Several plasma or serum markers were proposed in endometrial cancer diagnosis, however none has attained the high accuracy predisposing to the rank of a screening tool [4]. Despite new developments in cancer management a gradual increase of 1.6% a year in endometrial carcinoma death rates was noted over 2006–2015 [2, 3]. Beyond any doubt, early detection of endometrial cancer could increase survival rates and decrease morbidity connected to this disease.
Majority of the analytes chosen for the presented study has not been studied previously in the blood derived samples of endometrial cancer patients, however their roles in EC pathogenesis were suggested by immunohistochemistry (ICH) or gene expression studies [5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]. The multiplex ELISA-based approach was undertaken by our team, as it allows for simultaneous detection of several analytes in one sample. Such method reduces technical errors and increases accuracy. It can be also easily translated into clinical practice.
Accurate tool in endometrial cancer diagnosis needs to be sensitive but also specific, as this carcinoma rarely occurs as a lone disorder. It is often associated with obesity, hypertension and diabetes, but can be also accompanied by endometriosis [27]. In line with that assumption endometriosis samples were included in our study.
Analysis of data revealed that concentrations of CD44, EpCAM, and TGM2 were significantly increased in EC comparing to control samples. It was also found that concentrations of the two stem-cell markers, CD44 and TGM2 were significantly correlated in EC samples. These results correspond to results of increased CD44 expression in EC tissues reported by Elbasateeny et al. and Wojciechowski et al., as well as upregulation of EpCAM in serous uterine cancers presented by El-Sahwi et al. [8, 9, 11]. TGM2 has been previously indicated in the pathogenesis of a number of cancers including ovarian cancer, however our finding of its highly increased concentration in EC plasma samples seems to be a new discovery [25]. Concentration of CA9 was also increased in EC samples as compared to controls and the difference was very close to statistical significance. That result was in line with ICH studies performed by other authors [6, 7]. Kalikrein-6 levels were investigated in tissues and plasma samples of endometrioid adenocarcinoma by Santin et al., and similarly to our results, the authors did not find differences between cancer and healthy patients [15]. Kalikrein-6 levels were, however, highly increased in uterine serous papillary cancer, in both tissues and plasma samples [16]. L1CAM has been suggested as a strong prognostic marker by several ICH studies [17,18,19,20]. Unfortunately, we did not reproduce those results. We speculate, that it could be due to relatively small number of advanced EC cases in our study population. Our results could be also justified by the observation of Pasanen et al., who observed no difference between L1CAM serum levels of patients with an L1CAM-positive or L1CAM-negative endometrial carcinoma [18].
High tissue expression of mesothelin has been linked to aggressive tumor behavior and worse prognosis in ovarian and pancreatic cancers as well as in mesothelioma [28]. The diagnostic value of circulating mesothelin has been also suggested in those neoplasms [29, 30]. In our study, mesothelin plasma concentrations did not differ between EC and healthy samples. However, comparison of controls to more advanced cases (FIGO IB-III) revealed significantly lower concentration of that protein in plasma from EC patients. Moreover, mesothelin concentration in FIGO IB-III samples was decreased as compared to FIGO IA stage. There was a marked trend towards decreased mesothelin concentration with the progression of the disease. Few studies on mesothelin expression in EC tissues revealed a moderate expression level and to date there no available studies investigating soluble mesothelin levels in EC [21, 22, 28]. Therefore, the somewhat surprising results obtained in this study require further investigations combining tissue and blood derived samples from the same patients. The soluble form of mesothelin is likely due to an abnormal splicing event, but it is also possible, that it is a proteolytically cleaved fragment of membrane-bound mesothelin [31]. It has been recently revealed that mesothelin binds to Ca 125 and that this interaction mediates cell adhesion [32]. Therefore, one of the possible hypotheses behind the phenomenon of lower circulating mesothelin levels in advanced EC, could be connected to decreased release of the glycoprotein from the tumor cells due its enhanced utilization within the tumor microenvironment, including bounding to Ca 125, which expression is also increased in more aggressive ECs.
When concentrations of other analytes were compared according to clinicopathological characteristics the most important finding regarded EpCAM, which concentration was higher in FIGO IA as compared to more advanced stages (>FIGO IA). No differences in concentrations of other markers were found in regards to histological grading, myometrial invasion and FIGO staging. The lack of correlation of CD44 level with clinicopathological features despite its highly significant increase in EC samples corresponds to the results obtained by other authors [7, 8].
Further analyses including the subset of endometriosis samples revealed interesting findings. Concentration of seven analytes (ALDH1A, CA9, CD44, hepsin, midkine, TGM2, and kallikrein-6) differed in endometriosis samples as compared to control samples and levels of five markers (ALDH1A, CA9, CD44, hepsin, and midkine) were different between endometriosis and EC samples. We acknowledge that those results can only be regarded as preliminary findings, due to a small number of patients with endometriosis. However, they seem to explain the observation, that only three analytes were different (EpCAM, midkine and TGM2) in EC in comparison to non-EC group.
Regression analysis revealed panels of analytes characterized by high diagnostic accuracy in discriminating EC samples. Firstly, TGM2 was found to be a highly accurate, single marker able to differentiate between EC and healthy controls with the AUC of 0.901 (78% sensitivity, 100% specificity). Secondly, multivariable logistic regression models were constructed to evaluate, if the input of several biomarkers would improve the diagnostic accuracy. The backward analysis method retrieved the model consisting of EpCAM and TGM2 as most suitable model for discrimination between EC and controls, which yielded AUC of 0.909 with 84% sensitivity and 94% specificity. Subsequently, regression models were created using the enter method and different combinations of CD44, TGM2, and EpCAM. Out of the four models obtained in that analysis, the 3-marker model was characterized by the highest AUC of 0.937 with sensitivity of 84 and 100% specificity. Comparison of ROCs revealed that the AUC for CD44/TGM2/EpCAM model did not differ significantly from the AUC of TGM2 and all three 2-marker models. Analysis aiming at discrimination between EC samples and non-EC group, which included endometriosis samples, required input of least five markers to obtain a satisfactory AUC of 0.895, and the utilization of eight markers was necessary to increase the AUC to 0.945.
Comparing to the work of other authors, marker panels discovered in this study seem to offer similar or better accuracy. Yurkovetsky et al. reported prolactin as the strongest single biomarker for EC with 98.3% sensitivity and 98.0% specificity and the 5-marker panel (prolactin, GH, Eotaxin, E-selectin, and TSH), which yielded high sensitivity and specificity in discrimination between EC, ovarian and breast cancers [33]. Another study suggested that the panel of ApoA-I, TTR, and TF distinguished normal samples from early-stage endometrial cancer with a sensitivity of 71% (specificity, 88%) and normal samples from late stage endometrial cancer with a sensitivity of 82% [34]. CA 125 and HE4 were also extensively studied as single biomarkers or in combination and yielded no more than low to moderate accuracy depending on the study [35,36,37,38,39]. Although few studies have proven acceptable accuracy of CA 125/HE4 panel, it needs to be acknowledged, that both CA-125 and HE4 can be elevated in various malignancies and benign pathologies of reproductive tract [35]. Therefore, their specificity for endometrial tumors is questionable. What is more, the distinct effects of physiological factors on prolactin secretion shadow the credibility of this hormone in early diagnosis of endometrial tumors [40].
EC has a very good prognosis providing it is diagnosed at the early stage of clinical progression. Nowadays most EC are discovered only after the clinical signs are already present, and the screening method is still not available [2]. One of the research questions of the presented study was to verify, if one of the combinations of studied markers would be able to discern those early cases (FIGO IA) from healthy population. For that purpose, logistic regression analysis was conducted and retrieved a model consisting of TGM2 and CD44 with the 69% sensitivity and 94% specificity and AUC of 0.847. This finding is in line with the study of Elbasateeny et al., who reported that CD44 along with CD133 might participate in early-stage endometrial cancer carcinogenesis, and their overexpression might facilitate early diagnosis of endometrial cancers [9].
Accurate preoperative diagnosis of the disease progression has an important clinical significance in endometrial cancer management, as early EC does not require lymphadenectomy and radical surgery. A method with reasonable accuracy in depicting tumors at the earliest stage of progression could decrease morbidity and complications connected with unnecessary lymphadenectomies and extensive resection of pelvic tissues. In line with that notion a multivariable regression analysis was performed to discriminate between FIGO IA samples and more advanced stages of EC. That analysis incorporated histological grading as one of the variables and retrieved model consisting of mesothelin concentration and histological grade 1, as factors predicting early clinical stage of EC. The ROC curve plotted based on that model retrieved AUC of 0.911, with 95% sensitivity and 78% specificity.
These results present similar accuracy as compared to blood derived markers reported by other research groups [30, 35, 41].
There are some limitations to our study that need to be addressed. Firstly, although we investigated multiple markers the number of patients included in the study was only moderate. Therefore, we consider obtained results as preliminary, which need to be repeated in larger population, preferably in the multicenter setting. Secondly, we acknowledge that the research would benefit from tissue expression analysis. We plan to address these issues in the future studies.
