD1 (PD-1), and CTLA-4 for a possible association with PFS and OS in cc-mRCC patients that received sunitinib as first TKI. The most important finding of our study is the identification of CTLA-4 rs231775 as a potential prognostic biomarker for OS. Patients with the GG genotype showed an increased OS compared with those with the GA or AA genotype. To our knowledge, this is the first time that an association of the CTLA-4 rs231775 genetic polymorphism with OS in this specific patient population is reported.

CTLA-4 rs231775 is located at position +49 in exon 1 of the CTLA-4 gene and is a common nonsynonymous CTLA-4 polymorphism. The +49 G allele encodes a Thr to Ala substitution at codon 17 in the signal peptide of the CTLA-4 protein (25). Anjos and colleagues have found in a cell-free model of in vitro reconstitution of translation and endoplasmic reticulum processing that the G-allele of CTLA-4 rs231775 was associated with inefficient glycosylation, which leads to a decrease of CTLA-4 expression in cell surface (26). In addition, Sun and colleagues have demonstrated that CTLA-4-Ala (coded by the G-allele) had a lower capability to bind the CTLA-4 ligand and a weaker inhibitory effect on T-cell activation compared with the functional effects of CTLA-4-Thr (coded by the A-allele; ref. 27). In a large Chinese population, they also found that subjects carrying the AA genotype were more susceptible for developing cancer than those with the GG genotype (27). Either by the decrease of protein expression or through lower binding capability, both studies reach to the conclusion that the G-allele is associated with reduced inhibition of activated T cells (27). In addition, Ligers and colleagues have reported in patients with myasthenia gravis and multiple sclerosis that G-allele carriers showed a significant decrease of CTLA-4 mRNA and protein expression (28). Therefore, it seems likely that the G-allele of CTLA-4 rs231775 is associated with a decrease of CTLA-4 expression.

We sought evidence that the decreased CTLA-4 expression contributes to an improved OS. CTLA-4 is thought to play a negative regulatory role after binding with its ligands. Hence, it could be hypothesized that low CTLA-4 expression leads to low ligand binding, because of which T cells can only be weakly inhibited facilitating an enhanced autoimmune response of possible benefit for the patient with cancer. We, therefore, investigated the relationship of CTLA-4 mRNA expression levels with OS by using the TCGA-KIRC dataset (23, 24), which revealed that cc-RCC patients (regardless of tumor stage) with lower CTLA-4 mRNA expression had longer OS. Combining all findings, the G-allele of CTLA-4 rs231775 is indeed associated with lower CTLA-4 mRNA expression, and lower mRNA expression links to longer OS, which is consistent with our results.

Our finding provides insight into RCC prognosis and corroborates the current strategies of new drug development for RCC. In this respect, ipilimumab, which is designed to inhibit the CTLA-4 protein, has demonstrated clinical efficacy and manageable toxicity as monotherapy or combined with nivolumab in mRCC (29). To date, there are no studies investigating the role of CTLA-4 polymorphisms or CTLA-4 protein expression on ipilimumab outcome in mRCC patients. However, in patients with melanoma, Breunis and colleagues have assessed seven CTLA-4 polymorphisms with ipilimumab response. In responding patients, there were proportionally more rs231775 A-allele carriers than G-allele carriers, whereas there was no difference in nonresponding patients (OR, 0.39; 95% CI, 0.18–0.82; P = 0.009; ref. 30). If confirmed in mRCC patients on the basis of the results presented here, rs231775 might be important for the optimization of ipilimumab treatment of mRCC.

In the present study in cc-mRCC patients that received sunitinib as first TKI, we were able to demonstrate an association of CTLA-4 rs231775 with OS, but not with PFS. Interestingly, Song and colleagues have also identified CTLA-4 rs231775 is a prognostic biomarker in patients with advanced non–small cell lung cancer (31). Patients with a GG or GA genotype experienced a significantly longer OS than those with an AA genotype after correction for many covariates among which was treatment (31). These findings strongly suggest that the CTLA-4 genetic variant is more of influence on the biological behavior of the disease than with the effects of treatment, implying the prognostic role of this polymorphism. Introduction of checkpoint inhibitors in the treatment of cancer patients may, however, change the relevance of CTLA-4 rs231775 for PFS.

Polymorphisms in PDCD1 (PD-1) and CD274 (PD-L1) are thought to be promising genetic candidates to explain differences in immunosuppressive function. It has been reported that PDCD1 rs10204525 was associated with OS in 439 patients with locoregional gastric cancer (32) as well as in 668 patients with resectable colorectal cancer (33), suggesting that it may serve as a prognostic factor. In our patient population of metastatic disease, however, we did not observe any significant association of PDCD1 rs10204525 (which was captured by PDCD1 rs41386349) with survival outcome. CD274 rs4143815 has been reported to be significantly associated with worse OS in a total of 354 patients with non–small cell lung cancer who underwent curative resection (34). We observed an association of CD274 rs7866740 with OS in the discovery cohort, but failed confirmation in the final analysis. Owing to the fact that genotype of CD274 rs7866740 was imputed and the MAF was relatively small, the significant association with CD274 rs7866740 in the discovery cohort might be detected by chance. Future studies should focus on the relevance of these SNPs in patients with no/microscopic disease after surgery versus those with advanced disease.

Whereas previous studies have focused on SNPs in genes of the VEGF signaling pathway, we are the first to explore the potential association of the genetic variability in three genes encoding immune checkpoints with the outcome in sunitinib-treated patients. Although we did not formally confirm the association in the validation cohort, the meta-analysis showed similar results to those of the discovery cohort and with stronger statistical evidence. In addition, a biological and mechanistic rationale was provided using expression and survival data from GTEx (22) and TCGA (23). However, we should interpret results from GTEx and TCGA carefully. It is because the significant association of genotype with CTLA-4 expression is found in testis not in kidney tissue (only 39 samples available in GTEx) or T cells in microenvironment. Moreover, the TCGA-KIRC datasets include patients with different stages, and follow-up period is not long enough. As a result, the plot in Supplementary Fig. S3 might be changed when tumor stage is taken into account and after enough follow-up period. The possible reasons of the failed validation could be the relatively small sample size as well as differences in patient characteristics between the two cohorts. Due to the need for imputation, adjustment of PFS and OS might slightly be distorted. To reduce this chance, a multiple imputation procedure instead of a single imputation was used. Moreover, imputation was implemented in 16% and 10% patients in the discovery and validation cohort, respectively, which could be considered negligible. The significance threshold in the discovery study was not adjusted by the strict Bonferroni correction in this exploratory study, because Bonferroni correction will increase the type II error (the chances of false-negative results; ref. 35). In our opinion, an inflated type I error due to multiple comparisons is better solved by a validation study than by a Bonferroni correction.

In conclusion, data from the present study show that the G-allele of CTLA-4 rs231775 is associated with an improved OS in cc-mRCC patients receiving sunitinib treatment, suggesting this polymorphism may serve as a prognostic marker.

K. Junker reports receiving commercial research grants and speakers bureau honoraria from, and is a consultant/advisory board member for Novartis. T. Eisen is an employee of AstraZeneca, reports receiving commercial research grants from AstraZeneca and Pfizer, holds ownership interest (including patents) in AstraZeneca, and is a consultant/advisory board member for EUSA. B.I. Rini reports receiving commercial research grants from and is a consultant/advisory board member for Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: X. Liu, J.J. Swen, H. Gelderblom, E. Oosterwijk, T. Eisen, L.A.L.M. Kiemeney, B.I. Rini, H.-J. Guchelaar

Development of methodology: X. Liu, J.J. Swen, M.H.M. Diekstra, A. Sverrisdottir, H.-J. Guchelaar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.J. Swen, M.H.M. Diekstra, E. Boven, D. Castellano, H. Gelderblom, R.H.J. Mathijssen, E. Oosterwijk, K. Junker, M. Roessler, K. Alexiusdottir, A. Sverrisdottir, M.T. Radu, V. Ambert, T. Eisen, A. Warren, C. Rodríguez-Antona, J. García-Donas, L.A.L.M. Kiemeney

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Liu, J.J. Swen, E. Boven, D. Castellano, E. Oosterwijk, A. Sverrisdottir, C. Rodríguez-Antona, S. Böhringer, K.K.M. Koudijs, H.-J. Guchelaar

Writing, review, and/or revision of the manuscript: X. Liu, J.J. Swen, M.H.M. Diekstra, E. Boven, H. Gelderblom, R.H.J. Mathijssen, S.H. Vermeulen, E. Oosterwijk, A. Sverrisdottir, M.T. Radu, T. Eisen, C. Rodríguez-Antona, J. García-Donas, S. Böhringer, L.A.L.M. Kiemeney, B.I. Rini, H.-J. Guchelaar

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Liu, M.H.M. Diekstra, S.H. Vermeulen, E. Oosterwijk, L.A.L.M. Kiemeney

Study supervision: J.J. Swen, D. Castellano, T. Eisen, L.A.L.M. Kiemeney, H.-J. Guchelaar

Other (review of histology): A. Warren

The authors thank Renée Baak-Pablo and Danielle Klootwijk for their assistance with genotyping.

The research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 259939. X. Liu is supported by the Chinese Scholarship Council (No. 201406220186).

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.