The aim of our study was to define tissue and plasma miRNA signatures, which could potentially serve as diagnostic and prognostic markers in endometrioid endometrial cancer (EEC) and to investigate miRNA profiles in regard to clinicopathological characteristics. Tissue and plasma samples were collected from 122 women (77 EEC and 45 controls). Expression profiling of 866 human miRNAs and 89 human viral miRNAs was performed in 24 samples and was followed by qPCR validation in 104 patients. Expression of 16 miRNAs was analyzed in 48 plasma samples. Microarray study revealed regulation of 21 miRNAs in EEC tissues comparing to normal endometrium. Altered expression of 17 miRNAs was confirmed by qPCR performed in 104 tissue samples. Seven miRNAs were upregulated and two were downregulated in EEC plasma samples. Expression of a number of miRNAs was associated with International Federation of Gynecology and Obstetrics stage, grade, relapse and nodal metastases. Two miRNA signatures: miR-92a/miR-410 and miR-92a/miR-205/miR-410 classified tumor tissues with higher accuracy in comparison to single miRNAs (AUC: 0.977, 95% CI: 0.927–0.996 and 0.984, 95% CI: 0.938–0.999, respectively). miRNA signature composed of miR-205 and miR-200a predicted relapse with AUC of 0.854 (95% CI: 0.691–0.951). Tissue miRNA signatures were independent prognostic markers of overall (miR-1228/miR-200c/miR-429, HR: 2.98) and progression-free survival (miR-1228/miR-429, HR: 2.453). Plasma miRNA signatures: miR-9/miR-1228 and miR-9/miR-92a, classified EEC plasma samples with high accuracy yielding AUCs of 0.909 (95% CI: 0.789–973) and 0.913 (95% CI: 0.794–0.976), respectively. We conclude that miRNA signatures hold a great promise to become noninvasive biomarkers for early EEC detection and prognosis.
Endometrial cancer is the fourth most often diagnosed malignancy in the female population of the developed countries and the most common cancer of the female reproductive tract. In 2008, 82,530 endometrial cancer cases were recorded in Europe.1 Estimated number of new cases diagnosed in 2011 in the United States was 46,470 and was higher than in 2010.2 Data deposited in the Polish Cancer Registry indicated a progressive increase in endometrial cancer incidence over the period between 1999 and 2008.3 Similar trend has been observed in Great Britain.4
Endometrial cancer is a heterogeneous malignancy comprising many histological types, which differ significantly in terms of pathogenesis, clinical presentation and prognosis. Such heterogeneity impedes development of screening and treatment strategies. Despite a great improvement in endometrial cancer treatment and diagnosis, advanced stages of the disease are still difficult to manage with the 5-year survival rate of ∼10–29%.5 Increase in the incidence of endometrial cancer and the lack of powerful yet nontoxic treatment strategies indicate the need of developing novel diagnostic, prognostic and treatment strategies for this malignancy.
Endometrioid endometrial carcinoma (EEC) is the most common histological type of endometrial cancer and constitutes ∼80% of newly diagnosed cases. EEC pathogenesis has been connected to mutations of several genes including PTEN, AKT, mTOR, KRAS and CTNNB1. However, such alterations are not uniformly found in all EEC cases.5–9 Therefore, alternative oncogenic mechanisms have been studied. A new insight into cancer pathogenesis emerged with the discovery of small (19–25 nt), single-stranded, noncoding RNA molecules, called micro-RNAs (miRNAs), which are thought to regulate gene expression at the posttranscriptional level.10,11 Since their discovery miRNAs were found to influence key cellular processes and to regulate development, cell cycle, cell proliferation, differentiation and apoptosis.10 Because of their physical and chemical characteristics including significant stability in body fluids and formalin-fixed paraffin-embedded (FFPE) tissues miRNA molecules comprise a promising group of potential diagnostic and prognostic markers.12–15 Moreover, the possibility of modifying miRNA expression both in vitro and in vivo offers a great opportunity of using these small RNAs in the gene-targeted therapy.16
Although a number of studies suggested miRNAs involvement in EEC pathogenesis, only few of them involved large groups of patients and addressed issues concerning the influence of clinicopathological features on miRNAs expression.17–20 Similarly, diagnostic or prognostic accuracy of miRNA signatures in tissue and plasma of EEC patients has not been systematically studied. Thus, the aim of our study was to define tissue and plasma miRNA signatures, which could potentially serve as diagnostic and prognostic markers in EEC and to investigate miRNA profiles in regard to EEC clinicopathological characteristics.
Material and Methods
One hundred and twenty-two female patients were included in the study, of those 77 were EEC patients, 31 were patients operated due benign gynecological pathologies other than of endometrial origin (control group for the tissues part of the study) and the remaining group of 14 included females addressing gynecologist for a regular check-up. The latter group was used as a control group in the plasma part of the study. None of the females enrolled in the control groups had a history of cancer or endometrial pathology.
Fresh tissues and plasma samples used in the study were collected from patients hospitalized in gynecological departments of Medical University of Lublin. FFPE specimens were obtained from patients of the Gynecologic Oncology Division and Department of Pathology, Ospedale Sacro Cuore Don Calabria (Negrar, Italy). Medical University of Lublin Ethical Committee approved the study design (KE-0254/201/2008). Informed consent was obtained from each study participant. The EEC patients underwent total hysterectomy with bilateral salpingoophorectomy due to ECC diagnosed with endometrial biopsy prior to operation. None of EEC patients had a history of other malignant disease or was submitted to neoadjuvant therapy. After the operations patients were submitted to radiotherapy and/or chemotherapy according to International Federation of Gynecology and Obstetrics (FIGO) guidelines. Lymphadenectomy was performed in 57% of cases. Clinical stage of the disease was determined according to 2009 revised FIGO classification.21 Clinicopathological characteristics of the patients were presented in Table 1 and in Supporting Information Table S22.
Table 1. Clinicopathological characteristics of 77 EEC patients included in the study
Average age of EEC patients included in the tissue part of the study was 62.8 years (95% CI: 59.75–64.38) and was significantly higher comparing to control group (44.78, 95% CI: 42.51–47.05; t-test, p < 0.001). Median BMI in that group was 28.2 kg/m2, and was higher in comparison to BMI of control patients 24.95 kg/m2 (Mann–Whitney, p = 0.012). There was no significant difference in BMI and age between plasma control group and EEC cases (median BMI in controls 28.37 kg/m2, 95% CI: 24.7–30.09 vs. EEC: 31.3, 95% CI: 25.38–34.48, Mann–Whitney p = 0.21; average age of controls 54.7 years, 95% CI: 49.51–59.95 vs. EEC 60.2 years, 95% CI: 56.68–63.7, t-test p = 0.06). A total of 85.3% of patients with EEC were postmenopausal, whereas most women in the tissue control group were premenopausal (81.6%). The plasma control group comprised equal numbers of pre- and postmenopausal cases. Although there were differences in the BMI and age between groups used in the tissue part of the study, we did not find any significant correlations between expression of studied miRNAs and age and BMI of patients (Supporting Information Table S29).
Fresh tissue samples (EEC n = 30, NE n = 16) were collected during the surgery within 15 min from the uterus removal and immediately immersed in RNAlater (Ambion). After 24-hr incubation in RNAlater in 4°C, tissues were stored in −80°C until RNA extraction. FFPE tissues (EEC n = 43, NE n = 15) used for the study were fixed with 10% formalin and stored for maximum 10 years. Healthy endometrium included both proliferative and secretory phase epithelium. There was no overlap between fresh and FFPE tissue samples
Venous blood was collected from 48 females (34 EEC and 14 controls) from the antecubital vein (5 ml, EDTA) and was centrifuged for 15 min (1800g, 19°C). Plasma was collected and stored in −80°C. Thirty of 34 EEC plasma samples matched EEC samples used in the tissue part of the study and four were derived from females included only in the plasma part of the study.
RNA isolation from tissues stored in RNAlater was performed using mirVANATM miRNA Isolation Kit (Ambion) and 40–80 mg of macrodissected tissue according to manufacturer's protocol.
Before RNA isolation from FFPE tissues 10-μm sections containing at least 70% of cancer cells were prepared by microdissection. RNA isolation from FFPE specimens was performed using RecoverAll™ Total Nucleic Acid Isolation Kit for FFPE Tissues (Ambion) according to the protocol provided by the manufacturer. After extraction RNA underwent DNAse treatment using Turbo DNAase Kit (Ambion).
Isolation of RNA from plasma was performed using mirVANATM PARIS Kit (Ambion) and 400 ml of plasma. A total of 5 fmol/μl of each of the following synthetic Caenorhabditis elegans oligonucleotides: cel-miR-39, cel-miR-54 and cel-miR-238 were spiked into plasma sample after addition of 2 × denaturating solution. From that point isolation was performed according to manufacturer's protocol. Elution was performed with 52.5 μl of RNase, DNase-free water. Isolation of RNA from plasma was conducted in duplicates. RNA was stored in −80°C.
Concentration and purity of RNA was inspected using spectrophotometry (Biophotometer with Hellma TrayCell, Eppendorf) and RNA integrity was checked using Bioanalyzer 2100 (Agilent Technologies) and Agilent RNA Nano kit. In case of RNA isolated from fresh frozen tissues only samples with RIN ≥ 6 were used in downstream applications. In case of RNA extracted from FFPE the average RIN was 3.45. A total of 260/280 ratio of all tissue samples ranged between 1.8 and 2.2.
Expression of 866 human miRNAs and 89 human viral miRNAs was investigated using Agilent Human miRNA Microarray V.3 in 24 fresh tissue samples (EEC n = 18, NE n = 6) according to the manufacturer's protocol. Raw data were deposited in Gene Expression Omnibus database (accession number GSE35794). Normalization and statistical analysis of microarray data were performed using GeneSpring GX software (Agilent). Detailed description of microarray protocol is provided in the Supporting Information.
Reverse transcription and qPCR
Validation of the microarray experiment was performed using qPCR and specific hydrolysis probes.
Reverse transcription (RT) was performed using TaqMan® MicroRNA Reverse Transcription Kit and specific primers (Applied Biosystems). All RT reactions were carried out in triplicates in Mastercycler ep gradient S (Eppendorf) and stored in −20°C.
qPCR was performed using single tube TaqMan® MicroRNA Assays and TaqMan® 2 × Universal polymerase chain reaction PCR Master Mix, No AmpErase® UNG (Applied Biosystems) in 10 -μl reactions.
All qPCR reactions were performed in duplicates in ViiA7 Real–Time PCR System (Applied Biosystems) using protocol suggested by manufacturer. Positive and negative control reactions as well as interplate calibrator (IPC) reactions were carried out on each plate. Detailed description of RT and qPCR protocols is provided in the Supporting Information.
Normalization and statistical analysis of microarray data
Microarray raw data were normalized to 75th percentile using percentile shift normalization method. Comparisons between the groups were performed with Mann–Whitney and Kruskal–Wallis tests with Benjamini–Hochberg correction for multiple comparisons. Normalization, unsupervised hierarchical clustering and statistical tests were performed using GeneSpring GX software (Agilent). Significant difference was assumed, when the fold change was ≥2, and p value was <0.05.
Normalization and statistical analysis of qPCR data
Raw qPCR data obtained from tissue samples were initially normalized with IPCs and adjusted for reaction efficiency. To normalize for the variation in RNA extraction step qPCR data obtained in plasma samples were additionally normalized to expression of three synthetic oligonucleotides matching the sequence of Caenorhabditis elegans miRNAs: cel-miR-39, cel-miR-54, cel-miR-238 using median normalization procedure as it was described by Mitchell et al. and Kroh et al.22,23
Efficiencies of primer/probe sets were determined by performing standard qPCR with the sixfold dilution of a pool of 10 randomly chosen cDNA templates. Efficiencies for all amplicons were calculated using the equation E = 10(−1/slope) − 1.
For relative quantification of miRNA expression in tissues data were normalized using geometric mean of expression of three experimentally chosen, stable endogenous controls RNU48, RNU44 and U75. The three snRNA were chosen from the panel consisting of 12 noncoding RNAs (RNU48, RNU44, U75, RNU6B, U6, U54, RNU38B, U18, U49, miR-26b, miR-92a and miR-16), which were previously described to be stably expressed in tissues or were used in endometrial cancer qPCR studies. In case of plasma samples qPCR data were normalized using geometric mean of expression of five experimentally chosen endogenous controls: miR-93, miR-26b, miR-192, miR-103a and miR-142-3p. Reference miRNAs were chosen from the panel comprising nine miRNAs, which were reported to be highly expressed in plasma or were used as endogenous control in other studies. The panel included: miR-93, miR-26b, miR-192, miR-103a, miR-142-3p, miR-92a, miR-638, miR-16 and miR-451.
The procedures of experimental validation of candidate endogenous controls were performed in 45 tissue samples and 48 plasma samples using NormFinder and geNorm algorithms and were described in details in another submitted article.24,25
Data were log transformed before statistical analysis. Results are presented as mean with 95% confidence intervals (95% CIs) or fold change (FC) with 95% CI. Depending on the results of the tests assessing normal distribution (D'Agostino–Pearson test) and equality of variances (Fisher test), parametric (Student's t-test with or without Welch correction and ANOVA) or nonparametric (Mann–Whitney and Kruskal–Wallis) tests were used to compare between the groups.
In addition, unsupervised expression profiling was performed on qPCR data using principal component analysis (PCA). Spearman coefficient was applied to determine correlation between expressions of miRNAs. Correlation was considered relevant with coefficient larger than 0.5.
To determine the ability of miRNAs to classify EEC and control samples receiver-operating characteristics (ROC) curves were constructed and sensitivity, specificity as well as positive and negative predictive values were determined. To develop miRNA signatures featuring the best accuracy in distinguishing between tumor and control samples a multivariate logistic regression model was utilized. The model is as follows: p = 1/1 + e−z, where e is a base of natural logarithm equal to 2.71828…, and z is a linear combination of miRNAs expression values (xi) weighted by the regression coefficients (bi) derived from the multivariate regression analysis: z = b0 + b1x1 + b2x2 +… + bnxn. In the applied regression model the cut-off probability value of 0.5 was used to classify samples according to studied criteria. Evaluation of obtained regression models was performed with χ2Wald test and Hosmer–Lemeshow test.
Disease-free survival and overall survival were defined as the period between initial surgery and recurrence and/or death or last contact. Survival times of patients still alive or lost during follow-up were censored in December 2010. Survival analysis was performed using univariate and multivariate Cox proportional hazard models and Kaplan–Meier estimator. To generate Kaplan–Meier curves log transformed normalized Cq values were converted into discrete variables by splitting the samples into “high” and “low” expression group using quartile method. Log-rank test was utilized for comparison of survival curves.
All statistical tests were two-sided and p < 0.05 was considered to indicate statistically significant difference. Data analysis was performed using GenEx 5.3.4. (MultiD) and MedCalc (MedCalc Software) version 12.2.1.
miRNA subsets identified in the microarray study
Of 24 samples analyzed by microarrays 22 (18 EEC and 4 NE) were available for final analysis. One sample was excluded based on PCA data quality assessment and the other due to technical problems during hybridization. Twenty-one miRNAs were regulated in EEC tissues comparing to control samples (Fig. 1a and Table 2).
Table 2. miRNAs regulated in EEC tissues in microarray study and in qPCR validation
Analysis performed in groups distinguished based on FIGO staging, histological grading, and myometrial invasion revealed regulation of 77, 20 and 25 miRNAs, respectively (Supporting Information Tables S1–S3 and Figs. S1–S3). Comparison of the four sets of regulated miRNAs revealed 16 molecules that were consistently distinguished in all four analyses: miR-9, miR-1305, miR-410, miR-141, miR-22*, miR-1228, miR-205, miR-203, miR-183, miR-200a*, miR-200b*, miR-200a, miR-200c, miR-141*, miR-429 and miR-200b.
Validation of microarray data
All 21 miRNAs distinguished by the microarray study were subjected to qPCR validation, which was performed in 73 EEC samples (including 18 used for microarray analysis) and 31 NE samples. In addition, expression of miR-92a, which showed a trend toward regulation in NormFinder analysis, was also analyzed in that group of samples. qPCR confirmed microarray results in case of 15 miRNAs (Figs. 1b and 1c). Moreover, miR-92a was found highly upregulated. Expression profiling performed with unsupervised PCA and based on differentially expressed miRNAs clearly distinguished between EEC and control samples (Supporting Information Fig. S4).
Relationship between miRNA expression and clinicopathological characteristics
To investigate potential role of miRNAs in tumor development and progression we compared miRNA expression in groups distinguished based on clinical and pathological characteristics. Analysis of the microarray study data revealed different panels of regulated miRNA in groups distinguished based on clinical stage, histological grade and myometrial invasion. The largest set of regulated miRNAs (77) was attributed to clinically more advanced tumors (Supporting Information Table S2). When tissue qPCR data were analyzed a significant relationship was found between expression of miR-200a, miR-200b*, miR-429, miR-9, miR-92, miR-449 a, and histological grade (Fig. 2a). Analysis performed based on FIGO staging revealed differences in expression of six miRNAs: miR-92a, miR-96, miR-200a, miR-203, miR-429 and miR-449a (Fig. 2b). Comparison of early EEC cases (stage FIGO IA Grade 1 tumors) with NE unveiled a set of 18 regulated miRNAs, which included miR-449a, upregulated in FIGO IA(G1) group (Supporting Information Figs. S5–S6). Four miRNAs (miR-96, miR-183, miR-205 and miR-449a) were significantly related to occurrence of relapse and two miRNAs (miR-203 and miR-429) to lymph node metastases (Figs. 3a and 3b). Expression of miR-92a and miR-200c in tissues was upregulated in tumors invading at least half of myometrial thickness (Fig. 3c).
miRNA signatures classify tumor tissue samples wit high sensitivity and specificity
Receiver operating characteristic (ROC) curves for discriminating EEC samples from normal endometrium were constructed based on miRNAs expression in tissues. Analysis of the ROCs revealed high area under curve (AUC) values for a number of single miRNAs with highest appointed to miR-92a, miR-205 and miR-182 (Fig. 4a, Supporting Information Table S4). Logistic regression analysis revealed two miRNA signatures: miR-92a/miR-410 and miR-92a/miR-205/miR-410, which classified tumor tissues with significantly higher accuracy in comparison to single miRNAs and yielded AUCs of 0.977 (95% CI: 0.927–0.996, p < 0.001) and 0.984 (95% CI: 0.938–0.999, p < 0.001), respectively (Fig. 4a, Supporting Information Tables S5–S8).
Prognostic value of miRNA signatures in tissues of EEC patients
None of the single miRNAs was accurate in prediction of relapse. However, a logistic regression model based on expression of miR-205 and miR-200a yielded high AUC value (0.854, 95% CI: 0.691–0.951, p = 0.002) in relapse prediction (Fig. 4b, Supporting Information Tables S9–S10). miRNA signatures: miR-200a*/miR-203/miR-429 (AUC: 0.818, 95% CI: 0.673–0.918, p = 0.002) and miR-92a/miR-182/miR-200a/miR-205 (AUC: 0.981, 95% CI: 0.887–1.000, p < 0.001) were found accurate in prediction of lymph node metastases and early EEC (FIGO IA(G1)), respectively (Figs. 4c and 4d; Supporting Information Tables S11–S14). Two miRNA signatures: miR-92a/miR-141* and miR-301b/miR-410 were specifically characteristic for Grade 3 tumors (Supporting Information Figs. S10 and S15–S17).
Lower expression of miR-141 (HR: 0.456, 95% CI: 0.243–0.854, p = 0.01), miR-203 (HR: 0.514, 95% CI: 0.274–0.966, p = 0.03) and miR-301b (HR: 0.299, 95% CI: 0.151–0.595, p = 0.006) was correlated with decreased overall survival and downregulation of miR-1228 was close to statistical significance (log-rank test, p = 0.057). Lower expression of miR-141 (HR: 0.461, 95% CI: 0.246–0.865, p = 0.01), miR-203 (HR: 0.504, 95% CI: 0.268–0.947, p = 0.029), miR-301b (HR: 0.295, 95% CI: 0.149–0.583, p = 0.005), miR-429 (HR: 0.504, 95% CI: 0.269–0.946, p = 0.037) and miR-1228 (HR: 0.498, 95% CI: 0.262–0.947, p = 0.02) was associated with shorter progression-free survival. Moreover, a multivariate Cox proportional hazard model revealed miRNA signatures, which proved to be the only independent prognostic markers of overall (miR-1228/miR-200c/miR-429, HR: 2.978, 95% CI: 1.580–5.614, p < 0.001; Fig. 5a) and progression-free survival (miR-1228/miR-429, HR: 4.149, 95% CI: 2.193–7.852, p < 0.001; Fig. 5b) in multivariate analyses, which included FIGO stage, histological grade, myometrial invasion and lymph node metastases (Supporting Information Tables S18–S21).
A set of miRNAs is regulated in plasma of EEC patients
Fourteen miRNAs were chosen for expression analysis in plasma of 34 EEC patients and 14 controls based on their high upregulation in tissues revealed in the microarrays analysis. Two additional miRNAs were also chosen: miR-1290, which was highly upregulated in high-grade EECs in the microarray analysis and miR-92a, which increased expression was revealed in the qPCR part of the study. Of 16 examined miRNAs, nine showed significant regulation in EEC plasma samples. Among them two were down regulated (miR-9 and miR-301b) and seven were upregulated (miR-92a, miR-141, miR-200a, miR-203, miR-449a, miR-1228 and miR-1290) (Fig. 6c).
The only associations between miRNAs expression in plasma and clinicopathological characteristics were attributed to miR-9 and miR-449a. Expression of miR-9 was lower in Grade 1 samples comparing to Grades 2 and 3 and miR-449a was upregulated in clinically more advanced samples (group FIGO > IA) (Supporting Information Figs. S7–S9).
Diagnostic value of plasma miRNA signatures
Analyses of ROC curves showed that two single miRNAs: miR-449a and miR-1228 could differentiate between normal and EEC tissues with high AUC values of 0.879 and 0.890, respectively. Moreover, logistic regression analysis revealed two miRNA signatures: miR-9/miR-1228 (AUC: 0.909, 95% CI: 0.789–0.973, p < 0.001) and miR-9/miR-92a (AUC: 0.913, 95% CI: 0.794–0.976, p < 0.001), which classified EEC plasma samples with higher accuracy in comparison to single miRNAs (Figs. 6a and 6b; Supporting Information Tables S23–S26). A miRNA signature based on expression of four miRNAs (miR-200b/miR-200c/miR-203/miR-449a) could differentiate between tumors invading <0.5 and ≥0.5 of myometrial thickness and yielded AUC of 0.851 (95% CI: 0.687–0.949, p = 0.005) (Supporting Information Fig. S11 and Tables S27–S28).
Although a number of studies addressed the issue of miRNAs profiling in endometrial cancer tissues, several questions have not been answered and only a few miRNA molecules were consistently reported by more than two research groups.26–30 That might be due to different platform and normalization strategies used, as well as utilization of sample sets with various histological types, grade and clinical stage layouts. To reveal miRNA signatures characteristic exclusively for EEC, which is the most common type of endometrial malignancy, our study comprised cases with histologically proven endometrioid histology. In addition, low and high clinical stage and histological grade tumors were represented in comparable numbers. Several authors suggested that technical pitfalls, which are likely to occur during miRNA profiling in tissues and especially in body fluids, could significantly influence the results and conclusions of qPCR studies.31–33 Therefore, we put a substantial effort into experimental design and validation of endogenous controls used for normalization. It is in fact the first study performed in EEC to use experimentally validated set of endogenous controls for relative expression analysis both in tissue and plasma. The potential drawback of our study could be connected with the differences in BMI and age, which were found between EEC and normal endometrium groups in the tissue part of the study. However, as no significant correlations were found between expressions of all studied miRNAs and age and BMI of patients we can speculate that these characteristics would have only minor impact on the local miRNA expression in comparison to cancerous transformation.
Our study revealed that a number of miRNAs were up or downregulated in EEC tissues. All miR-200 family members were significantly upregulated, which confirms results reported in few recently published studies.26,29,30 Moreover, expression of those miRNAs was most pronounced in early clinical stages of EEC and a systematic decrease of their expression was noted in higher FIGO stages and in poorly differentiated tumors, which was significant in case of miR-200a, miR-200b* and miR-429. The role of miR-200 family (miR-141, miR-200a, miR-200b, miR-200c and miR-429) in EEC cancer has not been elucidated so far. Members of that family were found downregulated in many malignancies and were connected to epithelial-to-mesenchymal transition.34 They were therefore attributed an oncosuppressor role in carcinogenesis. Contrary to this thesis upregulation of miR-200 family seem to constitute the most characteristic signature of EEC.26,29,30 Moreover, an inhibition of miR-200 family members with anti-miRs resulted in attenuation of cell proliferation and cytotoxic effect in endometrial cancer HEC-1A and Ishikawa cell lines.26 Such observations implicate an oncogenic potential of miR-200 in EEC. In our study, the highest expression of miR-200 family members was noted in early phase of the disease and all members were upregulated in samples featuring histological Grade 1 and clinical FIGO stage IA. Corresponding with our findings Snowdon et al. reported increased expression of miR-200 family members in EEC precursor lesions.27 Taken together, these results suggest that miR-200 family may facilitate oncogenesis in EEC and its role could be important during initial phase of the disease. miR-203 and miR-205 were also upregulated in EEC, whereas their decreased expression was observed in other malignancies and was connected with more advanced disease and worse prognosis.17,20,30,35 In our study, miR-205 was highly upregulated and along with miR-183, it was significantly increased in cases with disease recurrence. Apart from miR-183, two other miRNAs belonging to the same family, miR-96 and miR-182, were also highly increased in EEC samples. In addition, we found an upregulation of miR-9. These results correspond with observations presented in other EEC studies and implicate oncogenic roles of miR-183, miR-96, miR-182 and miR-9 in EEC.17,19,36
Decreased expression of miR-410, which was revealed in our set of EEC samples, was previously reported in only one study.30 Interestingly, Ratner et al. observed downregulation of miR-487b, which occupies the same locus on chromosome 14 (14q32.31). Decreased expression of those two miRNAs seems to be particularly significant in view of the fact that region 14q32 is often deleted in endometrial cancers and is connected with worse prognosis.37
Altered expression of miR-92a and miR-1305 in EEC has not been reported to date. miR-92a, showed the highest degree of upregulation in the qPCR study, albeit it was not regulated in the microarray experiment. Such discrepancy might be due to smaller population included in the microarray part of the study and different normalization strategies applied in the two methods. Expression of miR-92a was higher in clinically and histologically more advanced tumors. Moreover, it tended to be increased in patients with disease relapse. Such observations suggest oncogenic role of miR-92a in EEC and its association with more aggressive tumors. miR-1305 was upregulated in the microarray study and downregulated in qPCR analysis. Due to inconsistent results miR-1305 was not considered in analyses regarding diagnostic and prognostic significance of miRNAs.
To investigate possible involvement of regulated miRNAs in EEC pathogenesis we applied DIANA-mirPath search and found that they targeted a large number of genes involved in a PTEN-PI3K-AKT-mTOR and MAPK pathways, which roles have been implicated in the development of EEC.38–40
To our best knowledge, our study is the first one to systematically assess diagnostic and prognostic value of miRNAs in EEC tissue and matched plasma samples. ROC curves analysis revealed that a number of single regulated miRNAs could distinguish between EEC and control tissue samples yielding high AUCs. In addition, two miRNAs signatures, miR-92a/miR-410 and miR-92a/miR-205/miR-410, classified EEC tumor tissues with higher accuracy in comparison to single miRNAs. We also found that distinct miRNA signatures yielded high AUCs for distinguishing early EEC cases (92a/miR-182/miR-200a/miR-205) and poorly differentiated tumors (miR-92a/miR-141* and miR-301b/miR-410). High sensitivity and specificity of such miRNA signatures could potentially facilitate histological diagnostics in difficult cases of poorly differentiated or metastatic tumors and scanty biopsy material. As miRNAs expression alterations are present during early stages of EEC development, miRNA based diagnostic test could potentially depict cases of atypical hyperplasia prone to faster progression into cancer.
If diagnosed in the early stage endometrial cancer is usually managed surgically with good results. However, patients with advanced disease comprise a difficult and very heterogeneous population in regards to treatment and prognosis. To date, no good molecular prognostic marker has been developed in EEC. In our study, we aimed to find a miRNA signature, which could potentially serve as a marker for distinguishing patients with worse prognosis. Along with clinical characteristics such signature could facilitate identification of patients, who could benefit from experimental or more aggressive therapeutic methods. By the means of a multivariate logistic regression analysis we have identified a miRNA signature (miR-205/miR-200a) capable of distinguishing patients with relapse with high AUC value (0.854, 95% CI: 0.691–0.951). In addition, we found that a model developed based on the expression of three miRNAs (miR-200a*/miR-203/miR-429) distinguished cases with nodal metastases with 80% sensitivity and 79% specificity, when compared to EEC without nodal involvement.
We also found that miRNA signatures were associated with overall (miR-1228/miR-200c/miR-429) and disease-free (miR-1228/miR-429) survival and were independent of clinical stage, grade, myometrial invasion and lymph node status. We suggest that such miRNA signatures after evaluation in other populations and in the prospective studies could serve as predictors for prognosis in EEC.
Diagnostic and prognostic significance of serum/plasma miRNA signatures has been suggested in various types of cancer.40–43 However to date no systematic evaluation of miRNA profiles in matched tissue and plasma samples has been performed in EEC patients. The source of plasma miRNAs has not been determined so far. Some authors suggested that they might derive from exosomes shed from normal and tumor cells.44 On the basis of such hypothesis, we chose a set of miRNAs upregulated in EEC tissues for qPCR expression profiling in plasma samples. RNA content in plasma may vary significantly between samples in normal and especially in disease states. Therefore, proper normalization for sample-to-sample variation in RNA isolation step as well as for variance introduced during downstream procedures is vital for acquiring biologically relevant results. In our study, we paid special attention to data normalization strategies. A median normalization procedure using expression of three synthetic spiked-in Caenorhabditis elegans miRNAs was applied for normalization of RNA isolation step as it was previously described.22 Experimentally validated set of five endogenous controls was applied to normalize for RNA quality, as well as variance in reverse transcription and qPCR steps. Data analysis revealed upregulation of seven and downregulation of two miRNAs in plasma of EEC patients. Decreased plasma expression of miR-9 and miR-301b was an unexpected finding and at present it is difficult to explain that phenomena. However, discrepant expression of another miRNA—miR-92a, was observed in tissue and plasma samples obtained from patients with leukemia, lymphoma and hepatocellular carcinoma. One hypothesis describing such phenomena could be that apart from transcribing essential miRNAs, cancer cell might specifically take in exosomes containing those miRNAs from the blood. As a result decreased expression of miRNAs would be observed in plasma.45–47 Two single miRNAs, miR-449a and miR-1228 showed high sensitivity and specificity in classifying EEC and control samples. Moreover, logistic regression analysis revealed two miRNA signatures: miR-9/miR-1228 and miR-9/miR-92a, which classified EEC plasma samples with higher accuracy in comparison to single miRNAs. In addition expression of miR-9 was associated with histological grade and miR-449a upregulation correlated with higher clinical stage. Other serum/plasma biomarkers studied in endometrial cancer were reported to yield lower sensitivity values in comparison to miRNA signatures developed in our study.48,49
Greater invasion of myometrium has been connected with higher risk of lymph node metastases and persistent or recurrent disease. However, intraoperative gross examination of myometrial invasion is often inadequate and may lead to wrong clinical decisions.50 Reliable preoperative markers of greater myometrial invasion could facilitate optimal surgical approach. We therefore attempted to develop a plasma miRNA signature, which could discriminate between EEC samples with different degree of myometrial invasion. A logistic regression model based on expression on four miRNAs (miR-200b/miR-200c/miR-203/miR-449a) yielded AUC of 0.851 with 94% sensitivity and 67% specificity in discerning patients with tumors invading at least half of myometrial thickness. Obtained results need to be validated in a larger population of patients but implicate the potential value of miRNAs in preoperative detection of EEC cases, which could benefit from more radical surgery. They are also promising in comparison to other blood derived biomarkers like Ca 125, imaging methods and clinicopathological risk factors which were reported to predict more advanced disease.51–53
In conclusion, our study revealed that a set of miRNAs was regulated in tissues and plasma of EEC patients. We confirmed expression alterations of several miRNAs that were previously indicated in EEC, which helps to establish a miRNA panel characteristic for this malignancy. At the same time we found deregulation of several miRNAs that were not reported so far in EEC. Moreover, our study revealed that expression of various miRNAs was associated with clinicopathological characteristics, which contributes to the understanding of miRNAs involvement in EEC pathogenesis. Lastly, we have assessed diagnostic and prognostic value of miRNA signatures both in tissues and plasma and found that they were accurate in distinguishing between EEC and control samples and were significantly associated with survival. Although our results need to be confirmed in larger, prospective studies we conclude that miRNA signatures hold a great promise to become noninvasive biomarkers for early EEC detection and prognosis.
We thank Drs. Paulina Wdowiak and Joanna Kozak as well as Mrs. Dominika Wilkołek, M.Sc., for their excellent technical help. This work was supported by grant from Polish Ministry of Science and Higher Education (to A.T.).