High expression of GEM and EDNRA is associated with metastasis and poor outcome in patients with advanced bladder cancer

Written informed consent was obtained from all patients and the study was approved by the Central Denmark Region Committees on Biomedical Research Ethics (1994/2920). All patients were cystectomized at Department of Urology at Aarhus University Hospital between 1998 and 2008, and surviving patients had at least 36 months of follow-up, and were censored after a maximum of 96 months. Tumor stage was determined using the American Joint Committee on Cancer recommendations from 2002 and WHO 2004 classification was used to determine tumor grade. All patients were clinically free of metastasis before surgery and no patients received neoadjuvant or adjuvant treatment in terms of chemotherapy or radiotherapy.

All patient specimens collected at the time of surgery were split into tissue for pathology and tissue for the biobank. Tissue for the biobank was embedded in Tissue-Tek® O.C.T™ Compound and snap frozen in liquid nitrogen before storage at -80°C. Sections were examined by a genitourinary pathologist to identify carcinoma cell content. Following, cresyl violet stained tissue was microdissected using the PALM laser microbeam system. RNA extraction was performed using RNeasy Micro Kits (Qiagen) according to manufacturer protocols. RNA quality was assessed using an Agilent Bioanalyzer 2100 (RIN: 2.4-8.8; median 5.9). Total RNA was amplified and converted to cDNA using Nugen Pico-RNA system. The two-round amplification kit is optimized to amplify low volumes and poor quality RNA for Affymetrix array analysis. After amplification, the cDNA was fragmented and labeled using NuGen FL-Ovation kit, loaded onto the Affymetrix U133 Plus 2.0 Gene Chip according to the manufacturer’s protocol, and scanned using the Affymetrix 3000 7G Scanner.

Raw microarray data was normalized and intensity measures generated by RMA [21] using GeneSpring version 11 software. Unsupervised hierarchical cluster analysis of all transcripts with a variance above 1.5 was performed using Cluster 3.0 and Java tree-view software [22]. Gene Set Enrichment Analysis (GSEA) v2.07 software was used to test if previously published gene signatures and curated pathways were enriched in the data. We used the inbuilt KEGG, BIOCARTA, REACTOME, gene ontology, and oncogenic signatures in MsigDB database and supplemented with curated signatures containing “cancer”, “metastasis”, “cell cycle”, “repair”, “DNA damage”, and “hypoxia”. We used the default significance levels to test if significant enrichment was reached with normalized p-values below 0.05 and with false discovery rates below 0.25. A previously published dataset (GEO ID: GSE31684; U133 Plus 2.0 GeneChip) from laser microdissected tumors from 93 cystectomized patients was retrieved. A total of 69 patients were included in the analysis, after exclusion of all patients without reported lymph node status, and all node negative patients without 24 months of follow-up.

Biopsies from a total of 368 tumors from cystectomy specimens and from 41 lymph node metastases were incorporated into a TMA. All tumors were reevaluated regarding T-stage and grade by the same uro-pathologist prior to placement on the TMA. The patients included and the TMA construction is described earlier [11].

The immunohistochemichal staining procedure was carried out based on the EnVision + TM System HRP (Dako) as previously described [23]. Antibodies against GEM (Novus Biologicals # NBP1-58906) diluted 1:150 and against EDNRA (Abcam #ab76259) diluted 1:800 were used. The specificity of the antibodies against GEM and EDNRA was validated by Western blotting using T24 cell line essentially as described earlier [24].

A Hamamatsu Nanozoomer scanner (Hamamatsu Corporation, Hamamatsu City, Japan) was used to scan the TMA slides, and VIS visualization software (Visiopharm A/S, Hørsholm, Denmark) was used for visualization of IHC staining during scoring of the protein expression intensities. Percentage of positive carcinoma cells was scored on a continuous scale for each core, and optimal cut-off values were afterwards defined by ROC curves. Scoring was performed by two observers blinded to outcome. The first observer scored on a continuous scale, and the second scored according to the dichotomized cutoff value generated. Differences in the dichotomized scorings were reviewed and consensus was reached.

Comparisons between the metastatic and non-metastatic groups were performed using two-sided t-test statistics. Categorical data was compared in univariate analysis using the χ2 test and censored data was compared using log-rank test. Hazard ratios (HR) were estimated using Cox proportional hazard models. Multivariate analysis was performed separately for each biomarker including only significant clinical parameter from the univariate analysis. All analyses were performed using STATA (version 11).

Initially, data was filtered, selecting only transcripts with a variance above 1.5 across all samples (11046 transcripts). We performed unsupervised hierarchical cluster analysis to investigate if tumors clustered based on stage or metastatic abilities, and if lymph nodes showed a high degree of similarity to the matched primary tumors (Figure 1). Cluster analysis separated the tumors into two main clusters; one cluster (cluster A) contained seven primary metastasizing tumors, three primary non-metastasizing tumors, and eight lymph nodes, and among these were six of the seven matched pairs. The other cluster (cluster B) contained five primary non-metastasizing tumors, five metastasizing primary tumors, and four lymph nodes. Seven of the lymph nodes clustered together with their matched primary tumor, indicating a large degree of intra-patient similarity in these patients. However, the overall expression patterns did not show significant separation of the tumors based on metastatic ability. Most of the muscle-invasive tumors clustered together in cluster A – as expected.

To investigate the differences between the metastatic and non-metastatic tumors more specifically, we applied GSEA for investigating enrichment for previously published signatures regarding key elements in the metastatic process together with enrichment for pathway elements (Table 2). Interestingly, all signatures regarding extracellular function, metastasis, hypoxia, proliferation, and survival were exclusively enriched in metastatic tumors while all signatures regarding repair and cell cycle were enriched in non-metastatic tumors. Cell signaling was primarily enriched in metastatic tumors while metabolism was primarily enriched in non-metastatic tumors. In addition, we investigated enrichment for previously published signatures comparing primary tumors and metastasis [25‐27]; both signatures containing tumors from many different tissues were significantly enriched in our dataset (Ramaswamy et al., P = 0.02 and Daves et al., P = 0.03), while the signature from metastatic malignant melanoma was borderline significantly enriched (Daves et al., P = 0.06).

We used the paired tumors and lymph node metastases to investigate the intra- and inter-patient similarity. When comparing differences in transcript levels between the matched primary tumors and metastases using two-fold difference as cut-off, we did not find any transcripts that were differentially expressed in all 12 tumor-lymph node comparisons (Figure 2). MMP2 was the only gene that was down-regulated in 11 lymph node metastases, while 18 transcripts were up or down regulated in 10 lymph node metastases. In general, as observed in the cluster analysis, the patients show a large heterogeneity in expression patterns between primary tumors and lymph node metastases. Using Ingenuity Pathway Analysis we did not identify any general pathway changes between primary tumors and lymph node metastases, probably because of this large heterogeneity observed between patients.

Because of the large heterogeneity observed and because of the limited sample size we included a previously published dataset for delineation of markers associated with outcome (GEO ID: GSE31684). The dataset contained Affymetrix U133 Plus 2.0 GeneChip data from 69 patients with known lymph node status and at least 24 months of follow-up if no lymph node metastasis was present at surgery. Separately, for both datasets, we delineated transcripts associated with the presence or absence of metastasis; only transcripts with a mean fold change difference > 2 and with a P < 0.05 (student’s t-test) were selected. Twelve transcripts up-regulated in metastasizing tumors passed our selection criteria in both datasets (Table 3). We selected EDNRA and GEM (Figure 3) for further validation using immunohistochemistry (IHC). For this we used a tissue microarray containing 409 core biopsies from both primary tumors (n = 368) and lymph node metastases (n = 41). Both GEM and EDNRA protein expression was localized in the cytoplasm of the cells, and no staining was observed in normal urothelium or connective tissue cells. IHC scoring was performed by two observers independently, with an inter-observer agreement of 0.70 (GEM) and of 0.81 (EDNRA), using Cohen’s kappa. The clinical and histopathological characteristics for the patients included in this cohort are listed in Table 4. High expression of GEM (P = 0.033; HR = 1.46) and EDNRA (P = 0.046; HR = 1.60) were significantly associated with decreased cancer-specific survival (Figure 4). Furthermore, after performing multivariate analysis high EDRNA expression showed significantly association with decreased cancer-specific survival (P = 0.046), while GEM showed no significance (P = 0.11). Finally we investigated the similarity in protein expression between matched primary tumors and lymph node metastases; 94% of the lymph nodes showed similar expression as in the primary tumors for EDNRA and 71% for GEM.