DNA Methylation of the SLC16A3 Promoter Regulates Expression of the Human Lactate Transporter MCT4 in Renal Cancer with Consequences for Clinical Outcome

Clear cell renal cell carcinoma (ccRCC) accounts for approximately 80% of renal carcinomas (1–3). The only curative treatment for localized ccRCC is partial or radical nephrectomy. However, patients remain at a considerable risk of recurrence and development of metastases after surgery (4), subsequently resulting in an overall survival for metastatic ccRCC of about 20% after 5-year follow-up. This poor survival rate is determined by resistance to radiation and chemotherapy and limited applicability of immunotherapy due to low response rates and severe toxicity (3, 5). Despite remarkable progress in the field of targeted therapies, identification of tumor-associated markers, which are able to predict patient outcome and at best may serve as target for ccRCC therapy, is of major interest.

Because several findings suggest that ccRCC is mainly a metabolic disease (6), tumor-specific changes in metabolic enzymes or transport proteins may provide novel therapeutic opportunities. Recent studies identified the monocarboxylate transporter 4 (MCT4) as metabolic target in a variety of tumors (e.g., colon cancer, breast cancer; refs. 7, 8). MCT4 (encoded by the SLC16A3 gene) mediates the H⁺-coupled transport of monocarboxylates, preferentially lactate, across the plasma membrane. Especially in highly glycolytic cells, such as tumor cells, lactate efflux is essential for energy metabolism to ensure the progression of glycolysis and for pH regulation to prevent intracellular acidification (9, 10). Genome-wide siRNA screening studies in a ccRCC cell line (11) showed that MCT4 may represent a promising metabolic key protein in ccRCC. Moreover, MCT4-mediated lactate efflux, explored in RCC cells by hyperpolarized ¹³C-pyruvate magnetic resonance, was significantly higher in metastatic RCC cells (12). However, detailed analyses not only of its expression, but more importantly of the regulation of MCT4 in ccRCC are limited.

Interestingly, in addition to the well-established relationship of HIF deregulation through VHL loss, genetic studies of ccRCC tumors identified mutations in genes encoding epigenetic regulators, such as PBRM1 or JARID1C (13). Accordingly, epigenetic modifications, such as DNA methylation seem to be important for gene regulation in ccRCC (14, 15). Although upregulation of MCT4 by hypoxia through an HIF-1α-mediated mechanism has been showed in HeLa cell line (16), the epigenetic regulation of MCT4 in ccRCC has not yet been reported and may offer clue to novel therapeutic strategy.

We therefore conducted comprehensive analyses of MCT4 expression on protein and mRNA level in independent cohorts of ccRCC patients and showed significant associations with clinicopathologic features and patient outcome. We elucidated DNA methylation in the SLC16A3 promoter region as underlying mechanism for expression differences of MCT4 in ccRCC, which was confirmed by in vitro cell culture experiments. Based on an independent ccRCC validation cohort, we provided first evidence for single CpG sites in the SLC16A3 promoter as novel epigenetic markers for clinical outcome in ccRCC.

A detailed description of the materials and methods used in the present study is given in the Supplementary Data.

Two independent retrospective cohorts (TuCP1 and TuCP2) of ccRCC patients treated at the Department of Urology, University Hospital Tuebingen, Germany, were investigated. The TuCP1 cohort consists of 207 ccRCC patients of Caucasian origin, who underwent partial or radical nephrectomy between 1993 and 2007 with respective paraffin-embedded ccRCC and corresponding nontumor tissues. For the TuCP2 cohort of 64 consecutive ccRCC patients, who underwent partial or radical nephrectomy between 2007 and 2010, in addition to paraffin-embedded tissues, surgically specimens of resected fresh-frozen ccRCC and matching nontumor tissues were available. Patients' characteristics, clinicopathologic features as well as survival data are given in Table 1. For further details about sample collection and definition of clinical endpoints, see Supplementary Data. Informed written consent was provided by each subject before the surgical resection and the use of the tissue was approved by the ethics committee of the University of Tuebingen. In addition, an independent cohort of ccRCC patients from The Cancer Genome Atlas (TCGA) with publically available, open-access clinical data as well as mRNA expression and genome-wide DNA methylation data was investigated. The cohort consists of 502 patients, for whom DNA methylation data (n = 283; Illumina Infinium Human Methylation 450 Bead Chip) and RNA-seq data (n = 482) were available for ccRCC or nontumor tissue samples at the time of our analyses. Details of the TCGA cohort are summarized in Table 1. Processed datasets were obtained from the TCGA Data Portal (http://tcga-data.nci.nih.gov/tcga/) using the Data Browser tool.

Tissue microarray (TMA) slides containing ccRCC and corresponding nontumor tissue sections were processed as described (17) and immunostained using an antibody against MCT4 (Santa Cruz Biotechnology). The slides were scanned using a digital Mirax scanner (Carl Zeiss, MicroImaging). For the evaluation of immunohistochemical staining, MCT4 immunoreactivity of all tumor and benign renal areas were classified according to a composite IHC score. This semiquantitative grading of membranous MCT4 staining was based on a comparison of all scanned TMA dots using a picture based software platform (Mirax Viewer, Carl Zeiss, Light Microscopy). Staining intensity was graded into 0, 1, 2, or 3 corresponding to the presence of negative, weak, intermediate, and strong MCT4 staining. In addition, the percentage of MCT4 stained cells was determined (0–100%). The composite IHC score was obtained by multiplying the staining intensity and the percentage of stained cells (0–100%), resulting in an immunoreactivity score value of 0 to 300. The score values 0 to 300 were finally converted to percentage 0% to 100% (see Supplementary Data for detailed description).

For validation of the MCT4 antibody, MCT4 siRNA knockdown was conducted in the renal carcinoma cell lines 786-O and A-498. The cell lines used were purchased from CLS and passaged for less than 6 months. Further authentication was conducted using the AmpFLSTR Identifiler Plus PCR Amplification Kit (Applied Biosystems). Details are provided in Supplementary Data.

For details see Supplementary Data.

For quantitative DNA methylation analyses, matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) was applied (18). Mass spectra were obtained using the MassARRAY compact system and evaluated using the EpiTYPER 1.0 software. Details see Supplementary Data.

The influence of DNA methylation on SLC16A3 promoter activity was analyzed in an in vitro cell culture system using a dual-reporter assay. In brief, SLC16A3 promoter reporter constructs, cloned into the pCpGfree basic reporter vector (Invivogen), were either methylated by incubation with CpG methylase M. SssI (Zymo Research) in the presence of SAM or mock-methylated through incubation under equal conditions in absence of CpG methylase M. SssI. Methylated and mock-methylated SLC16A3 promoter reporter constructs were cotransfected with the control vector pGL4.13, constitutively expressing firefly luciferase for normalization, into 4 different RCC cell lines (Caki1, Caki2, A498, and 786-O) using FuGene HD transfection reagent (Promega). SEAP and luciferase activity were measured after 24 hours with the Phospha-Light System (Applied Biosystems) and the Dual-Luciferase Reporter Assay System (Promega) respectively, according to manufacturer's instructions. For details, see Supplementary Data.

Statistical analyses were conducted with R-2.15.0 (ref. 19; http://www.r-project.org) as well as GraphPad Prism 5.0 (GraphPad Software, Inc.). Statistical significance was defined as P < 0.05, where indicated P values were adjusted for multiple testing by Holm's procedure (20). Detailed description is given in Supplementary Data.

Two independent cohorts of ccRCC patients (TuCP1 and TuCP2) from the University Hospital Tuebingen were investigated. Characteristics of these 2 cohorts TuCP1 (n = 207) and TuCP2 (n = 64) are summarized in Table 1. Median follow-up times of patients in TuCP1 and TuCP2 were 5.9 and 2.3 years, respectively. In addition, we investigated a third, independent cohort of ccRCC patients from TCGA (n = 502; Table 1). Overall, the cohorts were similar, but some differences did exist, such as grading (G) (TuCP1/TuCP2 vs. TCGA: P < 0.001, respectively), due to different grading systems in the study cohorts, involvement of regional lymph nodes (N) (TuCP1 vs. TuCP2: P < 0.001; TuCP2 vs. TCGA: P = 0.005) or follow-up times (TuCP1 vs. TuCP2/TCGA: P < 0.001, respectively; TuCP2 vs. TCGA: P = 0.02). Staging of tumors (T) did not differ significantly between the cohorts (Supplementary Table S1).

MCT4 protein expression in the TuCP1 cohort was investigated by immunohistochemical staining of tissue microarrays. Representative staining is shown in Fig. 1A. A scoring system (details see Supplementary Data) was assigned to semiquantitatively determine MCT4 protein expression. Statistical analyses revealed a significantly increased MCT4 protein expression in ccRCC compared to corresponding normal tissue (P < 0.0001; Fig. 1B, left), confirmed by direct comparison of matched samples (Fig. 1B, right). MCT4 protein expression levels in ccRCC correlated strongly with clinicopathologic features, for example T, N, M, G, or necrosis (Supplementary Table S2). In addition, patients with higher MCT4 expression in ccRCC had a significantly higher risk for cancer-related death. Kaplan–Meier analysis based on optimal cutoffs determined by conditional interference tree models revealed that patients with low MCT4 expression in ccRCC had an increased cancer-specific survival time compared to patients with high expression levels (Fig. 1C; HR = 4.23 (95% confidence interval (CI), 2.29–7.80); P[log-rank] = 5.36e⁻⁰⁷]. Moreover, nonmetastatic patients with higher MCT4 expression in ccRCC tissue [HR = 4.94 (95% CI, 2.19–11.13); P[log-rank] = 2.05e⁻⁰⁵] had a significantly higher risk for cancer-related death.

For validation of the expression differences between ccRCC and nontumor kidney tissue on mRNA level, mRNA expression of MCT4 was evaluated using RNA-seq data from the independent ccRCC TCGA cohort. As illustrated in Fig. 2A, MCT4 mRNA expression was significantly increased in ccRCC tissue compared to adjacent nontumor tissue (P < 0.00001). Based on the hypothesis that altered mRNA expression may be caused by aberrant DNA methylation, we investigated DNA methylation in the SLC16A3 promoter region in the TCGA tissue samples, for which genome-wide DNA methylation data were available. DNA methylation of SLC16A3 in the promoter region of the TCGA cohort was significantly decreased in the tumor tissue compared to histologically normal kidney tissue (Fig. 2B). DNA methylation of single CpG sites in the promoter correlated significantly inverse with mRNA expression in ccRCC, especially the CpG site cg18345635 (r_s = −0.31, P < 0.00001).

By correlation analysis of DNA methylation of the tumor tissue with clinicopathologic parameters, significant associations were again found particularly for the CpG site cg18345635 (Table 2). Moreover, DNA methylation of single CpG sites correlated with patients' survival in the TCGA cohort. Kaplan–Meier analysis based on optimal cutoffs determined by conditional interference tree models showed that patients with high methylation at the CpG site cg18345635 in the SLC16A3 promoter had a significantly prolonged overall survival compared to patients with low methylation [Fig. 2C; HR = 0.39 (95% CI, 0.24–0.64); P[log-rank] = 1.23e⁻⁰⁴]. Again, restriction of analyses to patients without metastatic ccRCC did not substantially change the observed results (data not shown). For the TCGA-cohort, only data for overall survival and not for cancer-specific survival were available.

Finally, we validated our results in a third independent cohort of 64 matching ccRCC and nontumor tissues (TuCP2) on mRNA, and protein expression level, as well as on DNA methylation level. Again, MCT4 protein as well as mRNA expression were significantly higher in ccRCC compared with nontumor kidney (Fig. 3A and B). MCT4 mRNA expression was normalized to Dipeptidyl-peptidase 4. The highly variable MCT4 mRNA and protein expression correlated significantly in ccRCC tissue (r_s = 0.34, P = 0.0099; Supplementary Fig. S2C). Moreover, the same trends for the associations between MCT4 protein expression and clinicopathologic parameters as in the TuCP1 cohort were observed, but correlations did not reach comparable high statistical significance due to the smaller sample size (Supplementary Table S4). With respect to clinical outcome, high MCT4 protein expression in tumor tissue was moderately associated with inferior cancer-specific survival (P[log-rank] = 0.175, cutoff determined by conditional interference tree; Supplementary Fig. S3).

DNA methylation at the SLC16A3 promoter was assessed in the TuCP2-cohort using MALDI-TOF MS (Fig. 3C). The DNA methylation pattern quantified at each analyzed CpG site in nontumor (red) and ccRCC (blue) tissue showed significantly decreased DNA methylation levels in specific areas of the SLC16A3 promoter in ccRCC compared with nontumor tissue (Fig. 3C). Of note, investigation of DNA methylation by MALDI-TOF MS allowed a more comprehensive coverage and quantification of CpG sites in the SLC16A3 promoter region compared to the genome-wide array analysis conducted in the TCGA cohort (see Supplementary Table S3). Again, DNA methylation at individual CpG sites in the validation cohort (TuCP2), quantified by MALDI-TOF MS correlated significantly inverse with MCT4 mRNA and protein expression (Supplementary Fig. S2C and S2D). In particular, the percentage of DNA methylation at 4 specific CpG sites showed significant inverse correlation with MCT4 protein expression in ccRCC even after adjustment for multiple testing (Supplementary Fig. S2D).

By correlation analysis of the percentage of DNA methylation of the tumor tissue with clinicopathologic parameters, the CpG unit 05_CpG_8.9 revealed significant associations (Table 2). This CpG unit is located in close proximity to cg18345635 (Supplementary Table S3), which was associated with overall survival and clinicopathologic features in the TCGA-cohort (Table 2). Higher methylation values at 05_CpG_8.9 in tumor tissue were significantly associated with prolonged survival [Fig. 3D; HR = 0.05 (95% CI, 0.01–0.40); P[log-rank] = 6.91e⁻⁰⁵]. Restriction of analyses to patients without metastatic ccRCC did not substantially change the observed results (data not shown).

To determine the influence of DNA methylation on SLC16A3 promoter activity, we used different promoter/reporter fusion plasmids containing either methylated or mock-methylated SLC16A3 promoter fragments for promoter reporter assays (Fig. 4). These constructs were then transfected together with the control vector, constitutively expressing firefly luciferase for normalization, into 4 different RCC cell lines (Caki1, Caki2, A498, 786-O) and subsequently promoter activities were measured. The influence of DNA methylation on SLC16A3 promoter activity by a dual-reporter gene assay confirmed that reporter constructs with a mock-methylated SLC16A3 promoter fragment showed significantly higher promoter activity than constructs with a methylated SLC16A3 promoter, even in the smaller construct and independent of the VHL status of the cells (Fig. 4).

To further investigate, if aberrant DNA methylation of SLC16A3 is independent of the VHL genotype, we analyzed data of the TCGA cohort (available for 199 of the 283 samples; 70%). DNA methylation was not significantly dependent on the VHL status (carriers vs. noncarriers of mutations) of the patients (Supplementary Fig. S4).

Univariate Cox regression analyses were conducted and the predictive ability of different models was estimated by calculating Harrell's c-index (Supplementary Table S5). The pathologic features T, M, and G were able to significantly predict cancer-specific survival in both the TuCP1 and TuCP2 cohort, whereas N and necrosis were only predictive in one of the cohorts (Supplementary Table S5). In line, overall survival could be predicted in the TCGA cohort by T, M, G, and necrosis. DNA methylation at specific CpG sites in ccRCC significantly predicted overall survival in the TCGA cohort (cg18345635: P[log-rank] = 0.00065, c-index: 61.3) as well as cancer-related death (which is identical to overall survival) in the TuCP2-cohort (05_CpG_8.9: P[log-rank] = 0.0125, c-index: 75.1). Moreover, in the subgroup of nonmetastatic patients of the TuCP2-cohort DNA methylation in ccRCC significantly predicted cancer-related death (05_CpG_8.9: P[log-rank] = 0.036, c-index: 86.4).

DNA methylation in the tumor was only moderately, but not significantly predictive for cancer-specific survival and overall survival in multivariate analyses in the TuCP2 and TCGA patients (P[Wald] = 0.31 or P[Wald] = 0.31), and in nonmetastatic patients only (P[Wald] = 0.14 or P[Wald] = 0.46).

ROC curve analysis further supported our results that DNA methylation in the tumor at a single CpG unit is feasible as novel biomarker for identifying patients at risk for cancer-related death and death (Supplementary Table S6) with AUCs of 76.3% in TuCP2 and 63.9% in TCGA, respectively. In comparison, AUCs for T, N, M, G, or necrosis ranged from 56.7% to 83.7% (TuCP2) or 51.6% to 74.7% (TCGA). Of note, the AUCs for the clinicopathologic parameters T, N, G, and necrosis were lower than the AUC for DNA methylation at 05_CpG 8.9 in TuCP2. The sensitivity, specificity, and negative and positive predictive values of DNA methylation at 05_CpG_8.9 in TuCP2 for prediction of death were 91.7% (95% CI, 33.1–100%), 68.8% (95% CI, 22.2%–83.3%), 97.1%, and 42.3%, respectively, versus 52.1% (95% CI, 39.4%–62.6%), 69.7% (95% CI, 54.6%–80.6%), 74.4%, and 46.2% for methylation at cg18345635 in the TCGA dataset.

Because ccRCC is estimated to be mainly a metabolic disease and a clear example of the Warburg effect in tumor metabolism, we aimed to comprehensively study MCT4 expression as well as regulation by epigenetic mechanisms in ccRCC to evaluate its relevance as metabolic target for diagnosis, prognosis, and potential therapy. In this study, we found that MCT4 protein is significantly overexpressed in >85% of all ccRCCs in 2 independent cohorts. DNA methylation at specific CpGs in the SLC16A3 promoter correlated significantly inverse with MCT4 expression in ccRCC, which was also replicated. In addition, the regulation of MCT4 by DNA methylation was confirmed by reporter gene assays in 4 different renal carcinoma cell lines, strongly supporting the importance of DNA methylation for MCT4 expression.

Previous results from adenocarcinomas of the lung showed that high MCT4 expression is associated with aggressive tumor behavior (21). In addition, in mouse xenograft models of human colorectal and breast cancer, release of lactate from tumor cells through MCT4 (and not the isoform MCT1) is sufficient to increase tumor growth (22). Moreover, lactate efflux mediated by MCT4, which can be explored in RCC cells by hyperpolarized ¹³C-pyruvate magnetic resonance, was significantly higher in a metastatic RCC cell line with increased MCT4 expression (12). In line with these data, MCT4 expression in ccRCC of our cohorts as well as DNA methylation at specific CpG units was strongly associated with various clinicopathologic features, including sarcomatoid differentiation. Thus, both factors may be valuable parameters for diagnosis of high-grade tumors, and highly aggressive variants, for example ccRCC with sarcomatoid differentiation and metastatic tumors.

Moreover, our data showed that MCT4 expression and DNA methylation were significantly associated with patient outcome and were able to predict cancer-related death in ccRCC patients. Compared to MCT4 expression, DNA methylation was more strongly associated with cancer-specific survival in TuCP2 indicating that quantification by immunohistochemistry may have some limitations in accuracy (e.g., protein degradation due to fixation or small tissue cores in TMA). The fact that the association of cancer-related death with MCT4 expression was superior in TuCP1 compared to TuCP2 is most likely explained by the larger sample size and longer follow-up times in the TuCP1 cohort. Interestingly, the association of low SLC16A3 DNA methylation at specific CpG sites with worse patient survival could also be detected in the large independent ccRCC cohort from TCGA with comparable low P values, supporting the clinical relevance of SLC16A3 DNA methylation for ccRCC. A similar approach, using the TCGA datasets as an independent confirmatory cohort has recently been successfully conducted for the association of BAP1 and PBRM1 mutations with survival in ccRCC (23, 24).

MCT4 protein expression and DNA methylation were associated with all important and validated prognostic factors (tumor stage, grade, tumor necrosis, or metastasis) known for ccRCC outcome, which further supports the prognostic relevance of MCT4 for cancer-specific survival. Because of the strong relationship of DNA methylation at specific CpG sites with clinicopathologic parameters in ccRCC, not surprisingly SLC16A3 DNA methylation at these sites did not independently predict either risk of cancer-related death or overall survival in the TuCP2 and TCGA cohort after correction for prognostic factors (T, N, M, G, and necrosis) in multivariate Cox regression analysis. However, it showed a similar (TCGA) or even larger (TuCP2) Harrell's c-index compared to the established pathologic risk factors (Supplementary Table S5). Moreover, several reasons exist that the use of quantitative DNA methylation of SLC16A3 in ccRCC as potential novel prognostic predictor is more applicable compared to clinicopathologic features. Quantification of DNA methylation by MALDI-TOF MS, but also other technologies (e.g., pyrosequencing), is a highly accurate and reliable method (18), which has already been showed for other biomarkers (25). Potential inaccuracies in the determination of clinicopathologic features are inevitable depending on the investigator and/or based on the absence of uniform grading systems for ccRCC resulting in inter- and intraindividual variability even for the most widely applied Fuhrmann grading system (26).

Taken together, the association of DNA methylation and patient survival was observed not only in our cohort but also in the TCGA cohort, despite patients from different centers were included and some differences in the patient cohorts did exist. Hence the association is apparently not dependent on the site of biospecimen's recruitment, ethnicity of patients, or the method used for quantification of DNA methylation, which further supports that MCT4 DNA methylation may serve as a novel biomarker in clinical practice. However, to definitely prove the importance of DNA methylation as a novel biomarker for prognosis, large-scale prospective studies are needed (27).

Apart from its prognostic potential, therapeutic targeting of MCT4 in ccRCC may be attractive because of the high MCT4 expression only in tumor and not in the corresponding nontumor tissue. Gerlinger and colleagues showed that silencing of MCT4 in ccRCC cells has an influence on cell proliferation and induction of apoptosis (11). Interestingly, the effect was independent from the VHL genotype, indicating that MCT4 targeting would be a therapeutic option irrespective of patients' VHL status. Theoretically, MCT4 silencing could be achieved by MCT4 inhibitors. However, so far, only isoform specific MCT1 inhibitors (e.g., AR-C155858; ref. 28) have been evaluated in phase I/II trials. Interestingly, treatment of patients with statins, which are known to inhibit MCT4 (29), reduced the risk of RCC (30). Thus, the development of isoform-specific MCT4 inhibitors seems to be a promising therapeutic approach beneficial for several cancer types (e.g., breast cancer; ref. 8).

An alternative approach is targeting of MCT4 expression and thus lactate efflux by modulating DNA methylation. In contrast to the approved therapy of the myelodysplastic syndrome using DNA demethylating agents such as 5-Azacytidine, specific MCT4 silencing by DNA methylation may be an attractive strategy for ccRCC treatment. It has recently been shown that zinc-finger proteins can target the DNA methyltransferase DNMT3a to specific promoter sites, resulting in stable reprogramming of breast cancer cells (31).

Our hypothesis generating, retrospective study has some limitations. First, we do not have complete data on treatment regimens for the study cohorts. However, when we compared patients with chemotherapy and targeted therapy in a small subgroup of the TCGA cohort where treatment data were available, DNA methylation of SLC16A3 in ccRCC tissue was not significantly different. Moreover, a substantial proportion of patients in the TuCP1 cohort (86.5%) and TCGA cohort (43.8%) was diagnosed before 2006, the time when targeted therapies were introduced into clinical practice (3), indicating that confounding of our results by novel therapeutic strategies is most unlikely. Second, genetic variants in SLC16A3 that might also alter variability of MCT4 expression, have not been investigated.

In summary, we identified DNA methylation as a so far unknown underlying epigenetic mechanism for MCT4 regulation in ccRCC. Quantification of DNA methylation of specific CpG sites in the SLC16A3 promoter offers considerable clinical potential for diagnosis and prognosis of ccRCC. Large-scale studies are warranted to elucidate its clinical utility.

No potential conflicts of interest were disclosed.

Conception and design: P. Fisel, S. Kruck, M. Schwab, E. Schaeffeler

Development of methodology: P. Fisel, S. Kruck

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Fisel, S. Kruck, J. Bedke, J. Hennenlotter, M. Scharpf, A. Stenzl

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Fisel, S. Kruck, S. Winter, J. Bedke, A.T. Nies, M. Schwab, E. Schaeffeler

Writing, review, and/or revision of the manuscript: P. Fisel, S. Kruck, S. Winter, J. Bedke, M. Scharpf, F. Fend, A. Stenzl, M. Schwab, E. Schaeffeler

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Kruck, J. Hennenlotter, M. Scharpf, F. Fend, A. Stenzl

Study supervision: M. Schwab

The authors thank M. Elbl, A. Jarmuth, H. Köhler, and U. Waldherr for excellent technical assistance. The authors thank The Cancer Genome Atlas initiative, all tissue donors, and investigators who contributed to the acquisition and analyses of the samples used in this study.

This work was supported by the Robert Bosch Foundation (Stuttgart, Germany), the Federal Ministry for Education and Research (BMBF, Berlin, Germany) grant (03 IS 2061C and 0315755) and the FP7-grant (PITN-GA-2009-238132), DFG SFB685 A09 and C05, and Ferdinand Eisenberger grant of the Deutsche Gesellschaft für Urologie (German Society of Urology), grant ID [KrS1/FE-10].

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.