Construction and comprehensive analysis of a ceRNA network to reveal potential prognostic biomarkers for hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is a severe cancer with an increasing incidence and is the fifth most prevalent cancer worldwide [1‐3]. Viral infections, such as hepatitis B and hepatitis C, are usually associated with liver cirrhosis and HCC tumorigenesis [2, 4]. HCC remains an important global clinical challenge due to its high incidence, limited treatment strategies, and poor prognosis [5]. Therefore, strategies based on a personal need to treat HCC, such as discovering potential biomarkers and therapeutic targets, are urgently needed. The present study explored how HCC-related long noncoding RNAs (lncRNAs) serve as competitive endogenous RNAs (ceRNAs) to regulate target genes and how they affect the pathogenesis and prognosis of HCC.

Although noncoding RNAs (ncRNAs) lack protein-encoding abilities, they are ubiquitous in organisms [6]. As a subtype of ncRNA greater than 200 nucleotides in length, lncRNAs were once considered transcriptional noise. Several studies have indicated that lncRNAs have many pivotal functions in tumor-related processes, including proliferation, invasion, and metastasis [7‐9]. However, verification of the regulatory roles of lncRNAs in gene expression remains difficult. Currently, many researchers are working to reveal the different biological functions of lncRNAs in malignant tumors.

One ceRNA hypothesis was proposed by Salmena et al. [10]. A complicated posttranscriptional regulatory network was described that allowed lncRNAs, mRNAs, and other RNAs to compete with microRNAs (miRNAs) via acting as natural miRNA sponges by virtue of sharing no less than one miRNA response element (MRE). These ncRNAs act as ceRNAs to modulate mRNA expression and regulate protein levels, which contributes to the occurrence and development of tumors [11, 12]. Studies have shown that each miRNA can control the transcriptional expression levels of hundreds of proteins, and each mRNA contains different MREs and thus may be targeted by multiple miRNAs [12].

Recently, a growing number of studies have verified that the lncRNA–miRNA–mRNA regulatory network plays a critical role in the progression and pathogenesis of several tumors, including liver cancer, gallbladder cancer, and other malignant tumors [13‐15]. lncRNAs with sequences similar to their target miRNAs can separate miRNAs from mRNAs. Wang et al. demonstrated that the lncRNA HULC influenced PRKACB gene expression by competitively combining with the miRNA miR-372 and thus participated in liver cancer pathogenesis [13]. Wang et al. confirmed that lncRNA H19 acts as a molecular sponge to absorb miR-342-3p in gallbladder cancer and upregulate FOXM1 gene expression [15].

Therefore, lncRNAs acting as ceRNAs have diverse biological functions that deserve further exploration. Here, we investigated differences in RNA expression patterns between 43 HCC tumor tissues and 43 paired nontumorous tissues and constructed a ceRNA network associated with HCC, including 26 lncRNAs, four miRNAs, and six mRNAs. The survival analysis showed that four lncRNAs (MYCNOS, DLX6-AS1, LINC00221, and CRNDE) and two mRNAs (CCNB1 and SHCBP1) were prognostic biomarkers for patients with HCC in both the Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. These candidate genes involved in the ceRNA network may become potential therapeutic targets or diagnostic biomarkers for HCC.

Approximately 782,500 new cases of HCC occur each year, making HCC the second greatest cause of cancer mortality worldwide [24]. Although many treatment strategies (i.e., radiofrequency ablation, surgical resection, and liver transplantation) have been adopted, the prognoses of HCC patients who undergo these treatments are still not satisfactory [25]. Most patients are diagnosed at advanced stages due to the asymptomatic nature of the disease. Elucidating the molecular mechanisms and processes underlying HCC is of great importance for identifying new therapeutic targets and improving the clinical outcomes of patients with this disease. More and more studies indicate that lncRNAs play a vital role in biological functions through multiple levels of regulation, which involve transcriptional, posttranscriptional, and epigenetic regulation [26, 27]. A great deal of research has shown that there is a complex and closely related regulatory network between miRNAs and lncRNAs. In addition, lncRNAs and miRNAs have crucial functions in the pathogenesis and progression of cancers [27, 28]. Thus, these lncRNAs may serve as tumor-related prognostic indicators. Recently, Robinson and Henderson constructed a network model to identify hypothetical mRNA–miRNA interaction networks associated with epithelial function [29]. A mRNA–miRNA interaction list was generated using functional-molecular databases and network modeling. The authors quantified and visualized inherent network structures by using R code and identified a subnetwork containing a large number of shared, targeting miRNAs, of genes related to cancer and cellular proliferation, including cyclin D and c-MYC. The complexity of the miRNA–mRNA interaction network represents an obstacle for predicting and verifying the function of the ceRNA network [29]. The ceRNA hypothesis was proposed to explain the mechanism of tumorigenesis; this hypothesis provides a novel guiding theory and suggests valuable strategies and research directions for the diagnosis and treatment of malignancies [10]. lncRNAs with sequences similar to those of their target miRNAs are able to regulate the expression of mRNAs by acting as sponges for miRNAs [10].

To the best of our knowledge, few studies on ceRNAs have focused on predicting HCC prognosis. Additionally, rare yet reliable lncRNAs, miRNAs, and mRNAs related to HCC can be treated as molecular biomarkers to detect HCC and stratify the HCC risk. Under this background and hypothesis, lncRNA, mRNA, and miRNA data from 371 HCC samples were collected from the TCGA. Based on comprehensive integration of the lncRNA, mRNA, and miRNA data from the 43 paired HCC samples, a ceRNA network associated with lncRNAs was constructed to study the regulatory mechanism of the ceRNAs. Although the biogenesis of HCC is extremely complex, the complexity of tumor growth and metastasis dissemination can be largely represented by DEGs; therefore, we used DEGs to construct the ceRNA network. Here, we develop a ceRNA network including 26 lncRNAs, four miRNAs, and six mRNAs that are specific to HCC. Among them, four lncRNAs (MYCNOS, DLX6-AS1, LINC00221, and CRNDE) and two mRNAs (CCNB1 and SHCBP1) presented an obvious relevance to overall survival using patient data obtained from both the TCGA and GEO databases.

In the ceRNA network, the lncRNA LINC00221 (connection degree = 2) had the highest connection degree among the prognostic DElncRNAs (MYCNOS, DLX6-AS1, LINC00221, and CRNDE) (Additional file 5: Table S5). Therefore, we concluded that it might exert a strong influence on HCC pathogenesis. Russell et al. showed that the expression of LINC00221 is 2.4-fold higher in high-risk neuroblastomas than in low-risk neuroblastomas [30]. LINC00221 was involved in ceRNA networks in several tumor types. Fan et al. reported that LINC00221 is upregulated in breast cancer compared to adjacent-normal breast tissues and is involved in the lncRNA–miRNA–mRNA ceRNA network of breast cancer [31]. He et al. demonstrated that LINC00221 is dysregulated in gastric cancer and competes with six miRNAs (hsa‑miR‑96, hsa‑miR‑143, hsa‑miR‑204, hsa‑miR‑372, hsa‑miR‑373 and hsa‑miR‑519d) to mediate target mRNA expression in gastric cancer [32]. Zhang et al. found that LINC00221 is associated with a poor prognosis in patients with gastric cancer, and an 11‑lncRNA signature consisting of LINC00221 can predict the survival rate for gastric cancer [33]. Wang et al. noted that LINC00221 is involved in the lncRNA–miRNA–mRNA ceRNA network of muscle-invasive bladder cancer, and its high expression level was significantly related to the progression of muscle-invasive bladder cancer [34]. Our study also showed that LINC00221 expression is 77-fold higher in HCC tissues than in paired nontumorous tissues. Increased expression of LINC00221 was associated with a poor prognosis in patients with HCC. Moreover, we found that LINC00221 might compete with two key DEmiRNAs (hsa-miR-182 and hsa-miR-96) to mediate target DEmRNA expression in HCC.

In addition to lncRNAs, miRNAs should also receive extensive attention. Undoubtedly, research related to tumorigenicity in terms of miRNA regulation is critically needed. miRNAs are RNA molecules of approximately 22 nucleotides in length that bind to the 3′-untranslated region (3′-UTR) of their respective target genes and exert their influence on gene expression by either inhibiting protein translation or degrading mRNA [35]. miRNAs are known to be an integral component of cancer development [35]. We noted that the DEmiRNA hsa-miR-182 (connection degree = 13) had the highest connection degree among the prognostic DEmiRNAs (hsa-miR-182 and hsa-miR-183) in the ceRNA network, suggesting an obvious influence of hsa-miR-424 on HCC pathogenesis and prognosis (Additional file 5: Table S5). miR-182 is located on human chromosome 7q31-34 [36]. miR-182 promotes tumorigenesis in a variety of tumors and is one of the most frequently studied cancer-associated miRNAs [36]. In glioblastoma, miR-182-5p targets protein phosphatase 1 regulatory inhibitor subunit 1C [37]. By downregulating RAB27A expression, miR-182-5p improves the migration, mitosis, viability, and invasion capabilities of human gastric cancer cells [37]. Inhibiting miR-182-5p by regulating CASP9 expression confers pro-apoptotic and anti-proliferative effects in human breast cancer [38]. Activated STAT3 induces miR-182-5p expression, which enhances the growth of gliomas [39]. Tang et al. used real-time quantitative PCR to detect the expression of miR-182 in normal cervical epithelial cells and primary cervical cancer tissues and found that miR-182 was significantly upregulated in primary cervical cancer and that the expression level of miR-182 was significantly associated with the patient’s International Federation of Gynecology and Obstetrics (FIGO) cancer stage level [40]. Aberrant expression of miR-182 promotes melanoma metastasis by inhibiting microphthalmia-associated transcription factor and FOXO3 [41, 42]. In HCC, Yu et al. reported that miR-182 was one of the most significantly overexpressed miRNAs in HCC [43]. Moreover, the upregulation of miR-182 expression was associated with unfavorable prognosis of HCC patients and intrahepatic metastasis. In cisplatin-treated HCC cells, upregulated miR-182-5p promotes drug resistance by targeting tumor protein 53-induced nuclear protein 1 (TP53INP1) [42]. By regulating metastasis suppressor 1 (MTSS1), miR-182-5p contributes to HCC metastasis [43]. Furthermore, upregulated expression of miR-182-5p may be a diagnostic and prognostic indicator for HCC patients [41]. Consistent with previous studies, our study showed that miR-182 is upregulated by sixfold in HCC, is associated with poor prognosis, and may act through downstream targets (Additional file 1: Table S1).

In addition, we investigated whether the ratios of hsa-miR-182 to its target RNAs (THBS1 and CHL1) and the ratio of hsa-miR-183 to its target RNA (CCNB1) have an impact on the prognosis of patients with HCC. We found that the hsa-miR-182/CHL1 ratio and the hsa-miR-183/CCNB1 ratio were significantly associated with the overall survival of patients with HCC. Zhu et al. reported that upregulation of miR-182 was significantly associated with CHL1 downregulation in papillary thyroid carcinoma (PTC) cell lines and human PTC tissues [44]. miR-182 inhibits the expression of CHL1 by directly targeting the 3’-UTR. Downregulation of miR-182 inhibited the invasion and growth of PTC cells. They concluded that miR-182 in PTC promotes cell invasion and proliferation by directly inhibiting CHL1. In glioblastoma, anti-miR-182 expression results in increased expression of biomarkers for proliferation and stem cell biomarkers (e.g., CCNB1, CD44, Sox2, and Nestin) [45]. Xu et al. reported that hsa-miR-183/CCNB1 may be related to the effect of ribavirin on HCC [46]. Therefore, we believe that the hsa-miR-182/CHL1 and hsa-miR-183/CCNB1 axes may have a strong influence on HCC pathogenesis and may provide novel therapeutic targets for HCC treatment.

Among the prognostic DEmRNAs, CCNB1 and SHCBP1 had the same connection degree (connection degree = 1) in the ceRNA network (Additional file 5: Table S5). CCNB1, a rigorous quality control regulator and an important initiator of mitosis, is a key member of the cyclin family [47, 48]. By promoting the transition of the cell cycle from the G2 phase to mitosis, CCNB1 plays a key role in forming cyclin‐dependent kinase 1 (CDK1) complexes [47, 48]. The dysregulated expression of CCNB1 is observed in many different cancers, including melanoma and esophageal squamous cell carcinoma [49, 50]. There is increasing evidence that CCNB1 is involved in checkpoint control and that its dysfunction is an early event in tumorigenesis [49, 50]. At the same time, there is evidence that inhibition of CCNB1 expression makes breast cancer cells more sensitive to the chemotherapy drug paclitaxel [51]. CCNB1 also has significant predictive power in monitoring hormone therapy efficacy and the prognosis of patients with ER+ breast cancer [52]. CCNB1 is also a biomarker for HBV-related HCC recurrence [53]. The overexpression of CCNB1 was an independent factor for unfavorable disease-free survival and had an unfavorable prognosis in patients with lung adenocarcinoma [54]. Gu et al. found that CCNB1 overexpression is closely related to poor survival in patients with HCC [55]. Knockdown of CCNB1 by RNA interference significantly suppressed HCC cell invasion, migration, and proliferation. Furthermore, they also found that miR-144 inhibits CCNB1 expression by directly targeting CCNB1. Moreover, miR-144 negatively regulates CCNB1 to delay tumor formation. The study concluded that the miR-144/CCNB1 axis plays a vital role in HCC. In the present study, our results showed that CCNB1 was upregulated in HCC samples compared to the paired adjacent nontumorous samples from both the TCGA and GEO databases. Furthermore, the survival analysis indicated that patients with HCC exhibiting high CCNB1 expression levels presented a poor prognosis in both the TCGA and GEO databases.

SHC SH2-domain binding protein 1 (SHCBP1) is a cytoplasmic protein that couples signaling pathways to activated growth factor receptors [56]. The expression of SHCBP1 mRNA levels has significant individual differences in human normal tissues. SHCBP1 mRNA and protein are absent in normal, quiescent tissues but are selectively expressed in tissues containing proliferating cells or even cancer cells, demonstrating that SHCBP1 may be involved in tumor development [56]. Immunohistochemical analysis revealed that SHCBP1 was upregulated in breast cancer [57]. High expression level of SHCBP1 was related to poorer survival and advanced clinical stage [57]. In gliomas, by activating the NF-κB signaling pathway, increased expression of SHCBP1 contributed to invasion and migration [58]. In HCC, SHCBP1 was significantly overregulated [59]. Upregulation of SHCBP1 significantly promoted colony formation and survival and cell proliferation in HCC cell lines [59]. In parallel, knockdown of SHCBP1 suppressed cell proliferation and induced cell cycle delay [59]. Our findings indicate that SHCBP1 was upregulated in HCC and that its overexpression was related to poor prognosis in patients with HCC.

Although our ceRNA network identifies many HCC-related lncRNAs, miRNAs, and mRNAs, the correlation and the extent of ceRNA effects in vivo remain poorly understood. Recent experimental studies have shown that miRNA-mediated competition between ceRNAs plays a vital role in many biological contexts by constituting additional levels of posttranscriptional regulation [60]. Sensitivity analysis demonstrates that repression mechanisms and binding free energy are vital factors for cross-talk between ceRNAs. Interactions that occur within a particular range of inhibitory values can be asymmetric (one ceRNA influences another but not the reverse) or symmetrical (one ceRNA influences another and vice versa) and can be limited by noise; at the same time, the interactions can be highly selective [60]. All in all, there are many criteria for validating ceRNA networks, such as cellular concentrations of RNA-binding proteins (RBPs) and miRNAs, timescales, steady-state parameters, kinetic parameters, the absolute concentration of the effective target pool, the miRNA:target ratio, miRNA concentration, the size and affinities of the competing target pool, quantitative measurements of miRNA, target abundance, and so on [61‐65]. Therefore, our results still need to be verified through in vivo and in vitro experiments and clinical practice. Although the ceRNA network has many interference factors in experimental verification, this deficit did not hinder the reliability of the ceRNA network because our network was based on rigorous processes. First, we included only cancer-specific lncRNAs, miRNAs, and mRNAs that had an absolute fold change > 4 and an adjusted P-value < 0.01. Second, the interactions between DElncRNAs and DEmiRNAs and between DEmiRNAs and DEmRNAs were predicted by experiment-supported databases, such as miRTarBase. These two approaches guarantee that the interactions identified not only occur in silico but are also based on experimentally supported evidence. Therefore, we believe that the genes in the current ceRNA network are important for HCC. In the future, with the emergence of larger sample sizes, better databases, and better algorithms, a more comprehensive ceRNA network will be constructed. In addition, further research is warranted on the functions of key ceRNAs in vivo and in vitro.

In conclusion, a ceRNA network including 26 DElncRNAs, four DEmiRNAs, and six DEmRNAs was successfully built. Importantly, four lncRNAs (MYCNOS, DLX6-AS1, LINC00221, and CRNDE) and two mRNAs (CCNB1 and SHCBP1) were remarkably related to the prognosis of patients with HCC in both the TCGA and GEO databases. Our research provides novel insights that will increase our understanding of the prognosis-related ceRNA network in HCC. Furthermore, the candidate lncRNAs, miRNAs, and mRNAs involved in the ceRNA network can be further evaluated as potential therapeutic targets and prognostic biomarkers for HCC.