A circRNA–miRNA–mRNA network identification for exploring underlying pathogenesis and therapy strategy of hepatocellular carcinoma

Circular RNA (circRNA), with a complete closed loop structure, is firstly identified in 1976 [1]. However, these transcripts without poly-A tail are ignored for a long time due to the limitation of traditional RNA detection methods. In recent years, with the development of high-throughput sequencing technology, a myriad of circRNAs has been found in the eukaryotic transcriptome [2]. With the features of cell-type specific [3] and highly conserved across species [4], circRNAs are thought to be new star RNAs which play important roles in various diseases, including human cancers [5, 6].

Competing endogenous RNAs (ceRNAs) are transcripts that act as miRNA sponges, modulating each other at post-transcriptional level via competely binding to shared miRNAs [7]. Recently, circRNAs have become new hotspots in ceRNA family since they have been demonstrated to harbor abundant conserved miRNA response elements (MREs) [8]. Increasing study has revealed that some circRNAs are involved in tumor initiation and progression by the ceRNA mechanism. For example, a classic cirRNA, ciRs7, is deemed to be a miRNA sponge, absorbing miR-7 and liberating the latter inhibitory effect on its target gene in many human cancers [9, 10]. CircRNAs as ceRNAs mediating pathological processes has also been reported in HCC [11, 12]. However, many unknown circRNAs still remain to be explored.

In this study, we employed a combinative strategy of gene chip and computational biology to investigate novel circRNAs and their potential action mechanisms in HCC. The flow chart recapitulating the present work is shown in Fig. 1 as follows: First, we collected HCC-related microarray datasets providing expression profile of circRNAs from the Gene Expression Omnibus (GEO), obtaining differently expressed circRNAs (DECs) with RobustRankAggreg method and corroborating their expression using reverse transcription-quantitative polymerase chain reaction (RT-qPCR). To depict whether the DECs function as ceRNAs in HCC, we collected their sponge miRNAs and miRNA target genes, constructing a circRNA–miRNA–mRNA network. A protein–protein interaction (PPI) network was subsequently established and the hubgenes were identified. Then, Gene Oncology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome enrichment analyses on the hubgenes were performed to elucidate the potential pathogenesis of HCC. Furthermore, we conducted a connectivity map (CMap) analysis to acquire bioactive compounds for the treatment of HCC, which provide a new insight into the latent therapeutic capacity of circRNAs in HCC.

Due to the lack of 5′ caps and 3′polyadenylated tails, circRNAs are ignored in classical polyadenylated transcriptome studies for a long time. In the past few decades, with the development of high-throughput sequencing, biochemical and computational biology methods, a large number of circRNAs are unmasked the veil in various tissues and cells [30]. Increasing study has unveiled the important roles of circRNAs in a myriad of human diseases, including malignant tumors [31‐33]. CircRNAs usually serve as diagnostic and prognostic biomarkers because of their relative tolerance to exonucleases, which benefits from the covalently closed loop structures [34, 35]. In addition, owing to the high cell- and tissue- specificity of circRNAs, their roles in different neoplasms are not complete accord. In HCC, an increasing number of circRNAs, such as ciRS-7 [36], has_circ_0067934 [12], CDR1as [37] and circMTO1 [11], have been reported to exert momentous roles in regulating pathophysiological process and guiding clinical diagnosis and treatment. However, there are still a lot of circRNAs that need to be unearthed.

In this study, we firstly collected three gene chips (GSE78520, GSE94508 and GSE97332) from the GEO database and screened six DECs (hsa_circRNA_103510,hsa_circRNA_100542, hsa_circRNA_102166, hsa_circRNA_104515, hsa_circRNA_105031 and hsa_circRNA_100291) with the RobustRankAggreg approach, an algorithm for integration of different gene expression spectra, ranking genes by their fold changes and aggregating these rankings to achieve the final robust and rigorous results. Following the RT-qPCR validation of the six DECs, three circRNAs (hsa_circRNA_102166, hsa_circRNA_100291 and hsa_circRNA_104515) were selected for further analysis.

As highly conserved endogenous RNAs, many circRNAs harbors abundant miRNA binding sites, indicating that they can sponge corresponding miRNAs and thus function as ceRNAs to regulate gene expression [8, 38, 39]. To ascertain whether the aforementioned three circRNAs function as ceNAs in HCC, we predicted their MREs via two online tools, CSCD and CircInteractome. The former web tool predicted MREs within 50 bp upstream and downstream of circRNA junction point [16]. The latter web tool predicted MREs based on the TargetScane algorithm, which forecasts MREs by searching for 7mer or 8mer complementarity to the 3′ end and the seed region of each miRNA [17, 40]. We chose miRNA predicted by both algorithms as the putative target miRNA for the three circRNAs. Finally, five circRNA–miRNA interactions consisting of two circRNAs (hsa_circRNA_104515 and hsa_circRNA_100291) and five miRNAs (hsa-miR-1303, hsa-miR-142-5p, hsa-miR-877-5p, hsa-miR-583 and hsa-miR-1276) were determined. The expression of the five miRNAs was then verified based on the microarray and RNA-seq data from the GEO and TCGA. The results showed that miR-142-5p was down-regulated in HCC, which was consistent with previous studies [41, 42]. No relevant study reported the expression of miR-877, miR-1303, miR-1276 and miR-583 in HCC. According to our results, miR-877 was up-regulated in HCC. MiR-1303 and miR-1276 showed high expression tendencies in HCC, while the differences were not statistical significant. For miR-583, its expression in HCC tissues was similar to that in normal liver tissues. Due to the inter-study heterogeneity that cannot be ignored, the reliable of the pooled results is reduced. Thereby, further rigorous studies are necessary to validate these findings.

Following the collection of the 172 overlapped genes between the target genes of the five miRNAs and the DEGs in HCC, we constructed a circRNA–miRNA–mRNA regulatory network. We found that hsa_circRNA_104515 and hsa_circRNA_100291 may act as ceRNAs to capture miR-1276, miR-583, miR-877-5p or miR-142-5p, and subsequently regulate the 172 genes expression. Our results provide an evidence of the ceRNA regulatory mechanism of hsa_circRNA_104515 and hsa_circRNA_100291 in HCC. To further elucidate the action mechanism of the ceRNA network, we constructed a PPI network, screening seven hubgenes (JUN, MYCN, AR, ESR1, FOXO1, IGF1 and CD34) from the PPI network. The functional annotations and pathway analyses showed that the seven hubgenes genes were involved in many critical tumor-related biological functions and metabolic pathways, such as “epithelial cell proliferation”, “AMPK signaling pathway”, “FoxO signaling pathway”, “Estrogen signaling pathway”, “MAPK6/MAPK4 signaling” and “MAPK family signaling cascades”. The important roles of the seven genes in HCC have also been demonstrated previously [43‐49]. However, their links with circRNAs have not yet been explored. Here, we identified seven circRNA–miRNA–mRNA axes (hsa_circRNA_100291/miR-1276/FOXO1, hsa_circRNA_100291/miR-583/ESR1, hsa_circRNA_100291/miR-583/JUN, hsa_circRNA_100291/miR-583/AR, hsa_circRNA_104515/miR-877-5p/AR, hsa_circRNA_104515/miR-142-5p/MYC, hsa_circRNA_104515/miR-142-5p/MYCN, hsa_circRNA_104515/miR-142-5p/IGF1 and hsa_circRNA_104515/miR-142-5p/CD34), indicating competitive regulatory relationships of hsa_circRNA_100291 and hsa_circRNA_104515 with the seven genes in HCC. However, given that the results are on the basis of computational biology, further in-depth studies are indispensable to verify the possible roles of the seven axes in HCC.

To date drug control is an important treatment for patients with HCC [50]. Digging effective and sensitive drugs against HCC helps improve patients’ outcomes. We therefore implemented the CMap analysis of the seven hubgenes to explore usable drugs for the treatment of HCC. Based on the genome-wide expression profiling of transcripts technology, CMap provides a comprehensive and accurate data resource for exploration of novel drug or relocation of existing drug [51]. Drugs available in CMap are all licensed for human use by the Food and Drug Administration, thus it is an ideal and reliable approach to obtain therapeutic agents for human diseases [52]. Three chemicals (decitabine, BW-B70C and gefitinib) were identified as the treatment options for HCC. As a cytidine antimetabolite analogue, decitabine represses DNA methylation, arresting cells into G1/S phase and inhibiting tumor cell proliferation. Its antitumor activity in solid tumors including HCC has been elucidated in previous studies [53]. Xing et al. [54] have demonstrated that decitabine could facilitate the expression of miR-122 via inhibition of methylation in HCC cells. More importantly, Skårn et al. [55] have demonstrated that miR-142 is epigenetically restrained by DNA methylation. Treating decitabine in mesenchymal cells promotes mature miR-142-5p/3p expression and thus depresses cell proliferation. We thus hypothesize that decitabine may exert its anti-HCC effect by augmenting miR-142-5p via demethylation and subsequently regulating the downstream target genes of miR-142-5p. Further well-design study is necessary to validate the conclusion. BW-B70C is an inhibitor of arachidonic acid 5-lipoxygenase. Previous study has reported its anti-neoplastic activity in leukemic cells by suppressing the NOTCH1-P13K-AKT-eNOS axis [56]. However, its anti-HCC effect and action mechanism have not been elucidated as of yet. In this study, we found its potential as therapeutic agent for HCC. More studies are needed to verify this finding. Gefitinib is a selective inhibitor of tyrosine kinase receptor used in clinic for the treatment of locally advanced or metastatic non-small cell lung cancer (NSCLC) [57]. Its antineoplastic effect on the other solid tumors including HCC has also been reported [58, 59]. However, the responsiveness of different HCC patients to gefitinib varies greatly and most patients even develop gefitinib resistance [60]. Multiple non-coding RNAs including miRNAs and lncRNAs have been reported to contribute to gefitinib resistance and sensitivity in NSCLC [61‐63]. We speculate that gefitinib share the similar resistance mechanisms in HCC and lung cancer. Our study provides a theoretical basis for studying gefitinib resistance mechanism and enhancing gefitinib sensitivity in patients with HCC from the perspective of circRNA–miRNA–mRNA network.

In conclusion, by employing a comprehensive strategy of big data mining, RT-qPCR and computational biology, we constructed a circRNA–miRNA–mRNA network and found that hsa_circRNA_100291 and hsa_circRNA_104515 may function as ceRNAs to exert critical roles in HCC. In addition, three bioactive chemicals (decitabine, BW-B70C and gefitinib) based on the CMap analysis was determined as therapeutic agents for HCC. Our study provides a novel insight into the pathogenesis and therapy for HCC from the circRNA–miRNA–mRNA view.