Transcriptional Coactivator Drosophila Eyes Absent Homologue 2 Is Up-Regulated in Epithelial Ovarian Cancer and Promotes Tumor Growth

Epithelial ovarian cancer is the most frequent cause of gynecologic malignancy-related mortality in women. The lack of preventive strategies and early diagnostic tools as well as effective therapies to treat recurrent tumors creates a pressing need to understand its genetic basis and identify molecular targets for therapy (1, 2). To discover genes specifically up-regulated in epithelial ovarian cancer, PCR-select cDNA subtraction was done between mRNA pools from normal ovary and ovarian cancer samples. Drosophila Eyes Absent Homologue 2 (EYA2) was isolated as a promising candidate.

The transcriptional coactivator eya belongs to a set of evolutionally conserved genes termed the retinal determination gene network in Drosophila. It collectively encodes a cohort of nuclear transcriptional factors and/or cofactors whose expression is regulated by a conserved hierarchy of transcriptional regulation (3–5). The retinal determination gene network comprises twin-of eyeless (toy), eyeless (ey), eye absent (eya), sine oculis (so), and dachshund (dac), which are essential for eye fate specification in metazoans and which control essential cellular functions such as proliferation, differentiation, and cell death during organogenesis (6–8). Genetic studies in Drosophila have shown that mutations of members of the retinal determination gene network lead to failures in eye formation, whereas their ectopic expression leads to formation of additional eyes. Eya is thought to function as a transcriptional coactivator with no specific DNA-binding activity. The collective evidence gathered in Drosophila suggests that eya and so together constitute a functional transcription factor, with eya providing the activation domain and so contributing the DNA-binding moiety (3–5). Eya family members are defined by a conserved ∼275-amino-acid carboxyl-terminal motif, referred to as the eya domain (ED), which has been shown to bind two other retinal determination members, so and dac. In addition, the retinal determination gene network functions in a variety of other contexts, including gonadogenesis (9, 10), myogenesis (11), limb formation (12), neurogenesis (13), thymus (14), ear and kidney development (15) and cell cycle control (16–18), apoptosis (19), and the development of the branchio-oto-renal syndrome (20). Human homologues, EYA1-4, are strikingly similar in their eya domain as well as their NH₂ termini, with the exception of a small tyrosine-rich region, named the eya domain 2 (21–24).

In the present study, we show that EYA2 expression is up-regulated in epithelial ovarian cancer compared with normal ovary, in part owing to genomic amplification, and that high levels of EYA2 expression are significantly associated with short survival in late-stage ovarian cancer. In addition, EYA2 functions as a transcriptional coactivator in ovarian cancer cells, and its ectopic expression in xenograft tumors significantly promotes tumor growth in vivo. Our results suggest a new role of EYA2 and further elucidate the molecular alterations underlying ovarian oncogenes and tumor progression.

Patients and Specimens. The specimens used in this study were collected at the University of Pennsylvania and the University of Turin, Italy. Specimens were analyzed by quantitative real-time reverse transcription–PCR (RT-PCR; n = 53), immunostaining (n = 140) or comparative genomic hybridization (CGH) array (n = 88). Normal human tissues analyzed (n = 12) were provided by the Cooperative Human Tissue Network.

Cell Lines and Cell Culture. A total of 17 ovarian, 5 breast, and 13 colon cancer cell lines were used in this study (25), as listed in Fig. 1B. All cancer cell lines were cultured in DMEM medium (Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (Invitrogen). Human ovarian surface epithelium (HOSE) cells were isolated by our laboratory (26) or generously provided by Dr. Birrer (27). Four immortalized HOSEs [IOSE398 (28), ITOSE4, ITOSE6 (27) and HOSE6-14 (29)] were cultured in 1:1 media 199/MCDB 105 (Sigma, St. Louis, MO) supplemented with 15% fetal bovine serum.

RNA Isolation, RT-PCR, and Nested RT-PCR. Total RNA was isolated with TRIzol (Invitrogen). After treatment with RNase-free DNase (Invitrogen), RNA was further purified with RNeasy RNA isolation kit (Qiagen, Valencia, CA). Total RNA was reverse-transcribed using SuperScript First-Strand Synthesis Kit for RT-PCR (Invitrogen). Reverse-transcribed cDNA was amplified with Taq Core PCR kit (Roche, Indianapolis, IN). All PCR primers in this study are listed in Table 1.

Suppression Subtractive Hybridization and Differential Screening. Polyadenylic acid + RNA was purified from total RNA using the Oligotex Direct mRNA purification procedure (Qiagen) and double-stranded cDNA was synthesized from 2 μg of polyadenylic acid + RNA using the SMART cDNA synthesis kit (Clontech, Palo Alto, CA). Subtractive hybridization with cDNA samples from whole normal ovary and late-stage ovarian cancer samples and subsequent differential screening were done with the PCR-Select cDNA Subtraction kit and the PCR-Select differential screening kit (Clontech). The resulting clones were sequenced, and their sequences were compared with GenBank.

Quantitative Real-time RT-PCR. cDNA was quantified by real-time PCR on the ABI Prism 7900 Sequence Detection System (Applied Biosystems, Foster City, CA). PCR was done using Taqman or Sybr Green PCR Core reagents (Applied Biosystems). PCR amplification of the housekeeping gene, glyceraldehyde-3-phosphate dehydrogenase (GAPDH), was done for each sample as control for sample loading and to allow normalization among samples. A standard curve was constructed with PCR-II TOPO cloning vector (Invitrogen) containing the same inserted fragment and amplified by real-time PCR (30).

Immunohistochemistry, Double Immunofluorescence, and Image Analysis. Immunohistochemistry was done using the VECTASTAIN avidin-biotin complex kit (Vector, Burlingame, CA). Primary antibodies, goat anti-EYA2 (1:200, Santa Cruz Biotechnology, Santa Cruz, CA), mouse anti-human cytokeratin (1:500, Dako, Carpinteria, CA), and rabbit anti-human Ki-67 (1:200; Dako) were incubated for 2 hours at room temperature. Immunofluorescent double staining was done as previously described (25). The prevalence of immunoreactive EYA2 in tumors was scored semiquantitatively and blindly by two independent investigators and confirmed by a third investigator as follows: 0, all cells were negative; 1, <20% of cells in tumor islets were positive; 2, between 20% and 80% of the cells in tumor islets were positive; and 3, >80% of cells in tumor islets were positive. Ki-67 staining index was analyzed using Image-Pro Plus 4.1 software (Media Cybernetics, Silver Spring, MD).

DNA Isolation and Array CGH. Genomic DNA was isolated from late-stage (stages III and IV) frozen tumors by overnight digestion, phenol-chloroform extraction, and ethanol precipitation. BAC DNA was extracted using 96-well blocks (REAL prep kits, Qiagen). DNA was then amplified by degenerate oligonucleotide–primed PCR and was resuspended to a final concentration of 10 to 15 μg/mL. Arrays were printed on Corning CMT Ultra-Gap slides. One microgram of tumor DNA and reference DNA were labeled with Cy3 or Cy5, respectively (Amersham, Piscataway, NJ), using the BioPrime random-primed labeling kit (Invitrogen). Tumor DNA and reference DNA were labeled with the opposite dye as well to account for difference in dye incorporation and provide additional data for analysis. Labeled tumor and reference DNA were combined and precipitated with human Cot-1 DNA to reduce nonspecific binding. DNA was resuspended and applied to arrays. Arrays were hybridized for 72 hours at 37°C on a rotating platform. Images were scanned with a 428 microarray scanner (Affymetrix, Santa Clara, CA) and analyzed with GenePix software (Axon, Union City, CA). A Cy3/Cy5 (tumor/reference DNA) fluorescent intensity ratio >1.2 was considered as amplification. Analysis of array CGH data and determination of amplifications and losses were done using the software suite CGHAnalyzer (31).

Tansactivation Assay. pGal4DBD, which contains the Gal4 DNA-binding domain (1-147) but not the activation domain, binds to specific DNA sequence but cannot activate transcription (32). p(Gal4)4 E1B-Luc contains four tandem Gal4 DNA binding sequence and luciferase gene as a reporter. All pGal4 vectors were constructed by inserting corresponding PCR products into EcoRI and XhoI linearized pGal4DBD. Cells were seeded on six-well plates and grown overnight to 40% confluence before transfection. A total of 1.5 μg pGal4 and 1.5 μg p(Gal4)4E1B-Luc were cotransfected to cells using FuGene6 (Roche). pRL-TK vector (0.01 μg, Promega, Madison, WI) was used for normalizing the transfection efficiency. Luciferase analysis was done 48 hours post transfection on Luminoskan Ascent (Thermo-Labsystems, Franklin, MA) using Dual-Luciferase Reporter Assay System (Promega).

Cluster and Tree View. Hierarchical clustering analysis of real-time RT-PCR data was done using the Cluster and TreeView software programs that were originally developed for analyzing cDNA microarray data. Cluster and TreeView software were downloaded from http://rana.lbl.gov/EisenSoftware.htm. The data were analyzed according to the user's manual (33).

cDNA Cloning, Retroviral Vector Construction, and Infection. Human EYA2 cDNA was cloned from an ovarian cancer specimen and from human kidney using TOPO TA cloning kit (Invitrogen). EYA2 cDNA fragment was inserted into pMSCV (Clontech) to create pMSCV-Eya2. Retrovirus was then produced by PT67 cells (Clontech) and ovarian cancer cells were infected with retrovirus containing pMSCV alone or pMSCV-EYA2 as described. After 72 hours, infected cells were selected by neomycine (Invitrogen).

Proliferation Assay. Human ovarian cancer cells were seeded at a density of (1 to 5) × 10³ cells per well in a 96-well plate in complete culture medium. After 24 hours of culture, cell proliferation was assessed using CellTiter 96 (Promega). Cell proliferation was determined as the ratio of the absorbance of the treated cells to the density of control, untreated cells.

Cell Cycle Assay. Cell cycle was analyzed using Cellular DNA Flow Cytometric Analysis Kit (Roche). Briefly, after Rnase treatment, cell suspensions were incubated in propidium iodide and immediacy analyzed by flow cytometer.

In vivo Tumor Generation. Female BALB/c severe combined immunodeficient (SCID) mice (6 to 8 weeks old) were used for the experiments. A total of (1 to 5) × 10⁶ cells were injected s.c. into the flanks of SCID mice. Once tumors were detectable, tumor size was measured weekly. Mice were sacrificed 10 weeks after injection. Metastasis to other organs including the lungs and peritoneum were examined under a dissecting stereomicroscope.

Statistics. Statistical analysis was done using the SPSS statistics software package (Chicago, IL). All results were expressed as mean ± SD, and P < 0.05 was used for significance. Kaplan-Meier curves were used to estimate 5-year rates and were compared with the use of log rank statistics.

Expression of EYA2 Is Up-Regulated in Epithelial Ovarian Cancer. EYA2 was isolated as a promising candidate upon the first suppression subtractive hybridization screen. Quantitative real-time RT-PCR analysis validated that EYA2 mRNA expression was indeed up-regulated in ovarian cancer (n = 53) compared with normal ovary (n = 3) or HOSE (n = 7), and was significantly increased in late-stage (n = 33) versus early-stage ovarian cancer (n = 20, P = 0.029; Fig. 1A). EYA2 expression was further confirmed in established ovarian cancer cell lines: 92.3% (12 of 13) of ovarian cancer cell lines were found to express EYA2 at high levels by both RT-PCR and immunohistochemistry (Fig. 1B). In addition, high levels of EYA2 mRNA were detected in 4 of 5 breast cancer cell lines, whereas only 3 of 13 colon cancer cell lines were found to express EYA2 at low levels (Fig. 1B). By immunohistochemistry, only weak EYA2 protein expression was detected in the nucleus of either normal ovarian surface epithelial cells or ovarian stromal cells (Fig. 1C, to F). EYA2 staining was found in 93.6% of ovarian cancer samples (131 of 140, n=140) in which it localized both to the nucleus and cytoplasm (Fig. 1G and H). Double immunofluorescent staining further confirmed that EYA2-positive cells were also positive for cytokeratin, an epithelial cell marker, and were located within tumor islets (Fig. 1I, andJ). Few EYA2-positive but cytokeratin-negative cells were found in tumor stroma in less than 5% of patients (Fig. 1J).

To further screen the EYA2 expression pattern in other types of human solid cancers, a cDNA expression array data (34) was searched at Stanford Genomic Resources (http://source.stanford.edu). EYA2 mRNA was highly expressed in almost all ovarian cancer, prostate cancer, and lung adenocarcinoma samples examined, as well as in some breast cancer, urinary tract cancer, and lung squamous cancer samples. EYA2 was not found in other cancer types including colon cancer, which is consistent with the above results (Fig. 1K). In addition, the expression pattern of EYA2 was analyzed in 12 normal human organs by real-time RT-PCR. Significantly lower expression of EYA2 was found in normal ovary compared with other organs such as prostate, testis, thyroid, and thymus (Fig. 1L). Finally, a novel EYA2 expression variant lacking exon 9 was cloned from ovarian cancer samples (GenBank accession no. AY705349), which was detected in 28.6% ovarian cancer samples at low levels, but not detected in normal ovary (Fig. 1M).

EYA2 Genomic DNA Is Amplified in Ovarian Cancer. Human EYA2 gene is located in 20q13.1, which is a frequently amplified region (20q13-qter) in ovarian cancer (35, 36). Therefore, we analyzed the DNA copy number of EYA2 in late-stage ovarian cancer by high-resolution array CGH. EYA2 was found to be amplified in 14.8% of ovarian cancer specimens (n = 88, Fig. 2B). EYA2 mRNA expression was quantitated by real-time RT-PCR in 15 specimens, including 10 with a normal EYA2 copy number and 5 exhibiting EYA2 amplification, as revealed by array CGH. The average relative EYA2 mRNA expression level in tumors with EYA2 amplification (222.8 ± 83.0) was significantly higher than in tumors with normal Eya2 copy number (81.0 ± 58.7, P < 0.05; Fig. 2C).

High EYA2 Expression Is Significantly Associated with Short Survival in Ovarian Cancer. To investigate the significance of EYA2 in ovarian cancer, EYA2 expression and overall survival were analyzed in 140 ovarian cancer specimens, including 41 early-stage (I and II) and 99 late-stage (III and IV) patients. In patients with early-stage ovarian cancer, there was no significant difference in the distribution of overall survival based on EYA2 expression levels (n = 41, P = 0.6318; Fig. 3A). However, in patients with late-stage ovarian cancer, high EYA2 expression was significantly associated with short overall survival (n = 99, P = 0.0361; Fig. 3B). Patients with tumors expressing high levels of EYA2 had a median duration of overall survival of 31.7 months, whereas patients whose tumors expressed EYA2 at low level had a median duration of survival of 42.5 months.

EYA2 Serves as a Transcriptional Activator in Ovarian Cancer Cells. The collective evidence gathered in Drosophila suggests that eya and so together constitute a functional transcription factor, with eya providing the activation domain and so contributing the DNA binding moiety. We thus asked whether human EYA2 serves as a transcriptional activation factor in ovarian cancer. Transactivation assay was done to address this question. The DNA-binding domain of yeast transcription factor Gal4 (Gal4DBD) was fused in frame to the EYA2 coding region and used to analyze the ability of EYA2 to activate the expression of E1B-luciferase reporter gene Fig. 4A). As expected, the fusion product of full-length EYA2 coding region and Gal4DBD exhibited a significantly higher activity than the Gal4DBD control alone. It has been previously reported in Drosophila that a fusion protein expressing the eya NH₂ terminus but lacking the conserved eya domain displays an approximately 70-fold increase in transactivation potential relative to the full-length Eya (23). Interestingly, neither the NH₂ terminus nor the eya domain of human EYA2 exhibited any activity in human ovarian cancer cell lines. In addition, the EYA2 transcript lacking exon 9 isolated from ovarian cancer and its eya domain also exhibited no activity.

Because the transcriptional activation function of eya in Drosophila is mediated via so binding to specific DNA sequences (refs. 3 4 5; Fig. 4B), our next question was whether human SIXs, the mammalian homologue genes of so, were also expressed in ovarian cancer. To address this question, we analyzed the expression of all SIXs by real-time RT-PCR in 32 established human cancer cell lines (15 ovarian, 5 breast, and 13 colon) as well as in 54 primary human ovarian cancer specimens (19 early stage and 35 late stage). SIX1-5 were detected in most ovarian and breast cancer cell lines and ovarian cancer specimens (Fig. 4C and D). Only SIX5 was detected in colon cancer cell lines. SIX6 was not detected in any of these cancer cell lines or specimens. We next did cluster analysis to analyze the distribution of SIX expression. In cluster analysis of cultured cell lines, ovarian cancer lines clustered together with breast cancer lines, suggesting a similar pattern of expression, whereas colon cancer formed a distinct group. Interestingly, late-stage ovarian cancers clustered distinctly from early-stage cancer (Fig. 4D).

EYA2 Overexpression Promotes Ovarian Cancer Growth. To address the function of Eya2 in ovarian cancer, we cloned human EYA2 cDNA and overexpressed it in human ovarian cancer cell lines (A2780 and 2008). Tumors overexpressing EYA2 were significantly larger than control tumors (Fig. 5A). In 2008 model, the tumor volume of EYA2 transfected (2.31 ± 0.23 mm³) was significantly larger compared with control (1.52 ± 0.21 mm³, day 38, P < 0.05). In A2780 model, the tumor volume of EYA2 transfected (8.39 ± 1.76 mm³ ) was significantly larger compared with control (4.27 ± 0.79 mm³, day 28, P < 0.05). More importantly, tumors overexpressing EYA2 maintained a rapid growth rate 30 to 40 days after transplantation, whereas control tumors exhibited growth arrest and displayed considerable necrotic areas on the surface (Fig. 5B). We next analyzed cell proliferation in transplanted tumors in vivo by Ki-67 immunohistochemistry. The proliferation ratio, that is, the ratio of Ki-67+ cells to total cells in areas with high proliferating activity was not significantly different in EYA2-overexpressing tumors (2008: 91.0 ± 4.9, A2780: 75.8 ± 13.8) than in control tumors (2008: 88.3 ± 4.1, A2780: 71.7 ± 9.9, both P > 0.05; Fig. 5C). Areas with high proliferating activity and no proliferating activity could be clearly distinguished by Ki-67 staining in xenograft tumors (Fig. 5D and E). We found that the areas exhibiting high proliferating activity were significantly larger in EYA2-overexpressing tumors (28.28 ± 1.7%) than in control tumors (17.00 ± 1.6%, P = 0.01; Fig. 5F). To confirm these in vivo results, we analyzed cell proliferation and cell cycle in vitro. EYA2 overexpression did not significantly change either cell proliferation (Fig. 5G) or cell cycle in four ovarian cancer cells in vitro (Fig. 5H). The above results suggest that EYA2 affects the ability of tumor cells to sustain proliferation in vivo possibly through complex interactions.

Choosing the appropriate source for the normal counterpart compared with ovarian cancer has proven difficult (27). Epithelial ovarian cancer is thought to arise from the single cell layer of ovarian surface epithelium. Because the ovarian surface epithelium represents only a fraction of the total ovary, isolation of significant quantities of ovarian surface epithelium RNA has been difficult (27). Therefore, a variety of “normal” equivalents has been used in ovarian cancer. One method involves analyzing a whole normal ovary, resulting in a sample in which stroma constitutes the majority of the tissue. Other approaches rely on short-term cultures of ovarian surface epithelium scrapings, which are usually termed human ovarian surface epithelium (HOSE). An alternative method involves HOSE cells immortalized with SV40 large T-antigen (28), human papilloma virus-E6E7 (29) or telomerase (27). In this study, candidate genes were screened using whole normal ovary by suppression subtractive hybridization. To exclude the interference from stroma RNA, we further tested the expression of candidate genes by HOSE or immortalized HOSE.

The transcriptional activator eya belongs to a conserved regulatory network that plays an important role during development. This study is the first to report that EYA2 may play an important role in human cancer. Our data show that EYA2 is expressed at low levels in normal human ovary. The transcription and expression of EYA2 are, however, significantly up-regulated in human ovarian cancer. Moreover, EYA2 is significantly increased in late-stage ovarian cancer, where high expression is significantly associated with poor outcome. These results support a novel role of EYA2 in human ovarian cancer. We found that EYA2 gene was amplified in 14.8% of ovarian cancer specimens. Therefore, DNA amplification might partly provide an explanation for the observed EYA2 overexpression in ovarian cancer. On the other hand, studies in Drosophila have showed that eyeless (ey) induces expression of eya and so (4, 5), suggesting that ey is able to directly or indirectly activate eya expression (8). Increasing data indicate that human PAXs, the mammalian homologues of ey, are overexpressed or translocated in cancer and might play a critical role in tumorigenesis (37–39). Therefore, upstream transcriptional regulatory factors involved in EYA2 activation might also contribute to EYA2 overexpression in human ovarian cancer. In addition, EYA2 is likely posttranscriptionally modified in glioblastoma (40). Whether posttranscriptional modification occurs in ovarian cancer is still unknown.

Genetic studies in both Drosophila and mammals have showed that eya functions as a transcriptional activator that synergizes with so to activate specific gene targets, including those regulating precursor cell proliferation and survival during organogenesis. The fact that EYA2 protein is strongly expressed in the nuclei of ovarian cancer cells also indicates that Eya2 might function as a nuclear protein (e.g., transcription factor) in ovarian cancer. The transactivation assay further proved that EYA2 serves as a transcriptional activator in ovarian cancer. In Drosophila, eya and so together constitute a functional transcriptional regulatory complex, with eya providing the activation domain and so contributing the DNA-binding moiety (3–5). In mammals, SIX2, SIX4, and SIX5 are also able to synergize with EYAs to drive expression from a reporter construct (41). EYA2, SIX1, and DACH2 synergistically regulate myogenesis in chicken somite cultures (11). Although detailed information about the interactions between human EYAs and SIXs is still lacking, SIX expression is one of the necessary factors for transcriptional activation induced by EYA2.

Therefore, we analyzed the expression of all six human SIXs in ovarian, breast, and colon cancer. As expected, SIX 1-5 were widely expressed in ovarian and breast cancer. Accumulating evidence indicates that human SIXs are also implicated in cancer development (17, 18, 42). It is thus possible that the EYA-SIX transcriptional signal pathway may cooperatively contribute to human cancer development.

High expression of EYA2 was significantly associated with poor outcome in patients with ovarian cancer, whereas ectopic expression of EYA2 in human tumor xenografts promoted tumor growth. These data strongly suggest that EYA2 contributes to ovarian cancer growth. However, although EYA and human SIX1 have been reported to play a role in the control of cell cycle (16–18), our data indicate that the function of EYA2 in ovarian cancer is not directly dependent on cell proliferation or the cell cycle. Recent studies have showed that the eya family has a protein phosphatase function and that its enzymatic activity is required for gene expression regulation (43–45). In addition, EYA2 serves as a transcriptional activator of SIX, whereas genetic interactions between EYA and SIX factors regulate precursor cell proliferation and survival during organogenesis (3, 44). It is possible that in ovarian cancer EYA2 overexpression leads to sustained cell growth through complex interactions with the tumor microenvironment. Characterization of the downstream target genes of the EYA-SIX transcriptional complex would thus be critical to understanding the function of EYA2 in ovarian cancer.

Grant support: Ovarian Cancer Research Fund, the Abramson Cancer Center, and the Pennsylvania Department of Health.

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.

We thank Drs. Steven Johnson and Kang-Sheng Yao (University of Pennsylvania) for the human ovarian cancer cells and Drs. Michael J. Birrer (National Cancer Institute) and Nelly Auersperg (University of British Columbia) for HOSEs.