In recent years, researchers have observed that noncoding RNAs have a striking variety of biological functions[22].
Understanding the crosstalk among lncRNAs, miRNAs and mRNAs on the basis of the ceRNA hypothesis will lead to significant insights into gene regulatory networks with implications on tumorigenesis. However, there have been relatively few studies on the roles of lncRNAs, miRNAs and mRNAs as ceRNAs in EC.
In the present study, through bioinformatics miRNA target prediction and co-expression analysis of lncRNA-mRNA pairs, we constructed a ceRNA network with 15 DElncRNAs, 30 DEmRNAs and 8 DEmiRNAs in EC. In our ceRNA network, lncRNAs and mRNAs interacted with each other via miRNAs, forming a complex regulatory network.
In the EC ceRNA network, lncRNA MAGI2-AS3 was the most prominent lncRNA node that competed with some other lncRNAs (AC009093.1, TTTY14, SNX29P2, etc.) and sponged hsa-mir–143, hsa-mir–93, hsa-mir–204 and hsa-mir–372. Additionally, by interacting with these miRNAs, lncRNA MAGI2-AS3 regulated the expression of some mRNAs (RGMB, FSCN1, TMEM100, SLC2A4, etc.). We also found that the expression level of MAGI2-AS3 was associated with EC pathological stage. This implied that lncRNA MAGI2-AS3 might play a major role in EC pathogenesis. In recent studies, lncRNA MAGI2-AS3 has been shown to suppress bladder cancer progression by sponging miR–15b [26]. However, reports on lncRNA MAGI2-AS3 sponging miRNAs and interacting with mRNAs in EC are limited.
ceRNAs can affect the abundance of miRNAs and their impact on targets and impose an additional level of posttranscriptional regulation. MiRNAs act as key regulators in ceRNA networks[27]. In this study, hsa-mir–93 represented the node with the highest degree in the EC network and targeted 7 lncRNAs and 14 mRNAs (TGFBR2, OSR1, SLC2A4, etc.). These results were not consistent with the EC ceRNA network proposed by Li-Ping Chen et al., as hsa-mir–93 was not in their ceRNA network[28]. In previous studies, hsa-mir–93 was shown to be an important oncogene in prostate cancer[29] and to function as biomarkers for the diagnosis and prognosis of esophageal cancer.[30] Additionally, many studies have confirmed that hsa-mir–93 shares several lncRNA-binding sites [31–33]. Our results may shed new light on the interactions of hsa-mir–93 with lncRNAs and mRNAs in EC.
In our ceRNA network, SNX29P2-TGFBR2 pair was the common target of several miRNAs, such as hsa-mir–93, hsa-mir–372, hsa-mir–17 and hsa-mir–145. TGFBR2 was also the mRNA node with the highest degree in the EC network. TGFBR2 is a member of the TGFB receptor subfamily. In previous research, TGF-β signaling was shown to play an important role in carcinogenesis[34], and hsa-mir–17 can impede migration and invasion via TGF-β pathway in esophageal squamous cell carcinoma[35]. Our data implied that lncRNA SNX29P2 might function as a ceRNA to impede the TGF-beta pathway in EC.
In the EC ceRNA network, sponging of overexpressed lncRNA AC009093.1 and under expressed lncRNA MAGI2-AS3, relieved their targets COL1A1 and COL5A2 from being suppressed by hsa-mir–143. Previous studies have suggested that hsa-mir–143 functions as a tumor suppressor in several cancer types[36, 37], and COL1A1 has been shown to act as an oncogene to regulate tumors [38–40]. Our findings implied that hsa-mir–143 may function as a tumor suppressor by targeting COL1A1 and COL5A2 in EC and competing with lncRNA MAGI2-AS3/AC009093.1.
Most of the ceRNA triples included downregulated lncRNA-mRNA pairs and upregulated miRNAs, whereas in this study, downregulated hsa-mir–204 targeted many downregulated lncRNA-mRNAs. Many studies have demonstrated downregulation of hsa-mir–204 in the progression of various types of malignant tumors, including EC and have shown that hsa-mir–204 functions as a tumor suppressor[41]. This finding indicated that miRNA expression may not always be correlated with target ceRNAs because of the complexity of crosstalk among RNA transcripts.
Our study has provided a global view of regulatory ceRNA interaction networks in EC and laid a foundation for further functional research in EC. Because of their functional roles, lncRNAs have attracted much attention in the discovery of prognostic biomarkers [42]. Some studies have reported that lncRNAs may serve as biomarkers for EC, however, there are no commonly accepted prognostic biomarkers.
In survival analysis, the lncRNA expression data may have led to problems in predictive modeling because these data indicate an imbalance between a large number of covariates involving lncRNAs and a small number of samples, denoted as an issue with dimensionality [43].
In this study, to overcome the limitation involving high-dimensionality of expression data for 514 lncRNAs, we used a robust likelihood-based survival model utilizing a cross-validation technique to reduce dimensionality [44, 45]and finally developed a six-lncRNA signature to predict OS for EC patients. There have been very few studies on ZNF341-AS1, AC130324.2, AC027271.1, AL591212.1, AL732314.4 and LOC105372352. The six-gene signature was identified as an indicator for EC with high prognostic significance and demonstrated that lncRNAs have significant prognostic value for EC patients [46].
Our results were not consistent with the findings of Guo-Wei Huang et al. [47], in which a three-lncRNA signature including RP11–366H4.1.1, LINC00460 and AC093850.2 was identified as a potential prognostic indicator for esophageal squamous cell carcinoma. The discrepancy may be mainly ascribed to data and prediction method.