The ErbB signalling pathway: protein expression and prognostic value in epithelial ovarian cancer

Ovarian cancer is the most frequent cause of death from gynaecological cancer in the Western world. Current prognostic factors do not allow reliable prediction of response to chemotherapy and survival for individual ovarian cancer patients. Epidermal growth factor receptor (EGFR) and HER-2/neu are frequently expressed in ovarian cancer but their prognostic value remains unclear. In this study, we investigated the expression and prognostic value of EGFR, EGFR variant III (EGFRvIII), HER-2/neu and important downstream signalling components in a large series of epithelial ovarian cancer patients. Immunohistochemical staining of EGFR, pEGFR, EGFRvIII, Her-2/neu, PTEN (phosphatase and tensin homologue deleted on chromosome 10), total and phosphorylated AKT (pAKT) and phosphorylated ERK (pERK) was performed in 232 primary tumours using the tissue microarray platform and related to clinicopathological characteristics and survival. In addition, EGFRvIII expression was determined in 45 tumours by RT–PCR. Our results show that negative PTEN immunostaining was associated with stage I/II disease (P=0.006), non-serous tumour type (P=0.042) and in multivariate analysis with a longer progression-free survival (P=0.015). Negative PTEN staining also predicted improved progression-free survival in patients with grade III or undifferentiated serous carcinomas (P=0.011). Positive pAKT staining was associated with advanced-stage disease (P=0.006). Other proteins were expressed only at low levels, and were not associated with any clinicopathological parameter or survival. None of the tumours were positive for EGFRvIII. In conclusion, our results indicate that tumours showing negative PTEN staining could represent a subgroup of ovarian carcinomas with a relatively favourable prognosis.

in ovarian cancer, but remains unclear. A recent meta-analysis revealed that abnormal expression of these markers appears to be associated with poor 5-year survival, but this is not a uniform finding (Crijns et al, 2003).
The EGFR variant III (EGFRvIII) lacks exons 2 -7 of the extracellular domain of the receptor. Although EGFRvIII is unable to bind ligand, it is constitutively phosphorylated and able to activate downstream signalling pathways (Pedersen et al, 2001). EGFRvIII expression is thought to confer resistance to cisplatin and paclitaxel (Nagane et al, 1998;Montgomery et al, 2000). The two studies investigating EGFRvIII expression in ovarian cancer show conflicting results Lassus et al, 2006).
The aim of the present study was to investigate the prognostic significance of EGFR and HER-2/neu, and their downstream targets AKT, ERK and PTEN in a large series of 232 epithelial ovarian cancer patients using the tissue microarray (TMA) platform. In addition to immunostaining, we determined the expression of EGFRvIII in more detail in a subset of 45 ovarian tumours using the more sensitive method RT -PCR.

Patients
Since 1985 all clinicopathological and follow-up data of 329 epithelial ovarian cancer patients treated at the University Medical Centre Groningen have been prospectively stored in a database. All patients gave informed consent for data storage and tumour collection, and studies were conducted in accordance with the Declaration of Helsinki principles and institutional review board policies. For the current study all consecutive chemonaive ovarian cancer patients for whom sufficient paraffin-embedded tumour tissue and complete follow-up data were available were selected (n ¼ 232).
Patients were surgically staged according to FIGO (International Federation of Gynaecology and Obstetrics) criteria (Cancer Committee of the International Federation of Gynaecology and Obstetrics, 1986). Optimal and suboptimal debulking was defined as the largest residual tumour lesions having a diameter of o2 cm orX2 cm. The histology of all carcinomas was determined by a gynaecological pathologist according to WHO (World Health Organization) criteria (Scully, 2004).
Response to chemotherapy was evaluated according to WHO criteria (World Health Organization, 1979). When indicated, intervention surgery was performed after three cycles of chemotherapy, while until 1996 second-look surgery was regularly performed after six cycles of chemotherapy.

TMA construction and immunostaining
Tissue microarrays were constructed as described previously (de Graeff et al, 2006). In total, four tissue cores from 232 primary tumours and 45 paired tumours obtained at second-look surgery or surgery for recurrent disease were included on eight TMAs.
Antigen retrieval methods, primary antibodies and detection techniques are provided as supplementary data. Sections (4 mm) were de-paraffinised in xylene and endogenous peroxidase was blocked by incubation in 0.3% hydrogen peroxide for 30 min. After antigen retrieval, slides were incubated in normal goat serum (HER-2/neu), horse serum (EGFR, pEGFR), bovine serum (phosphorylated AKT (pAKT), phosphorylated ERK (pERK), PTEN, total AKT) or blocking solution (Dako, Cambridgeshire, UK) for EGFR. For pEGFR, pAKT, pERK and PTEN staining, endogenous avidin and biotin activity was blocked using a blocking kit (Vector Laboratories, Burlingame, UK). HER-2/neu staining was performed in a Dako autostainer (Dako). Staining was visualised by 3 0 3-diaminobenzidine tetrahydrochloride and sections were counterstained with haematoxylin. EGFRvIII staining was kindly performed by Dr A Jungbluth, Ludwig Institute for Cancer Research, New York, USA.
Positive controls included separate TMA slides containing multiple tumour and normal tissues for EGFR and pEGFR, sections from tumours with known marker expression for HER-2/neu and PTEN, ovarian cancer cell line A2780 for AKT, pAKT and ERK, and glioblastoma cell line U87 transfected with an EGFRvIII plasmid for EGFRvIII staining (Jungbluth et al, 2003). Negative controls were obtained by omission of the primary antibody, and by incubation with normal rabbit IgG for total AKT. All control experiments gave satisfactory results. Antigen preservation was verified by vimentin staining, which was positive in all tumour and control samples.
Evaluation of immunostaining was independently performed by two observers (KAH and PDG), blinded to clinical data. The agreement between the two observers was490%. Discordant cases were reviewed with a gynaecological pathologist and were re-assigned on consensus of opinion.
HER-2/neu staining was scored according to the HercepTest protocol (Lebeau et al, 2001), and was considered positive when410% of tumour cells showed moderate or strong membrane staining. For EGFR and EGFRvIII, tumours demonstrating410% membrane staining were considered to show overexpression (Elie et al, 2004;Skirnisdottir et al, 2004;Cunningham et al, 2005). Overexpression of p-EGFR was defined as 45% membrane or granular cytoplasmic staining (Han et al, 2004). Tumours were considered positive for AKT or ERK if 410% of tumour cells showed positive cytoplasmic and/or nuclear staining (Kurose et al, 2001). Phosphatase and tensin homologue deleted on chromosome 10 staining in tumour sample was scored relative to staining in vascular endothelium (Gimm et al, 2000;Choe et al, 2003), and was regarded as negative when staining was completely absent in tumour tissue but present in vascular endothelium.

RT -PCR for EGFRvIII
We performed RT -PCR analysis on a subset of 45 frozen tumour samples, of which 35 showed positive immunostaining for (p)EGFR or downstream targets and 10 were completely negative. Positive controls included a glioblastoma tumour sample expressing both the wild-type EGFR (wtEGFR) and EGFRvIII, and a cell line transfected with an EGFRvIII plasmid (Jurkat.EGFRvIII; Bremer et al, 2005).

Statistical analysis
Statistical analysis was carried out using the SPSS 12.01 software package. Cut-off points for positive marker expression were determined a priori. All cases with o2 evaluable cores were excluded from analysis.
Comparisons between paired tumour samples obtained before and after chemotherapy were made using the Wilcoxon rank sum test. Associations between markers, and between markers and clinicopathological characteristics were performed using the w 2 or Fisher's exact test, where appropriate.
The end points investigated were progression-free and diseasespecific overall survival (PFS and OS), defined as the time from primary surgery until progression/relapse of the disease or death of ovarian cancer, respectively. Response to platinum-based chemotherapy could only be evaluated in patients who had measurable disease after primary surgery and/or during first-line chemotherapy (n ¼ 130), and was defined according to WHO criteria (World Health Organization, 1979).
For univariate and multivariate survival analysis Cox proportional hazards model was used. Categorised covariates that were significant in univariate analysis were entered simultaneously into the multivariate model. Response to chemotherapy was analysed using logistic regression analysis. For this analysis, response was entered as a categorical variable (complete and partial response vs stable and progressive disease). P-values o0.05 were considered statistically significant.

Patients
A total of 232 patients (median age 57.8 years, range 22 -90) treated at the Groningen University Medical Centre between 1985 and 2002 were selected for the present study (Table 1). Of them 64 (27.6%) patients presented with stage I/II disease and 166 (71.5%) patients with stage III/IV disease. Optimal debulking was achieved in 61 (96.8%) stage I/II patients and 48 (31.0%) stage III/IV patients. First-line chemotherapy regimens were platinum based in 100 (43.1%) patients and platinum-and taxane-based in 72 (31.0%) patients. Total 25 (10.8%) patients were treated with other regimens, and 32 (13.8%) patients did not receive chemotherapy because of stage Ia disease, comorbidity or treatment refusal.

Immunostaining and RT -PCR
The number of non-evaluable primary tumours due to core loss during staining procedures or absence of tumour tissue ranged from 2 (0.9%) for HER-2/neu staining to 10 (4.3%) for pERK staining. Positive staining was present in 6.2% of tumours for EGFR, 5.1% of tumours for HER-2/neu, 11.8% tumours for pEGFR, 100% of tumours for total AKT, 8.3% of tumours for pAKT and 36.9% of tumours for pERK (Table 2; Figure 1). Of 224 tumours, 69 (30.8%) showed completely negative PTEN staining. None of the tumour samples stained positive for EGFRvIII, nor could EGFRvIII be detected by RT -PCR. Staining for pERK was more frequent in tumour samples obtained after three or six cycles of chemotherapy compared to paired primary tumour samples (65 vs 37%, P ¼ 0.020). For all other proteins, staining patterns in primary tumours were comparable to paired residual or recurrent tumour samples (Table 2).
Unexpectedly, PTEN staining was positively correlated with pAKT staining (P ¼ 0.034). No associations were found between other proteins (data not shown).

Clinicopathological characteristics
Overexpression of EGFR was more frequent in non-serous tumours (P ¼ 0.017; Table 3). Stage III/IV tumours more often showed overexpression of pAKT (P ¼ 0.029). Loss of PTEN was related to stage I/II disease (P ¼ 0.006). Furthermore, negative PTEN immunostaining was associated with non-serous tumour type (P ¼ 0.042), occurring in 25% of serous, 39% of endometrioid, 42% of mucinous and 56% of clear cell tumours. No other associations between protein expression and clinicopathological variables were found.

Response to chemotherapy and survival
Univariate Cox regression analysis revealed that patients with a PTEN-negative tumour had a better PFS and OS (Table 4; Po0.001 and P ¼ 0.037, respectively). On the basis of recent publications dividing ovarian carcinomas into subgroups with specific molecular alterations (Bell, 2005;Press et al, 2008), we performed subgroup analyses for early and late stage patients, and for patients with grade III and undifferentiated carcinomas. Subgroup analysis for stage I/II and stage III/IV patients showed that PTEN predicts PFS only in the early stage group (HR 0.29, 95% CI 0.095 -0.9, P ¼ 0.032 for stage I/II patients, HR 0.74, 95% CI 0.48 -1.15, P ¼ 0.18 for stage III/IV patients). Loss of PTEN also predicted improved PFS in 91 poorly differentiated serous carcinomas, of which 20 (22.0%) were PTEN negative (HR 0.43, 95% CI 0.23 -0.83, In multivariate analysis PTEN staining (P ¼ 0.015), FIGO stage (P ¼ 0.013) and residual tumour after primary surgery (Po0.001) independently predicted PFS (Table 5). Tumour stage (P ¼ 0.023) and residual tumour (Po0.001), but not PTEN staining (P ¼ 0.833) were significant prognostic factors in multivariate analysis for OS.

DISCUSSION
Our study in a large, well-defined series of epithelial ovarian cancer patients shows that PTEN-negative tumours might represent a subgroup of ovarian carcinomas with a relatively favourable prognosis. To our knowledge this is the first study describing a relationship between negative PTEN staining and improved survival in ovarian cancer. Although a relationship between negative PTEN staining and improved survival has been described for endometrial cancer patients , previous studies in ovarian cancer found no or an inverse relationship between PTEN and prognosis (Schondorf et al, 2003;Wang et al, 2005;Lee et al, 2005a). These contrasting results could be explained by the fact that previous studies either did not have the power to evaluate possible relations with survival, or restricted their analysis to stage III/IV ovarian cancer patients. In the current study PTEN staining was of prognostic significance mainly in the stage I/II group and in poorly differentiated serous carcinomas. We found negative PTEN expression in 30.8% of tumours, which is in agreement with previous studies Lee et al, 2005a;Hashiguchi et al, 2006). In ovarian cancer, loss-ofheterozygosity (LOH) at the PTEN locus (10q23.3) occurs in 31 -45% of tumours, whereas mutations of the second PTEN allele are relatively rare Obata et al, 1998;Kurose et al, 2001). Loss of protein expression is therefore also thought to arise through other mechanisms, such as DNA methylation (Sansal and Sellers, 2004).
Interestingly, we showed a high rate of negative PTEN staining in endometrioid and clear cell tumours. A high rate of PTEN loss in clear cell and endometrioid carcinomas has also been shown in previous, much smaller studies (Obata et al, 1998;Hashiguchi et al, 2006). Both cancers are thought to at least partly arise from endometriosis. Sato et al (2000) showed that in three out of five ovarian carcinomas associated with endometriosis, LOH at 10q23.3 occurs in both the carcinoma and in endometriotic lesions, implicating that LOH is an early event in carcinogenesis and that PTEN is involved in the progression from endometriotic precursor lesion to clear cell or endometrioid ovarian cancer.
Our results show that negative PTEN staining is strongly associated with early stage disease and a non-serous tumour type. Recent studies suggest that ovarian carcinomas could be divided in two categories. The first category, called type I, includes low-grade serous, mucinous, clear cell and endometrioid tumour with frequent alterations in BRAF, KRAS and PTEN. Type I tumours are thought to arise from precursor lesions such as endometriosis and have a relatively good prognosis. In contrast, type II tumours, including high-grade serous and undifferentiated carcinomas characterised by p53 mutations and overexpression/amplification of HER-2/neu and AKT2, tend to show a highly aggressive behaviour (Shih and Kurman, 2004;Bell, 2005). In the present study, we identified a relationship of pAKT expression with late stage disease. Moreover, our previous work showed that overexpression of p53 mostly occurs in high-grade, late stage, serous carcinomas (de Graeff et al, 2006). Our combined results therefore support this model of ovarian carcinogenesis.
A recent study by Press et al (2008) suggests that type II ovarian tumours can be subclassified into three groups based on their BRCA1 status. Their results indicate that poorly differentiated serous carcinomas with BRCA1 mutations frequently show loss of PTEN. The molecular mechanism underlying the relationship between loss of PTEN and BRCA1 mutations in ovarian cancer remains unknown. Possibly, ineffective DNA repair in BRCA1linked tumours results in specific mutations of the PTEN gene (Foulkes, 2008;Saal et al, 2008). On the basis of these observations we performed survival analysis in a subgroup of 91 poorly differentiated serous carcinomas. We were able to show that loss of PTEN was indeed associated with improved PFS in this subgroup of ovarian carcinomas. Patients with BRCA1-linked hereditary tumours have a favourable survival compared to sporadic tumours, possibly because of a good response to chemotherapy (Boyd et al, 2000;Chetrit et al, 2008). The link between PTEN and BRCA1 status might therefore explain an improved disease outcome in a subgroup of patients with an otherwise very poor prognosis. In that case, IHC staining of PTEN may be a rapid way of identifying tumours most likely to carry BRCA1 mutations. Subsequently, those patients might benefit from treatments with agents selectively targeting BRCA mutant tumour cells, such as poly(ADP-ribose) polymerase 1 inhibitors (Farmer et al, 2005).
In the current study, loss of PTEN was associated with improved PFS, but not OS. As PFS is closely related to response to chemotherapy, these results might indicate that patients with PTEN negative tumours respond favourably to first-line therapy. In the current study we did not observe a relationship between PTEN status and response to chemotherapy. However, this analysis was limited to patients who had measurable disease before start of chemotherapy or measurable disease progression during treatment. Response to chemotherapy could therefore only be analysed in a subset of advanced-stage patients with a very poor prognosis. One possible explanation for the lack of association between negative PTEN staining and OS might be explained by the fact that tumours can acquire secondary mutations during or after platinum-based chemotherapy (Sakai et al, 2008). Once a patient presents with progressive or recurrent disease, these mutations may render the tumour insensitive to platinum-based chemotherapy irrespective of the PTEN status. We did not observe any association between EGFR and HER-2/ neu immunostaining and disease outcome, confirming results of a previous study also from our institution (Van Der Zee et al, 1995). Previous studies on the relationship between EGFR or HER-2/neu overexpression and clinicopathological characteristics, response to chemotherapy and survival have shown conflicting results Elie et al, 2004;Nielsen et al, 2004;Psyrri et al, 2005). One of the most important reasons for these inconclusive data is the considerable methodological variability among studies (Hall et al, 2004). Techniques used to determine marker expression, antibodies and scoring systems used for immunostaining vary widely between studies. For the present investigation, we aimed to use well-characterised antibodies that have been extensively studied in other tumour types, and, if possible, used well-defined scoring criteria that have been shown to be reproducible. We have sought to adhere to the REMARK guidelines for publishing prognostic factor studies (McShane et al, 2005). The use of these guidelines and of standardised methods should aid in increasing transparency and reproducibility of prognostic factor studies in ovarian cancer and other tumour types.
As tumours showing evidence of strong signalling through a particular pathway are thought to have a high chance of responding to therapies directed against this pathway, the identification of reliable biomarkers could aid in selecting patients who are most likely to benefit from targeted therapy (Bild et al, 2006). Results of different clinical trials show that positive immunostaining for HER-2/neu or EGFR does not reliably predict response to ErbB-targeted therapy (Ciardiello and Tortora, 2008). A possible better marker of response to EGFR-and HER-2/neutargeted therapies is activation or downregulation of downstream pathways. Indeed, positive immunostaining for pAKT, pERK, PTEN and EGFRvIII has been reported to predict sensitivity to EGFR tyrosine kinase inhibitors in non-small-cell lung cancer and glioblastoma (Han et al, 2004;Mellinghoff et al, 2005). The association of pAKT and pERK in relation to response to ErbB-targeted therapy in ovarian cancer has not been studied yet, but expression of these proteins might be used as a marker of responsiveness to targeted therapies. Our results show that 8.3 and 36.9% of tumours show positive pAKT and pERK staining, respectively, indicating that only a subgroup of patients might benefit from agents directed against these pathways. As pERK is overexpressed in approximately one-third of primary ovarian tumours and 65% of tumour samples from primary chemoresistant tumours obtained after chemotherapy, treatment of patients with Ras/Raf/MEK/Erk-targeted agents appears to be an interesting therapeutic option (Messersmith et al, 2006).
In contrast to previous studies, we show a low percentage of pAKT-positive tumours (Altomare et al, 2004;Wang et al, 2005). The discrepancy between our results and those obtained in previous studies is not likely to be due to methodological variability. We have used the same well-characterised antibody   ErbB signalling pathway in ovarian cancer P de Graeff et al that was used in previous studies, with a comparable staining protocol. In all our experiments, the ovarian cancer cell line A2780 served as a positive control. Expression of pAKT in this cell line was confirmed by western blotting (data not shown). In agreement with previous large studies, we also show a relatively low percentage of EGFR-and HER-2/neu-overexpressing tumours (Bookman et al, 2003;Lassus et al, 2006). We therefore conclude that in this group of ovarian carcinomas, signalling of EGFRs via the AKT pathway might be important only in specific subgroups of ovarian tumours. Surprisingly, we identified a significant relationship between positive expression of AKT and positive expression of PTEN. The role of PTEN as a negative regulator of AKT is well documented in both cell line models and tumour samples (Stambolic et al, 1998;Sun et al, 1999;Kurose et al, 2001;Choe et al, 2003). However, others have also identified a positive correlation between expressions of the two proteins by immunostaining (Panigrahi et al, 2004;Slipicevic et al, 2005;Wang et al, 2005). This might mean that in tumours, the regulatory relationship between AKT and PTEN is not linear. In breast and ovarian cancer, it has been shown that aberrations of the PI3K and PTEN genes are mutually exclusive (Saal et al, 2005;Press et al, 2008), resulting in constitutive activation of the PI3K pathway in the presence of an intact PTEN. Loss of PTEN may also contribute to tumourigenesis and progression via AKT-independent pathways, such as the p53 pathway (Blanco-Aparicio et al, 2007).
In contrast to available data in literature we did not detect any EGFRvIII in this large group of ovarian carcinomas. Moscatello et al reported that EGFRvIII is expressed in 75% of ovarian tumours, but this high percentage could not be confirmed in subsequent studies (Jungbluth et al, 2003;Lassus et al, 2006). We determined EGFRvIII status by immunohistochemistry using the well-defined antibody DH8.3 and verified our results at the RNA level by RT -PCR on a subset of 45 tumours showing positive immunostaining for EGFR or downstream targets. As EGFRvIII heterodimerises with wtEGFR, is constitutively phosphorylated and activates AKT and to a lesser extent ERK, we hypothesised that the chance of finding EGFRvIII-positive tumours was largest in this subgroup (Montgomery et al, 1995;Li et al, 2004;Luwor et al, 2004). As we did not detect any EGFRvIII positivity in this subgroup, nor in 10 tumours that did not overexpress any of the studied markers, our data strongly suggest that EGFRvIII signalling does not play a major role in ovarian cancer.
In the current retrospective study we investigated protein expression in a large well-defined patient population. However, our results showed that protein expression was mainly important in specific patient groups. Unfortunately, these subgroups were too small to perform valid multivariate analysis. Furthermore, not all patients received the same chemotherapeutic treatment. Future studies should determine the prognostic value of PTEN staining, especially in early stage patients and poorly differentiated serous tumours, in large prospective studies including homogeneously treated patients.
In summary, we demonstrated that negative PTEN staining is associated with favourable patient and tumour characteristics, and independently predicts improved PFS. The importance of pAKT and pERK expression as downstream markers of responsiveness to receptor tyrosine kinase-targeted therapies deserves to be evaluated in clinical trials. A better understanding of these pathways and their role in ovarian cancer will enable us to use targeted drugs more efficiently, and to identify (groups of) genes that predict prognosis more accurately.
Supplementary Information accompanies the paper on British Journal of Cancer website (http://www.nature.com/bjc)