3.1 Identification of DECs
Two circRNA expression profiles GSE83521 and GSE93541 were obtained from GEO, and GEO2R method was applied to analysis DECs. The GSE83521 dataset derived from gastric cancer tissues and GSE93541 dataset derived from plasmas. Differential circRNAs co-expressed in tissues and plasma of gastric cancer patients are our target circRNAs. We found that 53 circRNAs were identified to be differentially expressed in GSE83521, including 39 up-regulated and 14 down-regulated circRNAs; while 267 differentially expressed circRNAs were identified in GSE93541, including 138 up-regulated and 129 down-regulated circRNAs. Among them, 3 up-regulated and 0 down- regulated circRNAs were observed in both circRNA expression profiles. A Venn diagram of the results is shown in Figure 2A and B. The up-regulated circRNAs that overlapped in the two datasets (hsa_circ_0001013, hsa_circ_0007376, hsa_circ_0043947) were selected for further analysis. Details of the overlapped up-regulated circRNAs are listed in Table 1, and the basic structural features of the three selected circRNAs are shown in Figure 2C.
3.2 Expression of circRNAs in datasets and cell lines
As shown in Figure 3A and B, the expression patterns of the three selected circRNAs were upregulated in both tissues and plasmas according to the datasets. We also detected the expression of selected circRNAs in gastric cancer cell line SGC‐7901 and human gastric epithelial cell line GES‐1 by qRT‐PCR. The results showed that all three selected circRNAs had higher expression levels in SGC‐7901 than in GES‐1 as shown in Figure 3C.
3.3 Prediction of circRNA-miRNA and their function analysis
An increasing number of evidence demonstrate that circRNAs might function as competing endogenous RNAs (ceRNAs) that operate by competitively binding common microRNAs (miRNAs) and increase the expression of the target genes of these miRNAs. Target miRNAs of the three selected circRNAs were predicted by two online tools circBank and circInteractome. A total of 43 consensus miRNAs from both prediction tools were identified and DECs potentially bind to these miRNAs were presented in Table 2. Results showed that one specific circRNA might bind to more miRNAs, while different circRNAs could interact with one specific miRNA. Rich Fun software was used to GO analysis for the 43 miRNAs. The top five enrichment items were shown respectively in Figure 4: ‘Regulation of nucleobase, nucleoside, nucleotide and nucleic acid metabolism’ , ‘Regulation of cell growth’, ‘Cell cycle’, ‘Regulation of enzyme activity’ and ‘Cell-cell adhesion’ for biological progress (BP), ‘Cytoplasm and Nucleus’, ‘Nucleus’, ‘Lysosome’, ‘Actin cytoskeleton’ and ‘Endosome’ for cellular component (CC), and ‘Transcription factor activity’, ‘Receptor signaling complex scaffold activity’, ‘Translation regulator activity’, ‘Protein binding’ and ‘RNA binding’ for molecular function(MF). All of them indicated that circRNAs might impact on GC progression by modulating various miRNAs.
3.4 Construction of the ceRNA network
We identified 119 experimentally strongly supported target genes of 43 miRNAs by mirTarBase on line tool (Table 2). Then we used 3 circRNAs, 43 miRNAs and 119 mRNAs in Cytoscape 3.6.1 to construct a circRNA-miRNA-mRNA visualization network (Figure 5).
3.5 Functional and pathway enrichment analysis and PPI network
GO analysis indicated that the 119 mRNAs were mainly enriched in ‘regulation of apoptotic signaling pathway’, ‘autophagy’, and ‘process utilizing autophagic mechanism’ (BPs); ‘glutamatergic synapse’, ‘nuclear chromatin’ and ‘external side of plasma membrane’ (CCs); and ‘DNA-binding transcription activator activity, RNA polymeraseⅡ-specific’ (MFs) (Figure 6A). KEGG pathway analysis revealed strong enrichment in the ‘PI3K-Akt signaling pathway’ (Figure 6B). After obtaining the target genes of candidate miRNAs, we created a PPI network composed of 165 nodes and 170 edges (Figure 7A). Following the identification of the vital functions of hub genes in the network, 18 hub genes (CCND2, STAT3, TP53, MCL1, MYC, FOXO1, FOXO3, BCL2L11, PTEN, MTOR, CDH1, CASP3, IL6, GSK3B, CDKN1A, MAPK1, SMAD4, CDC42) were identified in GC using the MCODE plugin, MCODE_Score = 13.76. These hub genes were predicted target genes for hsa-miR-197-3p, hsa-miR-451a, hsa-miR-136-5p, hsa-miR-337-3p, hsa-miR-654-3p, hsa-miR-182-5p, hsa-miR-1228-3p, hsa-miR-942-5p, hsa-miR-488-3p and hsa-miR-876-3p, and all these 10 miRNAs were predicted miRNAs for hsa_circ_0001013. So a core circRNA–miRNA–mRNA network based on hub genes was displayed in Figure 7B.