Genetic Clustering of Clear Cell Renal Cell Carcinoma Based on Array-Comparative Genomic Hybridization: Its Association with DNA Methylation Alteration and Patient Outcome

Renal cell carcinoma (RCC) is the most common malignant tumor of the adult kidney and frequently affects working-age adults in mid life (1). In general, RCCs at an early stage are curable by nephrectomy. However, some RCCs relapse and metastasize to distant organs, even if the resection has been considered complete. Metastatic RCCs are resistant to conventional chemotherapy and radiotherapy and have a poor outcome (2). Recently, immunotherapy has been established (3) and novel targeting agents have been developed (4) for treatment of RCC. However, unless relapsed or metastasized tumors are diagnosed early by close follow-up, the effectiveness of any therapy is very restricted. Therefore, to assist close follow-up of patients who have undergone nephrectomy and are still at risk of recurrence and metastasis, a prognostic indicator should be established based on an understanding of the molecular mechanisms of renal carcinogenesis.

Although the classification of RCC is based largely on histology, the WHO classification has introduced genetic alterations as a hallmark corresponding to the histologic subtypes of RCC, e.g., clear cell RCC, the most common histologic subtype, is characterized by loss of chromosome 3p and inactivation of the von Hippel-Lindau (VHL) gene at 3p25.3 (1). Somatic inactivation of the VHL gene occurs in only 60% to 70% of sporadic clear cell RCCs (5), and aberrations of the VHL gene alone cannot explain the development of all clear cell RCCs. However, only a few studies using recently developed array-based technology and demonstrating copy number alterations in clinical tissue samples of clear cell RCC have been reported (6, 7). In such previous array-based analyses, the numbers of examined clear cell RCCs and the numbers of clones on the arrays used were low, and significant prognostic factors for clear cell RCC based on array-based comparative genomic hybridization (CGH) profiles have never been proposed.

We have reported that DNA methylation alterations are another important change occurring during renal carcinogenesis (8, 9). Accumulation of DNA methylation on C-type CpG islands in a cancer-specific but not age-dependent manner (10) in clear cell RCCs was significantly correlated with tumor aggressiveness and poorer patient outcome (8, 9). DNA methylation alterations were observed even in noncancerous renal tissues obtained from patients with clear cell RCCs, and such renal tissues were considered to be at the precancerous stage. Accumulation of DNA methylation on C-type CpG islands in such noncancerous renal tissues was significantly correlated with higher histologic grades of the corresponding clear cell RCCs developing in individual patients, suggesting that regional DNA hypermethylation in precancerous conditions generates more malignant RCCs (8, 9). However, to our knowledge, no published systematic reports have examined the correlation between copy number alterations and changes in DNA methylation. Therefore, current understanding of the genetic and epigenetic alterations occurring during renal carcinogenesis is far from complete.

In this study, we analyzed copy number alterations by array-CGH using a microarray of 4,361 bacterial artificial chromosome (BAC) clones, allowing high-resolution genome-wide analysis, and DNA methylation alterations on 9 CpG islands by bisulfite modification in 51 clear cell RCCs. Correlations between copy number alterations and changes in DNA methylation, and the clinicopathologic significance and prognostic effect of the copy number alterations, were examined.

Patients and tissue samples. Tumor tissues were obtained from materials surgically resected from 51 patients (RCC01 to RCC51) with primary clear cell RCC. These patients did not receive preoperative treatment and underwent nephrectomy at the National Cancer Center Hospital, Tokyo, Japan, between 1999 and 2006. There were 34 men and 17 women with a mean (±SD) age of 59 ± 10 years (range 31-81 years). Histologic diagnosis was made in accordance with the WHO classification (1). All the tumors were graded on the basis of previously described criteria (11) and classified according to the pathologic tumor-node-metastasis (TNM) classification (12). The presence or absence of vascular involvement was examined microscopically on slides stained with H&E and elastica van Gieson. The presence or absence of tumor thrombi in the main trunk of the renal vein was examined macroscopically. This study was approved by the Ethics Committee of the National Cancer Center, Tokyo, Japan.

Clear cell RCC is usually enclosed by a fibrous capsule and is well demarcated, hardly ever containing a fibrous stroma between the cancer cells. Therefore, we were able to obtain cancer cells of high purity from surgical specimens, avoiding contamination with both noncancerous epithelial cells and stromal cells. High-molecular-weight DNA from fresh frozen tumor samples was extracted with phenol-chloroform, followed by dialysis.

Array-CGH analysis. Copy number alterations were analyzed by array-CGH using a custom-made array (MCG Whole Genome Array-4500) harboring 4,361 BAC clones throughout chromosomes 1 to 22 and X and Y, providing a resolution of ∼0.7 Mb (13), as described previously (14, 15). Briefly, gender-matched Human Genomic DNA (Promega) was used as reference. DpnII-restricted test and reference genomic DNAs were labeled by random priming with Cy3- and Cy5-dCTP (GE Healthcare), respectively, using a BioPrime array CGH genomic labeling system (Invitrogen) and precipitated together with ethanol in the presence of Cot-I DNA. The mixture was applied to array slides and incubated at 43°C for 72 h. Arrays were scanned with a GenePix Personal 4100A (Axon Instruments) and analyzed using GenePix Pro 5.0 imaging software (Axon Instruments) and Acue 2 software (Mitsui Knowledge Industry).

The results of array-CGH were validated by fluorescence in situ hybridization (FISH) analysis in representative RCCs as described previously (16). BAC clones, RP11-115G3 (3p25.3), RP11-4E3 (5q31.1), and RP11-79I6 (5q32), which are included in the MCG Whole Genome Array-4500, were labeled with SpectrumOrange-dUTP (Abbott Laboratories) using a nick translation kit (Abbott Laboratories) and hybridized to 5-μm-thick sections of formalin-fixed, paraffin-embedded tissue specimens taken from a region immediately adjoining that from which the corresponding fresh frozen sample had been obtained within the same RCC. Nuclei were stained with 4,5-diamidino-2-phenylindole.

Methylation-specific PCR and combined bisulfite restriction enzyme analysis. DNA methylation status on 9 CpG islands (8 C-type CpG islands plus the CpG island of the VHL gene) was analyzed by methylation-specific PCR and combined bisulfite restriction enzyme analysis as described previously (17, 18). Briefly, bisulfite conversion was carried out using a CpGenome DNA Modification Kit (Chemicon International). DNA methylation status on CpG islands of the p16, human MutL homologue 1 (hMLH1), and VHL genes was determined by methylation-specific PCR using the primers described previously (19, 20). The DNA methylation status of the thrombospondin (THBS) 1 gene and the methylated in tumor (MINT)-1, MINT-2, MINT-12, MINT-25, and MINT-31 clones was determined by combined bisulfite restriction enzyme analysis using previously described primers (10) and restriction enzymes (8).

Statistics. Unsupervised hierarchical clustering analysis of the RCCs was done using Impressionist software (Gene Data) as described previously (15, 21, 22). The average number of BAC clones for which copy number alterations (loss and gain) were observed in clear cell RCCs belonging to clusters A and B yielded by the unsupervised hierarchical clustering was analyzed using the Mann-Whitney U test. The frequency of copy number alterations (loss and gain) on each BAC clone, DNA methylation on each CpG island, and CpG island methylator phenotype in clusters A and B were analyzed using the χ² test. Correlations between genetic clustering of clear cell RCC (clusters A and B) and clinicopathologic variables were analyzed using the χ² test. Survival curves were calculated by the Kaplan-Meier method according to genetic clustering of clear cell RCC (clusters A and B), and the differences were analyzed by the log-rank test. The Cox proportional hazards multivariate model was used to examine the prognostic effect of genetic clustering of clear cell RCC (clusters A and B), histologic grade, and pathologic TNM stage. Differences with P values of <0.05 were considered significant.

Array-CGH analysis. Examples of array-CGH profiles of the two representative clear cell RCCs (RCC01 and RCC02) are shown in Fig. 1A to D (Fig. 1A and B for RCC01 and Fig. 1C and D for RCC02). Fig. 1A and C are scattergrams of the signal ratios (test signal/reference signal) and Fig. 1B and D are the corresponding histograms. As shown in Fig. 1B and D, the histograms of the signal ratios showed multiple distinct peaks. The thresholds of the signal ratios for copy numbers 0, 1, 2, 3, and 4 or more were determined by the troughs between the peaks on the histogram of each RCC (red bars, Fig. 1B and D for RCC01 and RCC02, respectively). For example, the signal ratio of the RP11-115G3 clone in RCC01, shown as an asterisk in Fig. 1A and B, corresponded to copy number 1 (Fig. 1B). Similarly, the signal ratio of the RP11-4E3 clone in RCC01, shown as double asterisks in Fig. 1A and B, corresponded to copy number 3 (Fig. 1B). The signal ratios of the RP11-115G3 and RP11-4E3 clones in RCC02, shown as an asterisk and double asterisks in Fig. 1C and D, corresponded to copy numbers 1 and 4 or more, respectively (Fig. 1D). These data for copy numbers were validated by FISH analysis: clones RP11-115G3 and RP11-4E3 actually revealed 1 and 3 signals in RCC01 (Fig. 1E and F, respectively) and clones RP11-115G3 and RP11-4E3 actually revealed 1 and 4 signals in RCC02 (Fig. 1G and H, respectively).

Unsupervised hierarchical clustering of clear cell RCC based on array-CGH data. By unsupervised hierarchical clustering analysis of the RCCs based on array-CGH data, 51 clear cell RCCs were clustered into the two subclasses, clusters A and B (Fig. 2A), which contained 34 and 17 clear cell RCCs, respectively.

The copy number on each BAC clone was defined as a loss, no change, and a gain when it was 0 or 1, 2, and 3 or 4 or more. In clear cell RCCs, the average number of BAC clones on which loss was detected was significantly higher in cluster B (563 ± 568) than in cluster A (164 ± 116, P = 0.0174). The average number of BAC clones on which gain was detected tended to be higher in cluster B (319 ± 306) than in cluster A (206 ± 243). The average number of BAC clones on which loss or gain was detected was significantly higher in cluster B (881 ± 831) than in cluster A (370 ± 274, P = 0.0368), indicating that copy number alterations were significantly accumulated in cluster B compared with cluster A. Figure 2B shows the copy number alteration profiles of clusters A and B. In both clusters, loss or gain of an entire chromosome or an entire chromosome arm was frequent (Fig. 2B). Table 1 shows an overview of the copy number alterations in clusters A and B, and the frequency of loss and gain on a BAC clone, which represented the tendency for the entire chromosome arm, per chromosome arm.

Figure 2B shows that loss of chromosome 3p and gain of chromosomes 5q and 7 were frequent in both clusters A and B. These data are confirmed in Table 1: The frequency of loss on clone RP11-865O5 (3p14.3-3p21.1) in clusters A and B was high (91% and 76%, respectively) and there was no significant difference in the frequency between the clusters (P = 0.3139; Table 1). Similarly, the frequency of gain on clones RP11-125L2 (5q35.2), RP11-90J13 (7p14), and RP11-451M8 (7q11.22) in clusters A and B was high and there was no significant difference in the frequency between the clusters (P = 0.7660, P = 0.2935, P = 0.1855 for RP11-125L2, RP11-90J13, RP11-451M8 clones, respectively, Table 1). Figure 2B shows that loss of chromosomes 1p, 4, 9, 13q, and 14q was frequent only in cluster B, but not in cluster A. These data are confirmed in Table 1: The frequency of loss on clones RP11-4O6 (1p35.3), RP11-38M16 (4p12), RP11-11P20 (4q28.1), RP11-327L3 (9p13.1-9p13.3), RP11-65C15 (9q21.3-9q22.1), RP11-8C15 (13q12.1), and RP11-140I4 (14q32.1) in cluster B (29%, 29%, 29%, 41%, 47%, 35%, and 71%) was significantly (P = 0.0047, P = 0.0047, P = 0.0047, P = 0.0003, P < 0.0001, P = 0.0013 and P < 0.0001) higher than that in cluster A (0%, 0%, 0%, 0%, 0%, 0%, and 0%, respectively, Table 1). Figure 2B shows that gain on 1q31-ter, 3q, and 8q was frequent only in cluster B, whereas loss at the same loci was observed in cluster A. These data are confirmed in Table 1: The frequency of gain on clones RP11-135J2 (1q41), RP11-91B3 (3q13.1), and RP11-90N3 (8q21.3-8q22) in cluster B was 24%, 18%, and 29%, whereas that in cluster A was 0%, 0%, and 0% and the frequency of loss on the same clones in cluster A was 6%, 24%, and 6%, respectively (Table 1).

Correlation between genetic clustering of clear cell RCC and accumulation of DNA methylation on CpG islands. DNA methylation status on the nine CpG islands in 41 clear cell RCCs of the present cohort had been already analyzed and was included in our previous article (8). DNA methylation status in the remaining 10 clear cell RCCs was analyzed here. The frequency of DNA methylation on CpG islands of the p16, hMLH1, VHL, and THBS1 genes and the MINT-1, MINT-2, MINT-12, MINT-25, and MINT-31 clones was 23 of 34 (detected/analyzed, 68%), 4 of 34 (12%), 0 of 34 (0%), 5 of 34 (15%), 2 of 34 (6%), 0 of 34 (0%), 5 of 34 (15%), 10 of 34 (29%), and 0 of 34 (0%) in cluster A and 9 of 17 (53%), 1 of 17 (6%), 2 of 17 (12%), 10 of 17 (59%), 5 of 17 (29%), 2 of 17 (12%), 1 of 17 (6%), 9 of 17 (53%), and 0 of 17 (0%) in cluster B, respectively. The frequency of DNA methylation on CpG islands of the THBS1 gene in cluster B was significantly higher than that in cluster A (P = 0.0034). The average number of methylated CpG islands was significantly higher in cluster B (2.29 ± 1.49) than in cluster A (1.44 ± 0.89, P = 0.0279). Patients were considered CpG island methylator phenotype–positive when DNA methylation was seen on three or more examined CpG islands, based on previously described criteria (10). The frequency of CpG island methylator phenotype in cluster B (47%) was significantly higher than that in cluster A (12%, P = 0.0142). Figure 3 shows a histogram of the number of methylated CpG islands in clusters A and B. DNA methylation on CpG islands was accumulated in cluster B compared with cluster A, and DNA methylation on 4 or more CpG islands was observed only in cluster B and never in cluster A (Fig. 3).

Clinicopathologic significance and prognostic effect of genetic clustering of clear cell RCC.Table 2 shows the clinicopathologic variables of clear cell RCCs belonging to clusters A and B. Clear cell RCCs in cluster B showed significantly higher histologic grades (P = 0.0063) and more frequently showed vascular involvement (P = 0.0045), renal vein tumor thrombi (P = 0.0064), and higher pathologic TNM stages (P = 0.0066) than those in cluster A (Table 2). Figure 4 shows the Kaplan-Meier survival curves of patients based on genetic clustering of clear cell RCC (clusters A and B). The period covered ranged from 88 to 2,801 days (mean, 1,679 days). After nephrectomy, none of the patients received any adjuvant therapy before recurrence or metastasis was revealed. Recurrence or metastasis was observed in 6 (40%) of 15 patients who underwent curative resection in cluster B, but in only 3 (9%) of 34 patients who underwent curative resection in cluster A. The recurrence-free survival rate of patients in cluster B was significantly lower than that of patients in cluster A (Fig. 4A; P = 0.0018). Four (24%) of the total 17 patients in cluster B died as a result, whereas none (0%) of the patients in cluster A died. The overall survival rate of patients in cluster B was significantly lower than that of patients in cluster A (Fig. 4B; P = 0.0009). Multivariate analysis revealed that genetic clustering was a predictor of recurrence-free survival (P = 0.0297) and was independent of histologic grade and pathologic TNM stage (Table 3), although the effects of these three variables on overall survival were not independent from each other (data not shown).

Only a few studies using recently developed array-based technology and demonstrating copy number alterations in clinical samples of clear cell RCCs have been reported (6, 7). Therefore, the genetic pathway of renal carcinogenesis has not been fully clarified. To our knowledge, the resolution achieved in the present study was higher than that of any other previous array-CGH analyses (6, 7) of clear cell RCCs. We used MCG Whole Genome Array-4500, which harbors 4361 BAC clones distributed throughout the human genome and has successfully revealed copy number alteration profiles in human cancers derived from various organs (13). Our array-CGH analysis was performed using RCC cells of high purity from fresh good-quality specimens and the results were carefully validated by FISH using the same BAC clones as probes.

First, our genome-wide analysis revealed that loss or gain of an entire chromosome or an entire chromosome arm was frequent in clear cell RCCs. Other than the well-studied VHL gene (3p25.3), some tumor-related genes may be identified in BAC regions where copy number alterations were observed. Unsupervised hierarchical clustering analysis based on array-CGH data grouped the 51 clear cell RCCs into two subclasses, clusters A and B, and copy number alterations were found to be significantly accumulated in cluster B. Moreover, we identified distinct copy number alteration profiles in the two clusters: (a) loss of chromosome 3p and gain of chromosomes 5q and 7 were frequent in both clusters A and B; (b) loss of chromosomes 1p, 4, 9, 13q, and 14q was frequent only in cluster B but not in cluster A; and (c) gain on 1q31-ter, 3q, and 8q was frequent only in cluster B, whereas loss on the same loci was observed in cluster A, although the frequency was rather low. Therefore, our clusters A and B, by unsupervised hierarchical clustering analysis, can be considered valid. Although the genetic profiles of clear cell RCCs obtained by CGH (23–25) and FISH (25–27) analyses without array-based technology have been compared with those of other histologic types of RCC (such as papillary RCC and chromophobe RCC), the subclasses of clear cell RCC itself, which may reflect the distinct genetic pathways of carcinogenesis, have not been defined. The present high-resolution, genome-wide analysis indicated that loss of chromosome 3p and gain of 5q and 7 may be indispensable copy number aberrations for the development of clear cell RCCs, regardless of genetic clustering. Additional loss of chromosomes 1p, 4, 9, 13q, and 14q may promote the genetic pathway to cluster B, although the order of occurrence of indispensable copy number aberrations and additional losses for cluster B during renal carcinogenesis has not yet been defined.

We have reported that DNA methylation alterations are an important epigenetic change during renal carcinogenesis (8, 9) and are significantly correlated with tumor aggressiveness and poorer patient outcome. However, to our knowledge, there have been no published systematic reports about the correlation between copy number alterations and DNA methylation alterations in clear cell RCCs. We therefore examined the correlation between array-CGH data and DNA methylation status. DNA methylation on CpG islands was accumulated in clear cell RCCs belonging to cluster B. The histogram of the number of methylated CpG islands (Fig. 3) revealed a tendency for biphasic accumulation only in cluster B, whereas a monophasic peak was observed in cluster A, and DNA methylation on 6 CpG islands, which corresponded to the second peak indicating severe accumulation, was detected only in cluster B. These data suggest that genetic and epigenetic alterations were not mutually exclusive during renal carcinogenesis and that our genetic clustering of clear cell RCC based on copy number alterations is significantly correlated with DNA methylation alterations. In cluster B, showing simultaneous accumulation of both copy number and DNA methylation alterations, tumor suppressor genes may be effectively silenced by a combination of chromosome loss and DNA hypermethylation on CpG islands. Genes that participate in chromosome integrity may be silenced by regional DNA hypermethylation. Such mechanisms linking genetic and epigenetic alterations should be further examined, especially in clear cell RCCs belonging to cluster B.

Finally, we examined the clinicopathologic significance and prognostic effect of our genetic clustering of clear cell RCC. Clear cell RCCs showing higher histologic grades, vascular involvement, renal vein tumor thrombi, and higher pathologic TNM stages were accumulated in cluster B. Because there was no significant difference of tumor size between clusters A and B, it was not feasible for clear cell RCCs in cluster B to have simply been in existence longer than those in cluster A: genetic and epigenetic alterations in cluster B may not simply accumulate at the same constant speed as those in cluster A. Although it has been reported that careful dissection of large clear cell RCCs almost always reveals renal vein tumor thrombi (28), because there was no significant difference in tumor size between clusters A and B, accumulation of clear cell RCCs with renal vein tumor thrombi in cluster B was not attributable to the difference in tumor size. Recurrence-free and overall survival rates were significantly lower in cluster B patients than in cluster A patients. Previous studies using FISH (29) and microsatellite loss of heterozygosity (30) analyses and focusing on chromosome 9p have revealed that loss of 9p is associated with tumor recurrence and poorer outcome. A previous study that used the inter-Alu long PCR method focusing on chromosome 14q showed that loss of 14q was associated with poorer outcome (31). These separately examined variables are consistent with our results because loss of chromosomes 9 and 14q was frequent in cluster B. Our data suggest that our genetic clustering of clear cell RCC based on copy number alterations is significantly correlated with tumor aggressiveness and is a significant prognostic indicator. Accumulated genetic and epigenetic alterations may play a significant role in the malignant potential of clear cell RCCs belonging to cluster B. Multivariate analysis revealed that genetic clustering was a predictor of recurrence-free survival and was independent of histologic grade and pathologic TNM stage. Gain at some BAC clones on chromosomes 3q and 8q was observed only in clear cell RCCs showing recurrence. Genes that participate in malignant progression and are activated by chromosome gain may be identified in such BAC regions.

Although low-stage clear cell RCCs are generally curable by nephrectomy, occasional relapse or metastasis can lead to death in middle-aged adults belonging to the working population. For effective follow-up and adjuvant therapy, an indicator for prognostication of clear cell RCCs using nephrectomy specimens should be established. Our genetic clustering of clear cell RCC seems to be a significant prognostic factor. In addition, as shown in the present study, a sufficient quantity of good quality DNA can be obtained from each nephrectomy specimen. Therefore, array-based analysis is applicable to routine laboratory examinations for prognostication after nephrectomy. We are currently attempting to make a mini-array harboring a set of BAC clones that can effectively discriminate cluster B from cluster A for prognostication of clear cell RCCs. The reliability of such prognostication will need to be validated in a prospective study.

In summary, indispensable copy number aberrations, loss of chromosome 3p and gain of chromosomes 5q and 7, may solely promote the development of RCCs belonging to cluster A, which show a favorable outcome. When loss of chromosomes 1p, 4, 9, 13q, and 14q and DNA methylation alterations are added to loss of 3p and gain of 5q and 7, more malignant RCCs belonging to cluster B may develop. The order of occurrence of each genetic and epigenetic event during renal carcinogenesis has not yet been defined. Our genetic clustering associated with regional DNA hypermethylation may also become a prognostic indicator for clear cell RCCs.

No potential conflicts of interest were disclosed.