Prognostic Value of microRNA Signature in Patients with Gastric Cancers

The occurrence of lymph node metastases (LNM) after endoscopic submucosal dissection (ESD) in patients with gastric cancer (GC) leads to poor prognosis. However, few biomarkers are available to predict LNM in GC patients. Thus, we measured expression of 6 cancer-related miRNAs using real-time RT-PCR in 102 GC samples that were randomized into a training set and a testing set (each, 51 cases). Using logistic regression, we identified 4-miRNA (miR-27b, miR-128, miR-100 and miR-214) signatures for predicting LNM in GC patients. Patients with high-risk scores for the 4-miRNA signature tended to have higher LNM than those with low-risk scores. Meanwhile, the ROC curve of the 4-miRNA signature was better for predicting LNM in GC patients. In addition, Cox regression analysis indicated that a 2-miRNA signature (miR-27b and miR-214) or a miR-214/N stage signature was predictive of survival for GC patients. This work describes a previously unrecognized 4-miRNA signature involved in LNM and a 2-miRNA signature or miR-214/N stage signature related to GC patients’ survival.

MUC1 25 , Catenin-δ 1 26 and N-cadherin 9 . As well, miR-214 inhibited tumor invasion by reducing expression of β -catenin 27 and FGFR1 28 in hepatocellular carcinoma cells. So, these six miRNAs play effective tumor-regulatory function by modulating multiple target genes, and newly discovered target genes are increasing. Thus, in our study, we chose these six miRNAs for investigation and analyzed any correlation between them and patient prognosis or lymphatic metastasis.
Specifically, we measured expression of the 6 miRNAs in 102 cases of GC and identified a 4-miRNA signature to predict LNM and a 2-miRNA signature or miR-214/N stage for predicting overall survival for GC patients.

Methods
Patients and clinical clinicopathological parameters. One hundred two formalin-fixed paraffin-embedded (FFPE) tissue samples of GC obtained from Qilu Hospital, Shandong University, Jinan, China, between 2004 and 2007, were collected for miRNA expression analysis. GC tissues were confirmed by 2 pathologists. Of the 102 patients, the median follow-up time was 63.4 months (ranged 1 to 67 months). Among the 102 participants, there were 65 patients with LNM and 37 patients without LNM. Subject characteristics are summarized in Supplementary Table 3. Methods were performed according to the approved guidelines. RNA extraction. Paraffin tissues were cut into 4-mm thick sections, then dewaxed, rehydrated and lightly stained with hematoxylin. Tumor tissues were inspected and microdissected with a 25 G needle under a dissecting microscope. MiRNA extraction was performed with a miRNeasy FFPE kit (Bioteke, Beijing, China) according to the manufacturer's protocol, which isolated miRNA from FFPE tissue sections.
Real-time quantitative reverse transcription-PCR. The reverse transcription (RT) was conducted with 100 ng total miRNA and quantitative PCR reactions were done by the 7900HT system (Applied Biosystems, Foster City, CA) using an All-In-OneTM qRT-PCR detection kit (Genecopeia, Rockville, MD) following the manufacturer's protocol. Primers for measuring 6 miRNAs were synthesized by Genecopeia. We calculated the relative amounts of selected miRNAs using the equation 2 −ΔΔCt . U6 and RNU44 were detected by qRT-PCR as an endogenous control. Based on the median score of individual miRNA of the GC patients, expressions of the 6 miRNAs (miR-27b, miR-101, miR-128, miR-100, miR-145 and miR-214) were classified as a high-or low-expression.

Statistical analysis.
Associations between miRNA expression and clinical characteristics were assessed with either the Student's t test, the Chi-squared test or Fisher's exact test. A stepwise logistic regression model was used to select predictive miRNA markers. The predicted probability of positive LNM with GC was used as a surrogate marker to establish the ROC curve. The area under the curve (AUC) was used as an accuracy index for evaluating the predictive performance of the selected miRNA signature. Univariate Cox regression analysis was used to evaluate the hazard ratio (HR) of miRNA and clinical variables for patient survival. Multivariate Cox regression analysis was conducted to test for independent prognostic factors of OS. A prognostic score model was constructed to compare the miRNA signature prognostic validity with the individual miRNA model using ROC analysis. Overall survival (OS) was defined as the length from the operation date to the date of death or the final follow-up. The Student's t test, the Chi-squared test or Fisher's exact test were performed using the statistical software Prism 5 (GraphPad Software, La Jolla, CA), and all the other statistical tests were performed with SPSS version 20.0. Statistical significance was defined p < 0.05.

Results
Cancer-related miRNA expression of the GC training set. Specimens (N = 102) randomized into a training and a testing set (both, N = 51). In the training set, of the 6 miRNAs, miR-27b, miR-101, miR-128, miR-100 and miR-214 in expression for patients with LNM was significantly lower than for patients without LNM (Fig. 1), indicating a correlation between expression of 5 miRNAs and LNM. No correlation was found between miR-145 expression and LNM. Additionally, lower expression of miR-101, miR-128, miR-145 and miR-214 was associated with larger tumor size (p = 0.0395, p = 0.0179, p = 0.00395, and p = 0.0008, respectively). According to the American Joint Committee on Cancer 29 , tumor, node and metastasis (TNM) classification T stage was categorized into T1a (mucosa), T1b (submucosa), T2 (muscularis propria), T3 (subserosa) and T4 (tumor invades adjacent structures). A correlation was found between less miR-145 or miR-214 and positive T stage of the TNM classification (p = 0.0365 and p = 0.0203, respectively). To discover LNM-specific miRNAs in GC, the predictive accuracy of one single miRNA to distinguish between patients with and without LNM was assessed using a receiver operating characteristic (ROC) curve. Considered individually, apart from miR-145 ( Fig. 1E; AUC = 0.6161, p = 0.1649), miR-27b, miR-101, miR-128, miR-100, and miR-214 had high AUC ( Fig. 1A-D,F). A correlation between expression of 6 miRNAs and patient clinicopathologic characteristics appear in Supplementary Table 1.
Validation of cancer-related miRNAs in the testing set and for both GC sets. We measured expression of 6 miRNAs in the testing set and in both GC sets and miR-27b, miR-128, miR-100 and miR-214 for the patients with LNM were significantly lower than for patients without LNM (Figs 2 and 3). No association was found between expression miR-101 or miR-145 and LNM. Additionally, to validate potential miRNAs to predict LNM in GC patients, the ROC curve in the testing set was investigated and results show that miR-27b, miR-128, miR-100, and miR-214 ( Fig. 2A,C,D and F) had significantly high AUC scores excluding miR-101 and miR-145 ( Fig. 2B and E). Distribution of the expression of 6-miRNAs in the 65 GC samples with LNM and 37 GC samples without LNM (training and testing sets) by hierarchical clustering showed a relative separation between the two groups apart from miR-101 (Fig. 4). Expressions of miR-27b, miR-128, miR-100 and miR-214 were down-regulated in GC samples with LNM; but in GC samples without LNM, expression of the 4 miRNAs increased. Although clustering of miR-145 indicated a relative separation, its expression was not correlated with LNM (Figs 1E,2E and 3E). Therefore, miR-101 and miR-145 were eliminated from the candidate miRNAs. In conclusion, ROC curve analysis in the training and testing sets and clustering analysis indicated that miR-27b, miR-128, miR-100, and miR-214 had stable predictive value for LNM. Associations between the 6 miRNA expression and patient clinicopathologic characteristics in the testing set are shown in Supplementary Table 2. In addition, the association between the 4 miRNA and GC patients with LNM in combination of the two sets was demonstrated by ROC curve. Our data revealed that miR-27b, miR-128, miR-100 and miR-214 had dramatically high AUC scores (Fig. 3A,C,D,F). Thus, a combination (miR-27b, miR-128, miR-100 and  miR-214) was selected to achieve the best prediction of LNM development of GC patients. The correlation between 6 single miRNA expression and patient clinicopathologic characteristics using the 2 sets is shown in Supplementary Table 3.
It was reported that RNU44 was more suitable than U6 as the endogenous control to normalize relative expression of miRNAs 30,31 . So, we analyzed the relationship between 6 miRNAs and LNM based on expression of 6 miRNAs normalized by RNU44. Expression of miR-27b, miR-128, miR-100 and miR-214 in patients with LNM were significantly lower than for those without LNM. No correlation was found between expressions of miR-101 or miR-145 and LNM. In addition, the ROC curve demonstrated that 4 miRNAs (miR-27b, miR-128, miR-100, and miR-214; Supplementary Figure 1A,C,D and F) had significantly high AUC scores except miR-101 and miR-145 (Supplementary Figure 1B and E). These data agree with results using U6 as a normalizer. To validate the 4-miRNA signature for predicting LNM for patients with GC, patients were classified into high-or low-risk groups based on a threshold of − 1.237, the maximum value of sensitivity and specificity plus 1. As expected, patients with high-risk scores tended to have higher LNM than did those with low-risk scores (Fig. 5A, p < 0.0001). We assessed the predictive accuracy of the 4-miRNA-based classifier with ROC analysis and the miRNA marker set can predict LNM in GC patients as indicated in Fig. 5B and Supplementary Table 4, which was superior to that of the individual miRNA markers. Additionally, compared with patients in the low-risk group, patients in the high-risk group had advanced T and N stages (Table 1). Subsequently, training or testing set subjects were classified into high-or low-risk groups. In the training set, compared with the low-risk group, the high-risk group had higher LNM and larger tumor size (Supplementary Table 6). Additionally, in the testing set, patients in the high-risk group had increased LNM and advanced N stage compared with the low-risk group (Supplementary Table 7).  (Fig. 6A and F). However, there was no significant association between expression of miR-101, miR-128, miR-100 or miR-145 and OS (Fig. 6B-E; p > 0.05). To investigate whether single miRNA markers may function as an independent prognostic factor for OS in GC, we evaluated the association between the 6 miRNAs or clinicopathological parameters and prognosis by Cox regression analysis. Univariate Cox regression confirmed that miR-27b, miR-214 and 4 clinical features, such as LNM, tumor size, clinical stage and TNM, were significant predictors (Table 2). Moreover, multivariate Cox regression confirmed that miR-214 and N stage (TNM classification) were independent prognostic factors for OS (Table 2). Next, to identify a comprehensive predictive value of miR-27b and miR-214, we integrated the 2 miRNAs into a comprehensive factor using Cox regression analysis to develop a new risk score formula: Risk score = 0.655 × expression of miR-27b + 0.687 × expression of miR-214, weighted by regression coefficient.
Using ROC analysis to compare the accuracy of survival prediction, the established miRNA panel better predicted survival than did the individual markers with regard to OS (Fig. 5D-F, Supplementary Table 5). In conclusion, the 2-miRNA signature had better prognostic value than that of individual miRNA markers for GC patients.
Next, we analyzed a combination of low expression of miR-214 and N stage with patient survival. We integrated the miR-214 and N stage into a comprehensive factor by Cox regression analysis to develop a new risk score formula: Risk score = 0.566 × the expression of miR-214 + 0.711 × N stage (N0 = 0; N1, N2 or N3 = 1), weighted by regression coefficient. The established miR-214/N stage signature better predicted survival than did the individual marker with regard to OS (Fig. 5C, Supplementary Table 8). Thus, the miR-214/N stage signature was better for prediction than the miR-214 or N stage alone for GC patients.   also indicated that high expression of the 5-gene signature was strongly correlated with patients' poor prognosis, and may be potential prognostic predictors in GC 33 . However, miRNA profiles have been indicated to have greater utility than mRNA profiles as prognostic biomarkers on account of miRNA stability within clinical specimens 34 . Differential miRNA expression signatures have been explored in various human cancers and miRNA expression alterations were associated closely with progression and prognosis of human malignant cancers 35,36 . We hypothesized that miRNA signatures may predict LNM and GC patients' prognosis.
In this study, we describe associations between aberrant expression of specific miRNA signatures and LNM and miRNA signatures and OS of GC patients. We developed a 4-miRNA signature (miR-27b, miR-128, miR-100 and miR-214) related to LNM. Patients with high-risk scores of this 4-miRNA signature tended to have higher LNM than did those with low-risk scores. We assessed the predictive accuracy of the 4-miRNA based classifier with ROC analysis. The ROC curve of the 4-miRNA signature had higher AUC than did individual miRNA markers. In addition, our results showed that the 4-miRNA-based classifier can predict LNM in GC patients with a significantly higher sensitivity of 89.19 than did individual miRNA markers. Thus, these data suggested that the combined detection of miR-27b, miR-128, miR-100 and miR-214 had more considerable clinical value to predict LNM in GC patients. Furthermore, high-risk scores of this 4-miRNA signature tended to have advanced TNM classification for GC patients compared to those with low-risk scores.
Also, univariate Cox regression demonstrated that miR-27b, miR-214 and 4 clinical features, such as LNM, tumor size, clinical stage and TNM were significant survival predictors for GC patients. Consequently, multivariate Cox regression suggested that miR-214 and N stage (TNM classification) were independent prognostic factors for OS. Thus, we developed a 2-miRNA signature (miR-27b and miR-214) and miRNA/N stage signature that were predictive of OS in GC patients. The ROC curve of the 2-miRNA signature or miR-214/N stage had higher AUC than did individual markers. The combination of miRNAs (miR-27b and miR-214) or miR-214/N stage better predicted survival than did individual markers with respect to OS. These results may add supportive evidence that miRNAs have roles for prediction of LNM in GC patients and may improve our understanding of clinical cancer progression and prognosis of GC.
At this time, studies indicate a relationship between 4 miRNAs and LNM or patient prognosis in various human cancers. However, no report is available for miR-27b, miR-128, miR-100 and miR-214 as a signature for predicting LNM in patients with human cancers or the integration of miR-27b and miR-214 or miR-214/N stage as a signature for predicting patient prognosis in human cancers. To our knowledge, this is the first report indicating that the 4-miRNA signature (miR-27b, miR-128, miR-100 and miR-214) possess the prognostic value for LNM in GC patients, contributing to potentially personalized therapy. In addition, our data showed that the 2-miRNA signature or miR-214/N stage could be a useful predictive biomarker for patient's OS, suggesting a prognostic value for GC patients. However, the present study was limited as only 102 GC patients were involved. The conclusion will need to be validated with large-scale clinical cases in different areas and involving multicenter studies before the 4-miRNA signature and 2-miRNA signature or miR-214/N stage could be clinically applicable for GC patients.
Ethics approval and consent to participate. The collection of tissue samples was obtained with informed consent and approval to conduct this study was obtained from the Ethics Committee of Shandong University, China.