DNA Copy Numbers Profiles in Affinity-Purified Ovarian Clear Cell Carcinoma

Ovarian cancer is a heterogeneous group of neoplastic diseases that are thought to arise from the epithelial cells of the fallopian tube, ovarian surface inclusion cysts, or endometriosis (1). They are classified into serous, mucinous, endometrioid, and clear cell types corresponding to the different types of epithelia in the organs of the female reproductive tract (2–4). Correlated with their respective clinical behaviors, each of these histologic types is further divided into three groups: benign (cystadenoma), intermediate (borderline tumor), and malignant (carcinoma; ref. 2). Clear cell carcinoma of the ovary (CCC) is characterized by distinct morphologic features and clinical behavior. Although CCC represents <10% of ovarian cancers in the United States, it appears to occur more frequently in Asia. Previous clinicopathologic studies have indicated that CCC develops in a stepwise fashion from endometriosis, atypical endometriosis to CCC (5–7). In general, most CCCs are indolent, typically presenting at early stage when surgical intervention is effective in eradicating the disease. However, in advanced stage, CCC is usually resistant to conventional chemotherapy and highly aggressive (8, 9). Little is known about the pathogenesis of CCC, thus a better understanding of which holds the promise of development of new and more effective therapies is needed (9–15). At present one of the most significant molecular genetic changes described in CCC is the frequent somatic activating mutation of PIK3CA (16), which were detected in ∼48% of affinity-purified fresh tumors and cell lines. On the other hand, the frequency of PIK3CA mutations is relatively low in high-grade serous carcinoma and the other major types of ovarian cancer. In contrast, CCC has a significantly lower frequency of TP53 mutations than high-grade serous carcinomas, which harbor this mutation in over 50% of cases.

Besides these observations in a few specific genes, the genomic landscape of CCC is poorly understood. Alterations in DNA copy number is a cardinal feature of carcinogenesis and identification of amplified or deleted genes would be important in elucidating the molecular pathogenesis of CCC. In this study, we applied a high-resolution single nucleotide polymorphism (SNP) array to analyze genome-wide DNA copy number changes in CCC. To enhance the sensitivity and specificity of our analysis, we used affinity-purified tumor cells prepared from fresh surgical specimens in addition to established cell lines. Our data revealed a distinct pattern of DNA copy number changes in CCC compared with high-grade and low-grade ovarian serous carcinoma. More importantly, we identified the ZNF217 locus as the most commonly amplified region in CCC.

Tissue samples from ovarian CCCs were freshly collected from the Department of Pathology at the Johns Hopkins Hospital and the National Taiwan University Hospital. Tumor cells were affinity purified by anti-BerEP4–conjugated magnetic beads as previously described (17). The acquisition of the anonymous tissue specimens for this study was approved by the Institutional Review Boards of participant institutions. Their clinical disease stages and mutational profiles of the clinical samples were shown in Table 1. In addition, genomic DNA of the same batch of DNA isolation from 10 CCC cell lines used in our previous study were included as controls (16). The CCC cell lines were maintained in the University of California at Los Angles in Dr. R. Glas and Dr. D. Slamon's laboratory. The cell lines ES2 and TOV21G were obtained from the American Type Culture Collection. The cell lines OVISE, OVMANA, OVTOKO, and RMG1 were obtained from the Japanese Health Science Research Resources Bank. OV207 was a kind gift from Dr. V. Shridhar, Mayo Clinic, Rochester, MN. OVCA429 was a kind gift from Dr. B. Karlan, Cedars Sinai, Los Angeles, California and this line was initially described as a serous tumor, but later studies showed phenotypes close to CCC (18–20). JHOC5 was a kind gift of Dr. Kentaro Nakayama, Shimane University, Shimane, Japan. Individuality of each cell line was checked by mitochondrial DNA sequencing.

SNPs were genotyped using 250K StyI arrays (Affymetrix) in the Microarray Core Facility at the Dana-Farber Cancer Institute. A detailed protocol is available at the Core center Web page.⁸

dChip software (version 2006) was used to analyze the SNP array data as previously described (21). Data were normalized to a baseline array with median signal intensity at the probe intensity level using the invariant set normalization method. A model-based (PM/MM) method was used to obtain the signal value for each SNP in each array. Signal values for each SNP were compared with the average intensities from 12 normal samples. To infer the DNA copy number from the raw signal data, we used the Hidden Markov Model based on the assumption of diploidy for normal samples. Mapping information of SNP locations and cytogenetic band were based on the curation of Affymetrix and ensemble National Center for Biotechnology Information Build 36.1 (March 2006, hg18). Amplification was defined as a cutoff value of >3.0 copies in more than six consecutive SNPs. A cutoff of <0.69 copy was used to define homozygous deletion. If there were intermittent, nonconsecutive probes with inferred copy numbers that deviated less than ±0.1 copies from the cutoff criteria, this probe was considered within the cutoff range. The boundary of amplification and deletion was visually inspected and determined if difference between franking probes is >50%.

To facilitate the quantification of subchromosomal copy number alterations, we applied the Circular Binary Segmentation algorithm to determine the levels of DNA copy number changes and the outcome is represented as chromosome instability index (CIN). The computation for CIN index was described in one of our previous studies (22). Briefly, the chromosome-specific CIN index is defined as the sum of amplitudes of all gain/loss segments divided by the total number of SNPs in the chromosome, and the genome-wide CIN index was defined as log(C1+1) + … + log(Ci + 1) + … + log(C23 + 1), in which Ci is the CIN index of chromosome i. For a gain segment, the amplitude is the average intensity of SNP signals within the segment. For a loss segment, to match the effect of losses to the same scale of gains, the amplitude is calculated by 2.5 + (A − 2.5)(1.5 − a)/1.5, in which “a” is the average intensity of SNP signals within the loss segment and “A” is the maximum gain amplitude across all cases in the same chromosome.

The gDNA copy number of the candidate genes was validated by quantitative real-time PCR using an iCycler (Bio-Rad) with SYBR green dye (Molecular Probes). Averages in the threshold cycle number of triplicate measurements were obtained. The results were expressed as the difference between the threshold cycle number of the gene of interest and the threshold cycle number of a Line-1 gene for which gDNA copy number is relatively constant among tumor samples (23). cDNA copy number was measured using the same procedure except the relative copy number of each candidate gene was normalized to the copy number of APP, a gene that mRNA expression is constant among samples (24).The primer sequences are listed in Supplementary Tables S1 and S2.

Synthetic small interfering RNA (siRNA) targeting ZNF217 and Luciferase were purchased from Dharmacon and they were used to transfect CCC cell lines using Lipofectamine 2000 (Invitrogen). Two different siRNAs were found to efficiently knock down ZNF217, and their sequences are GAACAGAACCUCCCAAGGA and GAGGAUGCCUUGUCAAUGA. siRNA for Luciferase (UAAGGCUAUGAAGAGAUAC) was used as a control. Following transfection, cells were seeded into 96-well plates and the relative cell number was measured by the SYBR Green reagent (Invitrogen) using a fluorescent microplate reader. Data were expressed as mean ± 1 SD from five replicates in each experimental group. Early apoptosis was detected by measuring the caspase-3/7 activity using a kit purchased from Promega. The uptake of bromodeoxyuridine (BrdUrd) was used as an indicator for cell proliferation. In brief, 24 to 96 h after transfection, cells were incubated with 10 μmol/L BrdUrd for 1 h, fixed with methanol, and immunostained with an anti-BrdUrd antibody following the protocol provided in a kit (RPN202, GE Healthcare). The percentage of BrdUrd-positive cells was determined by counting ∼300 cells from each well. The data were expressed as mean ± 1 SD from triplicate wells.

The differences in parameters were determined by Mann-Whitney nonparametric test for data presented in Fig. 1B, and P value was determined by two-tailed analysis. Paired t test was done to determine the significance of difference in the CIN index between matched normal and tumor samples. Fisher's exact test was done to determine the correlation of ZNF217 amplification and stage. Unpaired t test was done for the data presented in Fig. 5. The significant level (α) was set at 0.05 (*, P < 0.05; **, P < 0.01; and ***, P < 0.001).

The level of CIN in CCC was assessed according to the degree and extent of changes in DNA copy number in all chromosomes. The CIN index was used to quantify the CIN level and was developed based on the sum of amplitude in each gain or loss region, normalized to the total SNP probe numbers for each chromosome (22). The CIN index was first compared between tumor stromal fibroblasts and the epithelium in the carcinomas isolated from seven CCCs. The results showed that the CIN index of the stromal cells consistently approached zero in all chromosomes, whereas the carcinoma showed an elevated CIN index. Paired t test indicated the CIN index was significantly elevated in carcinomatous component compared with the matched tumor-associated stromal fibroblasts (P < 0.01) as internal controls. Next, CIN indices of CCCs were compared with those from affinity-purified ovarian serous borderline tumors, low-grade serous carcinomas, and high-grade serous carcinomas previously reported (22). The results show that CIN indices in CCCs were similar to those in low-grade serous carcinomas but were higher than those in serous borderline tumors and lower than those in high-grade serous carcinomas. Figure 1 lists the genome-wide CIN indices in all CCCs analyzed together with those in different types of ovarian serous tumors. Specifically, Fig. 1A includes the CIN indices at each chromosome and Fig. 1B includes the genome-wide CIN indices (the combined CIN indices from all chromosomes) in each specimen. These findings indicate that although CCC as a group has an elevated level of genomic instability compared with associated nonneoplastic stromal tissue, its CIN level is relatively modest compared with that found in high-grade serous carcinoma.

Although the genomic landscape of DNA copy number in CCC is relatively “flat” when compared with high-grade serous carcinoma, there are several discrete DNA copy number gains and losses. Among them, the most remarkable amplification found in the affinity-purified CCC samples is the DNA copy number gain at the chr20q13.2 locus (Fig. 2A). Five of the 12 cases analyzed are found to harbor this genomic amplification with peak amplitude ranges from 3.8 to 5.1 copies. To delineate the minimal amplified region, we aligned the amplified locus in three cases with discrete amplification and the minimal amplified region is determined to be between chr20:51,428,300 and 52,180,900 bp (Fig. 2B). Only three genes (TSHZ2, ZNF217, and BCAS1) are located within this minimal amplified region, including a previous reported potential oncogene ZNF217, which is amplified in breast and gastric carcinomas (25, 26). To determine the amplification frequency at this locus, we performed quantitative PCR in an additional 21 CCCs and found 7 of them with ZNF217 DNA copy number gain (Table 1). Quantitative PCR was also done to determine the amplification frequency of the other two franking genes in 20 of the 21 samples. The results showed that TSHZ2 is amplified in 5 of 20 cases, whereas BCAS1 is amplified in 2 of 20 cases. In sum, by combining data from SNP array and qPCR analyses, ZNF217 showed the highest amplification frequency, which occurred in 12 (36%) of 33 CCCs.

To determine if the amplification of chr20q13.2 leads to transcript upregulation, we performed quantitative real-time PCR to analyze gDNA and cDNA copy number in 20 of the 33 CCC samples for all three genes within the minimal amplicon. The 20 cases were selected because both their gDNA and cDNA samples from the same cases were available for analysis. The results showed that there was a positive correlation between genomic DNA copy number and ZNF217 transcript number with an R value of 0.61 (P < 0.01, Spearman test; Table 2), whereas there was no significant correlation of either TSHZ2 or BCAS1.

In addition to the amplification in the chr20q13.2 locus, low levels of chromosomal gain encompassing large chromosomal regions were observed in chr8q (Fig. 3) as well as chr17q (Supplementary Fig. S1; Fig. 4). However, these regions are relatively large and contain numerous genes that preclude the identification of a candidate driver oncogene(s) at this moment.

One of the advantages of high-density SNP array is its exquisitely sensitive resolution that allows for the detection of discrete subchromosomal deletion. By using affinity-purified tumor cells, we were able to detect discrete homozygous deletions that have not been previously described. These include deletions at chr2q12.1, chr8p21.3, and chr9p21.3 (Fig. 3), each of which was identified in one of the 12 fresh samples analyzed. One additional case with copy neutral loss of heterozygosity (LOH) was identified in the chr2q12.1 and chr9p21.3 locus, respectively, but not in the chr8p21.3 locus. The deletion at chr2q12.1 encompasses 104,649,000 to 109,476,000 bp and harbors at least 25 known genes. There is no known tumor suppressor gene reported thus far in this region. Deletion at chr8p21.3 spans from 19,651,900 to 21,642,400 bp and includes leucine zipper, putative tumor suppressor 1 (LZTS1), a previously reported candidate tumor suppressor gene. The frequency of homozygous deletion at LZTS locus is low because it is only identified in 1 of 12 cases by SNP array analysis and in 1 of 21 additional cases analyzed by qPCR. Deletion at chr9p21.3 spans from 21,596,500 to 22,009,700 bp at chromosome 9 and includes a well-known tumor suppressor gene, CDKN2A/2B. Based on qPCR analysis in 21 additional cases of affinity-purified CCCs, we detected homozygous deletions of this locus in two of the samples, making a total homozygous deletion frequency of 9% (3 of 33 cases). Mutational analysis of the CDKN2A/2B gene failed to show somatic mutations in any of the 12 CCCs examined.

There were no subchromosomal deletions observed in chr1; however, we found two cases harboring copy neutral LOH at chr1p36.22 (Fig. 3). Our previous report has shown frequent hemizygous deletions and LOH in this region in low-grade ovarian serous tumors (16).

In addition to the 12 CCC tissue samples, we performed the same SNP array analysis using genomic DNA isolated from 10 CCC cell lines that were used in our previous study for mutational analysis of PIK3CA (16). Although our previous report showed a similar mutation frequency of PIK3CA between affinity-purified clinical CCC samples and cell lines, surprisingly, the DNA copy number alterations and LOH in the cell lines were very different from that detected in the CCC clinical samples (Supplementary Fig. S2; Fig. 4). For example, ZNF217 amplification was only observed in 1 of 10 CCC cell lines in contrast to 5 of 12 clinical CCC samples (Fig. 2A). In clinical samples, the CIN indices at chromosomes 6, 11, 13, and 18 approached zero, whereas those in CCC cell lines were significantly elevated (P < 0.05, Mann-Whitney test; Fig. 4B). Furthermore, the deletions detected in clinical samples were not the same as those detected in the cell lines except for the CDKN2A/B locus. In contrast, many amplifications, deletions, and LOHs were found only in cell lines but not in clinical specimens (Supplementary Fig. S2; Fig. 4). A list of the amplifications and deletions detected in CCC cell lines appears in Supplementary Tables S3 and S4.

Given the frequent amplification at the chr20q13.2 locus and the overexpression of ZNF217 in CCCs, we decided to examine if ZNF217 is required for the oncogenesis in CCCs. OVMANA, a CCC cell line that harbored ZNF217 amplification and overexpression, and OV207, a CCC line with ZNF217 overexpression were selected for functional studies (Fig. 5). The relative ZNF217 mRNA levels decrease to 45% in OVMANA and 35% in OV207 compared with control (luciferase) siRNA after treatment of ZNF217 siRNA (Supplementary Fig. S3), and significantly suppressed cell growth in CCC cell lines (Fig. 5A and B). We then used OV207 as a representative cell line to determine if the reduced cell number is due to a decrease in cellular proliferation and/or an increase in apoptosis. We found that ZNF217 siRNA treatment resulted in a significantly higher caspase-3/7 activity in CCC cells than the control group treated with luciferase siRNA (t test, P < 0.01). Moreover, we also observed that the cells with BrdUrd incorporation significantly decreased by ZNF217 siRNA compared with luciferase siRNA (P < 0.01; Fig. 5C and D).

This is the first study analyzing the ovarian CCC genome using high-density SNP arrays on affinity-purified tumor samples. Our approach offers the assessment of genome-wide DNA copy number alterations in CCCs with very high resolution because affinity purification minimizes contamination by stromal and inflammatory cells that may mask small discrete amplifications and homozygous deletions. Our data show that the level of CIN in CCC is significantly lower than that seen in high-grade serous carcinomas. Moreover, we identified discrete regions with either DNA copy number gain or loss with a resolution level that was never achieved. Lastly, we found evidence that CCC cell lines are distinctly different from fresh CCC tissues in their DNA copy number landscapes, attesting to the fact that many established cell lines have been selected through passages in vitro and may not represent their in vivo counterparts. We believe the information provided by this study has many important biological and clinical implications.

Among the amplified regions detected, the only one with a discrete amplicon with >2-fold amplification is located at chr20q13.2 containing three genes: TSHZ2, BCAS1, and ZNF217. The amplification occurs in up to 36% of CCCs, suggestive of an important role of this amplicon in CCC development. Teashirt zinc finger homeobox 2 (TSHZ2) encodes a zinc finger protein, and its homologue in Drosophila is involved in the specification of the embryonic trunk (27). It was also known as the ovarian cancer–related protein, indicating its abundant expression in ovarian cancer. Breast carcinoma–amplified sequence 1 (BCAS1) was previously identified as a gene within the 20q13 amplicon (28). However, this gene is located at the border of the amplicon and its expression was not detected in the breast cancer cell line, MCF7, in which this region is highly amplified (28). Zinc finger protein 217 (ZNF217) encodes a nuclear protein with Krüppel-like zinc finger domains and serves as a common protein in forming transcriptional repressor complexes with containing HDAC2 and LSD1 (29, 30). Previous reports have shown frequent amplification and overexpression of ZNF217 in other carcinomas including breast, colorectal, and prostate. The amplification frequency was found to range from 10% to 50% in these tumors (26, 31–34). In breast carcinomas, increased ZNF217 expression was found to be associated with a poor prognosis and ectopic expression of ZNF217-immortalized epithelial cells, and resulted in the loss of transforming growth factor-β responsiveness (35). In this study, we found a significant correlation of genomic DNA and mRNA copy numbers of ZNF217 in CCC samples. Furthermore, like in other cancer types (36, 37), we showed the essential role of ZNF217 in maintaining cellular proliferation and survival, an observation further supporting a role for ZNF217 in CCC tumorigenesis.

Among different types of ovarian carcinomas, discrete amplification of the ZNF217 locus seems specific to CCCs, as our previous SNP analysis in serous ovarian tumors has identified DNA copy number gain of the entire or large region of chr20q arm rather than a relative minute amplicon (22). Although we have only analyzed 33 CCCs thus far, there seems to be no statistically significant difference in the rate of ZNF217 amplification between early stage (stage I) and advanced stage (III and IV; P > 0.05).

In this study, we detected homozygous deletions at three different loci in CCCs. The first one is at 9p21.3 that harbors the CDKN2A/2B gene. The homozygous deletion frequency at the CDKN2A/2B locus is ∼9%. Deletion at the chr9p region encompassing CDKN2A/2B has been detected by CGH in 17% to 55% of CCCs (38–40). Our results confirm loss of this locus containing CDKN2A/2B in a significant number of CCC cases and implicate its role in the pathogenesis of CCC. Two other novel discrete homozygous deletions at chr2q12.1 and chr8p21.3 were also identified by us. The deleted locus at chr2q12.1 includes >25 genes in which the candidate tumor suppressor(s) is currently not known and will require further investigation. The deleted locus at chr8p21.3 contains a candidate tumor suppressor gene leucine zipper, putative tumor suppressor-1 (LZTS1). LZTS1 encodes a transcription factor that contains the leucine zipper motif and, when reintroduced into Lzts1-null cancer cells, it suppresses cell growth (41, 42). A recent study has shown that Lzts1 knockout mice are prone to develop spontaneous tumors and, furthermore, mouse embryonic fibroblasts from these animals display increased mitotic progression, decreased Cdk1 activity, and are resistant to Taxol- and Nocodazole-induced M-phase arrest (43). These findings may explain the long-time clinical observation that CCCs are resistant to conventional chemotherapeutic drugs such as Taxol.

Another observation in this study is that CCC cell lines harbor a genomic landscape distinct from that identified in fresh CCC tissues. For example, ZNF217 amplification was found in 36% of the 33 clinical samples as opposed to only 1 (10%) of the 10 CCC cell lines analyzed. In addition, we found that CCC cell lines do not have the high-frequency DNA copy number gain at 8q that is observed in CCC tissues nor do they contain deletions seen in CCC tissues except for CDKN2A/2B. On the other hand, CCC cell lines harbor many genomic alterations that are not detected in the clinical samples, making interpretation of their biological and clinical implications difficult. It should be noted that we used the same genomic DNA preparation of these CCC clinical samples and cell lines in a previous mutational analysis study (16), which interestingly we observed similar frequencies of somatic PIK3CA mutation in CCC tissues (45%) and CCC cell lines (40%). One explanation for this disparate finding is that different selection pressures exist under in vivo and in vitro conditions for different types of mutation. It is possible that point mutation of PIK3CA and deletion of CDKN2A/2B confer a survival advantage for both tumor cells in vivo and those maintained in vitro as immortalized cell lines, whereas amplification of ZNF217, among other DNA copy number changes, is advantageous only to tumor cells in vivo. If this is true, our observations add to many previous reports that cautions should be taken in interpreting the results of DNA copy number alterations identified only in cell lines. Another observation in this study is that we did not detect increased DNA copy numbers in either PIK3CA or AKT locus in all CCCs. This finding together with our previous study showing frequent PIK3CA mutation suggest that activating PIK3CA mutation represents the major mechanism in eliciting PIK3CA/AKT pathway in CCCs.

A dualistic model of ovarian carcinogenesis based on clinicopathologic and molecular genetic studies has been proposed for ovarian carcinomas (44–46). In this model, low-grade serous, mucinous, and endometrioid carcinomas are designated as “type I” tumors. These neoplasms develop in a stepwise fashion from well-recognized precursors termed “borderline” tumors that originate from cystadenomas/adenofibromas and endometriosis. They typically are low-grade, present as large cystic tumors confined to the ovary, and have an indolent behavior. In contrast, high-grade serous carcinoma, undifferentiated carcinoma, and malignant mixed mesodermal tumors (carcinosarcomas) are designated as “type II” tumors. In contrast to type I tumors, these high-grade malignancies are not associated with benign precursor lesions at pathologic evaluation. Recent studies have suggested that type II tumors develop from high-grade intraepithelial carcinomas in the fimbriated portion of the fallopian tube. Regardless of their origins, type II tumors are invariably highly aggressive and the disease often is in advanced stages at the time of diagnosis. Compared with type I tumors at the molecular level, type II tumors have high chromosome instability and are characterized by frequent TP53 mutations. In contrast, type I tumors have moderate chromosome instability and contain frequent somatic mutations of genes participating in signal transduction pathways, including KRAS/BRAF, PTEN/PIK3CA, and CTNNB1. In contrast, mutational analysis indicates that these genes are rarely mutated in type II tumors.

CCC appears to exhibit features of both type I and type II neoplasms. Similar to type I tumors, CCCs develop from well-characterized benign precursors (e.g., endometriosis) and present as large cystic tumors confined to the ovary. However, when the disease spreads beyond the ovary, CCC is typically high-grade, high-staged, and very aggressive. CCC was tentatively assigned to the type I group awaiting the results of molecular genetic studies that could assist in its more accurate categorization. In this study, we found that CCCs have a CIN index very similar to that of low-grade serous carcinomas and much lower than that of high-grade serous carcinomas. This finding together with our previous results showing rare TP53 mutations but frequent PIK3CA-activating mutations in CCCs (16) are in stark contrast to results obtained in high-grade serous carcinomas and strongly support classifying CCC as a type I tumor.

In summary, this is the first report describing DNA copy number changes in the CCC genome using high-density SNP array analysis of purified tumor cells isolated from fresh cancerous tissue. Based on the CIN index of CCCs analyzed and their previously reported mutational profiles, CCC is much more similar to type I tumors at the molecular level than to type II tumors. Our current analysis also identified novel amplification and deletion in CCC. The frequent amplification of the ZNF217 locus and deletion of the CDKN2A/2B locus suggest that pathways involving these two genes are important in CCC tumorigenesis. We believe that the analysis of CCCs described herein is a snapshot of the genomic landscape of this disease and provide a very useful foundation for future studies aimed at elucidating the molecular pathogenesis of CCC and at the development of target-based therapeutics for this highly aggressive ovarian malignancy.

No potential conflicts of interest were disclosed.

Grant Support: U.S. Department of Defense Research Council OC0400600 (T-L. Wang), American Cancer Society RSG-08-174-01-GMC (T-L. Wang), Ovarian Cancer Research Fund (T-L. Wang), NIH RO1CA116184 (R.J. Kurman), NIH RO1CA129080 (I-M. Shih), NIH RO1 R01CA103937 (I-M. Shih), NSC97-2320-B-002-047-MY3 (T.-L. Wang), and NIH R33CA109872 (Y. Wang).

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.