Identification and Prognostic Significance of an Epithelial-Mesenchymal Transition Expression Profile in Human Bladder Tumors

Epithelial to mesenchymal transition (EMT) is a process that was first observed in embryonic development (1–3) and has more recently been implicated as an underlying event in neoplastic progression (4–7). To date, the definition and occurrence of EMT in in vivo tumorigenesis remains controversial (6, 8); however, the conceptual framework embracing loss of epithelial markers and gain of mesenchymal markers has been reportedly associated with numerous cancers (9–11), often identified in cell lines established from tumors representing different grades and stage (10, 12). In this way, the discontinuous progression model observed within a panel of cell lines of common tissue origin may be closely linked to the differentiation status of cells. In most cases, the projected transition between epithelial and mesenchymal phenotypes is accompanied by increased motility and invasive potential.

With the advent of the establishment of in vitro models of EMT, elicited by the action of different factors on alternative cell types (13–18), the existence of EMT in progression becomes more credible. In addition, such models enable us to confirm events linked to EMT identified in discontinuous progression models and to discover further molecular events underlying this transition (7, 19). Alterations associated with the cadherin/catenin complex often feature at the center of EMT related to increased migration and invasion of cells (19). Such events will follow different courses in different models dependent on the expression profile of the cadherins in the targeted cell. In a discontinuous model of EMT in bladder cancer, we have identified loss or reduced expression of E-cadherin accompanied by the silencing of plakoglobin expression as late-stage events in bladder tumorigenesis (20). This change was linked to increased motility and invasion potential. In bladder carcinoma cell lines, loss of E-cadherin expression is accompanied by novel expression of N-cadherin, representing cadherin switching from an epithelial-specific classic cadherin to a mesenchymal-specific cadherin (21). Such EMT-related events accompany increased motility and invasion in bladder carcinoma cells. In a previous report, we have shown novel expression of N-cadherin in human bladder tumor tissue (22). In this study, using a tissue microarray format, we evaluated five biomarkers linked to EMT to determine the occurrence of expression of these molecules in vivo and the potential prognostic value of such events within a well-defined series of bladder carcinomas.

Cell culture. Human bladder cell lines RT4, RT112, HU456, BC16.1, CUBIII, 5637, PSI, HT1197, HT1376, SCaBER, EJ, KK47, J82, UM-UC-3, and TCCSUP were maintained in DMEM supplemented with 7.5% fetal bovine serum and penicillin/streptomycin.

Invasion assays.In vitro invasion assays were carried out using modified Boyden chambers consisting of Transwell (8-μm pore size; Corning Costar Corp., Cambridge, MA) membrane filter inserts in 24-well tissue culture plates. For invasion assays, the upper surfaces of the membranes were coated with Matrigel (Collaborative Biomedical Products, Bedford, MA) and placed into 24-well tissue culture plates containing 600 μL of NIH/3T3 conditioned medium (experimental) or plain DMEM (control). Cells (1 × 10⁵) were added to each Transwell chamber and allowed to invade toward the underside of the membrane for 24 h at 37°C. Cells that pass through the membrane were fixed in methanol, stained with crystal violet, and counted under a light microscope.

Immunocytochemical staining of bladder carcinoma cell lines. Cells were grown on glass slides, washed with PBS, and fixed in 3.7% formaldehyde for 15 min at room temperature. Cells were then rinsed in three changes of PBS and permeabilized in 0.5% Triton X-100 in PBS. Following three washes in PBS, cells were stained for EMT biomarkers using the antibodies listed below. Immunostaining was done on an automated stainer (Autostainer Plus, DAKO Corporation, Carpinteria, CA).

Western blot analysis. Subconfluent dishes of cells were washed in PBS followed by lysis in hot sample buffer [2× ESB-0.08 mol/L Tris (pH 6.8); 0.07 mol/L SDS, 10% glycerol, 0.001% bromophenol blue, and 1 mmol/L CaCl₂] and sheared through a 26-gauge needle. Lysates were then assayed for protein concentration using the bovine serum albumin method (Pierce, Rockford, IL). After determination of protein content, β-mercaptoethanol (1%) was added to each sample. Samples were boiled for 5 min, and protein was loaded in each lane of a 7.5% or 12.5% polyacrylamide gel. Proteins were transferred overnight onto nitrocellulose. Membranes were blocked in 10% milk in TBS with 0.05% Tween and incubated with primary antibody overnight at 4°C. Blots were washed in TBS with 0.05% Tween, thrice for 15 min each, and incubated with secondary antibody linked to horseradish peroxidase for 60 min at room temperature. Blots were then washed as described above and developed with an enhanced chemiluminescence kit (Amersham, Arlington Heights, IL).

Clinical samples. We have used tumor tissue samples from patients diagnosed with bladder cancer at the Lahey Clinic Medical Center between 1990 and 2005 under an institutional review board–approved protocol. Formalin-fixed, paraffin-embedded tumor tissues from patients were retrieved from the archives of the Pathology Department. The tumor stage was determined using tumor-node-metastasis classification and graded according to WHO guidelines.

Construction and immunohistochemical staining of tissue microarrays. Eight hundred and twenty-five tumor samples from 572 patients were assembled in 10 tissue microarrays. All tumor samples were transitional cell carcinomas. Tissue microarrays were designed with replicas for each tumor sample and each control. Controls included normal tissue from prostate, testis, tonsil, liver, cerebellum, kidney, lung, and bladder. Strategic placement of control cores in each tissue microarray enabled definitive orientation during the scoring of the tissue microarrays. Each tissue microarray consisted of 400 tissue cores (four cores per specimen) and control tissue cores for immunohistochemical validation and orientation. Multiple 4-μm sections were cut and stored at 4°C in the presence of a desiccant before immunohistochemical staining. Individual tissue microarray sections were deparaffinized and antigen retrieval in citrate buffer (pH 6.0; DAKO) was done. The antibodies were diluted with DAKO antibody diluent solution. Immunohistochemical staining was done on an automated stainer (Autostainer Plus, DAKO) using a high-sensitivity, polymer-based detection system (EnVision, DAKO). A negative control was included using a nonspecific mouse antibody solution (DAKO) substituting for the primary antibody.

Antibodies. Mouse monoclonal antibodies to cytokeratin, E-cadherin (DAKO), β-catenin, and plakoglobin (BD Transduction Laboratories, San Diego, CA) were diluted 1:25; 1:100, and 1:100 (working concentrations, 0.75, 7.56, and 2.5 μg/mL), respectively, for use in immunohistochemistry. Mouse monoclonal antibodies to mesenchymal cell markers N-cadherin (Zymed, San Francisco, CA) and vimentin (DAKO) were diluted 1:100 (working concentration 5 μg/mL) and 1:50 (working concentration 4.6 μg/mL), respectively. The negative control reagent (DAKO) is a cocktail of nonimmune mouse immunoglobulins (IgG and IgM) and was purchased prediluted. The specificity of each antibody in immunohistochemistry was determined using xenograft sections derived from bladder cell lines in which the expression profile of the antigen of interest had been previously identified.

Scoring of immunohistochemistry. The tissue sections were scored semiquantitatively, assessing staining intensity and protein localization, including membrane, cytoplasmic, or nuclear localization. In the case of E-cadherin, β-catenin, and plakoglobin, a staining intensity scale of 0 to 3 was applied. In normal bladder tissue cores, the staining intensity with each antibody was recorded as 3 with membrane localization throughout all epithelial layers of the mucosa. In tissue sections that displayed heterogeneous staining throughout the section or between cores from the same tumor, the worst-case-scenario score was assigned to that sample when >5% of the tumor cells displayed this phenotype. Greater than 90% concordance between scores from different cores of the same tumor was recorded. In samples where scoring differed between cores, the worst-case-scenario score was recorded for analysis. Specimens that exhibited a complete absence of staining or faint staining in <5% cells were scored negative. In the case of N-cadherin and vimentin, a positive or negative score was given for each tissue sample because the presence of either mesenchymal marker in bladder carcinoma cells represented novel expression. Each tissue microarray was scored independently (I.C.S. and M.L.). Where discordant results were obtained, both individuals rereviewed the stained cores to obtain a consensus.

Statistical analysis. Comparisons of groups were done by Pearson's χ² test. Correlation between markers was determined by Spearman's correlation coefficient. Univariate overall and cancer-specific survival analysis was done using the product-limit procedure (Kaplan-Meier method), with the surgery date as the entry date. The log-rank (Cox-Mantel) test was used to compare survival curves for different categories of each variable. Variables with effect on survival in univariate analyses (P ≤ 0.15) were examined by log-log plot to determine how these variables could be incorporated into a Cox proportional hazards regression models, and variables with P ≤ 0.1 were maintained in the model. Test for interactions were carried out for the variables that were significantly related to survival in Cox regression analysis. Data were analyzed using the SPSS software package, version 11.5. B. Silva Neto and A. Biolo did the statistical analysis.

Characterization of EMT markers in bladder carcinoma cells. To establish the occurrence of the putative EMT phenotype in bladder carcinoma cells, we initially screened a panel of 15 cell lines for the expression of the five biomarkers we propose to evaluate in tissue sections. Table 1 shows the expression profile of each component as determined in Western blot analysis and includes morphologic and in vitro invasion data for each cell line. No nuclear localization of proteins was detected in cell lines using immunocytochemistry (data not shown). Morphologic classification of epithelial morphology was defined by tightly adherent cuboidal cells growing as discrete colonies with mesenchymal morphology defined by poorly adherent carcinoma cells displaying a stellate morphology. From the aforementioned data, it is clear that the projected EMT phenotype is recorded within this panel of bladder carcinoma cell lines and is correlated with a mesenchymal morphology and invasive phenotype as assessed in an in vitro assay.

EMT phenotype in bladder tumor tissue. If the EMT phenotype identified in the bladder carcinoma cell panel is a valid paradigm in bladder cancer, we would expect to see the proposed modulations in epithelial and mesenchymal cell markers associated with more aggressive bladder disease. To test this postulate, E-cadherin, plakoglobin, β-catenin, N-cadherin, and vimentin expression levels and localization were assessed in tissue microarrays of primary bladder tumors. The construction of 10 microarrays included 825 transitional cell carcinoma samples from 572 patients. The tumor-node-metastasis and histologic grade of the samples is shown in Table 2. Analysis by number of patients as opposed to number of samples revealed no difference in the results. In this study, the analysis of the superficial tumor group included pT_a, pT_is, and pT₁ with the invasive group represented by pT₂-T₄ tumors. When T₁G₃ was included in the invasive group, the results did not change (data not shown).

E-cadherin expression. A reduction or loss of E-cadherin expression was significantly correlated with pathologic tumor stage (P < 0.001), histologic grade (P < 0.001), but not lymph node involvement or lymphovascular invasion in bladder carcinomas. Loss or reduction in E-cadherin expression was recorded in 37.7% of superficial tumors and 60.2% of invasive tumors with similar values recorded for cytoplasmic localization of E-cadherin, 26% and 58.1%, respectively (Table 3).

β-Catenin expression. A reduction or loss of expression, in addition to a cytoplasmic localization of β-catenin, showed significant correlation with tumor grade and stage (P < 0.001) in bladder carcinomas. Neither the intensity nor the localization was found associated with a positive lymph node status or vascular invasion (Table 3). A reduction or loss of β-catenin expression was recorded in 16.7% of superficial tumors and 47.6% of invasive tumors (Table 3). The cytoplasmic localization of β-catenin revealed a greater differential between superficial and invasive tumor groups, 17.9% and 58%, respectively. Nuclear localization of β-catenin was recorded in 27 of 709 (3.8%) patient tumors, including 1 T_a, 9 T₁, and 17 T₂-T₄ bladder tumors.

Plakoglobin expression. A reduction or loss of expression of plakoglobin and the cytoplasmic localization of plakoglobin was significantly associated with tumor grade and stage (P < 0.001). In addition, plakoglobin localization was also correlated with nodal status in bladder cancer patients (P = 0.04). Neither the level of plakoglobin expression nor the location was associated with vascular invasion. Loss or reduced plakoglobin expression was recorded in 15.2% of superficial tumors and 57.9% of invasive tumors. As with β-catenin localization, there existed a marked differential between the frequency of cytoplasmic localization of plakoglobin in the superficial and invasive tumor groups, 23.1% and 72.1%, respectively (Table 3). Nuclear localization of plakoglobin was recorded in 13 of 726 (1.8%) bladder carcinomas, including four T_a, four T₁, and five T₂-T₄ lesions.

N-cadherin expression. Novel expression of N-cadherin was recorded in 61 of 746 (8.2%) bladder carcinomas (Fig. 1A and B), including 32 superficial and 24 invasive tumors with 5 tumors that remained unstaged. Membranous N-cadherin staining was rarely recorded expressed throughout a tumor (Fig. 1B) but displayed a limited focal localization (Fig. 1A). Of the 61 tumor samples positive for N-cadherin expression, only 21 were scored as positive in all scorable tumor cores from a single patient. Novel expression of N-cadherin did not show a significant association with grade, stage, lymph node involvement, or vascular invasion.

Vimentin expression. Novel expression of vimentin in bladder carcinoma cells in vivo was recorded in 143 of 795 (18%) tumors, including 29 superficial and 97 invasive lesions with 17 tumors that remained unstaged. Expression of vimentin (Fig. 1C and D) was significantly associated with tumor grade and stage (<0.001) but not with nodal involvement or vascular invasion. Figure 2 shows immunostaining for all markers in the same tissue core, demonstrating coexpression of the epithelial and mesenchymal intermediate filament proteins keratin (Fig. 2A) and vimentin (Fig. 2E), respectively.

Associations among EMT markers. Loss or reduced expression of E-cadherin was significantly associated with low β-catenin (P < 0.01; Spearman correlation coefficient, 0.446), low plakoglobin (P < 0.01; Spearman correlation coefficient, 0.370), and novel expression of vimentin (P < 0.01; Spearman correlation coefficient, −0.109). Low expression of β-catenin was independently associated with low plakoglobin expression (P < 0.01; Spearman correlation coefficient, 0.441) and vimentin expression (P < 0.01; Spearman correlation coefficient, −0.181). In turn, low plakoglobin was found significantly associated with vimentin expression (P < 0.01; Spearman correlation coefficient, −0.231). No significant associations were found with any of the aforementioned EMT markers and N-cadherin.

Survival analysis. Because there was no complete follow-up data for the superficial group, survival analysis was confined to the cystectomy patient group. Univariate analysis involving each of the proposed EMT markers in overall survival showed that the reduction or loss of plakoglobin expression (P = 0.02) was significantly associated with patient survival with a trend for association with β-catenin (P = 0.11; Fig. 3A and C). In contrast, no significant difference was recorded for E-cadherin or N-cadherin levels of expression (Fig. 3E and G). Within this patient group, the clinical variables of T-stage, lymph node status, and vascular invasion showed a significant correlation with survival, as previously established. The 5-year survival rate for patients whose tumors displayed a loss or reduction in expression of plakoglobin was 29.4% compared with 41.5% when plakoglobin displayed normal levels of expression. Ten-year survival rates were 22.9% and 29.2%, respectively, when the same plakoglobin variables were analyzed. The median time to death for patients with loss or reduced plakoglobin expression was 2 years compared with 4 years when a normal plakoglobin level was recorded. The 5-year survival rate for patients whose tumors displayed a loss or reduction in expression of β-catenin was 29.1% compared with 40.8% when β-catenin displayed normal levels of membrane staining. Ten-year survival rates were 24.2% and 23.5%, respectively, when the same β-catenin variables were analyzed. The median time to death for loss or reduced β-catenin was 2.4 years compared with 3.8 years when β-catenin displayed high expression.

When the outcome was limited to cancer-specific survival, the loss or reduction in plakoglobin expression (P = 0.02) and the reduction or loss of expression of β-catenin (P = 0.02) were significant for survival (Fig. 3B and D). As expected, T-stage, lymph node status, and vascular invasion correlated with cancer-specific survival. The 5- and 10-year survival rates were 43.7% and 41%, respectively, for loss or reduction of plakoglobin compared with 55% and 52.7%, respectively, in the group with normal levels of expression. With β-catenin, 5- and 10-year survival rates were 41.9% and 39.4%, respectively, when a loss or reduction of expression was recorded compared with 56.3% and 53.7%, respectively, for normal levels of expression. E-cadherin and N-cadherin did not show a significant difference for cancer-specific survival (Fig. 3F and H).

Multivariate analysis was done using plakoglobin, β-catenin, tumor stage, age, and vascular invasion because all showed significance in univariate analysis as recorded in Tables 4 and 5. For overall survival, vascular invasion and age of diagnosis were strong independent prognostic predictors (hazard ratio; 1.83; P < 0.001 and hazard ratio; 1.03 per year increment, P = 0.006, respectively). Loss of plakoglobin had a hazard ratio of 1.31 (P = 0.11), and β-catenin was no longer associated with survival (P = 0.82).

For cancer-specific survival, plakoglobin and β-catenin sites were included in the model with the aforementioned variables. Vascular invasion was the only significant independent predictor in the model (hazard ratio, 2.57; P < 0.001).

In this study, we have assessed the expression and prognostic value of a panel of five markers associated with the EMT phenotype. Loss of E-cadherin and novel expression of N-cadherin represent defining features of EMT, both of which have been reported in bladder tumorigenesis (22, 23). Linked to this transition has been the expression of the mesenchymal intermediate filament protein vimentin, often accompanied by altered cytokeratin expression (19). In addition, delocalization of β-catenin has been implicated in the EMT process (9). Numerous studies have shown loss or reduced expression of plakoglobin in epithelial cells in the absence of E-cadherin expression and more recently this phenotypic change has been implicated as part of the EMT phenotype (24). Changes in the aforementioned proteins represent EMT end points in which the expression of each may be modulated by a growing number of molecules involved in transcriptional regulation, cell signaling, and modulation of the tumor microenvironment (for review, see refs. 7, 14, 19, 24). However, there remains some controversy over the occurrence of such events in vivo originating from the apparent rarity of the EMT-like morphologic changes observed by pathologists in primary tumor sections and secondary lesions (8).

The counter argument to this necessarily requires recording EMT-related events in vivo during tumor progression (25, 26). Evidence for such events is beginning to emerge and presently include reports of Snail 1 protein, α_vβ₆ integrin, and nuclear β-catenin visualized at the invading edge of colon cancer tumors (14, 19). A search for the expression of such EMT-related proteins may be best served from the analysis of total tumor sections because the reported limited localization of proteins to the invasive front of the tumor may be missed in alternative tissue sampling formats.

A reduction or loss in expression of E-cadherin has long been recognized as an important primary event in bladder tumorigenesis often linked to poor prognosis (23, 27–29). In this study, altered E-cadherin expression was significantly associated with low β-catenin and plakoglobin expression along with novel expression of vimentin. Consideration of changes associated with the cadherin complex proteins revealed a significant difference between the frequency of such events in the superficial and invasive tumor groups, linking such changes to tumor progression. Although altered expression or localization of E-cadherin were not found associated with lymph node involvement or vascular invasion, the cytoplasmic localization of plakoglobin was a significant indicator of lymph node involvement.

Nuclear localization of β-catenin and plakoglobin was recorded in 3.8% and 1.8% of tumor samples, respectively, and was always observed in the presence of cytoplasmic localization of the catenin member. The reported frequency of nuclear localization of β-catenin in bladder tumors ranges between 0% and 22% identified in study groups involving small patient numbers (30–33). Mutations associated with β-catenin have been shown to result in the nuclear localization of this catenin family member in bladder tumors (30). Nuclear localization of plakoglobin seems to be a less frequent finding and although nuclear plakoglobin has been reported in other cancers (34, 35), we are not aware of any reports of this finding in bladder tumors. The nuclear localization of either β-catenin or plakoglobin was observed throughout the tumor section in a majority of samples whereby isolated islands of cells representing <5% of tumor cells with apparent nuclear staining were not scored as positive in this study. Given the potential putative transient nature of EMT and the limited, possibly temporary, nuclear relocalization of catenins associated with the invasive edge of the tumor, we may have overlooked more minor representation of EMT events.

Novel expression of N-cadherin has also been reported in human bladder tumors (22) and was previously recorded in 39% (20 of 51) of tumors. In this more extensive study, we have recorded N-cadherin expression in only 8.2% (61 of 746) of bladder tumors using the same antibody and antigen retrieval technique. Consistent with our previous report, novel expression of N-cadherin was observed to be focal in nature throughout the tissue microarray, and although the presence of four cores of each tissue is calculated to be representative of the tumor phenotype we believe that the N-cadherin expression recorded in this study is an underrepresentation of the frequency of expression in bladder tumors. Support for this interpretation is found in the scores for each of the four cores from a single tumor. A review of the results revealed detection of N-cadherin expression in all of the cores from a single tumor in approximately one third of the cases (21 of 61). It is also of interest that N-cadherin expression was recorded approximately equally in the superficial and invasive tumor groups. Moreover, in one recent publication, N-cadherin was identified as a prognostic marker of progression in superficial urothelial tumors (36). Although we have shown that novel expression of N-cadherin in bladder carcinoma cells promotes invasion (37), the expression of N-cadherin throughout different tumor grades and stage makes identification of additional functions a necessary consideration.

Novel vimentin expression, the second mesenchymal marker linked to EMT, was frequently recorded in the bladder carcinoma cell lines that displayed the more stellate, fibroblast morphology. However, past experience has revealed that normal urothelium can express vimentin after an extended period of in vitro culture (38). In this study, we have shown novel expression of vimentin in bladder carcinoma cells in vivo mainly associated with invasive bladder lesions. Confirmation of the epithelial origin of vimentin-positive cells was established in neighboring sections stained for total keratin demonstrating coexpression of these two intermediate filament types. No association was found between expression of the two mesenchymal markers, N-cadherin and vimentin. Whereas N-cadherin was found not to be associated with any of the clinical variables studied, vimentin expression was significantly associated with tumor grade and stage. In addition, novel vimentin expression showed significant association with reduced expression of both β-catenin and plakoglobin. Expression of vimentin was recorded in 18% (143 of 795) of bladder tumors, establishing novel expression of this intermediate filament in vivo. It is interesting that a subgroup of superficial tumors (6.8%) was identified that displayed novel expression of vimentin, where, given the association found with tumor stage and vimentin in invasive tumors, it is desirable to establish whether this marker identifies superficial tumors at heightened risk to progress as has been reported for N-cadherin (36).

In this study, survival data involved only patients that underwent cystectomy. For overall survival, the reduction or loss of plakoglobin was the one putative EMT variable that showed significant association with patient survival. As expected within this group of patients, the clinical variables of tumor stage, lymph node status, and vascular invasion also showed a correlation with survival. The median time to death for patients with loss or reduced plakoglobin expression was half that (2 years) recorded for patients with high expression of plakoglobin (4 years). Indeed, a previous study within a small group of bladder cancer patients (39) also reported that altered plakoglobin expression correlated with poor survival. In addition, alterations in plakoglobin expression have recently been identified as a marker of progression in T₁ bladder tumors (40). A similar pattern of loss or reduced expression was recorded for β-catenin, although such differences did not reach statistical significance. When the outcome was limited to cancer-specific survival, both the expression level and delocalization of plakoglobin and β-catenin were found to have prognostic significance for survival.

In this study, we have identified changes linked to EMT in different models as events that occur in bladder tumors in vivo. Loss of E-cadherin in bladder tumorigenesis is well established and, more recently, we have shown novel expression of N-cadherin in human bladder tumors (22). However, reduced expression of plakoglobin associated with these cadherin events has only recently been linked to EMT (24). This combination of phenotypic changes has been observed in all bladder carcinoma cell lines that display the morphologic EMT phenotype accompanied by the enhancement of migration and invasion. In this study, we have shown that loss of E-cadherin expression is significantly associated with low plakoglobin expression even in the presence of N-cadherin expression. In addition, novel expression of vimentin was also found significantly associated with alterations of E-cadherin, β-catenin, and plakoglobin expression. The lack of association of N-cadherin with variables assessed as part of the EMT phenotype is surprising, although, as previously stated, we believe there is an underestimation of the frequency of novel expression of N-cadherin. However, despite this caveat, we have established the in vivo expression of putative EMT alterations in bladder cancer in which loss or reduced plakoglobin expression presented as a significant prognostic feature in these tumors.

We thank Barbara Rothman for the technical help in immunohistochemical staining of tissue microarrays, and Robert Kelley and Matthew Wszolek for help in collecting materials and data.