Biopsies from normal human ovaries were obtained from 18 women undergoing laparotomy or laparoscopy for non- ovarian diseases/benign pelvic conditions. Ovarian epithelial tumours tissues were obtained from 36 women (approved by the Ethics Committee of the Medical Faculty, Göteborg University) (Table 1). Tissues were immediately washed in ice-cold saline, snap-frozen in liquid nitrogen and stored at -70°C until analysis 2627. All samples were examined by an experienced pathologist and classified according to FIGO for diagnosis.

Soluble tissue extracts were prepared from 45 tissue biopsies (Table 1), and immunoblotting was performed as previously described 2627. Thirty-five μg of total protein was loaded into each well for 1D SDS PAGE (4–12% Bis-Tris NuPage gels, Invitrogen, Paisley, UK). The blotting-membranes were incubated with blocking buffer containing the following primary antibodies from Cayman Chemical Co. (Ann Arbor, MI, USA): COX-1 (cat. no. 160110, monoclonal, dilution 1:1,000), COX-2 (cat. no. 160126, rabbit polyclonal, dilution 1:1,000), mPGES-1 (cat. no. 160140, rabbit polyclonal, dilution 1:500), EP₁ receptor (cat. no. 101740, rabbit polyclonal, dilution 1:500) and EP₂ receptor (cat. no. 101750, rabbit polyclonal, dilution 1:500). COX-1 and COX-2 electrophoresis standards from Cayman Chemical Co. (cat. no. 360100, 360120) were used as positive controls. The specificity of the COX-2 antibody was verified by the addition of a blocking peptide to the antibody solution prior to immunodetection (Cayman Chemical Co,. cat. no. 360106). Ovaries obtained from immature rats stimulated with pregnant mare serum gonadotrophin (PMSG) and human chorionic gonadotropin (hCG), (preovulatory ovaries) were used as positive control for mPGES-1 and EP_1–2. Immunoreactive proteins were visualized by chemiluminescence using alkaline-phosphatase-conjugated secondary goat anti-mouse or goat anti-rabbit antibodies (dilution 1:40 000), and the CDP-Star^® substrate (Tropix, Bedford, UK). The conditions were kept constant during the analysis of the immunoblots and each blot contained an internal control sample (see below and Figure legend 1 for details) in order to compare the levels of expression between blots 2627.

Frozen biopsies from 18 normal ovaries and ovarian tumours (Table 1) were cryosectioned (5–7 μm) and mounted on electrically charged glass slides, fixed in cold acetone at -20°C for 10 min and air dried in room temperature. The endogenous peroxidase activity was blocked by treatment for 4 min with peroxidase block from DAKO EnVision™+System, Peroxidase (AEC) kit (DAKO Corporation, USA). Primary antibodies against COX-1 (1:100), COX-2 (Cayman Chemical Co., cat. no. 160112, monoclonal, 1:100), mPGES-1 (1:50), EP₁ (1:50) EP₂ (1:50) and CK8 (1:400) diluted in TBS were applied overnight at 4°C, followed by biotinylated goat anti-mouse immunoglobulin (1:60) (VECTASTAIN^® ABC kit, Vector Laboratories, Inc., Burlingame, CA) for 45 min at 37°C and finally avidin-biotin-peroxidase complex (VECTASTAIN^® ABC kit, Vector Laboratories) at 37°C for 30 min. The enzymatic reaction was developed using AEC Chromogen Substrate Buffer (DAKO EnVision™+System, Peroxidase (AEC) kit, DAKO). Slides were counterstained with haematoxylin and mounted in 10% glycerol. The monoclonal COX-2 antibody used here was evaluated previously for immunohistochemistry by blocking experiments using a specific peptide 28. Antibodies against cytokeratin 8 (CK8) (monoclonal, Low Molecular Weight, DAKO, Copenhagen, Denmark), was used as a marker for epithelial cells. Primary antibodies were replaced by equal amounts of TBS for negative controls. The sections were viewed in a Nikon microphot FX microscope and photographed with a Nikon Coolpix 990 digital camera.

Semi-quantitative measurements of proteins from the immunoblots were made by densitometry (Fluor-S™ Multimager, Quantity One ver. 4.1.0., BioRad, Hercules, CA, USA). The optical density (OD) of each band was measured. An ovarian sample was used as an internal standard on each blot, and the measured value of this sample was set to 100%. The signal from each band was then compared to the standard and the obtained relative value was used for statistical analysis 2627.

Values are given as mean ± SEM. ANOVA followed by Fischer's LSD post-hoc test was used for statistical analysis of the data obtained by densitometric scanning. A p-value less than 0.05 was considered to be significant. Data is presented in a histogram and with a typical immunoblot of selected samples for each analysed protein.

Significant (p < 0.05) increases of COX-1, COX-2, mPGES-1 and EP₁ were demonstrated in the adenocarcinomas (AC) in comparison with normal ovarian tissue, benign adenomas and borderline tumours (B/BL) (Figure 1, 2, 3, 4, 5). EP₂ was also increased in comparison to the B/BL tumours. An increase, although not significant, of COX-2, was noticed in the B/BL group, compared to specimens from normal ovaries.

To examine if the pattern of expression was influenced by the degree of differentiation (grade) in the malignant group of tumours (AC), these tumours were dived into three groups; i) borderline tumours and highly differentiated AC (BL+H); ii) high/moderately and moderately differentiated AC (H/M+M); and iii) poorly differentiated and undifferentiated AC (P+U). The rationale to include highly differentiated AC (H) with the borderline (BL) type tumours was that these tumours are clinically considered to have more benign prognosis. The expressions of COX-1, mPGES-1 and EP₁ were all significantly elevated in both the H/M+M and P+U group compared to the more differentiated tumours in the BL+M group. (Figure 6). There was also an increase of EP₂ in the same groups, although not significant. COX-2 expression differed between tumours in the same histological group, but the majority of the tumours exhibited a strong signal for COX-2 (18/22 of the AC, 21/36 of all tumours), while a few tumours demonstrated a weak band (4/22 of the AC, 15/36 of all tumours). Amongst the four AC with low levels of expression, one was highly differentiated (H), two were in the group of high/moderately differentiated (H/M) and one was moderately differentiated (M). The expression patterns were further analyzed in the AC according to FIGO-stage (Figure 7). COX-2 demonstrated significant increases in both stage II and III cancers, compared to stage I. An increase was also shown for EP₁ in stage III tumours compared to stage I. No differences were observed for COX-1, mPGES-1 or EP₂.

A representative blot from each protein analyzed is presented in Figure 1, 2, 3, 4, 5. The major band for all the examined proteins migrated at their expected molecular weights (M_r). For mPGES-1, EP₁, and EP₂, additional bands with higher and lower M_r were observed in the group of malignant tumours (AC) (Figure 3, 4, 5). The additional M_r species most likely represents phosphorylated forms of mPGES-1 and EP_1–2. However, these bands for EP_1–2 might also represent splice variants of the receptors 29.

Tissue biopsies from nine normal ovaries were used for immunohistochemistry. Four of the ovaries were from fertile women and five from post-menopausal women undergoing surgery for benign non-ovarian diseases. Each ovary was consecutively sectioned and stained for cytokeratin 8 (CK8), a marker for epithelial cells (Figure 8A, 9A) and the individual proteins in the PGE₂ synthesis- and signalling pathway. DAPI was used for nuclear staining.

COX-1 was present in both surface epithelial cells (OSE) and epithelial cells lining the inclusion cysts (Figure 8B, 9C). Some staining was also observed in the stroma (Figure 8B). The signal for COX-2 in the OSE was negligible. However, weak staining was noticed in some areas of the underlying stroma (Figure 8C). An interesting observation was the clear staining for COX-2 in the epithelial cells of the inclusion cysts (Figure 9D). Staining for mPGES-1 was predominantly localized to cells in the stroma. Some of the epithelial cells on the ovarian surface demonstrated staining for mPGES-1 (Figure 8D). Both EP₁ and EP₂ were expressed in OSE, while only EP₂ was present in stroma cells, an expression pattern similar to that observed for mPGES-1 (Figure 8E,F).

Five benign cystadenomas (three serous and two mucinous), three BL (two serous and one mucinous), and three serous ovarian AC (one highly, one moderately and one poorly differentiated AC) were used for immunohistochemistry (IHC). Serial sections from each tumour were stained for the expression of COX-1, COX-2, mPGES-1, EP₁ and EP₂. CK8 was used as a marker for epithelial derived cells (Figure 10A, 11A). The results are presented as representative pictures from a serous borderline type tumour (stage I, Figure 10A–F) and a poorly differentiated serous AC (stage II, Figure 11A–F).

Distinct staining of COX-1 was demonstrated mainly in the epithelial cells of the BL tumour (Figure 10B). COX-2 exhibited an even more epithelial concentrated staining compared to COX-1 (Figure 10C). Both COX-1 and COX-2 was confined to the cytoplasm of the epithelial cells (inserts, Figure 10B–C). Microsomal PGES-1 was expressed predominantly in epithelial cells, but to some degree also in the stroma, although the staining was much less homogenous compared to COX-1 and COX-2 (Figure 10D). EP₁ and EP₂ demonstrated almost identical staining patterns with both receptors localised to the epithelial cells of the tumour (Figure 10E,F).

The staining pattern for the AC was different compared to that of the BL tumours. One exception was COX-1, which was still predominantly localised to the epithelial cells, some expression was also present in the stroma (Figure 11B). An increased stromal content was also noticed for COX-2 (Figure 11C). Intense staining was observed for mPGES-1, EP₁ and EP₂ in clusters of tumour cells (Figure 1111D–F).

Numerous experimental models have convincingly demonstrated that the increased production of PGE₂ in various tumours, including EOC, is involved in tumorigenesis 19. The reduced incidence of tumours by NSAIDs, as shown in epidemiological studies, supports the concept of the PGE₂ signalling pathway as an important factor in tumour formation and growth 7. The present study demonstrates, for the first time, increased contents and cell-specific localisation of mPGES-1 and the PGE₂ receptors EP_1–2 in EOC of different grades and stages compared to normal ovaries. The study also confirms the earlier demonstrated expression patterns of COX-1 1421 and COX-2 in EOC 19.

The ovulatory process exhibits several signs related to an inflammatory reaction, e.g. hyperaemia, extravasation of leukocytes, proteolytic- and collagenolytic activity 430. One of the key mediators in inflammation, the prostanoids, are also released during ovulation from the ovulating follicle and surrounding ovarian stroma, together with other autocrine/paracrine factors 31. In the human ovary, COX-2 is present in the ovulating follicle during the preovulatory phase, followed by interstitial localisation of the enzyme after ovulation 32. These prostanoids, in particular PGE₂, are thought to contribute to proliferation and repair of OSE at the site of rupture, in addition to their crucial role(s) in the intrafollicular regulation of the ovulatory process 30. Repetitive trauma to the epithelial cell surface, followed by exposure to mitogenic factors, is suggested to be an initiating event in ovarian tumourigenesis 631. When the EOC was established, elevated levels of the prostacyclins (PGI₂) and tromboxane (TxA₂) in plasma, as well as increased tumour contents of PGE₂ and TxB₂ were reported 1617. Furthermore, ovarian tumours with high tissue contents of PGE₂, PGF₂α and 6-keto-PGF₁α demonstrated a reduced response to chemotherapy 33.

COX-1 is constitutively expressed and plays a "housekeeping role" in many cells and has previously been demonstrated in normal OSE and in inclusion cysts 34 which are in line with the results of the present study. Furthermore, the contents of COX-1 was unaltered in the group of benign and borderline type tumours compared to normal ovaries, and the cellular localisation remained confined to the epithelial cells of these tumours. An increase of COX-1 was significant in the malignant tumours, but only in the less differentiated groups (H/M+M and P+U). COX-1 staining was no longer limited to the epithelial tumour cells since localisation to the underlying stroma also was observed. No correlation was found between FIGO stage and COX-1. Earlier studies have also reported an increase of COX-1 at both the mRNA and protein levels in EOC 14. A stimulatory effect on neo-vascularisation was proposed since a correlation was demonstrated between COX-1 expression and the expressions of angiogenic factors (VEGF, HIF-1α, Flk1) 14. Studies by Spinella and coworkers 12 demonstrated that endothelin (ET-1) increased COX-1 in ovarian carcinoma cells in vitro. Since ET-1 is known regulator of VEGF, this effect might be mediated by COX-1/PGE₂ in these cells. Interestingly, several studies on ovarian cancer cell lines have emphasised the contribution of COX-1 to ovarian tumourigenesis. In fact, COX-1 was the predominant isoform reported to be expressed 203435. A recent study 9 used mouse EOC cells (deletion of p53, or overexpression of c-myc/K-ras or c-myc/Akt) to demonstrate increased expression of COX-1 in these genetically engineered EOC cells. The conclusion, in accordance with the study by Kino and co-workers 35, was that COX-1 is the major regulator of PGE₂ synthesis in vitro.

The observation in the present study of COX-2 as more or less absent from normal OSE confirms results reported earlier 21343637. A clear staining for COX-2 was first observed in the inclusion cysts, and some, although not significant, elevation was noticed in the B/BL group. Li et al. 34 reported a significantly higher expression level of COX-2 in BL compared to benign tumours (B). The majority of the analysed (by IHC) tumours in their material were BL (69 BL, 18 B and 27 AC), which potentially influenced their results. They found, however, that the content was lower in BL compared to AC, which is in line with the present study. The AC demonstrated a clear elevation of COX-2, in particular in tumours of lower grades (H/M+M, P+U) in the present study. In contrast to COX-1, a significant difference with elevated levels of COX-2 was also found for later stages of tumours (stages II-III). To our knowledge, this study is the first to quantify the COX-2 protein expression by immunoblotting in ovarian tumours. Several earlier studies have shown by IHC that COX-2 is up-regulated in ovarian neoplasm using different scoring systems of staining intensity 20. Our results confirm the IHC data on COX-2 shown by others. The observed increase of COX-2 expression in EOC was correlated to poor differentiation, shorter time to tumour progression, resistance to chemotherapy, poor prognosis and significantly shorter survival time compared to patients with tumours staining negative for COX-2 223638, suggesting COX-2 to be an independent prognostic factor 36. COX-2 expression was also significantly correlated with microvessel density and/or VEGF expression in advanced-stage ovarian serous carcinoma 373940. In our study, absence of or weak COX-2 expression was noticed in the majority of the B/BL tumours and in only four AC, all of them of high to moderately differentiated grades, and presumably better prognosis.A recent study describing positive COX-2 staining of preneoplastic OSE adjacent to COX-2 negative neoplastic cells, presumably of higher grades 41, supports our findings. A correlation to histological subtype was not found for COX-2 in the present study. The cell-specific localisation of COX-2 in the B/BL group was almost exclusive in the epithelial cell, similar to that of COX-1.

Interestingly, both COX-1 and COX-2 were expressed in OSE lining inclusion cysts in the present study. This observation might suggest that the appearance of COX-2 in these cells is an early sign of an altered phenotype (mesothelial to epithelial transition, 31) with malignant potential. We have previously shown that the expressions of the adhesion molecules E-cadherin 4243 and claudin-3/4 27 and the transcription factor C/EBPβ44 correlated with malignancy in EOC. These proteins were also expressed in epithelial cells of inclusion cysts. Notably, the expression of C/EBPβ is a putative requirement for the regulation of COX-2 expression45. Li and co-workers 34 reported also the expression of COX-1 in OSE and inclusion cysts of the normal ovary, but both these two locations lacked COX-2 expression. Interestingly, in two independent studies COX-2 was not expressed in any of the ovarian cancer cell lines used(nine 34 and five 41), although Denkert and co-workers 36 showed such expression in cell lines included in their study. The difference in COX-2 expression in cell-lines and tissue/tumour biopsies is further strengthened since positive COX-2 staining was more often found at the advancing margin of tumour invasion and new metastatic lesion 34. An explanation to the findings of ovarian cancer cells in cell culture, not expressing COX-2, can be that cell lines are devoid of their basement membranes as well as the surrounding and supporting stroma cells. Recent studies have demonstrated important roles for stromal cells in tumourigenesis 46. An autocrine- paracrine regulatory pathway was described for breast carcinoma and surrounding stromal cells, involving growth factors and cytokines 46. Zhou and co-workers 47 demonstrated that stromal-epithelial interactions resulted in induction of C/EBPβ in malignant breast epithelial cells. One possibility is therefore that the expression of COX-2 in early stages of cellular transformation, e.g. in inclusion cyst formation, might be initiated by paracrine factors produced by the stromal cells, resulting in increased transcription, e.g. induced by C/EBPβ, of the cox2 gene. The cox2 promotor has previously been suggested to be of potential value for gene therapy in EOC, due to its tumour-specific activation in ovarian cancer cell lines in vitro 48. Furthermore, studies on ovaries from women undergoing prophylactic oophorectomy (due to mutations in BRCA1, BRCA2, MSH1 or MLH1) demonstrated that the increase of COX-2 correlated to loss of epithelial basement membrane 49. Additional studies are needed to establish the nature of interactions between epithelial cells in inclusions cyst and the surrounding basement membrane and stromal environment. Furthermore, the co-expression of COX-1 and COX-2 in epithelial cells of inclusion cysts and B/BL tumours, opens also the possibility fort an autocrine/paracrine regulation of the two COX isoforms by PGE₂.

The expression and regulation of mPGES-1 share many similarities with that of COX-2, e.g. induced by pro-inflammatory stimuli and hormones 50. This is in contrast to the two other known members of this family, mPGES-2 and cPGES, which are constitutively expressed in various cells and tissues 50. Both these isoforms are functionally closer linked to COX-1, while mPGES-1 has a marked preference to COX-2 50. Staining of the normal ovary for mPGES-1 revealed expression in the stroma, while the OSE cells were mostly negative. This pattern was also observed in the BL with some staining of epithelial cells or cells associated to the epithelium. A significant increase was observed in the AC, but not in B/BL. This increase was correlated to grade, but not to stage. Intense staining was demonstrated in a cluster-like pattern, predominantly in the tumour stroma. This might to some extent reflect the presence of immune cells recruited to the tumour site as a part of an inflammatory reaction. A local, anti-inflammatory response to IL-1α was suggested 51 to involve production of cortisol in the normal OSE by an increase of 11β-HSD, while this enzyme was not expressed in cell lines derived from EOC 52. Infiltration of macrophages into EOC was described by Klimp and co-workers 53 and their study demonstrated increased levels of COX-2 in tumour-associated macrophages. The role of mPGES-1 for PGE₂ production in these cells are essential since isolated peritoneal macrophages from mPGES-1 null mice produced minimal amounts of this prostanoids in response to pro-inflammatory stimuli (LPS) 50. Similar results were obtained with macrophages from C/EBPβ-deficient mice as a result of ablated mPGES-1 expression. This emphasise the role of inflammation in EOC since C/EBPβ has been demonstrated to be major regulator of pro-inflammatory genes, in addition to COX-2 and mPGES-1 50. The cellular localization of mPGES-1 in the present study, suggested a close connection to the cells expressing EP_1–2 receptors, which also support an autocrine/paracrine regulatory pathway. An earlier study 35 showed the content of mPGES-1 in ovarian cancer cell lines, but the present study is the first, to our knowledge, to demonstrate the expression of mPGES-1 in EOC.

Previous studies have demonstrated that PGE₂ acted mainly via the EP₂ receptor in female reproduction and in tumourigenesis, while EP₁ was more involved in neuronal functions 29. Significant increases were found for the expressions of both EP₁ and EP₂ in AC compared to benign tumours in the present study. Interestingly, one of the findings in the present study was a difference in the distribution of EP-receptors between stroma and tumour tissue. The majority of the EP_1–2 immunoreativity was found in OSE while only EP₂ could be demonstrated in the stroma. First, this suggests target cells for PGE₂ in both these cell compartments and secondly a separate function, mediated by the EP₂ receptor in the stroma. In addition, the immunblotting experiment demonstrated multiple bands for the two receptors. These bands could represent phosphorylated forms of the receptors as well as splice variants 29, and therefore have important implications for cell signalling. The two receptors regulates different intracellular signalling pathways, i.e. IP₃/Ca²⁼ (EP₁) and cAMP (EP₂). Phosphorylated residues and splice variants can potentially alter sensitivity, responsiveness and preferred signalling system within the cells 29, all of which can contribute to enhanced tumour growth. However, additional experiments are needed to explore these possibilities.

Staining of cluster of cells in the AC, demonstrated for primarily for mPGES-1 and EP_1–2, were earlier reported for both COX-1 and COX-2 36. Previous studies have actually localised COX enzymes to a large variety of cells, including both normal and tumour epithelial cells, normal and tumour vascular endothelium and smooth muscle cells, macrophages and fibroblast cells. The clearly defined staining of populations of malignant cells, found here and by others, may implicate specific functions, such as angiogenesis and invasion or a more active anti-tumour response for mPGES-1 and EP_1–2 in different populations of the malignant cells. Cluster like expression of COX-2 was recently reported to correlate to angiogenesis in prostate cancer 54.