In this regional cohort with 5-year follow-up, the presence of MMP-14, CD44 and the EMT markers E-cadherin and vimentin and their interrelationships were investigated. The patients with double expression of MMP-14 and CD44 in their tumors had a poor prognosis despite complete debulking. Serous subtype in advanced-stage patients and CD44 expression were correlated with EMT markers and CD44 expression was found to be significantly correlated with complete debulking.
It is a challenge to interpret these findings pathophysiologically. A previous study found that CD44 expression is associated with peritoneal metastasis and EMT [16], but it is not clear how peritoneal metastasis and EMT are related to complete debulking. In our study, epithelial CD44 expression was a predictor for complete debulking, which is consistent with the better prognosis for CD44-positive tumors [20].
To determine CD44 expression in this study, a pan-CD44 antibody was used in order not to miss relevant CD44 expression in the tumors. Pan-CD44 expression is associated with a better prognosis in ovarian cancer, though some other isoforms of CD44 are associated with a poorer prognosis [7, 20, 21]. The first finding is consistent with the correlation of CD44 with complete debulking. These studies of somewhat older date correlated CD44-isoforms with prognosis in ovarian cancer: in a small series a low frequency of CD44v5 and CD44v6 were found [7]. In a larger series, CD44 was present in about half of the tumours and indicated a good prognosis [20]. In another series, CD44s, CD44v3 and CD446 were upregulated in ovarian cancer, but declined with advancing FIGO stage [21]. In a series with lymphnode metastasis, a high expression for CD44v5 and CD44v6 was found [22]. Recently, CD44v6 was found to correlate with prognosis while in another study, CD44v8-10 indicated a better prognosis and a more epithelial phenotype [23, 24]. A recent meta-analysis concluded that only CD44s is related to chemoresistance in ovarian cancer [25]. It may be that using an antibody for a specific isoform of CD44 would have resulted in a different relationship with EMT markers, because some isoforms of CD44 may be more important than others in EMT, like CD44v8-10.
However, this is not fully studied in ovarian cancer. In a study of colorectal cancer, the standard isoform CD44s seemed to be correlated with EMT, but higher CD44 expression predicted poor survival in that study in contrast to our results in ovarian cancer [26].
In other gynecological malignancies, the presence of EMT-like changes has been found to indicate poor prognosis [1]. Our study in ovarian cancer was not consistent with that finding.
A tentative explanation for this apparent discrepancy may be the pathology of ovarian tumors. Typically, a small proportion of ovarian tumors invade the ovary internally from their cystic lining, whereas the majority spread extensively on the peritoneum and to a lesser extent on the ovarian surface. Therefore, the invasive front is difficult to determine. Arguably, the superficial spreading on the peritoneal surface does not require the invasive properties of cells with EMT-like changes. However, deep invasion into other organs and metastasis into the liver for example do occur in ovarian cancer patients, albeit infrequently. In ovarian cancer, both changes to an EMT-like phenotype as well as a reverse to MET (Mesenchymal-to Epithelial Transition)-like changes are found [27].
The advantage of using an immunohistochemical panel for EMT, as in this study, is that various aspects of EMT in ovarian cancer are revealed.
One such aspect is the heterogeneous expression of the markers. The pattern is either positive expression of both E-cadherin and vimentin, negative expression of both these markers, or positive expression of one and negative expression of the other. Positive areas of E-cadherin expression indicate an epithelial phenotype, whereas positive expression of vimentin indicates a mesenchymal phenotype. Figure 2 shows an area where the expression patterns of E-cadherin and vimentin are heterogeneous. Although this heterogeneous pattern may be easy for the pathologist to interpret qualitatively, it makes it difficult to score the immunohistochemical results quantitatively, because in practice only the intensity and percentage of the cells are scored and heterogeneous expression is difficult to quantitate.
We studied primary ovarian tumor tissue. EMT markers in primary tumors may be expressed less than in metastatic tumors, especially in advanced-stage disease, as has been found for some CD44 isoforms. [22] It is the extent and progression of metastatic disease that eventually determines prognosis in a particular patient.
Although we have found that, for these reasons, EMT-like changes does not indicate prognosis in ovarian cancer, some of our results are consistent with the hypothesis that EMT is stimulated by CD44 and MMP-14 expression [5]. In the present study, the patients with double expression of MMP-14 and CD44 in their tumor had a poor prognosis. CD44 was found to be correlated not only with serous histology in advanced-stage patients but also with vimentin expression as a marker for EMT (Table 3). These correlations have a magnitude of 0.2-0.3, indicating a moderate effect. No correlation with EMT-like changes was found for MMP-14.
However, the small subset of tumors with both MMP-14 and CD44 expression seem to resemble type-I tumors with a lower percentage of serous histology and a higher percentage of clear cell and unspecified histology [28]. As previously stated, the choice of CD44 isoform to be measured may be important, as was described in a study on a small series of clear cell carcinomas where CD44-10v expression predicted recurrence and death [29].
The lack of significance of EMT for prognosis in ovarian cancer evident in the present study may also be due to the cross-sectional study design, in which the tumors were not followed in time. If a subset of the tumors with EMT-like changes were to result in a poorer prognosis for those patients, the statistical significance of this would be lost due to dilution with patients in which EMT is irrelevant to the clinical course. Solving this problem by collecting large series of tumor samples and performing immunohistochemistry on them requires preparation of such samples in one laboratory according to a uniform protocol. Bias may result in various ways, including missing samples and different protocols for tissue handling and fixation in different hospitals and laboratories. A shortcoming of our study was a relatively high number of missing samples for the various markers.
A strength of our study was the prospective data collection via the ROGY registry, which is used to register patients in all of the participating hospitals and sign them up for the tumor board meeting. By selecting patients via this registry and then collecting data from their hospital records, a thorough follow-up was achieved.