Tissue array analysis of expression microarray candidates identifies markers associated with tumor grade and outcome in serous epithelial ovarian cancer


  • Véronique Ouellet,

    1. Centre de Recherche du Centre hospitalier de l'Université de Montréal/ Institut du cancer de Montréal, Montreal, Canada
    Search for more papers by this author
  • Marie-Claude Guyot,

    1. Centre de Recherche du Centre hospitalier de l'Université de Montréal/ Institut du cancer de Montréal, Montreal, Canada
    Search for more papers by this author
  • Cécile Le Page,

    1. Centre de Recherche du Centre hospitalier de l'Université de Montréal/ Institut du cancer de Montréal, Montreal, Canada
    Search for more papers by this author
  • Abdelali Filali-Mouhim,

    1. Centre de Recherche du Centre hospitalier de l'Université de Montréal/ Institut du cancer de Montréal, Montreal, Canada
    Search for more papers by this author
  • Christian Lussier,

    1. Department of Pathology, Centre hospitalier de l'Université de Montréal, Montreal, Canada
    Search for more papers by this author
  • Patricia N. Tonin,

    1. Department of Medicine, McGill University, Montreal, Canada
    2. Department of Human Genetics, McGill University, Montreal, Canada
    3. The Research Institute of McGill University Health Centre, Montreal, Canada
    Search for more papers by this author
  • Diane M. Provencher,

    1. Centre de Recherche du Centre hospitalier de l'Université de Montréal/ Institut du cancer de Montréal, Montreal, Canada
    2. Division of Gynecologic Oncology, /Université de Montréal, Montreal, Canada
    Search for more papers by this author
  • Anne-Marie Mes-Masson

    Corresponding author
    1. Centre de Recherche du Centre hospitalier de l'Université de Montréal/ Institut du cancer de Montréal, Montreal, Canada
    2. Department of Medicine, Université de Montréal, Montreal, Canada
    • CR-CHUM/ICM, 1560, rue Sherbrooke est, Montreal, Quebec, Canada H2L 4M1
    Search for more papers by this author
    • Fax: +514-412-7703


Molecular profiling is a powerful approach to identify potential clinical markers for diagnosis and prognosis as well as providing a better understanding of the biology of epithelial ovarian cancer. On the basis of the analysis of HuFL expression data, we have previously identified genes that distinguish low malignant potential and invasive serous epithelial ovarian tumors. In this study, we used immunohistochemistry to monitor a subset of differently expressed candidates (Ahr, Paep, Madh3, Ran, Met, Mek1, Ccne1, Ccd20, Cks1 and Cas). A tissue array composed of 244 serous tumors of different grades (0–3) and stages (I–IV) was used in this analysis. All markers assayed presented differential protein expression between serous tumors of low and high grade. Significant differences in Ccne1 and Ran expression were observed in a comparison of low malignant potential and grade 1 tumor samples (p < 0.01). In addition, irrespective of the grade, Ccne1, Ran, Cdc20 and Cks1 showed significant differences of expression in association with the clinical stage of disease. While high level of Ccne1 have previously been associated with poor outcomes, here we found that high level of either Ran or Cdc20 appear to be more tightly associated with a poor prognosis (p < 0.001, 0.03, respectively). The application of these biomarkers in both the initial diagnosis and prognostic attributes of patients with epithelial ovarian tumors should prove to be useful in patient management. © 2006 Wiley-Liss, Inc.

Epithelial ovarian cancer (EOC) is the most lethal gynecological malignancy. Among the different histopathological subtypes present in ovarian tumors, serous is the most frequent and the second most lethal.1, 2 EOC tumors can be subdivided in 3 distinct groups: benign, borderline or low-malignant potential (LMP) and invasive ovarian tumors (TOV). Benign tumors are characterized by the proliferation of a single layer of cells without cellular atypia. LMPs and TOVs show epithelial proliferation as multilayers and present cytologic atypia. However, LMPs are morphologically distinct from TOVs, as they present no invasion of the ovarian stroma although microinvasion can be observed.3, 4, 5 EOCs are graded according to degree of differentiation: LMPs (referred to as either grade B or grade 0) are minimal deviation from their benign counterpart while well-differentiated tumors are grade 1 (G1), moderately differentiated are grade 2 (G2) and poorly differentiated are grade 3 (G3) carcinomas. Four stages are used according to the volume and extent of tumor spread. Stage I tumors are limited to one or both ovaries, stage II tumors are associated with pelvic extension, stage III tumors are characterized by spreading outside the pelvis into the abdominal cavity and patients with stage IV tumors present distant metastasis.1, 6

Approximately 20% of malignant epithelial tumors are LMPs. Serous LMPs are the most frequent subtype and have a better prognosis than serous TOVs. The 5-year survival rate of LMP patients reaches 90–95% in early stage disease and 55–70% for stage III disease compared to 30–40% for G3 TOVs.7 A portion of the LMPs that are considered as high-risk LMPs present a micropapillary structure, a stromal microinvasion or invasive implant and correlate to a worse prognosis compared to patients lacking these features.5, 8, 9 Recurrence, seen in less than 15% of patients with LMPs, generally remain LMPs, although in rarer cases, the disease becomes more invasive with a prognosis similar to that of TOVs.10

In EOC, several protein markers show differential expression between tumors of high and low grade, including Ccne1, CtnnB1, WT1, p53, c-kit, Bcl2, Her-2, Muc1, Krt7, clusterin, p21.11, 12, 13, 14, 15 However, none of these markers present differential expression between tumors of G0 and those of G1, the two most differentiated types of tumors. While many markers have been described, they have not been extensively exploited clinically, either for diagnosis or therapeutics. Few markers to date have been shown to correlate with the prognosis of patients with EOC. These markers, including Her-2, Ccne1, p27, bikunin, IGF-2, p53, p21 and Pgr, were assessed either by ELISA or immunohistochemistry16, 17, 18, 19, 20, 21, 22, 23 (and reviewed earlier24). However, no analysis on the sensitivity and specificity of these markers were performed in these studies.

In an earlier study, using cDNA microarrays to define the molecular profile of LMPs and G3 TOVs,25 candidate genes capable of discriminating between these 2 classes of EOCs were identified, and differential gene expression was validated for a subset of candidates, using quantitative PCR. On the basis of the notion that this differential gene expression may be reflected in protein expression levels, here we report an extensive immunohistochemical analysis of a subset of these candidate genes. For this purpose, we built a serous epithelial ovarian tumor tissue array composed of 244 specimens of G0, G1, G2 and G3. Immunohistochemistry results were related to various clinical parameters so as to determine their usefulness in the stratification of EOC tumors.

Material and methods

Patient tissue specimens and clinical data

After obtaining the appropriate consent, tumor samples were collected and banked following surgeries performed within the Division of Gynecologic Oncology at the Centre hospitalier de l'Université de Montréal (Hôpital Notre-Dame). Histopathology, tumor grade and stage were reviewed and scored by a pathologist, according to the International Federation of Gynecology and Obstetrics (FIGO).6 Clinical data were extracted from the Système d'Archivage des Données en Oncologie (SARDO), which includes entries on initial diagnosis, treatment, toxicity and clinical outcomes. We selected the samples of our tumor bank that corresponded to the following criteria: serous histopathology type, tumors of G0, G1, G2 and G3, all from chemotherapy naïve patients. Samples were collected between 1993 and 2003 and a Gynecologic Oncologist reviewed clinical data for all patients. Among the different grades, the mean age of patients at diagnosis were 50 (G0), 58 (G1), 59 (G2) and 63 (G3) years.

Serous epithelial ovarian cancer tissue array

Tissue samples were fixed in formalin and embedded in paraffin. A hematoxylin–eosin stained slide allowed the selection of 2 representative cores of 0.6 mm in diameter.26 The tissue array was composed of 58 LMPs, 8 tumors of G1, 59 of G2 and 119 of G3 serous EOC. The tissue array was then sectioned, stained with hematoxylin–eosin and received a final pathology review. For immunohistochemistry, the tissue array was cut in 5 μm sections and stained with appropriate antibodies and controls.


For immunohistochemistry analysis, the following antibodies were used: anti-Ahr rabbit polyclonal antibody (sc-5,579), anti-Cas goat polyclonal antibody (sc-1,709), anti-Ccne1 rabbit polyclonal antibody (sc-198), anti-Cdc20 mouse monoclonal (sc-5,296), anti-Cks1 goat polyclonal antibody (sc-12,986), anti-Glycodelin goat polyclonal (sc-12,290), anti-Madh3 rabbit polyclonal (sc-770), anti-Mek1 mouse monoclonal (sc-6,250), anti-Met rabbit polyclonal (sc-161) and anti-Ran goat polyclonal (sc-1,156). All antibodies were purchased from a single manufacturer (Santa Cruz Biotechnology, Santa Cruz, CA).


Tissue arrays were sectioned and stained by an immunoperoxidase method. Briefly, tissue sections were heated at 60°C for 30 min, deparaffinized in toluene and rehydrated in an ethanol gradient. Following a 3% H2O2 treatment to eliminate endogenous peroxidase activity, slides were submerged in boiling citrate buffer (0.01 M citric acid adjusted to pH 6.0) (J.T. Baker Philipsburg, NJ) for 15 min to unmask antigens. The sections were blocked with a protein blocking serum-free reagent (DakoCytomation, Mississauga, ON) and incubated with the different antibodies for 60 min at room temperature. The optimal concentration for each primary antibody was determined by serial dilutions. Tissues were incubated with either the secondary biotinylated antibody (DakoCytomation, Mississauga, ON) or with the rabbit anti-goat biotin-conjugated antibody (1:300) (sc-2774, Santa Cruz Biotechnology, Santa Cruz, CA) for 20 min, followed by incubation with a streptavidin-peroxidase complex (DakoCytomation, Mississauga, ON) for 20 min at room temperature. Reaction products were developed using diaminobenzidine containing 0.3% H2O2 as a peroxidase substrate. Nuclei were counterstained with hematoxylin, and all sections were observed by light microscopy at 20× magnification. Substitution of the primary antibody with phosphate-buffered saline served as a negative control. Protein expression was scored according to the extent (as a percentage of total malignant cells) and intensity (value of 0 for absence, 1 for low, 2 for moderate and 3 for high intensity) of staining based on manual visualization.25 All slides were independently analyzed in a blind study by 2 independent observers, and inter-rating was >90%. When strong differences in scoring between the 2 observers occurred, the core was re-evaluated to reach a consensus between the 2 observers. No decrease in staining intensity on older paraffin blocks has been observed.

Statistical analysis

Association between immunohistochemistry staining intensity and clinical variables (grades, stages, status of LMP tumors) was analyzed by logistic regression. Statistical significance of regression coefficients was monitored using Student t-test. The grades compared were 0, 1, 2 and 3. The LMP tumors were subdivided in 2 subgroups, representing LMPs of low and high risk of recurrence, according to the status of micropapillary structure, presence of invasive implants or stromal microinvasion. The stages were separated into 2 groups: early stage for stage I–II and advanced stage for III–IV. Significance of markers to predict survival of EOC patients was performed using Kaplan–Meier survival curve coupled to log-rank test. The threshold of the intensity used in the log-rank test was fixed based on the receiver operating characteristics (ROC) curves, and according to this value, 2 groups were assigned. ROC curves were developed for determining the optimal threshold of staining intensity that yielded the best possible sensitivity and specificity values of each marker to predict the invasive potential (G0 compared to G1, G2 and G3). Statistical significance was set at p < 0.05. Statistical analysis was performed using the STATS (logistic regression), ROC and SURVIVAL (Kaplan–Meier survival analysis, log-rank test, Cox analysis) packages from R version 2.01.


Staining of candidate markers protein in tumor tissue

Oligonucleotides microarray analysis were used to select a subset of candidate genes based on their ability to distinguish low grade from high grade EOC tumors and for which commercial antibodies were available (Supplementary Fig. 1).25 To study the implication of these candidates in the biology of ovarian cancer and their potential role as diagnostic and prognostic makers, we performed immunohistochemistry using a tissue microarray composed of 244 tumors of serous histopathology. This tissue array was stained using different antibodies representing 10 candidates, and the intensity of the staining was monitored. An image of a representative core of each grade of tumors (0–3) is presented in Figure 1. Nuclear and cytoplasmic staining was observed using antibodies against Ccne1, Cdc20, Cks1, Cas, Ahr, Paep, Madh3 or Ran proteins (Fig. 1). The staining seen for Met protein always involved the cytoplasm, but depending on the sample, could also include either cytoplasmic membrane or nuclear staining (Fig. 1), although alterations in the localization of staining did not correlate with any obvious clinical parameter (data not shown). Finally, an exclusive cytoplasmic staining of the Mek1 protein was observed (Fig. 1).

Figure 1.

Expression of markers on epithelial ovarian tumors tissues of different grades varying from G0 to G3. Representative images of the immunoperoxidase-stained slides are represented (20 × magnification). Staining was quantified based on both relative intensity and extent of staining. The markers Ahr, Paep, Madh3, Ran, Ccne1, Cdc20, Cks1 and Cas showed a nuclear and cytoplasmic staining; Met displayed a cytoplasmic membrane/cytoplasm or cytoplasm/nucleus staining; Mek1 presented cytoplasmic staining.

Statistical analyses of the correlation of staining intensity with histological grade

Expression of the different candidate proteins monitored by the frequency and the intensity of the staining was correlated with tumor grade (Table I). Ccne1 (p < 0.001) and Ran (p = 0.003) showed statistically differences in staining when tumors of G0 (LMP) were compared to G1. Differential expression of all candidates (except Cas) allowed the discrimination of G0 from G2 and G3. Statistically significant differences in staining were observed when the TOVs of G1 were compared to those of G2 and/or G3 respectively for Ahr (p = 0.040, 0.056), Paep (p = 0.003, 0.019), Mek1 (p = 0.003, 0.016), Cdc20 (p = 0.010, 0.008), Cks1 (p = 0.023, both) and Cas (p = 0.033 for G3). Finally, the TOVs of G2 showed an increase expression of Paep (p = 0.030) compared to G3. In this study, 23 samples came from high-risk LMP patients and 35 samples came from patients of low-risk LMPs. No marker tested in this study allowed the distinction of these 2 LMP subgroups (data not shown).

Table I. Statistical Analyses of Candidate Expression Determined by IHC and Association to Tumor Grade
ProteinTumorgradeMeanintensity ± SDSignificance, logistic regression
Ahr059 ± 160.3960.0040.0140.0400.0560.256
154 ± 13
269 ± 20
366 ± 17
Paep051 ± 130.469<0.001<0.0010.0030.0190.030
148 ± 17
267 ± 17
361 ± 16
Madh3044 ± 90.385<0.001<0.0010.1790.1600.949
147 ± 12
254 ± 14
354 ± 14
Ran065 ± 160.003<0.001<0.0010.6210.9860.249
182 ± 15
285 ± 16
382 ± 16
Met054 ± 120.224<0.001<0.0010.0880.2850.068
160 ± 15
271 ± 18
366 ± 17
Mek1043 ± 110.807<0.001<0.0010.0030.0160.440
144 ± 14
260 ± 16
358 ± 18
Ccne1044 ± 12<0.001<0.001<0.0010.2310.6500.081
170 ± 14
277 ± 16
372 ± 15
Cdc20048 ± 110.1100.0010.0010.0100.0080.680
142 ± 11
257 ± 18
356 ± 17
Cks1051 ± 130.940<0.001<0.0010.0230.0230.443
152 ± 11
265 ± 16
363 ± 14
Cas065 ± 200.2010.9360.0580.2090.0330.081
156 ± 17
265 ± 21
371 ± 19

Statistical analyses of the correlation of staining intensity and clinical stage

As the clinical staging reflects tumor spread, we were interested in determining whether the staining intensity of the different markers correlated with the stage. We grouped the clinical stages I–II in low-stage tumors (23 in LMPs, 17 in TOVs) and clinical stages III–IV as a high-stage tumor group (36 in LMPs, and 162 in TOVs). The expression of Ccne1 (p < 0.001), Ran (p = 0.001), Cdc20 (p = 0.016) and Cks1 (p = 0.007) showed a statistically significant difference (logistic regression, t-test) between early and advanced stage tumors.

Ran and Ccne1 are the most specific markers to definethe invasiveness of the tumor

To evaluate the prognosis potential of the markers, we determined the threshold of staining intensity by a ROC analysis, using the noninvasive tumors (LMPs) as one group and the invasive tumors (TOVs) as the other one. The threshold was fixed so as to give the best trade-off in term of specificity and sensitivity of the marker (Table II). The specificity represents the fraction of LMPs correctly classified as noninvasive tumors and the sensitivity defined the portion of invasive tumors correctly assigned as TOVs according to the staining intensity. Among the markers, Ccne1 and Ran are the ones that showed the best specificity, while Cks1 and Mek1 showed the best sensitivity (Table II).

Table II. Specificity and Sensitivity of the Markers for Diagnosis and Prognosis
MarkerROC thresholdDiagnosisSurvival >18 months
All tumors (G0, G1, G2, G3)All invasive tumors (G1, G2, G3)G0G2G3
Specificity (%)nSensitivity (%)nSpecificity (%)nSensitivity (%)nSpecificity (%)nSensitivity (%)nSpecificity (%)nSensitivity (%)nSpecificity (%)nSensitivity (%)n
  1. n, Number of patient. NI, noninformative, all patients survive.


Ran and Cdc20 staining correlates with patient outcome

Using the threshold of intensity determined by ROC analysis, we performed survival analyses based on Kaplan–Meier curve coupled to a log-rank test. Using tumors of all grades, the staining of all candidate proteins (except Cas and Ahr) correlated significantly with patient survival (p ≤ 0.01, except for Mek where p = 0.05). As grade and survival are related, we decided to determine the capacity of the markers to predict the survival of invasive tumors first in combining G1, G2 and G3 and second considering separately G2 and G3 (G1 tumors were excluded because of the small number of patients). Combining G1, G2 and G3, we were able to demonstrate that high expression of either Ran (p < 0.001) or Cdc20 (p = 0.03) was significantly correlated with a shorter survival in patients with invasive EOCs (Figs. 2 and 3). For the intragrade analysis in G2 tumors, Cks1 (p = 0.009) and Madh3 (p = 0.03) showed a correlation with the survival, while Ran showed a tendency (p = 0.06) (Figs. 2 and 3). Finally, the prognostic of G3 tumors correlated with the staining of Ran (p < 0.001) and Cdc20 (p = 0.04) (Figs. 2 and 3). We also performed a Cox analysis to determine whether the relation between intensity and survival depended of the ROC analysis threshold. Combining G1–G3, Cox analysis showed that high expression of Ran (p = 0.009) and Cdc20 (p = 0.003) was significantly correlated with poor patient survival.

Figure 2.

Relation between Ran expression and cumulative survival of patients with EOC. Kaplan–Meier graphical representation of survival curves demonstrated a poorer survival associated with high expression of Ran either when (a) all tumors (G0–G3) were analyzed (p < 0.001), (b) invasive (G1–G3) tumors only (p < 0.001) (c) G2 tumors only (p = 0.06) and (d) G3 tumors only (p < 0.001). Results are graphically represented as Kaplan–Meier curves. Patients with a follow up of less than 18 months were not used for this analysis. Log-rank test was used to verify the significance of the difference in survival.

Figure 3.

Relation between Cdc20, Madh3 and Cks1 expression and cumulative survival of patients with EOC. Kaplan–Meier graphical representation of survival curves demonstrated a poorer survival associated with high expression of Cdc20 in (a) invasive (G1–G3) tumors (p = 0.03) or in (b) G3 tumors only (p = 0.04) and with high expression of (c) Cks1 (p = 0.01) and (d) Madh3 (p = 0.03) in the G2 tumors only. Results are graphically represented as Kaplan–Meier curves. Patients with a follow up of less than 18 months were not used for this analysis. Log-rank test was used to verify the significance of the difference in survival.

Ran is a specific and sensitive marker to define the patient outcome

On the basis of the threshold of intensity previously set using ROC, we evaluated the specificity and sensitivity of each maker to predict those patients able to survive more than 18 months. When all TOVs were compared, Ccne1 and Ran (0.67 and 0.68 respectively) were the most specific markers to identify patients with a >18 months survival, while high Madh3, Mek1, Paep and Ran provided the highest sensitivity (0.72, 0.76, 0.72 and 0.72, respectively) and correlated with patients with a <18 month survival (Table II). Within the G0 group of tumors, Ccne1 and Ran were the most specific markers for prognosis (Table II). We were unable to evaluate the sensitivity of the markers, as all patients with G0 tumors have survived to date. Within the G2 tumors, Cas, Ran and Cdc20 were still the most specific markers, and Madh3, Met and Cks1 were the most sensitive (Table II). Finally, within the G3 tumors, Ahr and Ran were the most specific, and Ran, Paep and Mek1 were the most sensitive (Table II).


On the basis of our previous molecular profiling study,25 the purpose of this study was to identify protein markers whose expression varies with the malignancy potential of ovarian tumors and to correlate these differences in expression to relevant clinical parameters, such as tumor grade, stage and patient outcomes.

Since the difference between LMPs (G0) and G1 TOVs is primarily the presence of invasion, the finding of a molecular marker that distinguishes between these 2 groups of tumors is promising for its potential impact on clinical applications. We found that Ccne1 and Ran expression could be used to stratify G0 and G1 tumors while no markers to date were reported to allow this distinction. However, a study on a larger number of G1 tumors would be necessary to extend and confirm this finding. While all but one candidate (Cas) allowed the distinction between G0 and high grade (G2, G3) TOVs, overall most candidates showed little differences in marker expression between G0 and G1 or between G2 and G3 tumors. This may in part be due to the fact that in our microarray analysis we compared LMP tumors to the G3 tumors only, for distinction of aggressiveness.25 Subtle differences between G0 and G1 or G2 and G3 were not addressed in our previous analysis and may be difficult to identify based on immunohistochemistry that depends on strong differences in order to be significant. Furthermore, it has already been reported that closely related grades are difficult to separate based on their expression profile.15

In this study we show that the expression of all 10 tested markers (except Cas and Ahr) correlated with patient survival when all tumors (G0–G3) were analyzed. This is perhaps not surprising since the outcome of LMP disease is clearly superior. Among invasive tumors (G1–G3), we demonstrated that only high expression of Ran and Cdc20 was correlated with survival. Additionally, the expression of Ran, Cdc20, Madh3 and Cks1 were correlated with patient survival when tumors of the same grade were compared. Importantly, Ran was the only marker that correlated with the prognosis of the patient in each comparison. Together, these results are especially significant as there is a pressing need to define good molecular markers that predict patient outcomes in ovarian cancer, especially as clinical presentation is insufficient for proper stratification.

The differential expression of some proteins may also point to critical underlying alterations that contribute to EOC. For example, Ahr is a transcription factor that contributes to carcinogenesis associated with dioxin and aromatic hydrocarbons.27 Ahr is implicated in proliferation and cellular differentiation and can be associated with the estrogen receptors ER-α and ER-β, leading to estrogenic effects on sensitive cells.28, 29 The only previous report on Ahr RNA expression in ovarian tumors reports no differences in Ahr when normal ovarian surface epithelia were compared to TOVs.30 However, the fact that we see variable expression at the protein level appears to correlate with our previous observations of differential expression at the RNA level.25 The over-expression of Ahr in poorly differentiated tumors is consistent with its known activities, which include promoting proliferation as well as activating genes implicated in tumor progression, such as c-FOS and VEGF.29, 31

Paep, also known as Pp14 or glycodelin, is the major progesterone regulated lipocalin protein found at high levels in the plasma of patient with gynecologic malignancies and expressed by the tumors of reproductive organs.32, 33 In our study, we showed that Paep expression levels distinguished between low and high grades tumors. This result is consistent with the fact that this protein has immunosuppressive and angiogenic properties favoring the growth of the tumors expressing it.34 However, our results contradict a study showing that G1 serous ovarian tumors express higher levels of glycodelin protein than G2 and G3 tumors,35 although the source of antibody in that latter study was different from the one used here. While our study involved few G1 tumors, the lower grades LMPs are well represented and generally express less glycodelin.

Madh3, also known as Smad3, is a transcription factor involved in the TGF-β signaling pathway. In this study, we show that protein expression of Madh3 could distinguish LMPs from G2 and G3 tumors and correlates with survival. While Smad3 is commonly known as a suppressor of cellular proliferation, a recent study showed a dual role of Smad3 depending on the tumor grade in which the protein is expressed.36 Indeed, Smad3 has a tumor suppressor role in well-differentiated tumors, but confers a prometastatic ability in more aggressive tumor cells. This may explain the somewhat counterintuitive over-expression of the protein seen in the less differentiated tumors with a reduced expression in the patients with a longer survival.

During malignant transformation, the cell cycle progression is affected at different points and several of the candidates investigated in this report are implicated in this process (Ccne1, Cks1, Cdc20, Cas). Cyclin E (Ccne1) is required for the G1––S phase and has been reported to be over-expressed in many types of cancer. CCNE1 was shown to be over-expressed both at the mRNA and protein level in ovarian tumors compared to normal ovaries37 and was also associated with a poor prognosis.19, 20, 21, 22 These results are consistent with those seen in our analysis where Ccne1 represents the most specific marker for distinguishing noninvasive from invasive tumors and it also correlates with the stage of the disease. However, even though Ccne1 is one of the most specific marker for survival, its sensitivity remains very low (42%) in invasive tumors. In comparison, Ran had a similar specificity but was 30% more sensitive than Ccne1 to predict the survival of the patients. This further reinforces the potency of Ran as a prognosis marker in ovarian cancer.

Cks1 promotes the transition to anaphase through modulation of CDC20 transcription and favors cell cycle progression.8, 38 These two proteins have previously been implicated in different types of cancers.15, 39, 40, 41, 42, 43 We observed increased expression of these two proteins in the higher-grade tumors. Cks1 is a sensitive marker for diagnosis and prognosis although its specificity remains low. However, it showed a high significance in the survival of patients with G2 tumors, as all patients with low expression of Cks1 are still alive presently. We also demonstrated that expression of Cdc20 correlates with the survival of patient with TOVs. However, its sensitivity and specificity for prognosis is lower than Ran. Cas (for Cellular Apoptosis Susceptibility) is also involved in the cell cycle through the recycling of the transporter importin-α44 thus affecting the nuclear localization of cell cycle and apoptosis proteins, such as CDK, Cyclin/CDK complexes, as well as p53, Rb and NF-κB.45 Here, we report a higher expression of Cas in TOVs poorly differentiated compared to highly one. This is in line with previous reports showing that over-expression of Cas is associated with high stage and grade EOC tumors as well as with residual disease.45, 46, 47 Taken together these results confirm the importance of cell cycle regulation in cancer and point to key regulators for EOC.

Growth factors and hormones lead to proliferation and migration of cells by activating, among other pathways, the MAPK cascade. Mek1 is one of the components involved in this cascade and is known to be activated by Hgf signaling in ovarian cancer.48 Furthermore, over-expression of Mek1 has been demonstrated to be sufficient to promote proliferation.49 In our study, we showed that Mek1 and Hgf receptor (Met) were highly expressed in high compared with low grades tumors and that this protein is a sensitive marker for prognosis. However, our results are in disagreement with an earlier report showing higher expression of Met in G0 tumors compared to TOVs.50 This discrepancy might be related to either the small sample size and/or the varied histopathological subtypes included in this latter study. The MET oncogene is mutated or over-expressed in several types of tumors51 and its over-expression allows Hgf signaling, which leads to cellular proliferation, migration and invasion.52, 53 As LMP patients do not present tumor invasion, the lower expression of Met in LMP tumors may contribute to this feature. In primary culture, high MET levels in TOVs compared to normal ovarian surface epithelia has also been reported.48 Together, these results support the notion that Hgf signaling cascade may be involved in the deregulation of the proliferation of ovarian cancer cells of high grades. Functional assays would be necessary to confirm this hypothesis.

Met is known to localize either in the cytoplasm or at the cytoplasmic membrane. Surprisingly, we observed that Met was also localized in the nucleus of several tumors. While this phenomenon has been observed in melanocytic and lung tumors,54, 55 to our knowledge, this is the first report of a nuclear localization of Met in ovarian cells. Unexpected nuclear localization in cancer tissues has also been reported for other growth factor receptors, such as those of the ErbB family, and was associated with the proliferative status of the cells.56, 57, 58, 59, 60, 61, 62 Aberrant nuclear localization of membrane receptors in cancer cells may be due to an increase in nuclear core complex activity, since nuclear core complex proteins such as Ran are often over-expressed in several types of cancer.63, 64 Consistent with this hypothesis, we noticed an over-expression of Ran protein in TOVs of all grades compared to LMPs. Ran is a small GTP-binding protein essential for the nucleocytoplasmic transport through the nuclear pore complex and is also involved in nuclear organization and cell cycle progression.64 Its high expression in ovarian cancer cells also correlates with high stage disease. In addition to its capacity to distinguish between tumors of G0 and G1, Ran presented the highest combination of sensitivity and specificity for diagnosis (LMP or TOV) as well as prognosis.

In conclusion, this study provides another example of the usefulness of DNA microarrays to identify potential tumor markers able to stratify EOC tumors. The use of a serous tissue array composed of different grade tumors allowed the rapid study of different candidate proteins on a large number of samples. Here, we show the usefulness of Ahr, Paep, Madh3, Ran, Met, Mek1, Ccne1, Cdc20, Cks1 and Cas to stratify EOCs based on grade. Moreover, this study is the first to report that Ccne1 and Ran levels can distinguish between two groups of well-differentiated tumors (G0 LMPs and G1 TOVs). We also report on the potential use of Cdc20, Madh3, Cks1 and especially Ran, as prognostic markers in patient with invasive EOC. Our findings suggest that one or more of these candidate markers may eventually be useful clinically, and may point to important pathways that are deregulated in EOCs.


We are grateful to Louise Champoux, Lise Portelance, Manon de Ladurantaye, Jason Madore, Stéphanie Girard and Marise Roy for technical assistance. We are grateful to Dr. Christine Maugard for her help in statistical analysis and Dr. Lilia Sanchez for pathology review. We thank Dr. L. Masson for thoughtful comments. This work was supported by a grant from the Canadian Institutes of Health Research (CIHR) to A.-M.M.-M., P.N.T. and D.M.P. Tumor banking was supported by the Banque de tissus et de données of the Réseau de recherche sur le cancer of the Fonds de la recherche en santé du Québec (FRSQ). V.O. was supported by studentships from the CIHR and Canderel fund of the Institut du Cancer de Montréal. D.M.P. is a recipient of a Chercheur-Clinicien Senior and A.-M.M-M. is a recipient of a Chercheur National, both fellowships provided by the FRSQ.