Differential Gene Expression between Normal and Tumor-derived Ovarian Epithelial Cells¹

Ovarian cancer is the fourth leading cause of cancer-related deaths in the United States (1). Each year, ∼23,000 women will be diagnosed with this disease and close to 14,000 will die from it. A significant factor contributing to the high mortality rate of ovarian cancer is the relatively asymptomatic progression of this disease. As a consequence, most patients are diagnosed with advanced(stage III/IV) disease when widespread i.p. metastases are already present (2). Greater than 90% of ovarian malignancies arise from the transformation of the ovarian surface epithelium, a single continuous layer of epithelial cells surrounding the ovary.

There are estimated to be 20,000 genes expressed in a typical cell, and∼1% of those is differentially expressed in cancerous versus normal cells (3). A limited number of genes has been found to have elevated or depressed levels of expression in ovarian cancers when compared with normal tissue(4, 5, 6, 7, 8). As in other neoplasm, it is generally accepted that both activation of oncogenes and inactivation of tumor suppressor genes are involved in the etiology of ovarian carcinomas. Brca1, Brca2,and p53 mutations are associated with the development and progression of ovarian cancer (9, 10, 11). Because these proteins are involved in the maintenance of genomic integrity, loss of their functions is thought to result in the accumulation of genetic mutations, leading to extensive changes in gene expression(12, 13, 14, 15). Comparison between gene expression profiles of normal ovarian epithelial cells and ovarian tumors could identify candidate genes for biological markers of cellular transformation,possibly leading to earlier detection and new therapy.

Because ovarian epithelial cells represent a small proportion of the total cells found in the normal ovary, it is difficult to obtain primary material that is free of contaminating ovarian stromal cells in large enough quantities to conduct comparative gene expression studies. However, ovarian epithelial cells can be isolated and expanded in culture for ∼15 passages (16, 17). The ability to culture human ovarian epithelial cells from both normal ovaries and ovarian carcinomas provides an opportunity to study differential gene expression between relatively pure populations of normal versus tumor-derived epithelial cells. This type of comparison minimizes gene expression differences that reflect the presence of nonepithelial cells, such as stromal or germ cells of normal ovaries and host-derived immune cells in ovarian tumors.

We used cDNA-RDA³to identify a set of genes that are differentially expressed between primary cultures of normal and tumor-derived ovarian epithelial cells. cDNA-RDA was subsequently combined with cDNA filter array hybridization to identify a subset of genes that are aberrantly expressed in a large number of malignant ovarian epithelial cells. Direct gene expression profiles were obtained by Northern blot analysis on four differentially expressed genes to confirm the cDNA array analysis.

The conditions for growing normal HOSE cells in vitro were modifications of the method described by Auersperg et al.(18). Briefly, normal ovarian tissue was obtained from the operating room from consenting donors and placed in 199:MCDB 105 (1:1)medium (Sigma Chemical Co., St. Louis, MO) containing 10% FCS, 200 u/ml penicillin, and 200 μg/ml streptomycin. Epithelial cells were microdissected or scraped from the ovarian surface. The epithelial explants were placed in culture medium and allowed to attach and proliferate. Once the epithelial cells reached confluency, the explants were removed and the cells were subcultured. CSOC cultures were established from ovarian carcinomas in a similar manner. Fresh tumor tissue was finely minced with scissors and allowed to attach to culture dishes in McCoy’s 5A medium (Life Technologies, Inc., Grand Island,NY) supplemented with 10% FCS and penicillin/streptomycin. The epithelial nature of HOSE and CSOC cultures was verified by immunohistochemical analysis with antibodies against cytokeratin(AE1/AE3; Roche, Indianapolis, IN), vimentin (clone v9; Roche), and factor VIII (Factor VIIIC; Calbiochem), as described previously(19). p53 status of cultures was determined by immunostaining with Ab-6 antibody (Roche).

TfxH, an SV40 Large T antigen-immortalized HOSE cell line, was grown in 199:MCDB 105 (1:1) medium containing 10% FCS (17). Ovarian carcinoma-derived cell lines, Caov-3 and Sk-OV-3 (American Type Culture Collection, Manassas, VA), were grown in the presence of 10% FCS in DMEM or McCoy’s 5A medium, respectively.

cDNA-RDA was used to compare gene expression between two HOSE and two CSOC cultures (20). Total RNA was prepared from each culture, using RNA STAT-60 reagent (Tel-Test, Inc., Friendswood, TX). mRNA was purified from 120 μg of total RNA using Oligotex mRNA columns (Qiagen, Inc., Chatsworth, CA) and used for cDNA synthesis. cDNA from HOSE and CSOC samples was digested with Dpn II and PCR-amplified following the addition of RDA adapters. Subtractive hybridization was performed in 2.5 μl of 3× EEP buffer [10 mm EPPS[N-(2-hydroxyethyl)piperazine-N′-3-propanesulfonic acid], 1 mmEDTA, 1 M NaCl, and 10% polyethylene glycol] for 21 h at 67°C. Two rounds of subtraction were performed in both directions, using tester to driver ratios of 1:100 and 1:500 for the first and second rounds, respectively. Finally, the cDNA-RDA products were digested with Dpn II and cloned into the BamHI site of pBluescript KS⁺ (Stratagene, La Jolla, CA).

Individual bacterial transformants were isolated into 96-well microtiter plates containing 100 μl of LB-ampicillin (100 μg/ml)and incubated overnight at 37°C. Using a 96-well replicating tool(V&P Scientific, San Diego, CA), bacterial culture was transferred into 96-well thermowell plates (Costar, Cambridge, MA) to inoculate 50-μl PCR reactions containing 250 μm dNTPs, 1 ng/μl SK(GGCCGCTCTAGAACTAGTGGATC) and KS (TGATATCGAATTCCTGCAGCCCG) primer each,and 0.03 u/μl Taq polymerase (Qiagen, Inc.). Amplification was carried out for 35 cycles (94°C for 45 s, 68°C for 45 s, 72°C for 1 min), with a final 10-min extension at 72°C. The average size of the PCR-amplified fragments was ∼500 bp.

The PCR-amplified inserts were spotted onto 7.8 × 12.3-cm Hybond N+ membrane (Amersham Corp., Arlington Heights, IL),using a 96-well replicating tool. Redundant clones were eliminated by back hybridization, as described previously (21). The DNA sequences of nonredundant clones were determined using a commercially available sequencing kit (ABI PRISM Dye Terminator Cycle Sequencing Ready Reaction Kit; Perkin-Elmer Corp., Foster City, CA). The nucleotide sequences were analyzed using the BLASTN program and the GenBank database.

Nonredundant cDNA-RDA fragments were organized into 96-well plates and subsequently arrayed in duplicate on nylon membranes. cDNA array hybridization was carried out as described previously(21). To generate probes, ∼50 μg of total RNA from each of 5 HOSE and 10 CSOC cultures were poly(A) selected and used to synthesize cDNA. One-fifth of the total cDNA obtained was labeled with[³²P]-α-dCTP, using a random primer labeling kit (Prime-it II; Stratagene). The hybridization signal was quantified on a phosphorimager (Molecular Dynamics), using the ImageQuaNT software package.

The signal intensity per pixel within each square of the grid was calculated and corrected for the background. Each filter was also spotted with actin, tubulin, glyceraldehyde-3-phosphate dehyrdrogenase and EF-Tu cDNAs. These genes function in housekeeping activities in the cell and display little variability in expression between normal and transformed cells (22). EF-Tu, which showed the least variability in mRNA expression between all of the samples, was used as the internal standard in this study. The signal of each background-corrected spot was determined relative to average background-corrected signal for EF-Tu within a given filter, and averages of HOSE and CSOC standardized signal were compared for each of the 864 dots on the filter. In this study, only those with differences>2.5-fold were considered to be differentially expressed.

Total RNA (5 μg) was separated on 0.9% agarose formaldehyde gels and transferred to Hybond-N+ nylon membranes. Filters were hybridized with ³²P-cDNA probes corresponding to the differentially expressed genes, as described previously(21). The filters were then stripped and rehybridized with ³²P-labeled EF-Tu cDNA to control for mRNA loading.

Ovarian epithelial cells tend to assume atypical fibroblast-like morphology and a dual epithelio-mesenchymal phenotype that is characterized by the expression of both keratin (an epithelial marker)and vimentin (a mesenchymal marker) in culture (16). Cultures containing endothelial cells, identified by Factor VIII staining, and ovarian stromal cells, characterized by the lack of cytokeratin staining and low levels of vimentin staining, were excluded in this study. The staining patterns for HOSE and CSOC cells used in the cDNA-RDA and cDNA filter array hybridization are presented in Table 1.

There are no established molecular criteria to distinguish HOSE from CSOC cells. Therefore, we relied on cytogenetic studies to confirm that HOSE and CSOC cultures represent normal and malignant ovarian epithelial cells, respectively. Both HOSE cultures used in the cDNA-RDA contained normal female diploid cells; however, the cells from both CSOC cultures displayed an abnormal karyotype (Table 1). Sixty percent of CSOC 817 cells exhibited a loss of the X chromosome, which is commonly associated with ovarian carcinomas (23, 24, 25). Rearrangements in chromosomes 12 and 18 were also seen in CSOC 817 cells. Ten percent of CSOC 826 cells contained a small chromosome that may be isochromosome 21. A gain in small regions of chromosome 2 was also observed in CSOC 826 cells, which is consistent with a previous study showing frequent gains in discrete regions of chromosome 2 in some ovarian cancer cells (26). These abnormal karyotype profiles of CSOC 817 and CSOC 826 are consistent with the malignant origins of these cultures.

A total of 255 nonredundant genes were identified after two rounds of subtractive hybridization. Of these, 95 were preferentially expressed in CSOC cells and 160 in HOSE cells. We then used cDNA array hybridization to identify a subset of genes that were differentially expressed in a larger cohort of HOSE and CSOC cultures. Duplicate filters were spotted with genes identified by cDNA-RDA and hybridized with ³²P-labeled cDNA probes derived from additional 5 HOSE and 10 CSOC cultures. Forty-four HOSE-specific and 16 CSOC-specific genes displayed a >2.5-fold difference in expression(Tables 2 and 3). An example of cDNA filter arrays probed with ³²P-labeled cDNAs from HOSE 224 or CSOC 869 cells are shown in Fig. 1. cDNA array hybridization easily identified Cx43 and OSF-2 as HOSE- and CSOC-specific genes, respectively. The hybridization signal intensity of Doc-1, which was cloned as a HOSE-specific gene in cDNA-RDA, on the other hand, did not vary significantly between these two HOSE and CSOC cultures. The expression levels of EF-Tu, a housekeeping gene shown to be expressed at a constant level in normal and cancer cells (22), remained unchanged in HOSE 224 and CSOC 869 cells.

To assess the predictive value of cDNA array hybridization analysis,four differentially expressed genes were chosen at random and evaluated by Northern blot analysis. OSF-2 was cloned as a CSOC-specific gene by cDNA-RDA, whereas the other three(KRT19, Cx43, and STC) had been identified as HOSE-specific genes. The expression levels of OSF-2 were greatly elevated in the 2 CSOC samples used for cDNA-RDA, as well as in 6 of the 9 CSOC samples used in array hybridization (Fig. 2). On the other hand, all three HOSE-specific genes were expressed at lower levels in CSOC cells (Fig. 2). For example, there was an overall decrease in Cx43 mRNA expression in CSOC samples compared with HOSE samples. STC and KRT19 were down-regulated in all but two CSOC samples. Interestingly, there was very little correlation in the expression of these genes in established ovarian cell lines. Specifically, OSF-2 was not overexpressed in either of the two established ovarian cancer cell lines Sk-Ov-3 and Caov-3. In addition, Cx43, which was expressed at varying levels in both HOSE and CSOC cultures, was not detected in either ovarian cancer cell lines or an SV40 Large T-antigen immortalized TfxH cell line.

The majority of genes identified by cDNA-RDA was found to be differentially expressed by <2.5-fold in cDNA array hybridization and eliminated from further consideration. To ensure that this arbitrary cutoff level did not eliminate any genes of interest, we performed Northern analysis on two cDNA clones displaying an expression level difference of <2.5-fold (Fig. 3). Northern analysis demonstrated that Doc-1, which was identified as a HOSE-specific gene in our study, as well as in a previous study (4), was expressed at higher levels in HOSE samples. However, its expression was variable, accounting for the<2.5-fold difference in cDNA array hybridization, which used average signal intensity in data analysis. GDN, a CSOC-specific gene, was up-regulated in both the CSOC 826 and CSOC 817 cells used in the initial cDNA-RDA analysis. In the additional HOSE and CSOC samples,however, GDN was expressed at a variable level in both normal and tumor-derived epithelial cells.

We next studied the expression of OSF-2 in ovarian tumor samples from which the CSOC cells used in this study were derived. Total RNA was isolated from the quick frozen tumor samples corresponding to 6 of the 12 CSOC cultures used in this study. Northern analysis revealed expression of OSF-2 in all but one tumor sample (T949; Fig. 4). The OSF-2 expression was low in the CSOC culture (C824), derived from the same tumor (see Fig. 2 A). In T1040, the level of OSF-2 expression was lower than the level observed in the corresponding C871, but was easily detectable. Overall, there was a high degree of concordance in the OSF-2 expression in tumors and the corresponding CSOC cultures, indicating that the observed overexpression of OSF-2 in CSOC cultures is not a consequence of in vitro expansion of tumor-derived epithelial cells.

Identifying genes that are differentially expressed in ovarian tumors when compared with their normal counterpart is a challenging issue because of the scarcity of the normal ovarian epithelial cells. Ovarian surface epithelia from which ovarian cancers originate represent a minute cellular component of the ovary, compared with the more abundantly present stromal cells and germ cells. We used in vitro expanded ovarian epithelial cells derived from either normal ovaries or ovarian tumors to identify 60 genes that are either up- or down-regulated in ovarian cancer cells. In this study, only the cancer cells derived from papillary serous histology tumors, which is considered to be the most common histological type of ovarian cancer,were included to limit the complexity of gene expression analysis.

Because of the variability in signal intensities in array hybridization analysis, we focused our attention on genes displaying a >2.5 fold difference in the expression level. Whereas the reliance on cDNA array hybridization with this arbitrary cutoff may have eliminated some differentially expressed genes from our analysis, it was highly effective in identifying genes that display a consistent pattern of expression differences in a large number of CSOC samples. Among the subset of genes displaying an expression level difference of >2.5-fold there was strong qualitative correlation between cDNA array hybridization data and Northern analysis. Northern analysis confirmed KRT19, Cx43, and STC as HOSE-specific genes and OSF-2 as a CSOC-specific gene. Two genes, Doc-1 and GDN, with expression level differences of <2.5-fold in cDNA array hybridization, on the other hand, displayed only a moderate difference (e.g., GDN) or a high sample-to-sample variability (e.g., Doc-1) in Northern analyses.

The gene expression patterns seen in HOSE and CSOC cells were not consistent with those seen with TfxH cells, an immortalized HOSE cell line, or Caov-3 and Sk-OV-3, two ovarian carcinoma-derived cell lines. Because these cell lines were originally derived from normal epithelial or tumor-derived epithelial cells, similar to HOSE and CSOC cells(27), our results illustrate the divergence of gene expression that can occur as the result of long-term in vitro manipulation of these cells. Although cell lines provide a relatively simple model to examine gene expression in ovarian cancer-derived cells, our findings emphasize that the use of HOSE and CSOC cultures represents a better model system of normal and cancerous ovarian tissues in comparative gene expression analysis.

The use of cultured ovarian epithelial cells is not without concerns. Ovarian tumors frequently are histologically inhomogeneous(2). There are reports in the literature of loss of tumor markers associated with continuous tissue culture of some xenograft lines (28, 29, 30, 31). Despite the primary tumors containing areas of differentiated cells, in each instance, a selection of a poorly differentiated subpopulation had occurred during the propagation of these lines. Although we have used primary cultures to avoid a selection bias inherent in any long-term cultures, we cannot formally exclude the possibility that our cultured CSOC cells represent a subpopulation of cancer cells present in the original tumor, or in vitro expansion conditions may have modified gene expression. In the case of OSF-2, there was a high degree of concordance in its expression in tumors and their corresponding CSOC cells, indicating that despite the in vitro expansion process, the expression of this particular gene is preserved in the cultured cells.

The extent of gene expression differences between HOSE and CSOC cultures is not known. Numerous genes identified in this study,including Doc-1, have previously been shown to be differentially expressed in ovarian carcinomas (4, 5, 6, 7, 8). Comparing our data to previous studies on ovarian cancer-related genes revealed only a minor degree of overlap, indicating that the extent of gene expression differences far exceeds the number of genes identified in this or other previous studies. Genes associated with protein synthesis or mitochondrial metabolism are frequently identified in differential gene expression analysis of tumor tissues when compared with the normal tissue, and have been attributed to differences in proliferative and metabolic rates (8, 32). The absence of such genes in our analysis probably reflects the use of HOSE and CSOC cells, which are morphologically indistinguishable and display similar growth characteristics. This finding further emphasizes the use of HOSE cells as well-matched controls in comparative gene analysis to identify aberrant gene expression in CSOC cells.

Several of the genes identified in this study are noteworthy. OSF-2 was originally reported as a transforming growth factor β-inducible protein secreted in the extracellular matrix of the periosteum(33). The potential significance of OSF-2 in ovarian cancer is illustrated by the high degree of OSF-2 overexpression observed in both ovarian tumors and cultured CSOC cells. OSF-2 overexpression in CSOC cells may be a consequence of inappropriate transforming growth factor β signaling that can be seen in some CSOC cells (19, 34). Although the function of OSF-2 is not known, OSF-2 or a related protein, βig-H3, is believed to function as a matrix protein that promotes cell attachment (33, 35). One possibility is that OSF-2 expression may facilitate i.p. spread of cancer cells, which leads to a significant morbidity and mortality in women with ovarian cancer. STC is expressed at high levels in organs derived from the müllerian duct (36), and,therefore, the loss of STC expression in CSOC cells may reflect cellular de-differentiation. Cx43 and Cx40, which encode gap junction proteins, were cloned as HOSE-specific genes. Decreased gap junction communication and loss of Cx43 expression have been reported in ovarian cancer (37) and may be related to the loss of epithelial cell features in cancer cells, as well as decreased cellular communication that is seen in many types of cancers.

In conclusion, the availability of primary epithelial cultures from both normal and malignant ovaries has enabled the identification of 60 differentially expressed genes, using a combination of subtractive hybridization and cDNA array expression studies. Despite the lack of observable phenotypic or growth differences between HOSE and CSOC cells, reciprocal expression of OSF-2 with STC, KRT19, and Cx43 was seen, reflecting differences in these cells at the molecular level. A large number of genes identified in this study encode transmembrane or secreted proteins (see Tables 2 and 3) that may be present in serum and could be used as marker for ovarian cancer. In addition, the genes identified in this study may provide clues to the cellular changes responsible for metastatic progression of ovarian cancer.

We thank H. Tran for technical assistance with cell cultures;Drs. A. Carlson, C. Denny, T. Lane, and J. McAdera for critical review of the manuscript; and W. Aft for assistance with preparation of the manuscript.