28 GBM specimens and 23 normal brain tissues were obtained from patients who received surgery at The First Affiliated Hospital of the University of Science and Technology of China (USTC) from 2019 to 2020. As to the normal brain tissues, 3 samples were GBM adjacent surgical margin (paired samples), and the rest 20 samples were from epilepsy patients. This study was approved by the Ethics Committee of The First Affiliated Hospital of USTC. Informed consent was acquired from each patient. The clinical, surgical, imaging and pathological records were reviewed retrospectively. The pathological diagnosis was validated by two pathologists independently. The survival time was calculated from the date of diagnosis of GBM until October 16, 2020. The expression of potential circRNAs, miRNAs and mRNAs in resected tissues was tested through qRT-PCR. The primers used in this study were synthesized by Sangon Biotech.Ltd. (Shanghai, China) (Table 1). QRT-PCR was set at an initial denaturation step of 10 min at 95°C and 95°C for 10 s, 60°C for 20 s, and 72°C for 10 s for a total of 40 cycles. The relative circRNAs expression levels were normalized to GAPDH.

Chi-square test and Fisher’s test and plotting receiver operating characteristic curves (ROCs) was used to analyze relevant factors by SPSS20.0 (SPSS, Inc., Chicago). The Kaplan-Meier method was employed in estimating the survival rate. The Cox proportional hazards regression analysis was used in evaluating multivariate prognosis factors by SPSS20.0. Relation for circRNA-miRNA, miRNA-mRNA and circRNA-mRNA was performed by Person correlation coefficient using Graphpad Prism software (version 9). Results were regarded as statistically significant when p-value < 0.05.

The circRNA dataset in this study was obtained from NCBI GEO database (http://www.ncbi.nlm.nih.gov/) GSE147352 (15 normal brain tissues and 18 gliomas). The miRNA expression dataset derived from GEO GSE158284 (12 normal brain tissues and 20 gliomas) and mRNA expression profile were gained from TCGA database (http://cancergenome.nih.gov/), including 5 normal brain tissues and 169 gliomas.

All raw data were normalized and log2-transformed. The ID of the corresponding probe name was converted into gene symbols. We identified differentially expressed RNAs through R statistical software and the limma Bioconductor package. The criteria for screening differentially expressed circRNAs, miRNAs (DEmiRNAs) and mRNAs (DEmRNAs) was p-value <0.05 and |log2FC(fold change)|>1.

The Cancer-Specific circRNA (CSCD, http://gb.whu.edu.cn/CSCD/) and the circular RNA Interactome (https://circinteractome.nia.nih.gov/) were used to get the overlapping target miRNAs (TmiRNAs) as candidate targets of the selected circRNAs. The common members of TmiRNAs and DEmiRNAs were obtained the potential targeted miRNAs for further analyses. The interactions between miRNAs and mRNAs were predicted through TargetScan (http://www.targetscan.org), miRBD (http://www.mirdb.org/) and miRNATAR (http://microrna.gr/tarbase/). Overlapping mRNAs called TmRNAs were picked up and further intersected with DEmRNAs to obtain targeted mRNAs. The prognostic values of mRNAs in GBM were evaluated by Gliovis database (http://gliovis.bioinfo.cnio.es/) using Kaplan-Meier analysis.

The differentially expressed circRNAs were annotated with Circbase database (http://circrna.org/) to obtain informations of corresponding gene, then these genes were analyzed with Gene Ontology (GO) analysis. Differentially expressed mRNAs were studied with GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis. The function of circRNAs and mRNAs were presumed according to GO and KEGG analyses by using KOBAS website (http://kobas.cbi.pku.edu.cn/kobas3/).

To detect the circRNAs specifically expressed in gliomas, we analyzed the GEO dataset (GSE147352) that determined the expression of all circRNAs in 15 normal brain tissues and 18 gliomas through high throughput sequencing. Among the 8,184 detected circRNAs, 38 upregulated circRNAs and 186 downregulated circRNAs were identified in gliomas in accordance with the criteria |log2FC| > 1, padj <0.05 (Figures 1A,B; Supplementary Table S1). GO analysis showed that the differentially expressed circRNAs were significantly enriched in 8 GO categories, including nucleoplasm, cytoskeleton, postsynaptic density, membrane, cell junction, presynaptic membrane, ion channel binding and presynaptic active zone (Figure 1C). We picked up the top 5 up- and down-regulated circRNAs (Table 2) and tried to construct the circRNA-miRNA network through cytoscape software. Finally, we got the predicted circRNA-miRNA network for 8 circRNAs (hsa_circ_0073237, hsa_circ_0024085, hsa_circ_0009027, hsa_circ_007513, hsa_circ_0131934, hsa_circ_0055954, hsa_circ_0114014 and hsa_circ_0078784), whereas the rest 2 circRNAs (hsa_circ_0021350, and hsa_circ_0092798) were not able to be incorporated (Figure 1D). To investigate the expression pattern of these circRNAs in GBM, 3 pairs of samples of GBM and the matched adjacent surgical margins from our hospital were employed for qRT-PCR analyses. Among the 6 validated circRNAs (hsa_circ_0024085, hsa_circ_0009027, hsa_circ_007513, hsa_circ_0131934, hsa_circ_0114014 and hsa_circ_0092798) with consistent expression patterns in qRT-PCR and the RNA-sequencing, hsa_circ_0114014 (circLRRC7) demonstrated the most obvious downregulation in GBM samples (Figures 1E,F), suggesting an important role of this circRNA in GBM.

More samples (25 GBM and 20 normal brain tissue from epilepsy patients) from our hospital were used to examine the expression of circLRRC7 and its association with clinicopathologic characteristics. Indeed, circLRRC7 was strongly downregulated in GBM (relative fold change> 7, p < 0.0001) (Figure 2A). When the GBM patients were stratified into the circLRRC7-high and circLRRC7-low groups by using the median circLRRC7 level as the cut-off, no significant relationship was identified between circLRRC7 expression and the age, gender, tumor size, KPS score, tumor location or IDH mutation status of the patients (Table 3). In contrast, higher circLRRC7 expression was strongly associated with better 1-year progressive-free survival (PFS, p = 0.011) and 1-year overall survival (OS, p = 0.011) (Table 3). Consistently, Kaplan-Meier survival analysis showed that higher circLRRC7 expression was significantly correlated with better PFS (median: high-19.4 m vs low-6.5 m, p < 0.05) and OS (median: high-20.9 m vs low-8.5m, p < 0.05) of the GBM patients (Figures 2B,C), suggesting that circLRRC7 may function to alleviate GBM progression. To determine the values of different clinicopathological characteristics (age, gender and circLRRC7 expression) on predicting prognosis, the ROC curves were plotted and the area under the ROC curve (AUC) larger than 0.7 was regarded as the threshold of a diagnostic discriminatory value. Only circLRRC7 expression showed and AUC of 0.708, indicating its capacity to predict prognosis of GBM patients (Figure 2D). To further confirm the impact of circLRRC7 on prognosis, the multivariate Meta-analysis was performed and the result indicated that the relative expression of circLRRC7 was a protective factor for the survival of GBM patients, whereas other factors such as number of lesions, tumour size, KPS score, sex and age did not affect prognosis (Figure 2E). In summary, circLRRC7 is the most significantly down-regulated circRNA in GBM with potential prognostic values.

RNA-seq data showed that circLRRC7 is derived from the exons 23–26 of the LRRC7 gene and is transcribed into a 510nt circular RNA transcript (hsa_circ_0114014) in GBM (Figure 3A). We used forward primer for exon 26 and reverse primer for exon 23 of the LRRC7 gene to perform PCR analysis. The band of circLRRC7 was only detected by PCR using cDNA template transcribed with random Hexamer primers but not the oligo (dT) primers (Figure 3B), which confirmed the existence of circLRRC7 as a novel circular RNA. PCR using the genomic DNA (gDNA) as template did not generate the band of circLRRC7 (Figure 3C). Sanger sequencing of the PCR product confirmed the splice junction connecting exon 23 and exon 26 of the LRRC7 gene (Figure 3D). These results confirmed the circLRRC7 as a circular RNA transcribed from the LRRC7 gene in GBM.

As most circRNAs function as miRNA sponges in tumors, we predicted target miRNAs of circLRRC7 by using the CSCD database and the CircInteractome database, resulting in 21 common miRNAs in the two databases (Supplementary Table S2). After then, we screened for differentially expressed miRNAs (DEmiRNAs) in adult GBM and normal brain tissues by using GEO dataset (GSE158284) that determined the most differentially expressed miRNAs in 20 GBMs and 12 normal brain tissues (Figure 4A), resulting in 41 upregulated and 30 downregulated DEmiRNAs (Figure 4B; Supplementary Table S3). Intersection of the 21 potential target miRNAs of circLRRC7 and the 71 DEmiRNAs revealed hsa-miR-1281 as the only common miRNA with significant upregulation in GBM (Figure 4C). Furthermore, we determined the expression levels of miR-1281 in a batch of normal (n = 15) and GBM (n = 18) tissues collected in our hospital. The results showed that miR-1281 was highly expressed in tumors relative to normal tissues (Figure 4D). Therefore, hsa-miR-1281 might be the potential target miRNA of circLRRC7 in GBM.

miRNAs usually function by negatively regulating the expression of their target genes. We sought to identify the target mRNAs of hsa-miR-1281 via three databases (miRDB, miRTarBase and TargetScan) and found 97 candidate mRNAs (Supplementary Table S4). We then looked for differentially expressed mRNAs (DEmRNAs) in GBM relative to normal brain by analyzing the TCGA RNA-seq database (Supplementary Table S5), and found 1872 downregulated as well as 826 upregulated DEmRNAs (Figures 5A,B). Intersection of the 97 candidate mRNAs and the downregulated DEmRNAs resulted in 14 overlapping mRNAs (PDXP, PDE1C, CACNA1G, C6orf106, DUSP8, SLC1A4, RTN2HRH3, MAFG, PPP2R2D, EGR3, FAM20B, STOX2, PRNP) (Figure 5C). KEGG pathway analysis demonstrated that these genes were significantly enriched in calcium signaling pathway, vitamin B6 metabolism, MAPK signaling pathway and so on (Figure 5D). In GO analysis, these genes were annotated to certain biological processes, cellular components and molecular functions, such as synapse, phosphoprotein phosphatase activity, cognition, phosphatase activity, dendrite, and dephosphorylation (Figure 5E). Therefore, miR-1281 as a potential target of circLRRC7 may be involved in regulation of several signaling pathways in GBM.

The expression of the 14 candidate miR-1281 target genes in GBM was further checked by analyzing the datasets of TCGA Agilent 4502-A platform microarray assays. The results showed that 13 genes except for PDE1C were lowly expressed in GBM relative to normal brains with statistical significance (Figure 6A-N). In addition, Kaplan-Meier survival analyses demonstrated that higher expression of PDXP, DUSP8, HRH3, and FAM20B was associated with better OS in GBM patients (Figure 7A). CircRNAs as ceRNAs may function as miRNA sponges to offset miRNA-mediated mRNA inhibition. By using the Gliovis database, we found that PDXP among the four candidates was the only gene with a negative correlation with miR-1281 (Figure 7B). By using the same batch of samples as in Figure 2A, we determined the expression levels of PDXP in normal and GBM tumor samples. The results showed that PDXP had a lower expression in tumor relative to normal tissues (Figure 7C). We further tried to determine the potential association between circLRRC7, miR-1281 and PDXP in the samples collected in our hospital. Among all samples collected in our hospital at different time points, simultaneous detection of circLRRC7, miR-1281 and PDXP was achieved in 15 samples. Lineage regression analyses revealed a negative correlation between circLRRC7 and miR-1281 (Figure 7D). Meanwhile, a negative correlation was detected between miR-1281 and PDXP (Figure 7E). However, a positive correlation was detected between circLRRC7 and PDXP in these samples (Figure 7F). These data strongly suggest the association between circLRRC7, miR-1281 and PDXP in GBM progression.