Distinction between serous tumors of low malignant potential and serous carcinomas based on global mRNA expression profiling

YGYNO-70668; No. of pages: 11; 4C:

Gynecologic Oncology 96 (2005) 684 – 694 www.elsevier.com/locate/ygyno

Distinction between serous tumors of low malignant potential and serous carcinomas based on global mRNA expression profiling C. Blake Gilksa, Barbara C. Vanderhydenb, Shirley Zhuc, Matt van de Rijnc, Teri A. Longacrec,* a

Genetic Pathology Evaluation Centre of the Department of Pathology and Prostate Research Centre, Vancouver General Hospital, British Columbia Cancer Agency and University of British Columbia, Canada b University of Ottawa and Ottawa Regional Cancer Centre, Canada c Department of Pathology, Stanford University, 300 Pasteur Drive, Stanford, CA, 94305, USA Received 16 June 2004 Available online 30 December 2004

Abstract Objectives. The molecular pathogenesis of ovarian serous tumors of low malignant potential (S-LMP) is not well understood, although the collective data suggest that they arise through molecular mechanisms distinct from those leading to conventional serous carcinomas (S-Ca). To further examine the molecular differences between these two diseases, we studied the gene expression pattern of ovarian S-LMP and S-Ca using high-density spotted cDNA and tissue microarrays. Methods. Total RNA from 23 ovarian S-LMP and S-Ca was analyzed on 43,200 spot cDNA microarrays and the differential expression of proteins encoded by differentially expressed genes was validated using tissue microarrays. Results. Unsupervised hierarchical clustering analysis of filtered data showed a complete separation between S-LMP and S-Ca, based predominantly on a small set of genes expressed at higher levels in S-LMP than in S-Ca. Many genes previously identified as up-regulated in ovarian carcinoma relative to normal ovarian tissue were expressed at even higher levels in S-LMP. These genes included mucin-1, mesothelin, HE4, PAX 8, and apolipoprotein J/clusterin. Immunohistochemical staining of tissue microarrays confirmed higher expression of selected proteins encoded by these genes in the S-LMP. Few genes were expressed at a higher level in S-Ca; these included E2F1, topoisomerase IIa, and cyclin E, with higher levels of cyclin E protein confirmed by immunohistochemistry. Conclusions. S-LMP and S-Ca are distinguished at the molecular level by a relatively small gene set, suggesting the pathogenesis of SLMP as well as S-Ca may involve molecular pathways that escape detection by global gene expression profiling. In order to obtain biologically and clinically relevant information about the mechanisms involved in ovarian carcinogenesis, future studies based on molecular profiles of ovarian cancer should include analyses of low malignant potential tumors. Inclusion of such tumors is also critical to the evaluation of the efficacy of potential new diagnostic and/or therapeutic biomarkers. D 2004 Elsevier Inc. All rights reserved. Keywords: Ovarian cancer; Borderline tumor; Serous cancer; Serous tumor of low malignant potential; Gene expression profiling; Microarray

Introduction For a little over three decades, it has been recognized that in addition to the two traditional groups of benign and malignant serous ovarian neoplasms, there exists a third group of serous tumors. These tumors are characterized by a cellular proliferation that is atypical but not as morphologically deviant as in the usual serous carcinoma [1]. By * Corresponding author. Fax: +1 650 725 6902. E-mail address: [email protected] (T.A. Longacre). 0090-8258/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.ygyno.2004.11.039

definition, serous tumors of low malignant potential (also known as borderline tumors) lack stromal invasion but have other histologic features of malignancy (e.g., nuclear atypia, cellular stratification, mitotic activity). Not uncommonly, serous tumors of low malignant potential (S-LMP) are associated with advanced stage disease, with similar appearing lesions in the pelvis and intra-abdominal sites (extra-ovarian implants). Although the prognosis of most patients with serous ovarian cancer (S-Ca) is ultimately poor, patients with serous neoplasms of low malignant potential (S-LMP) have an unpredictable, but comparatively

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

more favorable prognosis, characterized by prolonged periods free of progression, despite high-stage disease. The category of tumor of low malignant potential was introduced by FIGO (International Federation of Gynecology and Obstetrics) in 1971 in recognition of these tumors [2] and this classification was subsequently adopted by the World Health Organization [3]. Despite a substantial body of literature concerning the clinical and pathologic features of these two disease processes, the pathogenesis and molecular events underlying the development of S-LMP and S-Ca are poorly understood. Several studies have found mutations in p53 and somatic or germline abnormalities of BRCA1 and/or BRCA2 in S-Ca but not in S-LMP, whereas point mutations in BRAF and microsatellite instability are more characteristic of S-LMP [4–7]. Data also indicate isolated loss of heterozygosity of chromosome X in S-LMP, whereas loss of heterozygosity of multiple chromosomes is found in S-Ca [8,9]. While most evidence indicates that these two diseases arise via separate pathways, the identification of trisomy 12 and K-ras mutations in both S-LMP and low-grade S-Ca suggests that S-LMP may not represent a truly independent entity and may, at least in some cases, form a continuum of tumor progression to carcinoma [10]. Indeed, it is now well documented that many of the patients with S-LMP who develop clinically progressive disease develop overtly invasive, albeit low grade, serous carcinoma [11–13]. The successes of DNA microarray technology in the subclassification of solid tumors based on characteristic gene expression patterns, and their stratification into prognostically relevant subgroups by the simultaneous evaluation of large numbers of genes has led to the tentative identification of sets of genes that are differentially expressed during development of specific tumor types [14–22]. Data garnered from microarray analyses have yielded new molecular classifications of lymphoma and breast cancer, among others [19,20,22]. Previous ovarian gene array studies have focused primarily on invasive ovarian cancer; no studies have expressly examined gene expression patterns of the low malignant potential tumors [21,23–32]. The present study was designed to determine the differences in gene expression levels among a series of

685

histologically well-defined primary ovarian serous tumors of low malignant potential (S-LMP) and ovarian serous carcinomas (S-Ca). Gene expression results were then correlated with the expression levels of selected proteins using tissue microarray analysis.

Materials and methods Tumor samples A total of twenty-three samples of snap frozen S-Ca (n = 13) or S-LMP (n = 10) was obtained from the Vancouver and Ottawa tissue banks of the National Ovarian Cancer Association of Canada and the Department of Pathology at Stanford University. These samples represented primary ovarian tumors from 21 patients. One patient had two samples taken from different areas of a S-LMP (OvCa2 and OvCa41). Another patient had samples taken from a grade 1 S-Ca (OvCa4) and a contralateral S-LMP (OvCa5) (Fig. 1). All cases were reviewed by two investigators (C.B.G. and T.A.L.) and classified by WHO criteria [3]. FIGO stage and grade (for the carcinomas) for each sample is provided in Table 1. Institutional Review Board approval was obtained by the respective institutions to analyze specimens by gene expression profiling and immunohistochemistry. Arrays Arrays were printed at the Stanford University School of Medicine according to the Brown laboratory protocols in the Stanford Functional Genomics Facility (http:// www.microarray.org/sfgf/jsp/home.jsp). For this study, all arrays used were from the same print run. These arrays have 43,000 elements or spots, representing more than 28,000 unique cDNAs; more than 9000 of these are characterized genes with the rest being ESTs. RNA isolation and gene expression profiles Prior to RNA isolation from the snap-frozen tissue, a frozen section was obtained and examined histologically to

Fig. 1. S-LMP exhibits papillary architecture, tubal-like epithelium, and cellular proliferation (A), but lacks significant cytologic atypia, stromal invasiveness, and aggressive metastatic potential of the usual S-Ca (B).

686

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

Table 1 Clinical and pathological features of S-LMP and S-Ca Case #

Age (years)

FIGO stage

Grade

% Tumor cell nuclei

OvCa1 OvCa2a OvCa4b OvCa5b OvCa7 OvCa8 OvCa9 OvCa10 OvCa11 OvCa14 OvCa19 OvCa22 OvCa23 OvCa27 OvCa29 OvCa30 OvCa32 OvCa34 OvCa38 OvCa41a OvCa43 OvCa44 OvCa45

75 31 37 37 48 62 83 22 50 50 34 27 52 77 51 52 65 65 63 31 55 57 36

III III III III III III III I III III I I III I III IV II III III III III I I

3 LMP 1 LMP 3 2 2 LMP 2 LMP LMP LMP 2 LMP 3 2 3 3 2 LMP 3 LMP LMP

75 58 83 61 87 79 88 83 82 87 73 88 81 83 94 89 51 70 85 57 65 96 66

LMP = low malignant potential. OvCa2 and OvCa41 are from different areas of a single patient’s unilateral ovarian tumor. b OvCa4 and OvCa5 are from right and left ovarian tumors, respectively, from a single patient. a

ensure that the tissue was representative of the diagnostic material on the permanent formalin-fixed and paraffinembedded tissue sections and that the tumor cells were the predominant constituent of the frozen tissue. The percent tumor cells in each sample was obtained by counting 100 cells from randomly selected fields of the frozen section slides; using an eyepiece graticle, the cells on intersecting lines of were scored as either tumor or not tumor (Table 1). In all cases, the tumor cells accounted for more than 50% of the total cells on the frozen section. Tissue was homogenized in TRIzol reagent (GibcoBRL/Invitrogen, Carlsbad, CA), centrifuged, washed with chloroform, and precipitated in isopropanol. Total RNA (50–75 Ag) was hybridized to anchored oligo-dT primers at 708C, snap-chilled, and incubated at 428C for 1 h in the presence of reverse transcriptase, Cy5-dUTP, and a deoxynucleotide mix deficient in dTTP. NaOH was used to destroy residual RNA. This process was repeated (using Cy3-dUTP) for the reference (Universal Human Reference RNA, Stratagene). Test and reference cDNA were pooled, and mixed with genomic repetitive DNA, yeast tRNA, and poly-dA to limit nonspecific binding. This mix was pipetted onto a microarray slide, coverslipped and hybridized overnight at 658C in 3xSSC/SDS. The next day, the slide was washed at increasing stringencies, and scanned in the red and green channels. After washing, the array slides were scanned on a GenePix 4000 microarray scanner (Axon Instruments, Foster

City, CA) and, after normalization of fluorescence intensities per array to control for experimental variation, fluorescence ratios (tumor/reference) were calculated using SCANALYZE software (Eisen, http://rana.lbl.gov). Spots with aberrant measurements due to array artifacts or poor quality were manually flagged and removed from further analysis. The primary data tables and the image files are stored in the Stanford Microarray Database (http://genome-www4.stanford.edu/MicroArray/SMD), and are available for downloading or analysis using SMD software. Data processing and statistical analysis After log transformation of the fluorescence ratios, gene spots were selected for further analysis using the following criteria: (1) only cDNA spots with a ratio of signal over background of at least 2.0 in either the Cy3 or Cy5 channel; (2) cDNA spots were measurable on at least 18 of 23 arrays (80% good quality data); (3) fluorescence ratios were presented as mean centered data, expressed relative to the mean fluorescence ratios for that spot from all 23 samples, and (in 4 separate analyses) only those cDNAs were selected that had an absolute value of the fluorescence ratio at least log2 1.75, log2 2.00, log2 2.25, or log2 2.50 times greater than the geometric mean ratio across all specimens, in at least three arrays. These criteria were established in order to: (1) identify genes for which the signal to background ratio was at least 2 (and eliminate genes expressed at such low level that expression could not be quantified accurately); (2) identify genes for which there was good-quality data in most samples (at least 80%); and (3) identify genes for which there was variation in gene expression among the samples, by selecting only those genes for which the absolute value of the red:green fluorescence ratio was log2 1.75–2.50 times the mean ratio for all specimens, in at least three samples. Hierarchical clustering analysis [33] and significance analysis of microarrays (SAM) [34] were performed as previously described [19,22]. Tissue microarray and immunohistochemistry Tissue microarrays were constructed from the archives of Surgical Pathology at Stanford University. Duplicate 0.6mm cores were obtained from formalin-fixed, paraffinembedded tissue blocks of 204 cases of S-LMP (n = 59) and S-Ca (n = 145). Four-micron-thick sections were cut from the array blocks and stained with antibodies to apolipoprotein J/clusterin (41D, Upstate Biotech), mucin-1 (VU4H5, Santa Cruz), cytokeratin 7 (OV-TL 12/30, Dako), and cyclin E (13A3, Novocastra) at dilutions of 1:400, 1:500, 1:40, and 1:50, respectively. Antigen retrieval was achieved by microwaving the slides in citrate buffer at pH 6.0. Staining was performed according to the manufacturer’s instructions for DAKO’s EnVision+ System, HRP (DAB) kit except that PBS, rather than Tris, was used as the wash buffer. This list of antibodies was chosen based on differential expression of

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

mRNA in S-LMP vs. S-Ca in the cDNA microarray experiments described previously, and the availability of antibodies suitable for immunohistochemical staining of formalin-fixed, paraffin-embedded samples. Immunohistochemical staining for mucin-1, apolipoprotein J (apo-J)/ clusterin, and cytokeratin 7 was scored as negative (fewer than 5% of tumor cells staining positively), weakly positive (5–50% of tumor cells showing weak to moderate immunoreactivity) or strongly positive (immunoreactivity of any intensity in N50% of tumor cells or N5% of tumor cells showing strong immunoreactivity). Immunostaining for cyclin E, a nuclear antigen, was scored as negative (b5% of tumor cell nuclei staining positively), weakly positive (5– 30% of nuclei positive), or strongly positive (N30% of nuclei positive). Cases in which immunoreactivity could not be assessed for technical reasons (failure of the tissue core to stick to the slide, no tumor in either of the cores, etc.) were excluded from further analysis. In cases where there was a discrepancy in the staining result recorded for the two cores, the higher score was accepted for that case (e.g., for a case with negative staining in one core and weak positive in the other, the case was considered to show weak positive immunoreactivity). Immunostaining results for S-LMP versus S-Ca were compared by chi-square test.

Results Global gene expression distinguishes S-LMP and S-Ca We used several different gene selection criteria for analysis of the expression profiles in S-LMP and S-CA. Data filtering resulted in the selection of 245 spots when a fluorescence ratio of log2 2.25 (sample fluorescence/mean fluorescence for all samples) in at least three spots was used as the selection criterion, and 541 spots using the less stringent fluorescence ratio of log2 2.00. The relationship between tumor type (i.e., S-LMP vs. S-Ca) and gene expression profile was analyzed using unsupervised hierarchical clustering analysis (Fig. 2). Hierarchical clustering analysis partitions genes into discrete groups and display of the clustered data with Treeview software (Eisen, http:// rana.lbl.gov) creates visually recognizable expression patterns, showing the cases separated into groups, with the relationship between cases or groups of cases presented as a dendrogram in which the length of the dendrogram arms is inversely proportional to the relatedness of gene expression patterns. This analysis revealed that the ten samples from SLMP clustered together on a single arm of the dendrogram, indicating that they were closely related to each other and were significantly different from the 13 samples of S-Ca, that clustered on a separate arm of the dendrogram (Figs. 2A and B). This clustering pattern was indistinguishable whether the clustering was done using expression data from 245 spots (filtered for fluorescence ratio Nlog2 2.25) or data from 541 spots (filtered for fluorescence ratio Nlog2 2.00).

687

When the filtering was repeated using higher stringency conditions, selecting only genes in which the fluorescence ratio was at least log2 2.5 times the mean of all samples in at least 3 arrays, only 74 spots satisfied this criterion. Unsupervised hierarchical clustering analysis based on these spots did not differentiate S-LMP from S-Ca (data not shown). When the filtering was repeated using much lower stringency conditions, selecting only genes in which the fluorescence ratio was at least log2 1.75 times the mean of all samples in at least 3 arrays, 1184 spots satisfied this criterion. Unsupervised hierarchical clustering analysis based on these spots separated S-LMP from S-Ca with the exception of a single case (OvCa5); this sample was from an S-LMP in a patient with a contralateral S-Ca (sample: OvCa4), and hierarchical clustering based on this gene set placed OvCa5 on the edge of the cluster of invasive carcinomas, more closely related to the S-Ca than the SLMP (data not shown). The remainder of this manuscript is based on the analysis resulting in 541 genes (i.e., data filtered for fluorescence ratio Nlog2 2.00). Unsupervised analysis of gene expression in S-LMP compared to S-Ca Visual inspection of the clustering results highlighted groups of genes responsible for the distinction between the two tumor types. One cluster of genes showed a high level of expression in S-CA and consisted predominantly of immunoglobulin genes expressed in a subset of the carcinomas (Fig. 2C). The immunoglobulin gene cluster was similar to that observed in other ovarian cancer classification studies [21] and was attributable to tumor infiltrating lymphocytes present in some of the S-Ca samples but not in the S-LMP samples. Most of the genes were expressed at higher levels in S-LMP than in S-Ca (Fig. 2D). Significance analysis of cDNA microarrays In addition to unsupervised hierarchical clustering, we used a supervised analytical method, SAM (Significance Analysis of Microarrays), to search for differentially expressed genes in S-LMP compared to S-Ca [34]. In this analysis, cases are identified as being either S-LMP or S-Ca and gene expression levels between the two groups are compared. Results of SAM analysis of the 541 spots identified using the filtering criteria noted above is shown graphically as Fig. 3A. A total of 217 spots showing significant differential expression was identified, with a median number of false significant results of 4.0. Some of the genes significantly differentially expressed in S-LMP, compared to S-Ca, are listed in Table 2 with a full listing provided online as Table2Supplemental.xls at http:// www.gpec.ubc.ca/index. The d score, or relative difference score, for a given spot is calculated as {Mean expression (SLMP) Mean expression (S-Ca)}/{S S o}, where S

688

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

689

Fig. 3. (A) SAM analysis of filtered data. Spots above the upper oblique line are expressed at significantly higher levels in S-LMP than S-Ca. Genes expressed at significantly higher levels in S-Ca than S-LMP would appear below the lower oblique line, but there were none identified. (B) SAM analysis of unfiltered data using a selected gene list composed of a set of genes specifically associated with ovarian cancer, proliferation, or cell cycle control. This shows that some genes on this list are expressed more highly in S-Ca than S-LMP, and appear below the lower oblique dotted line, but most of the genes are more highly expressed in the S-LMP than the S-Ca (shown as dots above the upper oblique dotted line).

reflects the standard deviation of the multiple expression measurements and S o is a small positive constant, as described by Tusher et al. [34]. This corrects for overestimation of the significance of differences in gene expression levels of genes expressed at low levels (compared to background) when only gene expression ratios (fold differences in expression levels) are used to compare the two groups. In Table 2, mRNA levels for apo-J and mucin-1 are seen to be higher in S-LMP than in S-Ca. Some genes appear more than once in this table, reflecting the presence of more than one dot corresponding to this cDNA on the microarray. Of note, the results for different dots corresponding to a single gene tend to show good correlation in expression levels. Mucin-5, subtype B, shows more variability in expression (6.5- to 46.6-fold greater expression in S-LMP than S-Ca) but also shows the greatest differential expression within this gene list. No significant differential gene expression was identified in a separate SAM analysis comparing low-stage S-LMP and high-stage S-LMP. We next performed SAM analysis of the unfiltered data set using a gene list consisting of genes previously suggested to be important in ovarian carcinoma or associated with cellular proliferation [24–32,35–45]. The latter were chosen based on the high mitotic rate of S-Ca compared to S-LMP. Parenthetically, the studies reporting on these genes had not included a comparison with S-LMP. In this analysis, 38 spots showing significant differential expression were identified, with a median number of false significant results of 4.7 (Fig. 3B). Some of the genes showing differential expression in S-Ca, compared to S-

LMP, are listed in Table 3, with a full listing provided online as Table3Supplemental.xls at http://www.gpec.ubc.ca/index. Genes that did not show significant differences in expression between S-Ca and S-LMP included bcl-2, prostasin, CD74, CD24, adenylate cyclase 5, and p16. As with results presented in Table 2, there is seen to be generally good correlation between expression levels of multiple spots measuring mRNA of the same gene. Immunohistochemical analyses of tissue microarrays In order to compare our RNA expression findings with protein expression data, immunohistochemistry was performed on a tissue microarray containing 145 ovarian carcinomas and 59 S-LMP with antibodies to four proteins that were differentially expressed in the S-LMP and S-CA groups (Table 3). Immunohistochemical staining generally paralleled mRNA expression, showing a significant trend to higher expression for three of these proteins in S-LMP than S-Ca, although the range of staining intensities in individual cases showed considerable overlap (Table 4). For example, mRNAs for keratin 7, mucin-1, and apo-J/clusterin were expressed at higher levels in S-LMP than S-Ca, while cyclin E was more highly expressed in S-CA than in S-LMP (Table 3). Analyzed by immunohistochemistry on a tissue microarray containing 145 ovarian carcinomas and 59 S-LMP, protein levels for keratin 7, mucin-1, and apo-J were greater in S-LMP than S-Ca, whereas protein levels for cyclin E were greater in S-Ca than S-LMP (Table 4). Thus, in each case, the protein expression levels in S-LMP vs. S-Ca, as

Fig. 2. Unsupervised hierarchical clustering of S-LMP and S-Ca, based on similarity of expression patterns of 541 genes (A). A closer examination of the dendrogram enlarged in (B) demonstrates the case to case variation by the length of the dendrogram arms leading to individual cases, at bottom. The length of the dendrogram arms to the S-LMP cluster and S-Ca cluster, at top, are relatively long, whereas the dendrogram arms linking cases within these clusters are shorter, reflecting lesser variation of gene expression within the cluster groups, compared to either case to case or cluster to cluster variation. An immunoglobulin cluster, present in most S-Ca, is highlighted by the blue vertical bar and expanded in (C). A separate representative gene cluster, highlighted by the yellow vertical bar and expanded in (D), depicts some of the genes overexpressed in S-LMP relative to S-Ca.

690

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

Table 2 Genes differentially expressed in S-LMP and S-Ca based on supervised (SAM) analysis of the filtered data seta Gene

Score (d)b

Fold differenceb (S-LMP/S-Ca)

mucin 5, subtype B claudin 10 mucin 5, subtype B mucin 5, subtype B kallikrein 6 B7 protein claudin 10 keratin 17 v-fos kallikrein 10 keratin 17 IGFBP7, insulin-like growth factor binding protein 7 integrin, beta 4 integrin, alpha 3 Apo J (clusterin) annexin A8 annexin A1 Apo J (clusterin) cyclin A1 jun B keratin 19 kallikrein 8 thrombospondin 1 heat shock 70kDa protein 8 kallikrein 3 (PSA) mucin 1

4.05 3.81 3.79 3.47 3.44 3.30 3.17 2.91 2.91 2.87 2.82 2.71

46.62 6.70 22.92 6.50 4.56 4.07 5.15 4.77 5.37 4.49 4.33 3.68

2.54 2.43 2.40 2.34 2.27 2.20 2.17 2.15 2.06 1.77 1.67 1.48 1.45 1.42

3.23 2.66 2.63 6.07 2.61 2.14 3.78 2.92 2.45 1.99 2.47 1.68 1.58 1.19

a A partial list is provided; the full list is provided online as Table2Supplemental.xls at http://www.gpec.ubc.ca/index. b The d score is a measure of difference of mean expression levels in SLMP vs. S-Ca for each gene, divided by the standard deviation of the expression levels, while the Fold difference is the ratio of the mean expression level of each gene in S-LMP vs. S-Ca.

assessed immunohistochemically, paralleled the mRNA expression levels.

Discussion Serous tumors of low malignant potential (S-LMP) and serous carcinomas (S-Ca) belong to the same histologic subtype of ovarian surface epithelial neoplasia, but the clinicopathologic features and biologic behavior of these two groups of tumors are distinctly different [1]. S-LMP are noninvasive and have a relatively favorable prognosis, characterized by an indolent clinical course, occasionally punctuated by recurrent disease after prolonged periods free of progression. A subset of S-LMP evolve over time to overtly invasive carcinoma, but this occurs infrequently and is almost always associated with an increased tempo of disease [11–13]. In contrast, S-Ca are frankly invasive, fully capable of metastatic behavior at the time of presentation (the vast majority are advanced stage at diagnosis), generally of intermediate or high histologic grade, and have an extremely poor prognosis.

The recent application of gene expression profiling studies in ovarian cancer research has led to the identification of a number of promising candidate ovarian cancer genes, but the expression patterns of these candidate genes have not been examined in tumors of low malignant potential. Our study, the first systematic examination of gene and protein expression in a series of histologically well-defined ovarian serous tumors of low malignant potential and serous carcinomas, demonstrates that many of the genes identified by previous gene expression studies to be of potential relevance to ovarian carcinogenesis are

Table 3 Genes differentially expressed in S-LMP and S-Ca based on supervised (SAM) analysis of unfiltered data using a selected gene seta Gene

Score (d)b

Fold differenceb (S-LMP/S-Ca)

S-LMP N S-Ca Rap1 keratin 17 keratin 17 annexin A1 cyclin D1 cyclin A1 cyclin D1 PAX8, paired box gene 8 ERBB2 (HER-2/neu) ERBB2 (HER-2/neu) mucin 1 ERBB2 (HER-2/neu) ERBB2 (HER-2/neu) WT1, Wilms tumor 1 MYC mucin 1 mesothelin cyclin C mucin 1 WFDC2, (HE-4) keratin 7 MYC p53 cyclin C p53

3.85 3.45 3.33 2.64 2.63 2.36 2.14 2.10 2.06 1.93 1.80 1.76 1.72 1.64 1.64 1.64 1.27 1.26 1.20 1.13 1.05 0.96 0.81 0.67 0.65

3.50 4.86 4.33 2.61 2.02 3.54 1.73 2.39 1.79 1.92 1.59 1.56 1.51 1.69 1.47 1.19 1.20 1.31 1.22 1.29 1.21 1.08 1.13 1.16 1.13

S-Ca N S-LMP E2F1 E2F1 MYBL2 topoisomerase II a OSF-2 topoisomerase II a cyclin E1 AKT2 topoisomerase II a CDK 2, cyclin dependent kinase 2

3.38 2.92 2.65 2.36 1.86 1.75 1.52 1.21 1.20 1.13

0.30 0.40 0.34 0.44 0.47 0.63 0.41 0.73 0.55 0.69

a A partial list is provided; the full list see provided online as Table3Supplemental.xls at http://www.gpec.ubc.ca/index.php?content= papers/papers.php. b The d score is a measure of difference of mean expression levels in SLMP vs. S-Ca for each gene, divided by the standard deviation of the expression levels, while the Fold difference is the ratio of the mean expression level of each gene in S-LMP vs. S-Ca.

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

691

Table 4 Selected proteins differentially expressed in S-LMP and S-Ca on immunostaining of tissue microarrays Antibody Apo-J/clusterin Mucin 1 Keratin 7 Cyclin E

S-LMP

S-Ca

P value

Negative

Weak

Strong

Negative

Weak

Strong

17 13 0 34

24 41 33 15

14 3 24 10

70 66 11 30

50 69 97 26

21 10 37 89

(31%) (23%) (0) (58%)

(44%) (72%) (58%) (25%)

(25%) (5%) (42%) (17%)

more highly expressed in S-LMP compared to S-Ca. Moreover, separation between these two biologically distinct tumor types is based predominantly on the differential expression of this gene set. S-LMP tumors exhibit comparatively higher levels of expression of many of the candidate ovarian cancer genes, while relatively few genes are significantly overexpressed in S-Ca compared to S-LMP. Prior gene expression studies of ovarian cancer have demonstrated increased expression of HE-4 protease inhibitor and mesothelin in ovarian carcinoma compared to normal ovary [46] and, utilizing a variety of other normal control samples and array templates, overexpression of mucin-1, keratin 7, PAX 8, apo-J, CD24, as well as other genes has been additionally identified [24,25,27,31]. On the basis of these data, several proteins encoded by these genes, including HE-4 and prostasin, have been targeted as potential serum biomarkers for ovarian carcinoma [27,35]. However, our gene expression profile of S-LMP and S-Ca demonstrates that HE-4, in addition to other candidate ovarian cancer genes (e.g., mucin-1, mesothelin, apo-J/clusterin, keratin 7, PAX 8) are expressed at even higher levels in tumors of low malignant potential than in the fully malignant carcinomas. Other genes initially reported as being highly expressed in ovarian carcinoma are either expressed at comparable levels in S-LMP and S-Ca (e.g., CD24, prostasin) or show only minimal up-regulation (OSF-2/periostin) in comparison to the tumors of low malignant potential [35]. A major problem in the design of virtually all of the ovarian gene expression profiling studies to date is the use of normal ovarian tissue to identify differential gene expression in ovarian cancer. Unlike other organ systems, the normal precursor cell of ovarian carcinoma is still under investigation. Most investigators favor a surface epithelial origin, but the evidence to support this hypothesis is circumstantial; indeed, whether carcinoma develops directly from the surface epithelium or occurs secondarily in epithelia that have undergone metaplasia is undetermined. The selection of the appropriate source of surface epithelium and method of isolation is also controversial. Because the ovarian surface epithelium constitutes a very small fraction of the normal ovary, it is difficult to obtain sufficient pure total RNA for good-quality expression studies, and that which is obtained is subject to frequent contamination by subjacent ovarian stroma [8]. In order to circumvent these problems, gene expression studies have utilized cell lines, short-term culture of surface epithelium, and enrichment procedures via RNA amplification, but each of these

(50%) (45%) (8%) (21%)

(35%) (48%) (67%) (18%)

(15%) (7%) (25%) (61%)

0.025 0.004 0.036 b0.001

techniques introduces bias and strongly influences the determination of differentially expressed genes in profiling studies [11–13,47,48]. Our study provides further indications of the limitations associated with the use of normal control tissue in ovarian cancer expression profiling and suggests that future molecular profiles of ovarian cancer include analyses of low malignant potential tumors before inferring any conclusions about carcinogenic pathways or the potential efficacy of new tumor markers. Based on our data, at least some of the gene subsets previously targeted as potentially important in ovarian carcinoma appear to be more specifically associated with serous differentiation than with malignant transformation. Previously, we and others have shown that HE-4 is relatively subtype-specific within the group of ovarian surface epithelial carcinomas, with an expression pattern predominantly limited to the serous subclass of ovarian tumors [21,26]. The down-regulated expression of this and similar genes in high-grade serous cancers relative to serous tumors of low malignant potential observed in this study likely reflects cellular dedifferentiation in association with increased malignant transformation and will pose significant limitations to the successful translation of these biomarkers into clinically useful screening and/or therapeutic tools. While most candidate ovarian cancer genes are overexpressed in S-LMP compared to S-Ca, a smaller list of genes is more highly expressed in S-Ca in comparison to the low malignant potential tumors. Not surprisingly, this gene set is composed of several proliferation-associated genes, including E2F-1, cyclin E, and CDK 2. Higher levels of cyclin E and CDK-2 expression are also seen at the protein level in immunohistochemical studies (Table 4) [43]. Within the group of ovarian carcinomas, cyclin E expression has been associated with a more favorable outcome [39,45] and topoisomerase II a has been cited as a potential target of chemotherapeutic inhibitors [49], while AKT2, which is amplified in 12% of ovarian cancers but not tumors of low malignant potential [36], has been associated with chemoresistance of ovarian carcinoma cells when overexpressed [50]. The occurrence of a relatively small number of differentially expressed genes in S-LMP in comparison to S-Ca may be related to the presence of tumor heterogeneity in SLMP and S-Ca. The morphologic heterogeneity of serous ovarian tumors has been recognized and well documented for many years. Mounting evidence indicates that the molecular pathogenesis of ovarian carcinoma is also

692

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

heterogeneous, giving rise to not one, but multiple diseases, each with differing molecular pathways. A high degree of genetic abnormalities in high-grade, advanced-stage carcinoma relative to low-grade, low-stage carcinomas has been consistently identified in many ovarian cancer studies and recent studies of serous low malignant potential tumors suggest that there may be molecular heterogeneity in these tumors as well [51]. This heterogeneity is particularly well illustrated by the different gene expression profiles of two separate samples obtained from two of the S-LMP in this study, one of which was associated with a low-grade serous carcinoma. An important corollary of this dmultiple molecular pathwayT model of ovarian surface epithelial transformation is the induction of multiple and diverse, possibly random genetic defects in these tumors, the latter of which may not be easily identified by a global genomic approach. Alternatively, and not necessarily exclusively, the critical gene(s) that ultimately prove to be biologically relevant to ovarian epithelial neoplasia may exert their effects through small differences in levels of expression that escape detection by global gene expression array analysis. In this regard, an inherent limitation of this study, as in most gene expression studies, is the relatively few numbers of samples examined. This limitation is partially circumvented by the utilization of a large number of samples for validation on tissue microarray, but large numbers of samples may be necessary in the initial screening profiles in order to pick out such small differences in gene expression. High throughput immunohistochemical expression profiling of four selected proteins (apolipoprotein-J/clusterin, mucin-1, keratin 7, and cyclin E) using ovarian tissue microarrays confirms the presence of differential mRNA and protein expression levels in S-LMP and S-Ca, even though differences in immunoreactivity are not sufficiently marked on a case-by-case basis to allow these markers to be used diagnostically to distinguish these two diseases. Positive correlations between mRNA and protein levels have been previously reported in other comparative mRNA and protein expression studies [21,26], although they are not invariably present; in a recent study, Ginestier et al. showed that there can be significant differences between mRNA expression levels as measured by hybridization to nylon membranes and protein levels as measured by immunohistochemistry [52]. This also appears to be the case for the two secreted proteins, prostasin and CD24, both of which show similar levels of mRNA expression in our series of S-LMP and S-Ca, but have been reported to show significantly increased protein expression in S-Ca compared to S-LMP, based on immunohistochemical staining of tissue arrays [53,54]. In conclusion, S-LMP and S-Ca are distinguished at the molecular level by a relatively small gene set. This set, chiefly composed of genes previously recognized as being overexpressed in ovarian carcinoma, is consistently expressed at higher levels in S-LMP than in S-Ca, whereas relatively few genes are significantly overexpressed in S-Ca relative to S-LMP. These results suggest that the molecular

pathogenesis of S-LMP as well as S-Ca may involve heterogeneous pathways and/or critical gene(s) that exert their effects through small differences in levels of expression that escape detection by global gene expression analyses. Further studies based on molecular profiles of serous ovarian cancer should include serous tumors of low malignant potential in order to obtain biologically relevant information about the molecular mechanisms involved in serous ovarian carcinogenesis. In addition, the expression patterns of the proteins whose presence has been confirmed in ovarian carcinoma should be studied in tumors of low malignant potential before drawing any conclusions about the potential efficacy of new tumor markers in the diagnosis and/or treatment of ovarian carcinoma.

Acknowledgments This work was supported by American Cancer Society Grant 96-50 (TAL) and the National Ovarian Cancer Association (CBG and BCV).

References [1] Longacre TA, Kempson RL, Hendrickson MR. Well differentiated serous neoplasms of the ovary. In: Hendrickson MR, editor. Surface epithelial neoplasms of the ovary. Philadelphia7 Hanley and Belfus, Inc.; 1993. p. 255 – 306. [2] Classification and staging of malignant tumours in the female pelvis. Acta Obstet Gynecol Scand 1971;50(1):1 – 7. [3] Serov SF, Scully RE, Sobin LH. Histologic typing of ovarian tumours. International histological classification of tumours. No. 9. Geneva7 World Health Organization; 1973. p. 17 – 54. [4] Singer G, Oldt 3rd R, Cohen Y, Wang BG, Sidransky D, Kurman RJ, et al. Mutations in BRAF and KRAS characterize the development of low-grade ovarian serous carcinoma. J Natl Cancer Inst 2003;95(6): 484 – 6. [5] Werness BA, Ramus SJ, Whittemore AS, Garlinghouse-Jones K, Oakley-Girvan I, Dicioccio RA, et al. Histopathology of familial ovarian tumors in women from families with and without germline BRCA1 mutations. Hum Pathol 2000;31(11):1420 – 4. [6] Haas CJ, Diebold J, Hirschmann A, Rohrbach H, Schmid S, Lohrs U. Microsatellite analysis in serous tumors of the ovary. Int J Gynecol Pathol 1999;18(2):158 – 62. [7] Caduff RF, Svoboda-Newman SM, Ferguson AW, Johnston CM, Frank TS. Comparison of mutations of Ki-RAS and p53 immunoreactivity in borderline and malignant epithelial ovarian tumors. Am J Surg Pathol 1999;23(3):323 – 8. [8] Dubeau L. Ovarian cancer. In: Vogelstein B, Kinzler KW, editors. The genetic basis of human cancer. 2nd ed. McGraw-Hill; 2002. p. 675 – 80. [9] Dodson MK, Hartmann LC, Cliby WA, DeLacey KA, Keeney GL, Ritland SR, et al. Comparison of loss of heterozygosity patterns in invasive low-grade and high-grade epithelial ovarian carcinomas. Cancer Res 1993;53(19):4456 – 60. [10] Haas CJ, Diebold J, Hirschmann A, Rohrbach H, Lohrs U. In serous ovarian neoplasms the frequency of Ki-ras mutations correlates with their malignant potential. Virchows Arch 1999; 434(2):117 – 20. [11] Zanetta G, Rota S, Chiari S, Bonazzi C, Bratina G, Mangioni C.

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28]

[29]

Behavior of borderline tumors with particular interest to persistence, recurrence, and progression to invasive carcinoma: a prospective study. J Clin Oncol 2001;19(10):2658 – 64. Silva EG, Tornos C, Zhuang Z, Merino MJ, Gershenson DM. Tumor recurrence in stage I ovarian serous neoplasms of low malignant potential. Int J Gynecol Pathol 1998;17(1):1 – 6. Longacre TA, McKenney JK, Tazelaar HD, Kempson RL, Hendrickson MR. Ovarian serous tumors of low malignant potential (borderline tumors): outcome-based study of 276 patients with long term (N5 year) follow-up. Am J Surg Pathol [in press]. van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Voskuil DW, et al. A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 2002;347(25):1999 – 2009. van ’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002;415(6871):530 – 6. Garber ME, Troyanskaya OG, Schluens K, Petersen S, Thaesler Z, Pacyna-Gengelbach M, et al. Diversity of gene expression in adenocarcinoma of the lung. Proc Natl Acad Sci U S A 2001; 98(24):13784 – 9. Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov JP, et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 1999;286(5439): 531 – 7. Nielsen TO, West RB, Linn SC, Alter O, Knowling MA, O’Connell JX, et al. Molecular characterisation of soft tissue tumours: a gene expression study. Lancet 2002;359(9314):1301 – 7. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, et al. Molecular portraits of human breast tumours. Nature 2000; 406(6797):747 – 52. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 2001; 98(19):10869 – 74. Schaner ME, Ross DT, Ciaravino G, Sorlie T, Troyanskaya O, Diehn M, et al. Gene expression patterns in ovarian carcinomas. Mol Biol Cell 2003;14(11):4376 – 86. Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 2000;403(6769):503 – 11. Bayani J, Brenton JD, Macgregor PF, Beheshti B, Albert M, Nallainathan D, et al. Parallel analysis of sporadic primary ovarian carcinomas by spectral karyotyping, comparative genomic hybridization, and expression microarrays. Cancer Res 2002;62(12): 3466 – 76. Ono K, Tanaka T, Tsunoda T, Kitahara O, Kihara C, Okamoto A, et al. Identification by cDNA microarray of genes involved in ovarian carcinogenesis. Cancer Res 2000;60(18):5007 – 11. Welsh JB, Sapinoso LM, Su AI, Kern SG, Wang-Rodriguez J, Moskaluk CA, et al. Analysis of gene expression identifies candidate markers and pharmacological targets in prostate cancer. Cancer Res 2001;61(16):5974 – 8. Schwartz DR, Kardia SL, Shedden KA, Kuick R, Michailidis G, Taylor JM, et al. Gene expression in ovarian cancer reflects both morphology and biological behavior, distinguishing clear cell from other poorprognosis ovarian carcinomas. Cancer Res 2002;62(16):4722 – 9. Schummer M, Ng WV, Bumgarner RE, Nelson PS, Schummer B, Bednarski DW, et al. Comparative hybridization of an array of 21,500 ovarian cDNAs for the discovery of genes overexpressed in ovarian carcinomas. Gene 1999;238(2):375 – 85. Sawiris GP, Sherman-Baust CA, Becker KG, Cheadle C, Teichberg D, Morin PJ. Development of a highly specialized cDNA array for the study and diagnosis of epithelial ovarian cancer. Cancer Res 2002; 62(10):2923 – 8. Tonin PN, Hudson TJ, Rodier F, Bossolasco M, Lee PD, Novak J, et al. Microarray analysis of gene expression mirrors the biology of an ovarian cancer model. Oncogene 2001;20(45):6617 – 26.

693

[30] Hough CD, Cho KR, Zonderman AB, Schwartz DR, Morin PJ. Coordinately up-regulated genes in ovarian cancer. Cancer Res 2001; 61(10):3869 – 76. [31] Hough CD, Sherman-Baust CA, Pizer ES, Montz FJ, Im DD, Rosenshein NB, et al. Large-scale serial analysis of gene expression reveals genes differentially expressed in ovarian cancer. Cancer Res 2000;60(22):6281 – 7. [32] Jazaeri AA, Lu K, Schmandt R, Harris CP, Rao PH, Sotiriou C, et al. Molecular determinants of tumor differentiation in papillary serous ovarian carcinoma. Mol Carcinog 2003;36(2):53 – 9. [33] Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 1998;95(25):14863 – 8. [34] Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 2001;98(9):5116 – 21. [35] Matei D, Graeber TG, Baldwin RL, Karlan BY, Rao J, Chang DD. Gene expression in epithelial ovarian carcinoma. Oncogene 2002; 21(41):6289 – 98. [36] Bellacosa A, de Feo D, Godwin AK, Bell DW, Cheng JQ, Altomare DA, et al. Molecular alterations of the AKT2 oncogene in ovarian and breast carcinomas. Int J Cancer 1995;64(4):280 – 5. [37] Bookman MA, Darcy KM, Clarke-Pearson D, Boothby RA, Horowitz IR. Evaluation of monoclonal humanized anti-HER2 antibody, trastuzumab, in patients with recurrent or refractory ovarian or primary peritoneal carcinoma with overexpression of HER2: a phase II trial of the Gynecologic Oncology Group. J Clin Oncol 2003;21(2): 283 – 90. [38] Cheng JQ, Godwin AK, Bellacosa A, Taguchi T, Franke TF, Hamilton TC, et al. AKT2, a putative oncogene encoding a member of a subfamily of protein-serine/threonine kinases, is amplified in human ovarian carcinomas. Proc Natl Acad Sci U S A 1992;89(19): 9267 – 71. [39] Farley J, Smith LM, Darcy KM, Sobel E, O’Connor D, Henderson B, et al. Cyclin E expression is a significant predictor of survival in advanced, suboptimally debulked ovarian epithelial cancers: a Gynecologic Oncology Group study. Cancer Res 2003;63(6):1235 – 41. [40] Hawes D, Liu PY, Muggia FM, Wilczynski S, Cote R, Felix J, et al. Correlation of p53 immunostaining in primary and residual ovarian cancer at the time of positive second-look laparotomy and its prognostic role: a Southwest Oncology Group ancillary study. Gynecol Oncol 2002;87(1):17 – 23. [41] Korkolopoulou P, Vassilopoulos I, Konstantinidou AE, Zorzos H, Patsouris E, Agapitos E, et al. The combined evaluation of p27Kip1 and Ki-67 expression provides independent information on overall survival of ovarian carcinoma patients. Gynecol Oncol 2002;85(3):404 – 14. [42] Page C, Lin HJ, Jin Y, Castle VP, Nunez G, Huang M, et al. Overexpression of Akt/AKT can modulate chemotherapy-induced apoptosis. AntiCancer Res 2000;20(1A):407 – 16. [43] Sui L, Dong Y, Ohno M, Sugimoto K, Tai Y, Hando T, et al. Implication of malignancy and prognosis of p27(kip1), E. Cyclin, and Cdk2 expression in epithelial ovarian tumors. Gynecol Oncol 2001; 83(1):56 – 63. [44] Thames HD, Petersen C, Petersen S, Nieder C, Baumann M. Immunohistochemically detected p53 mutations in epithelial tumors and results of treatment with chemotherapy and radiotherapy. A treatment-specific overview of the clinical data. Strahlenther Onkol 2002;178(8):411 – 21. [45] Rosenberg E, Demopoulos RI, Zeleniuch-Jacquotte A, Yee H, Sorich J, Speyer J, et al. Expression of cell cycle regulators p57(KIP2), cyclin D1, and cyclin E in epithelial ovarian tumors and survival. Hum Pathol 2001;32(8):808 – 13. [46] Wang K, Gan L, Jeffery E, Gayle M, Gown AM, Skelly M, et al. Monitoring gene expression profile changes in ovarian carcinomas using cDNA microarray. Gene 1999;229(1–2):101 – 8. [47] Zorn KK, Jazaeri AA, Awtrey CS, Gardner GJ, Mok SC, Boyd J, et al. Choice of normal ovarian control influences determination of differ-

694

[48] [49] [50]

[51]

C.B. Gilks et al. / Gynecologic Oncology 96 (2005) 684–694 entially expressed genes in ovarian cancer expression profiling studies. Clin Cancer Res 2003;9(13):4811 – 8. Haviv I, Campbell IG. DNA microarrays for assessing ovarian cancer gene expression. Mol Cell Endocrinol 2002;191(1):121 – 6. Herzog TJ. Update on the role of topotecan in the treatment of recurrent ovarian cancer. Oncologist 2002;7(Suppl 5):3 – 10. Yuan ZQ, Feldman RI, Sussman GE, Coppola D, Nicosia SV, Cheng JQ. AKT2 inhibition of cisplatin-induced JNK/p38 and Bax activation by phosphorylation of ASK1: implication of AKT2 in chemoresistance. J Biol Chem 2003;278(26):23432 – 40. Wenham RM, Lancaster JM, Berchuck A. Molecular aspects of ovarian cancer. Best Pract Res, Clin Obstet Gynaecol 2002;16(4):483 – 97.

[52] Ginestier C, Charafe-Jauffret E, Bertucci F, Eisinger F, Geneix J, Bechlian D, et al. Distinct and complementary information provided by use of tissue and DNA microarrays in the study of breast tumor markers. Am J Pathol 2002;161(4):1223 – 33. [53] Mok SC, Chao J, Skates S, Wong K, Yiu GK, Muto MG, et al. Prostasin, a potential serum marker for ovarian cancer: identification through microarray technology. J Natl Cancer Inst 2001;93(19): 1458 – 64. [54] Kristiansen G, Denkert C, Schluns K, Dahl E, Pilarsky C, Hauptmann S. CD24 is expressed in ovarian cancer and is a new independent prognostic marker of patient survival. Am J Pathol 2002;161(4): 1215 – 21.