Identification of a potentially functional circRNA–miRNA–mRNA regulatory network for investigating pathogenesis and providing possible biomarkers of bladder cancer

Microarray dataset GSE92675 (four pairs of BCa and four pairs of adjacent non-cancerous tissues) was retrieved from GEO (http://​www.​ncbi.​nlm.​nih.​gov/​gds/​), which is an international public repository for high-throughput microarray and next-generation sequence functional genomic data sets submitted by the research community [13]. The expression profile of circRNAs was determined from GSE92675, and the data were normalized and log10-transformed with the Limma bioconductor package. We selected upregulated circRNAs with the standard of |log10(foldchange) > 3|, adjusted P-value of < 0.01, and downregulated circRNAs with the standard of |log10(foldchange) > 2|, adjusted P-value of < 0.01.

We determined the expression levels of miRNAs and mRNAs from TCGA to select differentially expressed miRNAs and mRNAs on the basis of the criteria of |log10(foldchange) > 1|, adjusted P-value of < 0.05.

CircRNAs function as sponges for miRNAs with MREs. The DECs in BCa that act as sponges for miRNAs were predicted with the cancer-specific circRNA database (CSCD), an online database. Then, RNA sequencing data (composed of 411 BCa samples and 19 normal controls) were downloaded from TCGA [14] to detect differentially expressed miRNAs.

Three algorithms (miRanda, TargetScan, PITA) were used to depict miRNA target genes. Differentially expressed mRNAs were selected using the same methods for selecting differentially expressed miRNAs.

ClusterProfiler, an R package that automates the process of biological term classification and enrichment analysis of gene clusters, was used to conduct GO annotation and KEGG pathway analyses [15].

Sixteen pairs of BCa tissues and ANTs were obtained from patients who were diagnosed with BCa and underwent radical cystectomy at the First Affiliated Hospital of Nanjing Medical University, China. All tissue samples were frozen in liquid nitrogen until RNA extraction. The experiment was approved by the ethics committee of the First Affiliated Hospital of Nanjing Medical University. Informed consents were signed before the clinical materials were used for research purposes.

Total RNA was isolated from tissues by using TRIzol reagent (Invitrogen, USA). Then, cDNA was synthesized with HiScript II (Vazyme, China) for qRT-PCR. Primers for qRT-PCR, which was conducted on an AB7300 thermo-recycler (Applied Biosystems, USA) or LightCycler 480 (Roche, USA), were provided by TSINGKE Biological Technology. U6 and beta-actin were used as the internal reference. Expression levels of circRNAs were calculated using the 2^−ΔΔCT method.

The cytoHubba plugin from Cytoscape 3.7.1 was used to select the functional circRNA–miRNA–mRNA network according to the degree value.

A microarray dataset (GSE92675) from the GPL19978 platform was utilized in this study. Table 1 shows the fundamental information of the gene chip. To select DECs, R package Limma was adopted. The PCA plot showed the PC1 and PC2 coefficients of GSE92675 (Fig. 2a). As a result, 428 DECs were identified, which consisted of 261 upregulated and 167 downregulated circRNAs (Fig. 2b). We ranked all DECs using a robust method and an adjusted P-value of < 0.05. Twenty-one upregulated and 28 downregulated circRNAs were in the top rankings. The heat map for circRNAs in GSE92675 is shown in Fig. 2c.

On the basis of the above results, we conducted circRNA–miRNA pair prediction using CSCD. A total of 1861 target miRNAs were obtained. By intersecting target miRNAs and differentially expressed miRNAs obtained from TCGA, we confirmed 28 target miRNAs finally (Fig. 3). Using similar methods, 476 target mRNAs were selected, of which 79 were upregulated and 397 were downregulated (Fig. 4). Finally, the circRNA–miRNA–mRNA network containing circRNA–miRNA pairs and miRNA–mRNA pairs was generated (Fig. 5). Next, we conducted GO and KEGG analyses to detect the potential biological functions. The results are shown in Fig. 6. The MAPK and PI3K–AKT signaling pathways were statistically significant.

CircRNAs act as hub nodes in biological networks. Five downregulated circRNAs (hsa_circ_0001955, hsa_circ_0032821, hsa_circ_0060219, hsa_circ_0011385, hsa_circ_0084171) and four upregulated circRNAs (hsa_circ_0040039, hsa_circ_0082582, hsa_circ_0009172, hsa_circ_0077526) were identified on the basis of the degree value calculated by the cytoHubba plugin of Cytoscape. The heatmap for nine DECs in GSE92675 is shown in Fig. 7, and their basic features are displayed in Table 2. Sequences of specific primers for each circRNA were designed using Primer Premier 5. The expression levels of nine circRNAs in 16 pairs of BCa tissues and adjacent non-cancerous tissues were quantified by qRT-PCR. Among the qRT-PCR products, five circRNAs were unable to be verified by agarose gel electrophoresis. Two circRNAs could not be detected due to their relatively low expression. The expression of hsa_circ_0060219 showed no significant difference between BCa tissues and adjacent non-cancerous tissues. Hsa_circ_0011385 showed upregulated tendency in BCa tissues corroborated by qRT-PCR (Fig. 8a). On the basis of the above results, hsa_circ_0011385 was selected as the target for future research. The basic structural molds of hsa_circ_0011385 are displayed in Fig. 8b. Clinicopathological features in 16 BCa patients are displayed in Table 3. Four downregulated miRNAs (mir-211-5p, mir-204-5p, mir-182-5p, mir-96-5p) were confirmed. All four miRNAs were involved in cancer-related pathways as shown in Fig. 9. After obtaining the overlapping genes from target mRNAs in BCa and differentially expressed mRNAs from TCGA, a circRNA–miRNA–mRNA network was constructed around hsa_circ_0011385 (Fig. 10). Afterward, the expression of four miRNAs was validated by qRT-PCR as well. Mir-96-5p, mir-182-5p, and mir-211-5p showed no significant difference (Fig. 11a, d, g), whereas mir-204-5p showed downregulated tendency (Fig. 11j). The primer information is shown in Table 4. Correlation analysis revealed a moderate negative correlation between the expression of hsa_circ_0011385 and miR-204 (r = − 0.43, P = 0.01268) (Fig. 11n). Furthermore, according to the data from TCGA, we found that those miRNAs were differently expressed between high-grade and low-grade BCa except mir-96-5p (Fig. 11b, e, h, k). However, mir-96-5p, mir-182-5p, mir-211-5p, and mir-204-5p showed significant difference in different stages of BCa (Fig. 11c, f, i, l). High expression of mir-204-5p was correlated with poor prognosis in BCa (Fig. 11m).

After obtaining the target genes of hsa_circ_0011385, we created a PPI network composed of 67 nodes and 274 edges (Fig. 12a). Following the identification of the vital functions of hubgenes in the network, 18 hubgenes (ARHGAP11A, DEPDC1, DLGAP5, AURKB, CENPA, CDT1, CCNE1, RAD51, CHEK1, RACGAP1, ZWILCH, MKI67, CDCA8, DTL, FOXM1, SHCBP1, RRM2, MKI67) were identified in BCa using the MCODE algorithm, k-score = 2. The expression levels of the top 10 hubgenes (RRM2, MKI67, CENPA, AURKB, FOXM1, DLGAP5, DTL, RACGAP1, CHEK1, CDCA8) are shown in Fig. 13. A circRNA–miRNA–hubgene network is displayed in Fig. 12b. GO and KEGG analyses showed that hubgenes were involved in cell cycle activities and could be regulated by miRNAs and play an eventful role in BCa pathogenesis (Fig. 14).