The 2004 World Health Organization classification of renal tumors,1 which was built upon previous classification systems,2, 3 has refined the classification of adult renal epithelial neoplasms, taking into account morphological, clinicopathological and cytogenetic features. Several studies have shown that renal epithelial neoplasms have distinctive microscopic and molecular features.4–6 Although tumor stage at presentation, histological tumor grade and histological subtype are the principal prognostic factors of renal cell carcinoma (RCC), the clinical behavior, proclivity for metastasis, and potential response to therapy cannot be predicted thus far by histopathologic markers.4 Patients with clear cell RCC (ccRCC) have an increased risk of cancer specific death when compared to patients with papillary or chromophobe RCCs.7 However, significant heterogeneity is observed amongst ccRCCs as evidenced by variability in protein expression and disease progression.
Cytogenetic analysis of benign and malignant epithelial tumors of the kidney has identified associations between morphological features and chromosomal aberrations.8–10 Despite the frequent associations between morphological and karyotypic findings in epithelial kidney tumors,11–13 the association between protein expression of cytokeratins (CKs) and karyotypic alterations in ccRCC has not been addressed. Clear cell RCC is an extremely heterogeneous tumor with inconsistent immunohistochemical expression patterns.14, 15 We hypothesized that the variability observed for CK expression and other markers of differentiation may be best explained by genetic heterogeneity.
Cytokeratins are the fundamental markers of epithelial differentiation,16, 17 and CK expression patterns are largely retained during neoplastic transformation.18 The CKs found in simple epithelia (CK7, CK8, CK18, and CK19) are expressed in normal renal tubules and renal neoplasms. The majority of ccRCCs are positive for CK8 (about 40%) and CK18 (95–100%),14, 15 similar to the normal proximal renal tubule, suggesting that this is the site of origin for this subtype.19 As almost all ccRCCs express CK18,14, 15 we did not expect it to be a useful distinctive marker and therefore excluded this CK in the present study. CK7 and CK19 are also simple epithelial CKs but show a more restricted expression in ccRCC (each about 10–15%), making these 2 CKs useful markers to define certain ccRCC subgroups.14, 15 In the normal kidney, CK7 and CK19 are expressed by distal tubules and collecting ducts but are absent in proximal tubules. The expression of CK7 and CK19 in ccRCC subsets suggests that these ccRCCs may represent specific subtypes with potential origin from the collecting duct system or with neo-expression of these CKs.20
Here we demonstrate significant associations between the expression of specific CKs, the cell adhesion protein E-cadherin and cytogenetic alterations. Two ccRCC subtypes defined by CK7 and CK19 expression were determined to be associated with genetic stability, a distinct molecular signature as determined by expression array profiling and a more indolent clinical course.
Material and methods
Patients and tissue microarrays
Two hundred thirty-eight consecutive, unselected epithelial kidney tumors diagnosed between 1999 and 2004 were evaluated both by histology and cytogenetics at the Brigham and Women's Hospital, Boston, MA. All morphologic diagnoses were made on the basis of hematoxylin and eosin stained slides by 2 pathologists (K.D.M, M.S.H), according to the 2004 World Health Organization classification of renal tumors.1 Two RCC tissue microarrays (TMAs) were constructed of these tumors as previously described.21 Four cores (0.6 mm in diameter) were taken from each representative tissue block. Among the 238 kidney tumors 129 were ccRCCs, and karyotype data were available for 126 ccRCCs. Karyotypic studies were performed on all cases according to standard protocols. Cytogenetic aberrations were reviewed by a cytogeneticist (P.D.C.) who determined if the karyotypes were normal, diagnostic for a particular epithelial tumor, or nondiagnostic. Histologic and cytogenetic diagnoses were recorded independently and compared for concordance. All data is stored in a database (http://rubinlab.med.cornell.edu/supplemental_data/cytogenetics_data/index.jsp).
An independent cohort of 209 ccRCC patients with long-term follow-up and cancer-specific death as endpoint was identified in the files of the Institute for Surgical Pathology, University Hospital Zurich, Switzerland. The clinical and pathological parameters of this cohort are described elsewhere (mean follow-up 52 months, range 1–179 months).22
All samples were collected from consented patients with prior institutional review board approval at each respective institution.
All TMA sections were freshly cut and immunostained as previously described.23 Table I lists the monoclonal antibodies used in our study.
Table I. Monoclonal Antibodies Used
CK, cytokeratin; EMA, epithelial membrane antigen; WT-1, wilms tumor 1.
Reacts primarily with CK8 and weakly with CK7.
Reacts with CK 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 13, 14, 15, 16, 17, 18, 19.
Reacts with CK 1, 2, 3, 4, 5, 6, 7, 8, 10, 13, 14, 15, 16, 19.
Semiautomated quantitative image analysis by chromavision
Evaluation of all immunohistochemical stainings was performed with the Chromavision (Chromavision Medical Systems, San Juan Capistrano, CA) Automated Cellular Imaging System (ACIS II), an upgraded version of the previously described ACIS System.25 The system combines automated microscopy with computerized image analysis to generate continuous quantitative measurements of immunohistochemical stainings in terms of staining intensity (between 0 and 255) and percentage of stained tissue area (between 0 and 100). Each tissue microarray core was reviewed to ensure that protein expression measurements were taken from diagnostic regions and that staining scores obtained from nonrepresentative regions were excluded.
Unsupervised analysis of protein expression was performed by hierarchical clustering using average linkage method and correlation measure to evaluate case similarities. When analyzing protein expression data, single cluster enrichment for cytogenetic or histological characteristics was assessed by means. This analysis contains information from protein profiles of each tumor sample. We then focused our analysis on CK expression by correlating expression of each individual CK analyzed with either cytogenetic or histological characteristics of the tumors, by using unpaired student's t test (two-tailed p-value). For multiple hypothesis testing we applied Bonferroni correction, when appropriate.
For survival analysis, we used the Zurich cohort with 209 ccRCC cases. Kaplan–Meier analysis was used to present associations between cancer specific death and CK expression. Survival time was calculated from date of nephrectomy to date of death or to date of last clinical follow-up. Log-rank test was used to evaluate statistical significance of curve splitting. Cox proportional hazard regression analysis was used at univariate and multivariate level analyses. Backward Wald statistics was applied to identify the most parsimonious model. All statistics were performed using SPSS 16.0 for Windows (SPSS, Chicago, IL) with a significance level of 0.05 on two-tailed p-values.
For cDNA microarray analysis, frozen material of 21 primary ccRCCs and 14 ccRCC metastases were randomly selected from the archive of the Institute for Surgical Pathology, University Hospital Zurich, Switzerland. All tumors were histologically evaluated by one pathologist (H.M.) and selected for the study on the basis of hematoxylin and eosin stained tissue sections. Total RNA was isolated with Trizol (Invitrogen, Carlsbad, CA). RNA quantity was determined based on A260, and the integrity of RNA was confirmed by agarose gel electrophoresis.
cDNA array analysis
Total RNA was amplified, labeled, and hybridized to HT_HG-U133A High-Throughput Arrays (Affymetrix, Santa Clara, CA) which were scanned using the HT Scanner, allowing 96 samples to be processed in parallel with minimal operator intervention. The original CEL files were processed with dChip (http://biosun1.harvard.edu/complab/dchip/) to obtain single data files comprising 22277 probe-sets. For each array, probe level intensities were normalized against the median array. Low expression values were filtered out with a high-pass filter and the data was log-transformed. The gene expression was standardized such that the distribution in each sample had median zero and standard deviation equal to 1.
The array probe sets for CK7 and CK19 were identified and analyzed. The array design includes 2 probe sets for CK7: 209016_s_at (chr12:50913304-50928976 (+) // 96.24 // q13.13) and 214031_s_at (chr12:50913483-50923188 (-) // 96.79 // q13.13). We computed the mean of the 2 probe sets to obtain a representative CK7 expression for further analysis. CK19 is represented on the array by one probe set (chr17:36933395-36938054 (-) // 99.41 // q21.2).
Pearson correlation for gene selection
Pearson correlation was used to select genes according to their relation to CK7 or CK19 across samples. As we had 2 probe sets for CK7, we computed the mean of the probe sets to obtain a representative CK7 expression. For CK19 with 1 probe set only, we computed the Pearson correlation of CK19 against every probe set. Genes with a two-tailed p-value < 0.001 were selected. Clustering analysis on gene expression data was performed as described for protein expression data.
Associations between biomarker expression and cytogenetic profiles of ccRCC
The karyotypes of 126 ccRCCs from the Brigham and Women's Hospital Boston, obtained by conventional cytogenetic analysis, were compared with protein expression patterns. Sixteen immunohistochemical markers were selected (Table I), and the protein expression was evaluated by a semiquantitative image analysis system (Chromavision) (Fig. 1).
Within the group of 126 ccRCCs, we determined the number of alterations for each case and subdivided them in those without any cytogenetic alterations (30 cases, 23.8%), tumors with 1 or 2 alterations (37 cases, 29.4%), 3 or 4 alterations (31 cases, 24.6%), and more than 4 alterations (28 cases, 22.2%) (Supplemental Fig. 1). The most common cytogenetic alterations affected chromosome 3 in approximately one fourth of the cases (27%), most of them through translocation (42%), deletion (13.5%) or loss of the entire chromosome (26%) (Table II).
Table II. The Most Common Chromosomal Alterations in a Group of 126 ccRCCs as Determined by Conventional Cytogenetics
Alteration (on chromosome no.)
Number of cases (Frequency in %)
Add, addition; Del, deletion; t, translocation; t(3;a), translocation involving chromosome 3 and any other chromosome (a).
The frequencies are evaluated against the total number of alterations (365) within the set of ccRCCs.
One aim of this study was to classify ccRCCs on the basis of their differential expression of epithelial markers. Cytokeratins (Pan-CKa, Pan-CKb, CK7, CK8a, CK8 and CK19), other epithelial markers (EMA, E-cadherin), and the mesenchymal marker vimentin, all of which are known to be positive in subsets of ccRCC, were evaluated (Table I). Representative positive and negative cores for CK7, CK8 and CK19 immunostaining are depicted in Figure 1, illustrating the heterogeneity for these immunohistochemical markers in ccRCC. The mean immunohistochemical staining intensity of each protein, based on absence or presence of each common lesion, and the two-sided p-value of the t test corrected for multiple hypotheses were determined. To investigate if there were associations between protein expression patterns and genomic lesions, we ran hierarchical clustering and evaluated the clusters for enrichment for any particular lesion. The clustering analysis output is visualized by a dendrogram organizing the samples based on similarity of their protein expression profiles. The expression values of the entire cohort of 126 ccRCCs are shown in the heatmap (Fig. 2). Tumors are annotated according to main chromosomal aberrations as listed in Table II, to their total number of cytogenetic alterations and to Fuhrman nuclear grade (FNG).
Table III. Cox Regression Univariate Analysis of the Zurich Cohort (209 ccRCCs)
95.0% CI for RR
n.s., not significant.
For CK7, CK8a, CK8, and CK19, cases with expression above and below or equal to the median were compared.
A small group of cases with high expression of CKs (Pan-CKa, Pan-CKb, CK7 and CK8a) was enriched for absence of cytogenetic alterations (p = 0.02) and for FNG 2 tumors (p = 0.01). A second subset of cases consisting of E-cadherin expressing ccRCCs with FNG 2 was also lacking any cytogenetic abnormalities (p < 0.01). In contrast, tumors with a high number of cytogenetic changes tended to express lower levels of CKs (CK7, Pan-CKb) and E-cadherin compared to tumors with a low number of cytogenetic changes.
We identified another ccRCC group which does not express CKs but vimentin. This group showed higher frequency of cytogenetic alterations and higher FNGs. Vimentin-positive tumors, which are frequently negative for markers of epithelial differentiation, most likely represent a subset of advanced tumors with aggressive clinical behavior.26
A second independent cohort of 209 ccRCCs with clinical follow-up data was evaluated to test if CK expression in ccRCCs was associated with differences in prognosis. Sections of the TMA were stained for Pan-CKb, CK7, CK8a, CK8, CK19 and E-cadherin. The expression levels of each protein were determined by Chromavision and transformed into nominal variables, above and below the median. Cox regression analysis at univariate and multivariate levels considered histopathological and clinical parameters as well as protein expression.
Cox regression results are summarized in Table III (univariate level). Figure 3 shows the survival curves for selected biomarkers. CK7 expression was associated with better outcome (p < 0.01, Fig. 3a), whereas CK8 (Fig. 3b) and CK19 (Fig. 3d) expression did not show long-term association. However, within the first 5 years after initial diagnosis, CK19 expression and CK8a expression were associated with a more favorable outcome (p < 0.01, Table IV, Fig. 3d; p = 0.01; Table IV, Fig. 3c).
Table IV. Cox Regression Univariate Analysis of the Zurich Cohort (209 ccRCCs) within 5 Years from Diagnosis
95.0% CI for RR
ccRCC cases with CK7, CK8a, CK8 or CK19 expression above and below or equal to the median were compared.
Multivariate Cox proportional hazards regression analysis showed that metastatic status (RR = 7.28; 95%CI = 1.93–27.39; p < 0.01) and CK19 expression (RR = 0.23; 95%CI = 0.06–0.90; p = 0.04) emerged as independent prognostic factors for cancer-specific survival at a multivariate level.
When testing for association between expression of single proteins and histopathological characteristics of the 209 ccRCC tumors within this cohort, we detected a significant association between CK7 expression and low FNG (p < 0.05), low pathological stage (p = 0.02), absence of necrosis (p < 0.05), and absence of lymphocytic infiltration (p < 0.01) (Table V). All p-values are two-sided and corrected for multiple hypotheses.
Table V. For Each Immunohistochemical Marker (Percentage of Tumor area Expressing the Protein), we Tested for Significant Differences with Respect to Clinicopathological Variables
E-cadherin p-value (up-down) N = 199
Pan-CK, p-value (up-down), N = 178
CK7, p-value (up-down), N = 184
CK8, p-value (up-down), N = 174
CK19, p-value (up-down), N = 176
Mann–Whitney test was applied. All p-values are two-tailed. Uncorrected p-values are reported for all significant test results, and the reported results represent downregulations of the markers with respect to the clinicopathological features. Results in brackets are not significant, but revealed a tendency. If we consider multiple hypothesis correction, the bold printed p-values are still significant. Downregulation of CK7 expression appears to be strongly associated with aggressiveness.
Determination of CK7and CK19 specific signatures in ccRCC by microarray analysis
We further determined if CK7 and CK19 expression in ccRCCs is associated with specific gene expression signatures. Gene expression profiles were obtained from 21 primary ccRCCs and 14 ccRCC metastases by cDNA microarray analysis. Pearson correlation was used to estimate the relation between each single gene and CK7/CK19 expression, the 2 CKs identified as having prognostic relevance in ccRCC. Genes with a two-sided p-value <0.001 were selected. We identified 62 genes with a strong relation to CK7 (Fig. 4), and 182 genes related to CK19 (Supplemental Fig. 2).
Loss of CK7 expression in ccRCC was associated with significant dysregulation of genes involved in transcription, signal transduction, cell cycle regulation, DNA mismatch repair, cell adhesion and immune responses. To evaluate whether the transcripts identified could help distinguish between ccRCC subtypes defined by CK7 expression, we selected 2 candidates whose transcripts were up- (PMS2) or down- (MMP14) regulated in relation to CK7. For both PMS2 and MMP14, antibodies were commercially available.27, 28 PMS2 was strongly expressed in all ccRCCs positive for CK7. These cases had low or absent MMP14 expression. Representative TMA cores are depicted in Figure 5.
Thus, the data sets obtained by Pearson correlation for CK7 and CK19 represent a potential source of markers that may aid in distinguishing molecular ccRCC subclasses.
Markers of epithelial differentiation, and among those CKs as constituents of the cytoskeleton in epithelial cells, show variable expression in ccRCC. In our study, we define a correlation between the heterogeneous expression of CK subtypes and the heterogeneous clinical behaviour of ccRCC. We demonstrate that the distinctive CK expression patterns of ccRCCs can be a surrogate for underlying genomic alterations, have predictive value for patient outcome and reflect different gene expression patterns in ccRCC subgroups.
Our findings in ccRCC are consistent with CK expression studies in other malignancies. Schaller et al. have previously shown that loss of CK8 and CK18 in breast carcinomas was associated with a significantly worse prognosis, whereas the majority of patients with high CK18 remained recurrence-free.29 In breast cancer cell lines, forced re-expression of CK18 was associated with reduced invasiveness in vitro.30 A similar correlation between CK8/CK18 expression and reduced tumorigenicity was reported by Pankov et al. for murine pancreatic carcinoma cells.31 Similar results were also obtained by Sommers and Thompson who showed that breast cancer cell lines became more aggressive as keratin filaments were replaced by vimentin, the intermediate filament-protein of mesenchymal cells.32, 33
Our conventional cytogenetic analysis identified ccRCC subgroups with prognostically relevant CK expression patterns, indirectly suggesting that the number of cytogenetic alterations is linked to the biologic tumor behavior. It has been suggested that RCC progression is characterized by an accumulation of complex chromosomal alterations. Comparative genomic hybridization (CGH) analyses have shown that a high number of chromosomal copy number aberrations is associated with poor prognosis.5 Interestingly, 24% of ccRCCs in our study did not reveal any conventional cytogenetic aberrations. Conventional cytogenetic analysis as applied for the Brigham and Women's Hospital cohort has its limitations, and therefore could affect our results. One of these limitations is the low resolution of the method. Furthermore, conventional cytogenetics can result in a bias towards tumor cells with a growth advantage in vitro. For these reasons, we validated our hypotheses on a large independent cohort of ccRCCs with long-term follow-up data.
Cytogenetic studies have shown that 3p deletions are linked to ccRCC, occurring in 40–70% of sporadic tumors with this histological subtype.34–36 Previous molecular studies demonstrated that different genes on chromosome 3p are candidates for tumor suppressor genes (TSG), with VHL on 3p25.5 as the most frequently inactivated TSG. Although it is assumed that the inactivation of one or more genes on 3p plays a role in ccRCC initiation, it has been suggested that other cytogenetic alterations go along with ccRCC progression, but the nature and relevance of their interrelationship is poorly understood. Our data strongly indicate that there are ccRCC clusters defined by specific cytogenetic alterations, resulting in different CK expression patterns.
Microarray analysis of malignancies from other organ sites has revealed molecular subtypes of tumors that are histologically indistinguishable, and gene expression profiles have been identified that correlate with clinical outcomes.37–39 Global analysis of gene expression using cDNA microarray technology offers significant opportunities to identify novel markers that discriminate between classes of tumors and holds promise in identifying molecular subclasses of tumors with differing prognosis. Our data provide a starting point for identification of distinctive molecular signatures for ccRCC subtypes and may offer unique biological insights into these tumors. We demonstrate the potential benefits of 2 novel biomarkers in ccRCC with distinctive expression patterns that were identified in our analysis. The DNA mismatch repair protein PMS2 was found to be strongly expressed in CK7 expressing ccRCCs. Defects in DNA mismatch repair result in an accumulation of mutations that are associated with human cancers and diseases. While their contribution to gastrointestinal tumors is well studied, the role of mismatch repair defects in other tumor types is not well characterized. Norris et al. report a significant increase in the level of PMS2 in pre-cancerous and low grade prostate cancer as compared to higher grade tumors.28 Our finding of increased PMS2 expression in CK7 positive ccRCCs is in line with this observation. However, further validation will be necessary to determine the prognostic significance of PMS2 expression in ccRCC.
In our study, we show the absence of matrix metalloproteinase (MMP) MT1 (also known as MMP14) in CK7 positive ccRCCs. A crucial step in tumor progression and invasion is the proteolytic degradation of the extracellular matrix (ECM) and basal membranes.40 MMPs are zinc-dependent endopeptidases, which are largely involved in tissue remodeling, degradation of the ECM and tumor invasion.41 Although most MMPs are secreted, the membrane type 1 matrix metalloproteinase (MT1-MMP) is localized in the cell membrane.42 Its expression has been correlated with the invasive capacity of different tumors.41 Our findings suggest that the good prognosis of CK7 expressing ccRCC can be partially explained by absence of MT1-MMP expression.
In summary, we have identified ccRCC subgroups, which are characterized by specific CK expression patterns and different prognosis. These findings may also have therapeutic relevance in the near future. Targeted delivery of liposome-encapsulated CK18 expression vectors by conjugation to anti-HER2/neu antibodies seems a promising approach in the treatment of breast cancer.43 Similar treatment strategies could be envisioned for ccRCC. Biomarkers and pathways associated with CK7 or CK19 expression provide sources of novel therapeutic targets for ccRCC.
We thank Mr. Jeff Tang for generation of the cytogenetics database.