Differentially androgen-modulated genes in ovarian epithelial ... - Nature

Oncogene (2007) 26, 198–214

& 2007 Nature Publishing Group All rights reserved 0950-9232/07 $30.00 www.nature.com/onc

ORIGINAL ARTICLE

Differentially androgen-modulated genes in ovarian epithelial cells from BRCA mutation carriers and control patients predict ovarian cancer survival and disease progression A Motamed-Khorasani1,2, I Jurisica3,4,5, M Letarte5,6,7,8, PA Shaw7,9, RK Parkes1, X Zhang3, A Evangelou1,6, B Rosen7,10, KJ Murphy7,10 and TJ Brown1,2,7 1 The Samuel Lunenfeld Research Institute, Mt Sinai Hospital, Toronto, Ontario, Canada; 2Department of Physiology, University of Toronto, Toronto, Ontario, Canada; 3Division of Signaling Biology, Ontario Cancer Institute/Princess Margaret Hospital, University Health Network, Toronto, Ontario, Canada; 4Department of Computer Science, University of Toronto, Toronto, Ontario, Canada; 5 Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada; 6Cancer Research Program, The Hospital for Sick Children, University of Toronto, Toronto, Ontario, Canada; 7Department of Obstetrics and Gynecology, University of Toronto, Toronto, Ontario, Canada; 8Department of Immunology, University of Toronto, Toronto, Ontario, Canada; 9Division of Pathology, Ontario Cancer Institute/Princess Margaret Hospital, University Health Network, University of Toronto, Toronto, Ontario, Canada and 10Division of Gynecologic Oncology, Ontario Cancer Institute/Princess Margaret Hospital, University Health Network, University of Toronto, Toronto, Ontario, Canada

Epidemiological studies have implicated androgens in the etiology and progression of epithelial ovarian cancer. We previously reported that some androgen responses were dysregulated in malignant ovarian epithelial cells relative to control, non-malignant ovarian surface epithelial (OSE) cells. Moreover, dysregulated androgen responses were observed in OSE cells derived from patients with germline BRCA-1 or -2 mutations (OSEb), which account for the majority of familial ovarian cancer predisposition, and such altered responses may be involved in ovarian carcinogenesis or progression. In the present study, gene expression profiling using cDNA microarrays identified 17 genes differentially expressed in response to continuous androgen exposure in OSEb cells and ovarian cancer cells as compared to OSE cells derived from control patients. A subset of these differentially affected genes was selected and verified by quantitative real-time reverse transcription–polymerase chain reaction. Six of the gene products mapped to the OPHID protein–protein interaction database, and five were networked within two interacting partners. Basic leucine zipper transcription factor 2 (BACH2) and acetylcholinesterase (ACHE), which were upregulated by androgen in OSEb cells relative to OSE cells, were further investigated using an ovarian cancer tissue microarray from a separate set of 149 clinical samples. Both cytoplasmic ACHE and BACH2 immunostaining were significantly increased in ovarian cancer relative to benign cases. High levels of cytoplasmic ACHE staining correlated with decreased survival, whereas nuclear BACH2 staining correlated with decreased time to disease recurrence. The finding that Correspondence: Dr TJ Brown, Department of Obstetrics and Gynecology, Samuel Lunenfeld Research Institute, Mt Sinai Hospital, Room 876, 600 University Avenue, Toronto, Ontario, Canada M5G 1X5. E-mail: [email protected] Received 27 January 2006; revised 21 April 2006; accepted 21 April 2006; published online 10 July 2006

products of genes differentially responsive to androgen in OSEb cells may predict survival and disease progression supports a role for altered androgen effects in ovarian cancer. In addition to BACH2 and ACHE, this study highlights a set of potentially functionally related genes for further investigation in ovarian cancer. Oncogene (2007) 26, 198–214. doi:10.1038/sj.onc.1209773; published online 10 July 2006 Keywords: ovarian cancer; androgen; BRCA1; BRCA2; BACH2; acetylcholinesterase

Introduction Epithelial ovarian cancer is thought to be derived from the ovarian surface epithelium (OSE), a single layer of cells that line the surface of the ovary. However, the mechanisms causing malignant transformation of ovarian epithelium are largely unknown, and there are no reproducible precursor lesions to facilitate the study of early events associated with the disease or with predisposition. A favored hypothesis is that ovarian carcinomas arise from cortical epithelial inclusions formed by OSE invagination into the underlying stroma as the ovary undergoes changes in size or following ovulation (Cramer and Welch, 1983; Auersperg et al., 1997). Normally, the inclusions are short-lived, with cells undergoing apoptosis or conversion to mesenchymal (stromal) cells (Auersperg et al., 1997). The persistence of these epithelial inclusion cysts and their close proximity to a stroma rich in mitogenic factors, including steroids, have been hypothesized to contribute to susceptibility to neoplastic transformation (Kruk et al., 1994; Scully, 1995).

Androgen effects in ovarian epithelial cells A Motamed-Khorasani et al

199

At the present time, a family history of ovarian or breast cancer is associated with the greatest identifiable risk of ovarian cancer (Modugno et al., 2001). Mutations in breast cancer susceptibility genes, BRCA1 and 2, account for the majority of inherited predisposition, with carriers of germline mutations having an estimated lifetime risk of 28–54 and 21–28%, respectively (Easton et al., 1995; Ford et al., 1998; King et al., 2003). The structurally distinct proteins encoded by BRCA1 and 2 regulate numerous cellular functions, including DNA repair, chromosomal segregation, gene transcription, cell-cycle arrest and apoptosis (Deng and Brodie, 2000; Scully and Livingston, 2000; Venkitaraman, 2002). As OSE cells from BRCA mutation carriers are predisposed to malignant transformation, understanding the factors contributing to the risk of ovarian cancer in BRCA mutation carriers may provide further insight into early molecular processes involved in this disease. Several lines of evidence support a role for androgens in the etiology of epithelial ovarian cancer (reviewed by Risch, 1998). The majority of ovarian cancers are diagnosed post-menopause, when the balance of steroid produced in the ovary shifts from estrogens to androgens. Women with polycystic ovarian disease (i.e., associated with anovulation and overproduction of ovarian androgens) and those with pre- or postmenopausal high serum androgen levels have an increased risk of developing ovarian cancer (Helzlsouer et al., 1995; Schildkraut et al., 1996). Case report studies of women receiving high-dose androgens before genderaltering surgery have also highlighted a potential association between androgen and ovarian cancer (Hage et al., 2000). Up to 95% of ovarian epithelial tumors and the OSE are directly responsive to androgen as evidenced by androgen receptor expression (Ku¨hnel et al., 1987, 1988; Chada et al., 1993; Cardillo et al., 1998; Lau et al., 1999; Edmondson et al., 2002; Li et al., 2003) and by the ability of Mibolerone, a synthetic androgen, to increase DNA synthesis (Edmondson et al., 2002). We previously discovered a loss of coordinated androgen regulation of expression of transforming growth factor-beta (TGF-b) receptors and two steroid receptor co-activators, SRC-1 and ARA70, in ovarian cancer cells as compared to normal ovarian epithelial cells (Evangelou et al., 2003). Remarkably, this coordinated regulation was also lost in non-malignant epithelial cells derived from women bearing a germline BRCA1/2 mutation. These findings that ovarian epithelial cells with a genetic predisposition for malignant transformation resemble ovarian cancer cells in showing dysregulated responses to androgen, raised the possibility that altered responses to androgen influence the predisposition to cancer or reflect a consequence of regulatory changes brought about by BRCA heterozygosity and ovarian carcinogenesis. To identify differentially modulated genes in response to androgen treatment in control ovarian epithelial cells to that in non-malignant cells bearing a BRCA mutation and in ovarian cancer cells, we performed a high-density cDNA microarray analysis of cells maintained in culture in the

presence or absence of 10 nM 5a-dihydrotestosterone (DHT). Tissue arrays constructed from an independent set of 149 patients were used to validate the findings for two of the genes identified in the microarray analysis.

Results Patients and cell cultures Cell cultures were derived from surgical material obtained from 27 patients of the Gynecologic Oncology Division, University Health Network in Toronto, Ontario. Both the University Health Network and Mt Sinai Hospital Research Ethics Boards approved this study, and patients informed consent was obtained. Relevant clinical information on these patients is summarized in Table 1. Ten OSE cultures were derived from patients with non-malignant ovaries who were undergoing surgery for non-ovarian causes. Cells were obtained by lightly scraping the ovarian surface with a blunt scissors before oophorectomy, and were immediately processed as described previously (Evangelou et al., 2003). Most OSE cultures were derived from patients with nonmalignant ovaries who were undergoing surgery for uterine/cervical tumors. One case, OSE-137, underwent a prophylactic oophorectomy because of prior fertility drug usage despite successfully achieving full-term pregnancy. Eight OSEb samples were obtained from women undergoing prophylactic oophorectomy because of a germline BRCA1 or BRCA2 mutation: 6174delT (OSEb-7) and 983del4 (OSEb-16) for BRCA2; and 185delAG (OSEb-23 and OSEb-87), ins6kb (OSEb-29), 5382insC (OSEb-113 and OSEb-109) and 3450delCAA (OSEb-140) for BRCA1. Nine OVCAS cultures were derived from cells isolated from patient ascites as described by Hirte et al. (1992) and processed as previously reported (Evangelou et al., 2003). Eight patients were diagnosed with grade 2/3 papillary serous ovarian adenocarcinoma, and one with grade 3 endometrioid adenocarcinoma and in all cases the tumor was primary. Three cases (OVCAS-19, -27, and -29) were derived from patients who had received chemotherapy before surgery, a practice that is currently common in many clinics within Canada and Europe. The mean ages of OSE, OSEb and OVCAS patients were 55.779.9, 47.6710.6,and 60.3711.7 years (7s.d.), respectively. Fifteen patients were post-menopausal, whereas four of 10 OSE, five of eight OSEb and three of nine OVCAS cases were pre-menopausal. There was not a statistically significant difference between the groups with respect to age (P ¼ 0.11) or menopausal status (P ¼ 0.64). The epithelial nature of the cultures was confirmed by the cell surface expression of 2G3 and M2A antigens as determined by flow cytometry analysis (data not shown). 2G3 reacts with a carbohydrate epitope (Linsley et al., 1986) of MUC-1, which is present in both normal (Auersperg et al., 1994) and malignant ovarian epithelial Oncogene


200 Table 1 Culture OSE-21 OSE-22 OSE-32 OSE-70 OSE-84 OSE-89 OSE-92 OSE-96 OSE-98 OSE-137 OSEb-7 OSEb-16 OSEb-23 OSEb-29 OSEb-87 OSEb-109 OSEb-113 OSEb-140 OVCAS-19 OVCAS-24 OVCAS-26 OVCAS-27 OVCAS-29 OVCAS-32 OVCAS-34 OVCAS-37 OVCAS-43

Clinical characteristics of patients contributing tissue for culture derivation

Patient age (years)

Menopausal status

56 66 43 46 49 73 65 54 58 47 49 46 70 50 44 34 45 43 80 60 67 44 62 47 51 60 72

Pre Post Pre Post Pre Post Post Post Post Pre Post Pre Post Post Pre Pre Pre Pre Post Post Post Pre Post Pre Pre Post Post

Reason for surgery/other malignancies Endocervix mucinous adenocarcinoma cervix squamous carcinoma in-situ Benign ovarian dermoid cyst Benign mucinous cystadenoma/breast cancer Endometriosis Endometrial cancer Endometrial cancer Benign ovarian cystadenofibroma Uterine leiomyoma Fertility drug usage (prophylactic) Prophylactic (BRCA-2)/Breast cancer Prophylactic (BRCA-2) Prophylactic (FBOC, BRCA-1) Prophylactic (BRCA-1) Prophylactic (FBOC, BRCA-1) Prophylactic (FBOC, BRCA-1)/Breast cancer Prophylactic (FBOC, BRCA-1) Prophylactic (BRCA-1) OC (III, endometrioid) OC (III, papillary serous) OC (III, papillary serous) OC (III, papillary serous) OC (III, papillary serous) OC (III, papillary serous) OC (III, papillary serous) OC (III, papillary serous) OC (III, papillary serous)

Abbreviations: FBOC, familial breast/ovarian cancer history, III refers to stage; OC, ovarian cancer; OSE, ovarian surface epithelium cells derived from patients with apparently normal ovaries; OSEb, those derived from patients with BRCA1/2 mutations; OVCAS, ovarian cancer cells derived from ascites.

cells, whereas M2A reacts with a monomeric sialoglycoprotein expressed on ovarian epithelial cells (Marks et al., 1999). In all cultures used, 99% of the cells were positive for 2G3 and M2A staining. Identification of genes differentially expressed in OSEb and OVCAS relative to OSE The effect of DHT treatment on gene expression profiles was determined for OSE, OSEb and OVCAS cultures by hybridization to high-density cDNA microarrays containing 19 200 individual cDNAs or expressed sequence tags (ESTs) spotted in duplicate. All cultures were maintained in the presence or absence of 10 nM DHT. RNA was extracted, labeled directly with cyanin (Cy)-3 or -5, and the resulting labeled cDNA from each DHTtreated culture was co-hybridized to the array with its vehicle-treated control. The arrays were then scanned and the raw quantification data were subjected to serial normalizations and background subtraction. Two-class Significance Analysis of Microarrays (SAM) analysis (Tusher et al., 2001) was performed to identify a minimal set of genes differentially influenced by androgen in OSE vs OSEb and OSE vs OVCAS cultures. This analysis is based on the initial identification of the cultures as OSE, OSEb or OVCAS and reveals a minimal set of genes capable of differentiating the two culture types based upon the androgen response. In comparing OSE to OSEb cultures, seven differentiating genes emerged that were upregulated in OSEb in Oncogene

response to DHT, relative to OSE (Figure 1a). In comparing OSE to OVCAS, 10 genes were downregulated by androgen in OVCAS relative to OSE (Figure 1b). Two independent hierarchical clustering algorithms were used to illustrate the ability of these genes to discriminate the culture types. We first applied Eisen hierarchical cluster analysis (Eisen et al., 1998) to the cultures, selecting only the seven and 10 identified genes, respectively, as shown in Figure 1. Profiles from either OSE/OSEb or OSE/OVCAS comparisons showed an almost complete division at the first branch of the dendrogram, indicating the efficacy of these genes to differentiate between normal and predisposed or malignant cells. In clustering OSE and OSEb, all of the 10 OSE cultures partitioned together at the first branch of the dendrogram along with one of the OSEb cultures, OSEb-29 (Figure 1a). In clustering OSE and OVCAS, the two culture types partitioned to separate groups in the first branch of the dendrogram with one exception, OVCAS-34 (Figure 1b). Pseudocolor matrices of correlation coefficients were constructed to visualize the relationship among the 17 identified genes and 27 culture cases. The effect of androgen exposure on genes differentiating OSE and OSEb were highly correlated with one another, as was the case for genes differentiating OSE and OVCAS (Figure 1c). A lack of, or negative correlations were present comparing genes differentiating OSE/OSEb to those differentiating OSE/OVCAS. The patterns shown


201

Figure 1 Hierarchical clustering using Eisen cluster average linkage (Eisen et al., 1998) of OSE vs OSEb (panel a) and OSE vs OVCAS (panel b) cultures on the basis of seven and 10 classifier genes, respectively, identified by two-class SAM analysis (Tusher et al., 2001). The red (upregulated) and green (downregulated) color intensities represent log values of fold-changes over the expression in vehicletreated cultures. Black indicates no change from vehicle-treated levels. Gray indicates a missing value. UniGene accession numbers and the common gene name are provided for known genes. Numbers on the left indicate an arbitrary number assigned to the identified gene. (c and d) Pseudocolor correlation matrix representation for expression of the 17 genes identified by two-class SAM analysis (panel c) and for all 27 cultures based on expression of the 17 genes (panel d). In both panels, the color map corresponds to the scale of correlation coefficients. Uncorrelated data result in a correlation coefficient of zero; equivalent data sets have a correlation coefficient of one. The diagonal of the symmetric correlation matrix represents self-correlation and thus is equal to one. These images were generated using Matlab R12 (Mathworks Inc., Natick, MA, USA). The order of the cultures from left to right and bottom to top are: OSE-137, -21, -22, -32, -70, -84, -89, -92, -96, -98; OSEb-109, 113, 140, 16, 23, 29, 7, 87; OVCAS-19, -24, -26, -27, -29, -32, -34, 37 and 43. Oncogene


202

in Figure 1d represent the overall group pattern for each of the three culture types based on the responses of the 17 genes. OSE cultures were highly correlated with one another, whereas OSEb and OVCAS cultures were less correlated. To better determine how individual cultures are distributed relative to one another based upon their altered expression of all 17 differentially affected genes, a binary tree-structured vector quantization (BTSVQ) (Sultan et al., 2002) method of data clustering and visualization was used. Using the 17 differential gene responses, BTSVQ iteratively partitioned the 27 cultures by a k-means algorithm (k ¼ 2) into a binary cluster tree, and also clustered these genes and cultures by a selforganizing map (SOM) algorithm (Kohonen, 1995). In the first level of the cluster tree obtained by k-means, shown in Figure 2a, the cultures from control patients separated completely from the BRCA mutation carriers and from all but one (OVCAS-34) of the cancer patients, confirming that the effects of androgen on these 17 genes were distinct among the culture types. SOM clustering was next applied to both the 27 cultures and 17 genes. The resulting distances among the 17 genes, which reflect similarities/differences in androgen responses, are represented by a color scale in Figure 2b. Genes differentiating OSE and OSEb cultures (genes 1–7) clustered as a distinct group from those differentiating OSE from OVCAS cultures (genes 8–17). Individual SOM component planes of the 27 cultures, based on the altered expression of the 17 genes, are shown in Figure 2a. The location of the genes within each grid is as shown in Figure 2b. In these SOM component planes, color represents an average expression of the particular set of genes. The SOMs of the OVCAS and OSEb cases that clustered together (cluster a in Figure 2a) are more variable in contrast to those of the OSE cases (cluster b in Figure 2a). A SOM partitioning the cultures rather than the genes is shown in Figure 2c, with a color scale representing distances among cultures. Again, the control cases were more closely related to one another than either the BRCA carriers or cancer patients, which did not separate into distinct groups. Thus, three independent clustering algorithms have computationally verified the 17 genes identified by SAM analysis, as a differential signature. Validation of differential response to androgen by quantitative real-time reverse transcription–polymerase chain reaction Of the 17 genes, 15 were selected based upon their identity and the extent of their altered expression (fold-change) for confirmation by real-time reverse transcription–polymerase chain reaction (RT–PCR), which was performed on a total of 6–8 culture pairs (DHT- and vehicle-treated) (Table 2). All gene responses were in parallel with the microarray results except for H69632, which encodes a hypothetical protein. Of the genes differentially regulated by androgen in OSEb compared to OSE, three encode known proteins: acetylcholinesterase (ACHE), stromal interaction Oncogene

molecule 1 (STIM1) and BTB CNC homology 1 (BACH2). Of the genes differentially expressed in OSE and OVCAS cultures, four encode known proteins: glypican-3 (GPC3), cytochrome P450 IVB, polypeptide 1 (CYP4B1), iduronate 2-sulfatase (IDS), and embryonic development associated gene 1 (EDAG-1). Interpretation of differential genes using protein–protein interaction network As an initial step toward prioritizing differentially modulated genes for further study, the proteins encoded by these genes (Swiss Prot identifiers) were mapped to an OPHID (Online Predicted Human Protein Interaction Database; http://ophid.utoronto.ca) (Brown and Jurisica, 2005) and visualized using NAViGaTOR software (Network Analysis, Visualization and Graphing, Toronto, Canada; http://ophid.utoronto.ca/navigator/). Of the identified genes, six were contained in the database (as of 6 April 2006). Although considering these genes with their first-order interacting partners (immediate interactors, forming a network with 343 proteins and 411 interactions) failed to indicate a relationship, all but one of the six were linked when second-order interacting partners (interactors two steps away from the six genes, forming a network consisting of 374 proteins and 446 interactions) were included in the network. A subnetwork shown in Figure 3, which includes only the genes identified in the OSE-OSEb comparison, revealed a linkage between ACHE and BACH2. The node color represents Gene Ontology (Ashburner et al., 2000), as shown in the upper left corner. Known members of several pathways parsed from the KEGG database (Kanehisa et al., 2002; http:// www.genome.jp/kegg) implicated in ovarian cancer (Hu et al., 2000; Huang et al., 2000; Hirashima et al., 2003; Rask et al., 2003; Robertson et al., 2004; Steinmetz et al., 2004; Nishimoto et al., 2005), including TGF-b (five proteins), Wnt (13 proteins), mitogen-activated protein (MAP) kinase (nine proteins) and Jak-Stat (eight proteins) signaling systems, are present (nodes highlighted with red circles) in this network. On the basis of this subnetwork (Figure 3) and a previous report indicating amplification of ACHE in some ovarian cancers (Zakut et al., 1990), ACHE and BACH2 were selected for further investigation. Immunohistochemical detection of BACH2 and ACHE expression in ovarian cancer Ovarian cancer tissue microarray (TMA) slides containing duplicate sections from formalin-fixed, paraffinembedded ovarian tissue cores were stained with antibodies to ACHE or BACH2. Epithelial cytoplasmic and nuclear staining was scored in a blinded manner based on the intensity of expression (scale 0–3) and the percentage of positive cells (scale 0–3). A total of 129 ovarian carcinomas, 17 benign and three low-malignant potential (LMP) samples were included in the analysis. All patients were treated by members of the Division of Gynecological Oncology at either Princess Margaret Hospital or Sunnybrook Health Sciences Center in


203

Figure 2 Partitioning of cultures by the BSTVQ method (Sultan et al., 2002) based on the altered expression of the 17 classifier genes identified by two-class SAM analysis and listed in Figure 1a and b. (A) Dendrogram showing clustering of all 27 cultures by k-means algorithm (k ¼ 2). Green represents OSE, blue represents OSEb and red represents OVCAS cultures. In the first division, all OSE cases clustered separately (b) from OSEb and OVCAS cultures (a) with the exception of OVCAS-34. Individual gene expression SOM representations for each of the 27 cases are shown. Color reflects average gene expression levels with yellow to red indicating upregulated expression and blue indicating downregulated expression. Location of the individual 17 genes within each SOM is as indicated in panel B. Grouping (a) includes all OSEb and OVCAS cases that initially partitioned together in the first branch of the dendrogram whereas grouping (b) includes all OSE cases and OVCAS-34 that clustered together. The letters ‘n’ ‘b’ and ‘c’ before the numbers denote cultures derived from control patients, BRCA carriers, and ovarian cancer patients respectively. (B) SOM showing clustering of the 17 genes based upon their altered expression in all 27 cultures owing to androgen treatment. Color represents arbitrary distance units (dissimilarities) with genes showing most similar responses clustering closely within areas of blue. Red regions represent greater distances. For identification of a particular gene, refer to the numbers provided in Figure 1. (C) SOM showing partitioning of the cultures based upon their altered gene expression owing to androgen treatment. As in panel b, color represents arbitrary distance units with blue indicating short distances and red longer distances. The more similar the androgen response of the 17 discriminator genes within cultures, the closer the cultures are located to one another.

Toronto, Ontario. All patients were given platinumbased chemotherapy following surgery except for patients who were at stage 1, grade 1 (n ¼ 5) who received full abdominal radiation instead of chemotherapy. None of the patients included in the tissue array received neoadjuvant chemotherapy. ACHE staining was localized primarily to the cytoplasm (Figure 4), although some nuclear staining

was also observed (Table 3). Both nuclear and cytoplasmic epithelial cell expression of BACH2 was observed (Figure 4). A distribution of composite staining scores (the sum of intensity and percentage scores) for benign, low malignant potential (LMP) and the various histotypes of ovarian cancer is shown in Table 3. There were positive correlations between all of the scores, particularly between BACH2 and ACHE Oncogene


204 Table 2 Fold-changes for selected classifier genes determined by realtime RT–PCR OSE

OSEb

Gene Number

GeneBank Gene name accession no.

Mean7s.e.

Mean7s.e.

2 3 4 5 6 7

T98534 R85241 BF672232 AB037761 AI357692 R13871

0.3370.08 0.5170.06 0.5370.11 0.7270.12 0.4970.14 0.3070.06

1.7070.08 1.2870.05 1.7270.20 2.8070.85 1.8570.39 2.8071.02

OSE

OVCAS

Mean7s.e.

Mean7s.e.

1.3170.12 2.4170.51

0.4170.15 0.8770.04

1.9370.53 2.6170.32 2.4170.39 1.4870.14 1.3370.04 1.3770.14 5.7172.18

0.5270.09 0.4570.10 0.3670.10 1.2870.41 0.8570.05 0.7570.08 0.4070.12

8 9

R86925 BG673977

10 11 12 13 14 15 16

H59158 H16276 H57051 H69632 R96571 BI766823 AA036869

STIM1 ACHE NT5C2L1 KIAA1340 ASH1L BACH2

Glypican 3 Zinc-finger protein KIAA0543 GluR-2 EDAG-1 FLJ21928 EST IDS Cyp4B1

Abbreviations: DHT, 5a-dihydrotestosterone; OSE, ovarian surface epithelium cells derived from patients with apparently normal ovaries; OSEb, those derived from patients with BRCA1/2 mutations; OVCAS, ovarian cancer cells derived from ascites; RT–PCR, reverse transcription–polymerase chain reaction. Results are expressed as the mean ratio of the amount of mRNA detected in the DHT-treated culture to that of the conspecific vehicle-treated culture. In all cases, a significant difference between vehicle and DHT treatment was seen in three of the four cultures tested (Po0.05) (Pfaffl et al., 2002). Gene number is as indicated in Figure 1.

cytoplasmic scores (Po0.0001), and between the cytoplasmic and nuclear scores for BACH2 (Po0.0001; Table 4). In general, there was no evidence of an association between the clinical variables (age at diagnosis, stage, grade) and ACHE or BACH2 cytoplasmic or nuclear staining levels for the 129 carcinoma patients. There was strong evidence of differences in ACHE and BACH2 cytoplasmic staining among histologic types of carcinoma (P ¼ 0.003 for each), and between benign, LMP, and malignant tissues (P ¼ 0.021 and 0.006, respectively). Canonical discriminant analysis (Figure 5) indicated that the differences between groups could be summarized using only the first canonical variable (P ¼ 0.005), as the second (P ¼ 0.50) and higher dimensions did not significantly discriminate between group means. The largest contributions to the first canonical variable were from ACHE and BACH2 cytoplasmic scores, whereas stepwise discriminate analysis confirmed that ACHE and BACH2 nuclear scores were redundant (P ¼ 0.52 and 0.41, respectively, when the two cytoplasmic scores were also included). Therefore, the discriminant function obtained by simply adding together the ACHE and BACH2 total scores was determined to be optimal. Cluster analysis (BTSVQ) of the sample histotypes using cytoplasmic ACHE and BACH2 total scores Oncogene

indicated a separation of benign, mucinous and clear cell carcinoma cases from serous, endometrioid, and other ovarian carcinoma samples in the first level of the cluster tree (Figure 6). SOM component planes of the total scores (Figure 6) show their distribution, and illustrate their similarity at the terminal level determined by the k-means algorithm. The similarity is particularly striking for benign ACHE and BACH2 scores and for serous carcinoma ACHE and BACH2 scores. A Cox proportional hazards regression model was used to assess the association between overall survival and time to disease recurrence to each composite marker score after adjusting for appropriate clinical covariates: histological group, stage, tumor grade and surgical debulking success (optimal, suboptimal, unknown). Covariates were selected a priori based on a literature review of known prognostic factors, and all were confirmed to have prognostic value for both (stage and histology) or one (grade and debulking status for disease recurrence only) of the outcomes in the 129 carcinoma patients. Owing to the large number of model variables relative to the number of patients with events (48 deaths and 80 recurrences in 129 patients), the impact of model overfitting was minimized by using ridge regression to estimate parameters for the clinical covariates, with the likelihood penalty chosen to optimize Akaike’s information criteria before introducing the marker scores to the model (Gray, 1992). In the primary analysis, each of the four composite marker scores was analysed separately, to minimize overfitting. Secondary analyses were conducted to assess the performance of the models in the primary analysis, including univariate analyses (i.e. without clinical covariates) of each marker, and the full multivariate model (clinical covariates as well as all four marker scores in the same model). As shown in Table 5, there was evidence of a negative association between ACHE cytoplasmic staining and overall patient survival (P ¼ 0.056), and a positive association between BACH2 nuclear staining and disease recurrence (P ¼ 0.039) after adjusting for prognostic factors. Adjusted survival plots for these markers (in roughly equalized groups: total score 0, 2–4, and 5–6) were produced for both overall survival and disease recurrence (Figure 7). Each curve is adjusted via the Cox model to represent the expected survival experience of the hypothetical individual with mean levels of each of the clinical covariates. In both cases, the effect is confined to those with staining (scores 2–6) as compared to no staining (score 0).

Discussion The results of this study demonstrate that genes identified as differentially expressed in OSE from patients with BRCA mutations as compared to control patients, are increased in ovarian cancer, and are associated with a more aggressive disease, reflected as increased disease progression or shortened patient survival. These findings support the hypothesis that


205

Figure 3 Subnetwork of the interactome linking identified OSE–OSEb differentiating gene products. Nodes represent proteins identified through Swissprot numbers, and lines represent confirmed or high-confidence predicted PPIs between the linked nodes, indicating a potential functional relationship. Products of genes identified as having differentially altered expression in OSE vs OSEb in this study are indicated by triangular nodes, and include ACHE, BACH2, ASH1 and NT5C2L1 (Q96BV3). Node color indicates functional association based upon GO classification. Nodes representing proteins belonging to the TGF-b, WNT, MAP Kinase and Jak-Stat signaling pathways are highlighted with orange halos (total of 33 proteins).

altered androgen responses in BRCA mutation carriers could contribute to ovarian cancer predisposition or disease progression. Although expression of the remaining wild-type allele is usually lost in tumor cells, mutation in only one BRCA allele is sufficient to predispose a carrier to breast and/or ovarian cancer (Buchholz et al., 2002; Venkitaraman, 2002). Both BRCA1 and 2 proteins are involved in DNA repair

and homologous recombination, cell-cycle checkpoint progression and transcription (Deng and Brodie, 2000; Scully and Livingston, 2000; Venkitaraman, 2002). Brca1 has been reported to interact with the androgen receptor and increase its transactivational activity in prostate cancer and breast epithelial cell lines (Park et al., 2000; Yeh et al., 2000). Recent evidence also indicates that wild-type Brca2 interacts with the Oncogene


206

Figure 4 ACHE and BACH2 immunostaining in ovarian tissue microarray samples. (a) ACHE staining in an inclusion cyst with columnar metaplasia ( 400), (b) ACHE staining in a mucinous LMP borderline tumor ( 630), (c) ACHE staining in Clear cell adenocarcinoma ( 630), (d) lack of ACHE staining in a mucinous grade III adenocarcinoma ( 630), (e) BACH2 staining in a serous adenocarcinoma ( 630), (f) BACH2 staining in an endometriod adenocarcinoma ( 630), (g) BACH2 staining in a mixed mullerian adenocarcinoma ( 630), (h) BACH2 staining in a mucinous LMP tumor ( 630). Arrows indicate examples of cytoplasmic staining and arrowheads indicate examples of nuclear staining.

androgen receptor and the p160 co-activator GRIP1 to enhance androgen-mediated transcriptional activation (Shin and Verma, 2003). The differences observed in androgen effects on expression profiles in ovarian epithelial cells with BRCA mutations relative to OSE cells from control patients could reflect altered coactivator function because of diminished levels of Oncogene

functional Brca1 or 2. It remains possible that the altered response results from other regulatory factors that are modified in BRCA heterozygotes and affect normal hormonal mechanisms. The nature of androgen signaling in ovarian epithelial cells remains to be elucidated. Several investigators, including us have reported that ovarian cancer and OSE


207 Table 3

Distribution of cases based on scores derived from tissue array analysis Carcinomas

Score

Clear cell

Endometrioid

Mucinous

Serous

Other

All

LMP

Benign

0 2 3 4 5 6

0 1 0 1 6 3

3 0 2 6 2 1

4 0 0 1 2 0

22 12 13 20 11 9

0 0 2 2 2 1

29 13 17 30 23 14

1 0 0 1 1 0

11 0 0 3 3 0

0 2 3 4 5

9 0 0 1 1

12 0 0 1 1

8 0 0 0 0

66 5 7 7 3

6 0 0 0 1

101 5 7 9 6

2 0 0 1 0

17 0 0 0 0

0 2 3 4 5 6

1 0 0 1 3 6

0 2 2 3 7 0

2 2 0 1 2 0

18 8 15 20 20 7

0 0 3 1 3 0

21 12 20 26 35 13

2 0 0 1 0 0

8 1 3 2 1 2

0 2 3 4 5 6

2 1 2 3 2 0

2 0 3 2 4 3

2 0 1 2 1 1

28 7 7 19 22 4

1 0 1 3 2 0

35 8 14 29 31 8

0 0 2 0 1 0

9 0 1 1 4 1

ACHE cytoplasm

ACHE nucleus

BACH2 cytoplasm

BACH2 nucleus

Abbreviations: ACHE, acetylcholinesterase; LMP, low-malignant potential. Scores represent the sum of staining intensity (0-3) and percentage of cells stained (0, 1: 1–24%, 2: 25–49%, or 3: X50%).

Table 4 Spearman’s correlation coefficients between marker scores for 149 tissue samples (with P-values testing H0: r ¼ 0)

ACHE cytoplasm ACHE nucleus

ACHE cytoplasm

ACHE nucleus

BACH2 cytoplasm

BACH2 nucleus

1

0.147 P ¼ 0.076 1

+0.433 Po0.0001 +0.106 P ¼ 0.20 1

+0.233 P ¼ 0.005 +0.202 P ¼ 0.015 +0.367 Po0.0001 1

BACH2 cytoplasm BACH2 nucleus Abbreviation: ACHE, acetylcholinesterase.

cells express low levels of androgen receptor (Ku¨hnel et al., 1987, 1988; Chada et al., 1993; Cardillo et al., 1998; Lau et al., 1999; Evangelou et al., 2000, 2003; Edmondson et al., 2002; Li et al., 2003). We have shown previously that HEY and SKOV-3 cells respond to DHT with decreased transcript levels for TGF-b receptor type II, despite threshold expression levels of androgen receptor (Evangelou et al., 2000), raising the possibility of alternate mediators of signaling for androgens and their metabolites or the upregulation of androgen receptor co-activators potentiating androgen responses (Anzik et al., 1997; Shaw et al., 2001; Li et al., 2005). It is important to note that the cultures used in these studies were exposed chronically to DHT and thus the altered gene expression reflects the longstanding

Figure 5 Biplot of group means and canonical coefficients: canonical correlations with the groups are 0.44 (P ¼ 0.005) and 0.21 (P ¼ 0.50) for the first and second canonical variables, respectively.

effects of androgen exposure rather than direct responses to the hormone. Many of these responses likely also result from non-genomic mechanisms of androgen receptor action such as activation of SRC, Oncogene


208

Figure 6 Partitioning of histotype samples by the BSTVQ method (Sultan et al., 2002) based on the total cytoplasmic ACHE or BACH2 scores. The dendrogram shows clustering of the 12-score distributions by an iterative k-means algorithm (k ¼ 2). SOM component plane of score distribution for each sample histotype is shown at the terminal level. Color reflects the percentage of cases with a particular score, with blues representing low percentages and reds high percentages. Location of the individual scores within each SOM is as indicated in the inset at the lower left. The more similar the SOM component plane of the sample histotypes, the more similar is their score distribution.

phosphatidylinositol 3-kinase or MAP kinases that could affect the steady-state levels of gene transcripts. Based on experiments with Xenopus oocytes, Lutz et al. (2003) have concluded that activation of non-genomic androgen responses is dependent upon receptor localization, cell type and receptor ligand. In addition, the recent report that a G-protein-coupled receptor mediates some non-genomic activities of estradiol Oncogene

(Revankar et al., 2005) raises the possibility of similar receptors for androgens. The present data indicating increased expression of ACHE in ovarian cancers, particularly serous and endometrioid, is consistent with a previous report of ACHE gene amplification in ovarian cancer (Zakut et al., 1990). ACHE is best known for its activity as an esterase acting on several acetic esters, playing a key role


209 Table 5 Multivariate survival analysis for carcinomas with each markera entered into model individually (adjusted for stage, grade, histology, debulking success) Time to death

ACHE cytoplasm ACHE nuclear BACH2 cytoplasm BACH2 nuclear

Time to recurrence

Hazard ratio

P-value (Wald test)

Hazard ratio

P-value (Wald test)

1.153 0.963 1.053 1.061

0.06 0.72 0.53 0.41

1.035 1.103 0.982 1.138

0.41 0.25 0.90 0.04

Abbreviation: ACHE, acetylcholinesterase. aTreating composite scores (0–6) as linear variables.

in the rapid inactivation of acetylcholine at neuronal synapses and neuromuscular junctions, and in the extracellular fluid, to terminate signal transmission. However, ACHE has also been implicated in modulating cell–cell and cell–matrix interactions, and in the control of cell differentiation and proliferation (Johnson and Moore, 2000a, b; Vidal, 2005). The significance of upregulated ACHE expression in ovarian cancers is not presently known, but could involve these non-cholinergic actions. The finding that cytoplasmic BACH2 expression is increased in ovarian cancer, with increased nuclear BACH2 levels associating with shortened time to disease recurrence, is a novel finding. BACH2 is an oxidative stress-regulated transcription repressor at the antioxidant and 12-O-tetradecanoylphobol-13 acetate response elements, and could modulate expression of metallothionines, glutathione and glutathione S-transferase. NIH3T3 murine fibroblasts overexpressing BACH2 exhibit decreased proliferation and increased apoptosis when combined with oxidative stress (Muto et al., 2002). Although such studies suggest a potential antioncogenic role for upregulated BACH2 expression, the presence of other proteins that interact with the BACH2 BTB/POZ domain may modify its activity. For example, MAZ-related factor interacts with BACH2 to activate c-myc and FGF4 expression, at least in hematopoietic cells (Kobayashi et al., 2000). Upregulated c-myc expression may contribute to decreased response to the growth-inhibitory actions of TGF-b in ovarian cancer cells. Consistent with this, we observed increased myc expression in OSEb cultures treated with androgen. An increased level of cytoplasmic BACH2 immunostaining observed in ovarian cancers relative to benign epithelium could reflect sequestration of this nuclear transcriptional regulator within the cytoplasm, via a cytoplasmic localization signal located on the C-terminus of BACH2. Signals, such as oxidative stress, would result in nuclear translocation. Although we did not observe altered BACH2 nuclear staining in cancers relative to the benign cases, we did note an association between increased nuclear BACH2 staining and time to recurrence among ovarian cancer patients. At present, the full significance of altered BACH2 expression in

Figure 7 Association of cytoplasmic ACHE and nuclear BACH2 immunostaining levels with patient survival and disease progression. (a) Survival (all causes of death) in 129 cases of ovarian carcinoma by ACHE cytoplasm staining score, adjusted for stage, grade, histology and debulking status by Cox proportional hazards model to mean level of covariates. (b) Time to disease recurrence in 129 cases of ovarian carcinoma by BACH2 nuclear staining score, adjusted for stage, grade, histology and debulking status by Cox proportional hazards model to mean level of covariates.

ovarian cancer is unknown. It has been postulated that different BACH2 isoforms localize to different cellular compartments where they may participate in cell processes distinct from transcriptional regulation. The results from this study indicate further analysis of BACH2 isoform expression and potential cytoplasmic function in ovarian epithelial cells is warranted. In addition to ACHE and BACH2, the results of this study highlight a set of genes differentially affected by androgen in non-malignant OSE from control patients as compared to OSE from BRCA mutation carriers and malignant ovarian epithelial cells. Many of the identified genes have been implicated in cancer etiology or Oncogene


210

progression. For example, CYP4B1 is a mono-oxygenase that catalyses many reactions involved in drug metabolism and synthesis of cholesterol, steroids, fatty acids, xenobiotics, and other lipids. CYP4B1 also contributes to the conversion of aromatic amines (arylamines) present in cigarette smoke and of industrial chemicals to nucleophilic compounds capable of mutagenesis (Windmill et al., 1997). Expression of CYP4B1 mRNA in the rat bladder and mouse kidney is higher in males than in females and is regulated by androgen (Imaoka et al., 2001; Isern and Meseguer, 2003), leading to speculation that this regulation could contribute to gender differences in renal and bladder cancer. That expression of this gene is downregulated in most malignant cultures whereas upregulated in normal OSE could reflect differences in transcription factors and/or repressors. An upregulation of CYP4B1 in the ovarian epithelium by androgen could make these cells more susceptible to chemical carcinogens. EDAG-1 is the human homolog of the murine gene hemogen that encodes a novel nuclear factor of unknown function and is expressed almost exclusively in hematopoietic tissues (Yang et al., 2001). To our knowledge, the present data are the first demonstration of EDAG-1 expression in non-hematopoietic cells and of modulation by androgen. Further studies are required to determine its potential role in ovarian cancer. The overexpression of EDAG-1 in NIH3T3 fibroblasts led to a loss of contact inhibition and anchorage-independent growth, indicative of malignant transformation and suggestive of a proto-oncogene (Lu et al., 2002). Upregulation of this gene in OSE cells could thus contribute to carcinogenesis, whereas modulation of its expression in OVCAS cells may be less because the transforming event has already occurred or the gene is already overexpressed in the absence of androgen treatment. Based on the interactome generated through OPHID of the identified differentially expressed genes, we predict that many of the proteins encoded by these genes act in concert to modulate multiple signaling pathways and cell functions involved in ovarian cancer etiology or progression. As shown in Figure 3, ACHE and BACH2 are linked through interactions with b-amyloid (A4), SRC homology domain 2 containing transforming protein (SHC1), and phospholipase C (PIG2). Through these intermediaries, ACHE and BACH2 link to members of multiple signaling pathways implicated in ovarian cancer, such as the TGF-b (Hu et al., 2000; Hirashima et al., 2003; Robertson et al., 2004), Wnt (Rask et al., 2003), jak-stat (Huang et al., 2000; Nishimoto et al., 2005) and MAP kinase (Steinmetz et al., 2004) pathways, as well as to the other identified differentially expressed genes. As appropriate antibodies become available, investigation of protein expression of these differentially affected genes in ovarian cancer may provide further insight into the role of androgen in the predisposition and progression of this disease. The main strength of this study is that genes identified in one set of samples using microarray analysis have Oncogene

been validated in a separate set of clinical samples using tissue arrays. A criticism of the microarray analysis that might be raised is that OSE, OSEb and OVCAS samples were not matched for patient age and menopausal status. Although OSEb patients tended to be younger and premenopausal and OVCAS patients more likely to be older and postmenopausal, the differences were not statistically significant. Furthermore, in a secondary analysis of the tissue array data using analysis of covariance (not shown), controlling for patient age had no impact on the significance of the observed differences between tissue types and histologic groups.

Materials and methods Cell culture All cells were grown in Dulbecco’s modified Eagle’s medium/ F12 medium without phenol red and supplemented with 3% charcoal-stripped fetal bovine serum, 5 mg/ml insulin, 5 105 M ethanolamine, 5 ng/ml epidermal growth factor, 50 mM phosphoethanolamine, 10 mg/ml transferrin, 50 U/ml penicillin, 50 mg/ml streptomycin and 0.625 mg/ml amphotericin B (Fungizone) at 371C in a humidified 5% CO2 atmosphere. All cultures were maintained in the presence of 10 nM DHT (Sigma Chemical Co., St Louis, MO, USA), added every 48 h, or the ethanol diluent at a final concentration of 0.001% (vehicle control). This was performed to mimic an in vivo contact with this steroid, either in the ovary or in the ascites fluid. This concentration of DHT was chosen based on its optimal effect in previous studies with ovarian cancer cell lines (Evangelou et al., 2000). All experiments were carried out with low passage [1–5] cultures. Flow cytometry Cell monolayers were detached with 0.25% trypsin– ethylene diaminetetraacetic acid (EDTA), washed and tested for expression of HLA class I antigen and the epithelial markers MUC-1 (detected with monoclonal antibody 2G3) and the oncofetal antigen M2A or control non-immune murine IgG1 as described previously (Evangelou et al., 2003). Cells were stained with propidium iodide, to allow for gating of viable cells during analysis by FACScan with Cell QUEST software (Becton Dickinson, Mountain View, CA, USA). Mean fluorescence intensity and percent positive cells were determined by setting the gates such that o5% cells were positive when stained with control non-immune IgG1. RNA extraction and labeling Cultures were washed with ice-cold phosphate-buffered saline (PBS), and total cellular RNA was extracted using TRIZOL reagent (Invitrogen Corp., Carlsbad, CA, USA) according to manufacturer’s instructions. All RNA samples were treated with DNase-I (DNA-free kit; Ambion Inc., Austin, TX, USA) and RNA concentration was estimated in two independent determinations and at three dilutions. The integrity of all RNA samples was verified by 2% agarose gel electrophoresis before use in microarray experiments. Fluorescent-labeled cDNA was derived from RNA by incorporating cyanine (Cy)-3 or Cy5-dCTP during reverse transcription (direct labeling). Briefly, 10 mg total RNA were added to first-strand buffer (Gibco Life Technologies, Burlington, ON, Canada) containing 150 pmol AncT primer (50 -T20VN-30 ; Cortec DNA Service Laboratories Inc., Kingston, ON, Canada), 1.5 mM dNTP-dCTP (Amersham


211 Pharmacia, Oakville, ON, Canada), 50 mM dCTP (Amersham Pharmacia), 10 mM DTT (Gibco Life Technologies, Burlington, ON, Canada), 5 ng arabadopsis RNA, and 1 mM of either Cy-3 or Cy-5 dCTP (Amersham Pharmacia) in a total volume of 40 ml. The mixture was incubated at 651C for 5 min followed by 421C for 5 min. SuperScript II reverse transcriptase (Gibco Life Technologies) was then added (2 ml/reaction tube) and the reaction left for an additional 2–3 h at 421C. RNA was hydrolysed by 2 ml of 10 N NaOH and 4 ml of 50 nM EDTA (pH 8) followed by incubation at 651C for 20 min. The sample was subsequently neutralized by addition of 4 ml of 5 M acetic acid. cDNA microarray hybridization Each DHT-treated culture was referenced to its conspecific vehicle-treated control. The labeling reactions from the DHTand control-treated cultures of each case were mixed (1:1) and loaded onto Microcon PCR centrifugal filter devices (Millipore, Billerica, MA, USA) to remove unincorporated Cy-dyes. The cDNA was mixed with 100 ml Dig Easy Hyb solution (Roche, Laval, QC. Canada), 5 ml yeast tRNA (Gibco Life Technologies), and 5 ml calf thymus DNA (Sigma Chemical Co.) and hybridized to cDNA microarrays printed on glass slides (Microarray Center, University Health Network, Toronto, ON, Canada). Arrays were incubated in hybridization chambers at 371C for 18 h. The arrays consisted of 19 200 individual cDNAs or ESTs spotted in duplicate (see http:// www.microarrays.ca/support/glists.html for a list of the included genes). All cultures were run in duplicate, switching the Cy-dyes for labeling the DHT-treated and vehicle-treated RNA samples. Thus for each culture, each cDNA was represented four times. Following hybridization, the slides were washed 3 times (20 min) with 1 SSC supplemented with 0.1% SDS, and 3times with 0.1 SSC. The slides were centrifuged at 500 rpm for 5 min, scanned on an Axon Genepix 4000A Scanner (Axon Instruments Inc., Union City, CA, USA), and quantified using the GSI Lumonics Quant-Array 3.0 software package (Packard Bio Chip Technologies, Billerica, MA, USA). Scans were manually reviewed to remove anomalous features (spots) from subsequent quantification and analysis. The quantified spot information output files were saved for further analysis. Microarray data analysis The raw quantification data were initially subjected to a serial normalization process and background subtraction using Normalize Suite v1.5. (Beheshti et al., 2003) (available at http://www.utoronto.ca/cancyto). Briefly, each of the subgrids on each array was independently normalized by equalizing the Cy3 intensities with respect to the Cy5 intensities, whereas excluding spots flagged as anomalous or absent by the quantifying software. Other features excluded by the algorithm included saturated spots and those with a foreground to background intensity ratio o2. The Cy5:Cy3 normalized intensity ratio was determined for each spot, and averaged for the duplicate spots within each array. This was repeated for the replicate experiment of each sample and the ratios for each gene in each culture were averaged together as a single project file. The project files representing each sample and its replicates were then combined into a single text file for processing by the Eisen Cluster software package and the graphic output viewed by Tree View (Eisen et al., 1998) (both available at http:// rana.lbl.gov/EisenSoftware.htm). Two-class Significance Analysis of Microarrays (SAM; available at http://www-stat. stanford.edu/Btibs/SAM analysis) (Tusher et al., 2001) was performed for OSE vs OSEb, and OSE vs OVCAS cultures to

identify a minimal set of genes differentially altered in the two culture types. A BTSVQ algorithm (Sultan et al., 2002) was used to cluster the cultures based upon the effects of DHT on the identified discriminatory genes. BTSVQ combines a partitive k-means clustering and a SOM algorithm in a complementary way, to achieve unsupervised clustering of both samples and genes. Assuming that the 17 differential genes are functionally related sets of genes, we mapped them into OPHID ((Brown and Jurisica, 2005), a web-based database (http://ophid. utoronto.ca) of known and predicted human protein–protein interactions (PPI). It combines the literature-derived human PPIs from BIND, DIP, HPRD and MINT, with predictions made from Saccharomyces cerevisiae, Caenorhabditis elegans, Drosophila melanogaster and Mus musculus. The 23 889 predicted interactions currently listed in OPHID are evaluated using protein domains, gene co-expression, gene ontology terms and text mining (Otasek et al., 2006). OPHID also includes interactions derived from high-throughput screens in human (Barrios-Rodiles et al., 2005; Rual et al., 2005; Stelzl et al., 2005; Jones et al., 2006), and can be queried using single IDs or batch queries, using multiple supported identifiers. The results can be displayed as text, HTML or visualized using a custom graph visualization program. In addition, the entire database is available for download in tab-delimited text or PSI-compliant XML format.

Quantitative real-time RT–PCR The altered expression of selected genes owing to DHT treatment was verified by real-time RT–PCR. Residual DNA was removed from RNA extracted from randomly selected cases using a DNA-Free kit (Ambion, Austin, TX, USA). Reverse transcription was performed using 1 mg/ml extracted RNA and random hexamer primers to enable the use of 18S rRNA for normalization, as its expression is considered stable in various tissues, even under experimental treatments (Foss et al., 1998; Thellin et al., 1999). The RT reaction was carried out at 251C for 10 min, 371C for 60 min and 951C for 5 min using an ABI-PRISM 7900 (Applied Biosystem Inc., Foster City, CA, USA). The final cDNA concentration in each reaction (assuming 100% efficiency) was 20 ng/ml. The cDNA quality was verified by amplifying b-actin (300 bp) from this template in a conventional PCR reaction followed by electrophoresis in a 2% agarose gel. For real-time PCR, primer sets were designed based on the sequence-verified clones from the University Health Network Microarray Center. Primer Express software version 2.0 (Applied Biosystem Inc., Foster City, CA, USA) was used to select primer sequences with a melting temperature of 58–601C and a low probability of forming primer-dimers and secondary structures (Table 6). The cycle conditions were: 951C for 10 min, and 40 cycles of 951C for 15 s followed by 601C for 1 min as recommended by ABI. Serial dilutions of an ovarian cell line cDNA (OVCAR-3) were included in all runs to serve as a standard curve for the relative quantification method. PCR reactions were run for each template, both with the specific and 18S-rRNA primers for 6–8 culture pairs (DHTand vehicle-treated) for each of the 19 selected genes. RT minus controls were included in all runs to verify the absence of genomic DNA contamination. At the end of the PCR, a dissociation reaction (951C for 15 s, 601C for 15 s, and 951C for 15 s) was performed to ensure amplification of a single product at the appropriate melting temperature. The correct size of the product was validated by 2% agarose gel electrophoresis. Real-time RT–PCR data were analysed with Pair Wise Fixed Reallocation Randomization Oncogene


212 Table 6 Primer pairs for real-time RT–PCR Accession no.

Unigene ID

Primers sequencea

AA036869

Hs.436317

H16276

Hs.32763

T98534

Hs.501735

AB037761

Hs.505104

R86925

Hs.435036

R96571

Hs.33433

H59158

Hs.301277

H57051

Hs.176626

H69632

Hs.199763

BI766823

Hs.290481

BG673977

Hs.436317

AI357692

Hs.491060

BF672232

Hs.520341

R13871

Hs.269764

R85241

Hs.154495

F: TCACGGCTGCCCATCAA R: TGGAGGTGAAAGCCATTCTTG F: TGTGACTAGGTTGTTCCCGTGAT R: ATGAGAAGAGGTTATTAGGAAGGCAAA F: GATTATCTGCTAGGAGACTATTGAGACTTG R: TCAACTCCACATCTATTGGTAAATCC F: GAGTCAGTTGAATTAGTACTGGACAAACA R: AGGCAAATAGACCCAATGAATGTAA F: TTTAAAGTACGAGTGTCTAAGAGGAAGTTT R: TGAGAAAAAGGATGTCCCTGTTC F: CCCAGGTGGCCTGTTACG R: CCACATAAAAGTGCTGACAGATTCC F: TGGAGACCCCCTTGTTGGA R: CAGAGCCCTGTGGTTTTTCTTC F: GATGCTTATACTTTTCCTCAAGGTAAGG R: CCAGGATTGTCAGAGAAAGTTGTAAA F: GGAGTCCCGATAAGCAGATCAC R: GGAAGCAGCTATCTTCTCATGCA F: GACTCGCCCAGCCAGAA R: GGTGTCTGAATCCAGCATTCC F: GTGATCATGCCTACTTTCTTCTTTGTATC R: GCACCTTCCTTCCATCAGAGTCT F: TGGATGTGAAAGGTATAGGTTTGTTTAA R: AGATATTCATGACTTTTTGACAGAAACTG F: CACGAGGAGATGATCTAGAAGTCAAC R: TCCCCAATTGAGAATGTAATGCT F: ACACTGAGAAGACCCTGTTTTGAAT R: TTCCAGAACCAGTCTGATGCAA F: GGCGCCGCTAGTTAAATCTTT R: GGCATCACCTCCAGGATCTTT

Amplicon size 61 87 122 128 88 71 65 98 76 72 79 122 126 86 69

Abbreviation: RT–PCR, reverse transcription–polymerase chain reaction. aForward (F) and reverse (R) sequences are provided.

Test (REST-XL) v.2 software (Pfaffl et al., 2002) (available at http://www.wzw.tum.de/gene-quantification/download.html). The expression ratio for the culture was compared to that of the standard curve (OVCAR-3 cDNA) and normalized to 18SrRNA levels. The results are expressed as the ratio of transcript measured in the DHT culture to that measured in the ethanol-treated (control) culture. Expression ratios were tested for significance by a REST-XL (Pfaffl et al., 2002) and were considered significant at Po0.05.

Tissue microarray A tissue microarray (TA-02) was prepared from tissues collected by the UHN Ovarian Tissue Bank. The array contained 168 duplicate cores (0.6 mm diameter) taken from formalin-fixed, paraffin-embedded tissue blocks. Of the 165 ovarian samples on the array, 149 were assessable by immunostaining and included 129 ovarian carcinomas, three LMPs (two mucinous, one serous) and 17 benign (seven benign tumors, eight endometriosis, three normal ovaries). The positioning of cores in the array was in four sectors of 7 12 (84 sample cores in each sector), with two of these sectors serving as duplicates. Positioning samples (kidney tissue cores) were included in two sectors for orientation purposes. Sections (5 mm thick) from the array were mounted onto Superfrost Plus microscope slides (Fisher Scientific, Ottawa, ON, Canada) and immunostained for ACHE (ab2802, Abcam Inc., Cambridge, MA, USA) or BACH2 (sc-14702, Santa Cruz Biotechnology Inc., Santa Cruz, CA, USA) using standard peroxidase immunohistochemical procedures. Briefly, TA-02 slides were subjected to microwave heat retrieval with 10 mM citrate buffer (pH 6.0) for 20 min at high Oncogene

setting. The sections were washed with PBS buffer, blocked with 0.3% hydrogen peroxide and 10% normal serum, before overnight incubation with anti-ACHE (1/50) or anti-BACH2 (1/500) antibodies at room temperature. Both antibody concentrations were optimized on ovarian tissue before their use on the tissue arrays. The slides were washed and incubated with appropriate biotinylated secondary antibodies at 1/200 dilution for 30 min at room temperature. Following washes, the slides were incubated with strepavidin–peroxidase for 30 min and staining was visualized using the chromagen NovaRed (Vector Labs. Inc., Burlington, ON, Canada). The staining was scored in a blinded manner based on the site of localization (nuclear, cytoplasmic), types of cells, the intensity (0–3) and the percentage of stained cells (classified as 0: none, 1: 1–24, 2: 25–49 or 3: X50%). Total staining score was calculated as the sum of the intensity and percentage scores (Camp et al., 2000). The results were tabulated and combined with clinical and histologic diagnosis data. For all tissue samples, the Spearman rank correlation was calculated between each of the four composite marker scores (BACH2 and ACHE, in both the cytoplasm and nucleus). For the 129 carcinoma patients, correlations were calculated between each of the composite marker scores and age at diagnosis, stage (according to the International Federation of Gynecology and Obstetrics staging system), histologic grade and debulking success (optimal vs suboptimal), with optimal debulking defined as less than 1 cm total residual tumor mass remaining after attempted debulking surgery. The exact Jonkheere–Terpstra test was used to assess for differences between tissue types (treating carcinoma/LMP/benign as ordinal) for each of the marker scores. The exact Wilcoxon


213 test was used to test for differences between histologic types for each of the marker scores. Canonical discriminant analysis was applied to the four scores (cytoplasmic and nuclear ACHE and BACH2) and seven groups (five carcinoma histological types, LMP and benign). Several other classification methods (including quadratic and kernel-based discriminant analysis as well as treebased methods (CART)) were also applied, but none improved on the classification accuracy of linear discriminant analysis. Stepwise discriminant analysis was used to eliminate scores that lacked discriminatory value.

Acknowledgements We thank Heather Begley, of the Ovarian Tissue Bank and Database, for assistance in obtaining patient material and clinical information, and Alicia Tone and Marjan Rouzbahman for help in preparing the tissue arrays. This work was supported by grant MOP-42437 from CIHR. AMK is partially supported by a Kristi Piia Memorial Fellowship in Ovarian Cancer. IJ is supported by DOD Grant OC040038 and NSERC Grant #203833. XZ is supported by funds raised by the Toronto Women’s Fashion Show.

References Anzik SL, Kononen J, Walker RL, Azorsa DO, Tanner MM, Guan X-Y et al. (1997). Science 277: 965–968. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM et al. (2000). Nat Genet 25: 25–29. Auersperg N, Maines-Bandiera SL, Dyck HG. (1997). J Cell Physiol 173: 261–265. Auersperg N, Maines-Bandiera SL, Dyck HG, Kruk PA. (1994). Lab Invest 71: 510–518. Barrios-Rodiles M, Brown KR, Ozdamar B, Bose R, Liu Z, Donovan RS et al. (2005). Science 307: 1621–1625. Beheshti B, Braude I, Marrano P, Thorner P, Zielenska M, Squire JA. (2003). Neoplasia 5: 53–62. Brown KR, Jurisica I. (2005). Bioinformatics 21: 2076–2082. Buchholz TA, Wu X, Hussain A, Tucker SL, Mills GB, Haffty B et al. (2002). Int J Cancer 97: 557–561. Camp RL, Charette LA, Rimm DL. (2000). Lab Invest 80: 1943–1949. Cardillo MR, Petrangeli E, Aliotta N, Salvatori LRL, Chang C, Castagna G. (1998). J Exp Clin Cancer Res 17: 231–237. Chada S, Rao BR, Slotman BJ, van Vroonhoven CCJ, van der Kwast TH. (1993). Hum Pathol 24: 90–95. Cramer D, Welch W. (1983). J Nat Cancer Inst 71: 717–721. Deng CX, Brodie SG. (2000). BioEssays 22: 728–737. Easton DF, Ford D, Bishop DT. (1995). Am J Hum Genet 56: 265–271. Edmondson RJ, Monaghan JM, Davies BR. (2002). Br J Cancer 86: 879–885. Eisen MB, Spellman PT, Brown PO, Botstein D. (1998). Proc Natl Acad Sci USA 95: 14863–14868. Evangelou A, Jindal SK, Brown TJ, Letarte M. (2000). Cancer Res 60: 929–935. Evangelou A, Letarte M, Jurisica I, Sultan M, Murphy KJ, Rosen B et al. (2003). Cancer Res 63: 2416–2424. Ford D, Easton DF, Stratton M, Narod S, Goldgar D, Devilee P et al. (1998). Am J Hum Genet 62: 676–689. Foss DL, Baarsch MJ, Murtaugh MP. (1998). Anim Biotechnol 9: 67–78. Gray RJ. (1992). JASA 87: 942–951. Hage JJ, Dekker JJ, Karim RB, Verheijen RH, Bloemena E. (2000). Gynecol Oncol 76: 413–415. Helzlsouer KJ, Alberg AJ, Gordon JB, Longcope C, Bush TL, Hoffman SC et al. (1995). J Am Med Assoc 274: 1926–1930. Hirashima Y, Kobayashi H, Suzuki M, Tanaka Y, Kanayama N, Terao T. (2003). J Biol Chem 278: 26793–26802. Hirte HW, Clark DA, Mazurka J, O’Connell G, Rusthoven J. (1992). Gynecol Oncol 44: 223–226. Hu W, Wu W, Nash MA, Freedman RS, Kavanagh JJ, Verschraegen CF. (2000). Anticancer Res 20: 729–733. Huang M, Page C, Reynolds RK, Lin J. (2000). Gynecol Oncol 79: 67–73. Imaoka S, Yoneda Y, Sugimoto T, Ikemoto S, Hiroi T, Yamamoto K et al. (2001). Cancer Lett 166: 119–123.

Isern J, Meseguer A. (2003). Biochem Biophys Res Commun 307: 139–147. Johnson G, Moore SW. (2000a). Int J Dev Neurosci 18: 781–790. Johnson G, Moore SW. (2000b). Appl Biochem Biotechnol 83: 131–144. Jones RB, Gordus A, Krall JA, MacBeath G. (2006). Nature 439: 168–174. Kanehisa M, Goto S, Kawashima S, Nakaya A. (2002). Nucleic Acids Res 30: 42–46. King MC, Marks JH, Mandell JB. (2003). Science 302: 643–646. Kobayashi A, Yamagiwa H, Hoshino H, Muto A, Sato K, Morita M et al. (2000). Mol Cell Biol 20: 1733–1746. Kohonen T. (1995). Self-Organizing Maps Springer Series in Information Sciences, vol. 30 Springer-Verlag: Berlin, 362pp. Kruk PA, Uitto V-J, Firth JD, Dedhar S, Auersperg N. (1994). Exp Cell Res 215: 97–108. Ku¨hnel R, De Graaff J, Rao BR, Stolk JG. (1987). J Steroid Biochem 26: 393–397. Ku¨hnel R, Rao BR, Poels LG, Delemarre JFM, Kenemans P, Stolk JG. (1988). Anticancer Res 8: 281–286. Lau K-M, Mok SC, Ho S-M. (1999). Proc Natl Acad Sci USA 96: 5722–5727. Li AJ, Baldwin RL, Karlan BY. (2003). Clin Cancer Res 9: 3667–3673. Li AJ, Lerner DL, Gapuzan ME, Karlan BY. (2005). Cancer Epidemiol Biomarkers Prev 14: 2919–2922. Linsley PS, Ochs V, Horn D, Ring DB, Frankel AE. (1986). Cancer Res 46: 6380–6386. Lu J, Xu WX, Wang SY, Jiang Y, Li CY, Cai WM et al. (2002). Sheng Wu Hua Xue Yu Sheng Wu Wu Li Xue Bao (Shanghai) 34: 95–98. Lutz LB, Jamnongjit M, Yang WH, Jahani D, Gill A, Hammes SR. (2003). Mol Endocrinol 17: 1106–1116. Marks A, Sutherland DR, Bailey D, Iglesias J, Law J, Lei M et al. (1999). Br J Cancer 80: 569–578. Modugno F, Ness RB, Wheeler JE. (2001). Ann Epidemiol 11: 568–574. Muto A, Tashiro S, Tsuchiya H, Kume A, Kanno M, Ito E et al. (2002). J Biol Chem 277: 20724–20733. Nishimoto A, Yu Y, Lu Z, Mao X, Ren Z, Watowich SS et al. (2005). Cancer Res 65: 6701–6710. Otasek D, Brown KR, Jurisica I. (2006). Society for Industrial and Applied Mathematics (SIAM) Conference on Data Mining. 22 April 2006; Bethesda, MD. pp. 1–10. Park JJ, Irvine RA, Buchanan G, Koh SS, Park JM, Tilley WD et al. (2000). Cancer Res 60: 5946–5949. Pfaffl MW, Horgan GW, Dempfle L. (2002). Nucleic Acids Res 30: e36. Oncogene


214 Rask K, Nilsson A, Brannstrom M, Carlsson P, Hellberg P, Janson PO et al. (2003). Br J Cancer 89: 1298–1304. Revankar CM, Cimino DF, Sklar LA, Arterburn JB, Prossnitz ER. (2005). Science 307: 1625–1630. Risch HA. (1998). J Natl Cancer Inst 90: 1774–1786. Robertson DM, Burger HG, Fuller PJ. (2004). Endocr Relat Cancer 11: 35–49. Rual JF, Venkatesan K, Hao T, Hirozane-Kishikawa T, Dricot A, Li N et al. (2005). Nature 437: 1173–1178. Schildkraut JM, Schwingl PJ, Bastos E, Evanoff A, Hughes C. (1996). Obstet Gynecol 88: 554–559. Scully R, Livingston DM. (2000). Nature 408: 429–432. Scully RE. (1995). J Cell Biochem Suppl 23: 208–218. Shaw PA, Rittenberg PVC, Shaw P, Brown TJ. (2001). Gynecol Oncol 80: 132–138. Shin S, Verma IM. (2003). Proc Natl Acad Sci USA 100: 7201–7206. Steinmetz R, Wagoner HA, Zeng P, Hammond JR, Hannon TS, Meyers JL et al. (2004). Mol Endocrinol 18: 2570–2582.

Oncogene

Stelzl U, Worm U, Lalowski M, Haenig C, Brembeck FH, Goehler H et al. (2005). Cell 122: 957–968. Sultan M, Wigle D, Cumbaa CA, Maziarz M, Glasgow J, Tsao MS et al. (2002). Bioinformatics 18(Suppl 1): S111–S119. Thellin O, Zorzi W, Lakaye B, De Borman B, Coumans B, Hennen G et al. (1999). J Biotechnol 75: 291–295. Tusher VG, Tibshirani R, Chu G. (2001). Proc Natl Acad Sci USA 98: 5116–5121. Venkitaraman AR. (2002). Cell 108: 171–182. Vidal CJ. (2005). Chem Biol Interact 157–158: 227–232. Windmill KF, McKinnon RA, Zhu X, Gaedigk A, Grant DM, McManus ME. (1997). Mutat Res 376: 153–160. Yang LV, Nicholson RH, Kaplan J, Galy A, Li L. (2001). Mech Dev 104: 105–111. Yeh S, Hu YC, Rahman M, Lin HK, Hsu CL, Ting HJ et al. (2000). Proc Natl Acad Sci USA 97: 11256–11261. Zakut H, Ehrlich G, Ayalon A, Prody CA, Malinger G, Seidman S et al. (1990). J Clin Invest 86: 900–908.