EIF4A3-induced circular RNA MMP9 (circMMP9) acts as a sponge of miR-124 and promotes glioblastoma multiforme cell tumorigenesis

GBM tissues and corresponding normal tissues (2 cm away from the tumor) were obtained from 18 patients who were diagnosed with GBM at Characteristic Medical Center of Chinese People's Armed Police Force from March 2014 to September 2017. The detailed clinical patient information is shown in (Additional file 1: Table S1). Approval for this study was provided by the Ethics Committee of Characteristic Medical Center of Chinese People's Armed Police Force. Informed consent was obtained from each GBM patient. Tissues were immersed in liquid nitrogen immediately after removal from patients and were stored at − 80 °C until use.

Total RNA was extracted from GBM tissues, corresponding normal tissues and treated U87 and U251 cells using TRIzol reagent (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer’s protocol. A High-Capacity cDNA Reverse Transcription kit (Applied Biosystems, Foster City, CA, USA) was used to synthesize cDNA from RNA according to the manufacturer’s protocol. qRT-PCR was performed using a SYBR Green kit (Bio-Rad Laboratories, Inc., Hercules, CA, USA) in an ABI 7900 PCR Thermal Cycler according to the manufacturer’s protocol. The mRNA expression levels of genes were measured using the 2^-△△Ct method [25]. Small RNA RNU6 (U6) was used as the endogenous control to normalize miRNA expression, while GAPDH was used as the endogenous control to normalize circRNA expression.

Total RNA was extracted and purified using the mirVana™ miRNA Isolation Kit (cat.# AM1561; Ambion, Austin, TX, USA) following the manufacturer’s instructions and was evaluated for an RIN number to assess RNA integration using an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA). Total RNA was amplified and labeled using the Low Input Quick Amp WT Labeling Kit (cat.# 5190–2943; Agilent Technologies). Labeled cRNA was purified using the RNeasy mini kit (cat.# 74,106; Qiagen, Hilden, Germany).

Each slide was hybridized with 1.65 μg of Cy3-labeled cRNA using the Gene Expression Hybridization Kit (cat.# 5188–5242; Agilent Technologies) in a hybridization oven (cat.# G2545A; Agilent Technologies). After 17 h of hybridization, the slides were washed in staining dishes (cat.# 121; Thermo Scientific, Waltham, MA, USA) using the Gene Expression Wash Buffer Kit (cat.# 5188–5327; Agilent Technologies).

The slides were scanned using the Agilent Microarray Scanner (cat.#G2565CA; Agilent Technologies) using the default settings: dye channel, green; scan resolution, 3 μm; PMT, 100%; 20 bits. The data were extracted using Feature Extraction software 10.7 (Agilent Technologies). Raw data were normalized using the Quantile algorithm in the limma package in R. Microarray analysis was performed by Shanghai Biotechnology Cooperation (Shanghai, P.R. China).

MMP9 expression vectors, which were truncated to varying degrees (a1-a5), were constructed. A biotinylated MMP9 DNA probe was designed and synthesized by GenePharma Co., Ltd. (Shanghai, China). According to the protocol of the manufacturer, probe-coated beads were produced by dissolving the probe and incubating them with Dynabeads M-280 Streptavidin (Invitrogen, CA, USA) for 10 min at room temperature. U87 cell lysates were treated with the probe-coated beads, and the results were detected by qRT-PCR analysis.

According to the manufacturer’s protocol, the RIP assay was performed using an EZMagna RIP kit (Millipore, Billerica, MA, USA). U87 cells were collected and lysed with RIP lysis buffer. U87 cell lysates (100 μl) were treated with RIP buffer and were incubated with Proteinase K and magnetic beads conjugated with anti- eIF4A3 antibody or control (IgG). Next, the immunoprecipitated RNA was extracted. The results were measured by qRT-PCR.

Four-week-old male nude mice were purchased from the National Laboratory Animal Center (Shanghai, China). The animal studies were approved by the Institutional Animal Care and Use Committee of Pingjin Hospital. In total, 20 mice (n = 5 each group) were injected with the stable circMMP9 overexpressing U87 cells, circMMP9-knockdown U251 cells (1 × 10⁶) or parent control GBM cells resuspended in growth medium (150 μL) and Matrigel substrate (150 μL). Both the stable U87 and U251 cells contained a GFP marker. The mice were injected with 4.0 mg of luciferin (Gold Biotech) in 50 μl of saline. After 1 h, tumors were detected using an IVIS@ Lumina II system (Caliper Life Sciences, Hopkinton, MA). The animals were sacrificed 28 days after injection, and the tumors were collected to measure the tumor volume every 7 days. The tumor volume was calculated using the following formula: volume (mm³) = length × width²/2.

We selected 3 paired GBM tissues to carry out the microarray assay to determine the expression status of circRNAs in GBM tissues. Next, the host genes of differentially expressed circRNAs were subjected to GO analysis (Additional file 2: Figure S1A and S1B). The clustered heat map in Fig. 1a shows the top 20 upregulated and downregulated circRNAs. hsa_circ_0001162 (circMMP9) was the circRNA with the greatest differential expression (Fig. 1b). Using a bioinformatics method (UCSC Date), we then explored circMMP9 formation and found that circMMP9, with a molecular weight of 328 bp, was formed from exons 12 and 13 of MMP9 (Fig. 1c). PCR analysis indicated that divergent primers could produce the circular isoform of MMP9 with cDNA but not with genomic DNA (gDNA), while convergent primers could amplify the linear isoform of MMP9 from both cDNA and gDNA in the 3 GBM tissues, U87 cells and U251 cells (Additional file 2: Figure S1C). Furthermore, qRT-PCR showed that circMMP9 can resist RNase R, while MMP9 mRNA can be degraded by RNase R (Fig. 1c, Additional file 2: Figure S1D). Sanger sequencing of the PCR products using divergent primers also confirmed the presence of a splice junction in circMMP9 (Fig. 1c). To explore the circMMP9 expression level in GBM, we used a qRT-PCR assay to measure its expression in 18 pairs of GBM and adjacent normal brain tissues. The results indicated that the expression level of circMMP9 was significantly increased in GBM tissues compared with that in normal brain tissues (P < 0.05, Fig. 1d). FISH assay results also showed that circMMP9 was highly expressed in GBM tissues (P < 0.05, Fig. 1e-f). In addition, we explored the location of circMMP9 in U87 and U251 cells. Confocal FISH assay results revealed that circMMP9 was primarily expressed in the cytoplasm (Fig. 1g-h). The above results suggest that circMMP9 may play important roles in GBM pathology.

First, we detected the expression of circMMP9 in normal human astrocytes (NHAs) and GBM cells (U251, SHG44, A172, SNB19 and U87). We found that circMMP9 was highly expressed in U251 cells and was poorly expressed in U87 cells (P < 0.05, P < 0.001, Fig. 2a). Therefore, U251 and U87 cells were chosen for further study. Next, circMMP9 and mock expression vectors were constructed (Fig. 2b). U87 cells were transfected with circMMP9 or mock plasmids, and the total RNA was then treated with RNase R. The results indicated that the circMMP9 overexpression vector was efficient (P < 0.001, Fig. 2c-d). Simultaneously, we analyzed the sequences of siRNAs for circMMP9 at splice junctions and synthesized circMMP9 siRNA1 and siRNA2 (Fig. 2e). U251 cells were transfected with circMMP9 siRNA1, siRNA2 or mock for 48 h. The results revealed that circMMP9 expression was knocked down by siRNA1 or siRNA2 compared with that in the mock group (P < 0.001, Fig. 2f-g); FISH assay verified these results (Fig. 2h). Furthermore, we confirmed that overexpression of circMMP9 significantly promoted the proliferation ability of U87 cells (P < 0.001, Fig. 2i and k). Silencing of circMMP9 significantly inhibited the proliferation ability of U251 cells (P < 0.001, Fig. 2j and l). Next, we observed the influence of circMMP9 on GBM cell morphology and found that cells were spindle-shaped, irregular and in a disordered arrangement in the U87-circMMP9 group (Fig. 2m) and were relatively single, short shuttle-like or round in the U251-siRNA1 and U251-siRNA2 groups (Fig. 2n). We also found that overexpression of circMMP9 significantly promoted the migration and invasion abilities of U87 cells (P < 0.001, Fig. 2o). Silencing of circMMP9 significantly inhibited the migration and invasion abilities of U251 cells (P < 0.001, Fig. 2p).

The previous results indicated that circMMP9 was primarily expressed in the cytoplasm, indicating that circMMP9 may serve as a miRNA sponge in GBM. To identify the downstream miRNAs of circMMP9, a miRNA microarray was performed in the same GBM tissues. In Fig. 3a, the top 20 upregulated and downregulated miRNAs are shown. Based on the fold change and P value, we found that hsa-miR-129-1-3p, hsa-miR-124-3p and hsa-miR-129-5p were the three miRNAs with the greatest differential expression (Fig. 3b). Using bioinformatics analysis (RegRNA prediction, with a cutoff: mfe ≤ − 15, score ≥ 150), we found that miR-124 had a putative binding site with circMMP9 (Fig. 3c). These results indicated that circMMP9 may serve as a sponge of miR-124 in GBM. Indeed, our results showed that overexpression of circMMP9 in U87 cells markedly decreased miR-124 expression, while the FISH assay suggested that circMMP9 and miR-124 were colocalized in the cytoplasm of U87 cells (P < 0.001, Fig. 3 d and g); silencing of circMMP9 in U251 cells markedly increased miR-124 expression, while the FISH assay suggested that circMMP9 and miR-124 were colocalized in the cytoplasm of U87 cells (P < 0.001, Fig. 3e and g). Pull-down assay results also indicated that circMMP9 was enriched in the biotin-labeled miR-124 group (P < 0.001, Fig. 3f). Additionally, a dual-luciferase reporter assay was performed to confirm the interaction between circMMP9 and miR-124, and the data revealed that transfection with an miR-124 mimic observably attenuated the luciferase activity of wild-type (WT) circMMP9 compared with the scrambled control (P < 0.001, Fig. 3h and i). Meanwhile, the miR-124 mimic did not affect the luciferase activity of circMMP9-mut (Fig. 3j). Subsequently, we evaluated the expression level of miR-124 in GBM tissues, and the results showed that miR-124 was downregulated in GBM tissues compared with that in adjacent normal tissues (P < 0.05, Fig. 3k-l). There was also a negative correlation between circMMP9 and miR-124 (r² = − 0.5152, P = 0.0287, Fig. 3m).

To further investigate the roles of miR-124 and circMMP9 in GBM progression, we performed rescue assays to evaluate the effects of the circMMP9/miR-124 axis on the proliferation, migration and invasion abilities of GBM cells. The results indicated that circMMP9 promoted U87 cell proliferation and miR-124 weakened this promotion. circMMP9 also increased the expression levels of proliferation-associated markers (PCNA and Ki-67), and miR-124 reversed this tendency (P < 0.001, Fig. 4a and c). Meanwhile, silencing of circMMP9 inhibited U251 cell proliferation, and an miR-124 inhibitor (anti-miR-124) attenuated this inhibition. Additionally, silencing of circMMP9 decreased the expression levels of proliferation-associated markers (PCNA and Ki-67), and anti-miR-124 reversed this tendency (P < 0.001, Fig. 4b and d). Furthermore, we also found that circMMP9 accelerated U87 cell migration and invasion and miR-124 weakened this acceleration (P < 0.001, Fig. 4e and g). Meanwhile, silencing of circMMP9 suppressed U251 cell migration and invasion, and anti-miR-124 attenuated this suppression (P < 0.001, Fig. 4f and h). Western blot assays revealed that the circMMP9/miR-124 axis regulated EMT marker (E-cadherin, snail and vimentin) expression in GBM cells (Fig. 4e-right side and Fig. 4f-right side).

Next, we further explored which signaling pathway might be affected by the circMMP9/miR-124 axis to regulate the GBM phenotype. According to the ceRNA regulation model in tumor cells, an mRNA array was performed to analyze the differentially expressed mRNAs in the original three paired GBM tissues. In total, 655 upregulated mRNAs were found in GBM tissues compared with that in the adjacent normal brain tissues. Additionally, we used PubMed to screen miR-124 target genes and found 20 validated targets in GBM (Table 1). By Venn analysis, we found that CDK4 and AURKA were the target genes of miR-124 and may be regulated by circMMP9 in GBM (Fig. 5a). First, we confirmed the expression of CDK4 and AURKA in four paired GBM tissues via IHC and western blotting, and the results indicated that they exhibited higher expression in GBM tissues (P < 0.001, Fig. 5b, c and d). Additionally, their expression had the same tendency as that of circMMP9 (P < 0.001, Fig. 5e). These results suggest that CDK4 and AURKA may serve as downstream factors in the circMMP9/miR-124 axis. Indeed, western blot assays revealed that circMMP9 significantly regulates CDK4 and AURKA expression via miR-124 (P < 0.001; Fig. 5f and g).

Using a bioinformatics method (https://​circinteractome.​nia.​nih.​gov/​index.​html), we found that four binding sites for eIF4A3 are present in the upstream region of the circMMP9 mRNA transcript (Fig. 6a). The data from an RIP (RNA binding protein immunoprecipitation) assay using anti-eIF4A3 antibody indicated that eIF4A3 can bind with MMP9 mRNA through the four putative binding sites, which we named a, b, c and d, but not circMMP9 (we named e) in the corresponding RNA-protein complex (Fig. 6b). Next, we constructed 5 RNA transcripts that contained different MMP9 sequences (upstream of circMMP9), and the RNA pull-down assay was performed. The results revealed that the sequence located upstream of circMMP9, which contains eIF4A3 binding sites, is important for the interaction between eIF4A3 and MMP9 mRNA (Fig. 6c). We then knocked down the expression of eIF4A3 and found a reduction in circMMP9 expression (Fig. 6d), while overexpression of eIF4A3 increased circMMP9 expression (Fig. 6e).

In vivo, U87 cells were transfected with the mock or circMMP9 plasmid. U251 cells were transfected with mock or circMMP9 siRNA1 + 2. The stable transfected U87 and U251 cells (2 × 10⁶ cells) were injected in nude mice, and tumors were allowed to grow for 7, 14, 21 and 28 days. The results indicated that the overexpression of circMMP9 generated an outstanding increase in the rate of xenograft subcutaneous tumor growth (Fig. 7a-b). IHC assay data also showed that the overexpression of circMMP9 upregulated CDK4 and AURKA expression (Fig. 7c). We also found that the silencing of circMMP9 generated an marked decrease in the rate of xenograft subcutaneous tumor growth (Fig. 7d-e). The IHC assay data also showed that the silencing of circMMP9 downregulated CDK4 and AURKA expression (Fig. 7f).