A novel functional polymorphism of GFAP decrease glioblastoma susceptibility through inhibiting the binding of miR-139

Association of SNPs with GBM susceptibility

rs11558961, rs1042329, rs8067254, and rs17027 were consistent with HWE in healthy controls, with the P value of 0.620, 0.078, 0.080 and 0.261 respectively. Among the 4 candidate SNPs, rs11558961 variant was significantly associated with a decreased susceptibility of GBM. Compared with wild type CC, CG and CG+GG genotypes were significantly associated with reduced risk of GBM (CG vs. CC: adjusted OR=0.68, 95%CI=0.49-0.95, P=0.022; CG+GG vs. CC: adjusted OR=0.69, 95%CI=0.50-0.94, P=0.020). In allele comparing model, the rs11558961-G allele was significantly associated with a decreased susceptibility of GBM in comparison to C allele (adjusted OR=0.77, 95%CI=0.61-0.97, P=0.042) (Table 1).

Association of SNP rs11558961 Genotypes with miR-139 binding and GFAP expression

Bioinformatical analyses suggested that rs11558961 was located at miR-139 binding site, the 3'UTR of GFAP. rs11558961 C>G variant affected secondary structure and minimum free energy of the 3'UTR mRNA (Fig. 1A, B). To evaluate the different affinity of C and G allele to miR-139, three luciferase reporter constructs were created (Fig. 2A). The G allele was predicted to bind with miR-139 but C allele not. The binding site in mutant sequence was completely destroyed (Fig. 2B). In GBM cells transfected with rs11558961-G allele construct, miR-139 mimic attenuated the luciferase activity by 30.63% while compared with control group; however, miR-139 mimic only reduced the luciferase activity by 7.58% in rs11558961-C allele group (Fig. 2C). Antagomir-miR-139 increased the luciferase activity by 42.31% in G allele group than that by 9.86% in C allele group (Fig. 2D). Mutant construct lost the miR-139 binding site did not present any obvious change. Based on the above evidences, miR-139 mimic could interfere GFAP expression in both rs11558961-C and G allele groups. Antagomir-miR-139 increased GFAP expression in C and G allele groups. Under the three conditions (control, miR-139 mimic, antagomir-miR-139), cells carrying G allele construct showed lower luciferase activity than those with C allele. Destruction of the miR-139 binding site prevented the effects of miR-139, antagomir-miR-139 on GFAP expression.

Figure 1. rs11558961 affected secondary structure of GFAP mRNA. In silico prediction of rs11558961 impact on RNA folding structures. The structures and Mountain plots for free energy or entropy corresponding to rs11558961C (A) or G allele (B). With G allele, there was an emerging miR-139 binding site exposed.

Figure 2. SNP rs11558961 affects luciferase activities in U251 cells. (A) The schematic diagrams of reporter plasmids construct. It shows the luciferase reporters carrying two copies of rs11558961-C allele, G allele or mutant sequence at the 3'UTR of luciferase gene. (B) This plot represents the theoretical miRNA-mRNA duplex pairing between miR-139 and GFAP 3′UTR with C, G allele or mutant. The C, G alleles and mutant are highlighted with asterisks (*). (C, D) Relative luciferase activities (vs. Renila luciferase) were measured in U251 cells transfected with C allele, G allele or mutant construct. Cells in different groups were treated with control miR, miR-139 mimic or antagomir-miR-139. Six replicates for each group and the experiment were repeated three times. *P<0.05, ns: non-significant compared between treatment group and control group.

We further compared GFAP expression among GBM cells with different genotypes. The Western Blot data revealed that the expression level of GFAP were highest in CC group, and then in CG or GG group (Fig. 3A-B). In U251 cells transfected with miR-139 at different dose, GFAP expression was reduced dose-dependently by miR-139 mimic, and increased dose-dependently by antagomir-miR-139 (Fig. 3C-F).

Figure 3. miR-139 inhibited GFAP expression in U251 cells. (A, B) Western blot analysis for GFAP expression in GBM tissues from patients with different genotypes (A) and corresponding quantitative analysis (B). (C, D) Western blot assay of GFAP in U251 cells transfected with miR-139 mimic or antagomir-miR-139 in different doses. (E, F) Gray scale quantification of the GFAP protein expression levels. Each experiment was repeated three times. *P<0.05, **P<0.01, ns: non-significant compared between treatment group and control group.

rs11558961 C>G variant reduced chemoresistance and increased imatinib-induced apoptosis

To evaluate the biological effect of rs11558961 on malignant phenotype of GBM cells in vivo, we measured the chemoresistance in GBM cells from patients with CC, CG and GG genotypes. As shown in Fig. 4, the chemoresistance was notably different among patients with rs11558961 CC, CG and GG genotypes. With the treatment of imatinib (10μg/ml, 50μg/ml), the average apoptotic cell ratio was lowest in CC group, followed by CG and then highest in GG group (10μg/ml: 36.21±6.03, 53.19±10.32 and 62.51±11.34; 50μg/ml: 59.37±7.98, 72.59±11.57 and 75.42±12.53 respectively). These results suggested that rs11558961 variant, which was related with GFAP expression, indeed increased imatinib-induced apoptosis, and reduced chemoresistance.

Figure 4. rs11558961 C>G variant reduced chemoresistance and increased imatinib-induced apoptosis of GBM primary cells. (A) The apoptosis status of cells carrying different genotypes were examined under a fluorescence microscope after Hoechst 33258 staining. (B) Quantification analysis by Image J software. All experiments were performed in triplicate. Data was presented as mean ±SD. Scale Bar = 100μm. *P<0.05, **P<0.01.

rs11558961 variant suppressed the migration of GBM cells

It has been proven that GFAP is a pro-tumorigenic factor in brain tumor [5-7]. To confirm the association between rs11558961 and malignant progression of GBM, we detected the vimentin expression and migration of GBM cells with CC, CG and GG genotypes. After stained with Alexa flour 488 conjugated antibody, we found that the vimentin expression level in CC groups was significantly higher than that in CG or GG group (average integral optical density: 1869.05±57.32, 692.31±82.11, 387.62±42.76, respectively). Vimentin is a primordial component of class-III intermediate filaments. It is involved in increased cell motility and migration capability of GBM, which is a hint of malignant progression [12]. Then we further examined the migration ability of cells in different groups. The data clarified that migration cells trans filtration membranes were the most in CC group, then CG and GG group (172.39±18.75, 92.51±12.18 and 75.29±10.21, respectively). Taken together, rs11558961 C>G variant substantially reduced vimentin expression and suppressed the migration of GBM cells (Fig. 5).

Figure 5. Vimentin organization and migration of GBM primary cells with different genotypes of rs11558961. (A) Representative confocal-microscopy graphs of immunostaining for Vimentin in rs11558961 CC, CG and GG cells. Green: Vimentin; Blue: DAPI. (B) Quantitative analysis of vimentin fluorescence in different groups. (C) Migration of cells with different genotypes. The experiment was repeated three times. Data were presented as mean ± SD. Scale bar = 50μm. *P<0.05, **P<0.01.

Study population and postoperative follow-up

Consecutive GBM patients who received curative tumorectomy at Department of Neurosurgery in Zhenjiang first people's hospital, Changzhou first people's hospital, and the first affiliated hospital of Soochow University from Sep 2007 to Dec 2014. Healthy controls were randomly selected from people who received physical examination at Advanced Physical Examination Center. The demographics and clinical features of the GBM patients and controls involved in the study were listed in Table 2. Cases and controls were frequency-matched on gender and healthy controls were older than patients. All the patients were pathologically confirmed and signed the informed consent. Follow-up was performed via in person interview or telephone calls according to the standard epidemiologic procedure twice a year. The final date of follow-up was May 26, 2016. All the protocols conformed to the 1975 Declaration of Helsinki and were approved by the ethics committee.

Blood samples collection and GBM cells primary culture

Ethylene diamine tetraacetic acid (EDTA) anticoagulated peripheral blood samples were collected from patients before surgery and healthy controls. Genomic DNA was extracted using a RelaxGene Blood DNA System (TIANGEN biotech, Beijing, China) according to the manufacturer's instruction. The fresh surgical specimens of GBM from 11 patients were harvested and immediately immersed in ice-cold PBS containing 1% penicillin/streptomycin and 0.5% glutamine (Beyotime, Shanghai, China). Primary cell culture was performed within 60 min after surgery. As a variation of phenotype might occur at higher passages, we chose the cell cultures at the fourth passage for functional experiments.

SNP selecting and genotyping

We selected 4 SNPs by analyzing GFAP-related Han Chinese data from 1000 Genome Project resources (http://www.1000genomes.org). The candidate SNPs should meet these criteria: (1) the minor allele frequency (MAF) > 0.05; (2) r² < 0.80; (3) at the 3'UTR of GFAP gene. The significant SNP rs11558961 is located at a miR-139 binding site, with predicted proximal transcriptional regulatory potential (http://rsnp.psych.ac.cn/) (https://snpinfo.niehs.nih.gov/cgi-bin/snpinfo/snpfunc.cgi). rs1042329, rs8067254, and rs17027 were also predicted as functional SNPs, which were seated nearby other miRNA binding sites.

All SNPs of GFAP gene were genotyped using matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS). Amplification and single-base extension primers applied in multiple PCR were synthesized by Benegene (Benegene Biotechnology, Shanghai, China) (Table 3). The product of each sample was dispensed onto a 384-format SpectroCHIP with the MassARRAY Nanodispenser RS 1000. Then the MALDI-TOF MS assay was performed on a MassARRAY Compact Analyzer. Genotype calling was conducted using MassARRAY RT software version 3.0. Data was analyzed by MassARRAY Typer software 4.0.3 (Sequenom, San Diego, CA, USA). Genotyping quality was detected using Sanger sequencing of ~10% randomly selected samples, yielding a 100% concordance.

Bioinformatic prediction of candidate SNPs

The SNP-flanking region RNA was online analyzed using SNPfold (http://ribosnitch.bio.unc.edu/Downloads/SNPfold/) and RNAfold (http://rna.tbi.univie.ac.at/cgi-bin/RNAWebSuite/RNAfold.cgi). SNPs in the 3'UTR were predicted for putative miRNA binding site applying three algorithms: miRBase (http://www.mirbase.org/search.shtml), PicTar (https://pictar.mdc-berlin.de/- berlin.de/) and TargetScan (http://targetscan.org/).

Quantitative real-time PCR analysis of miR-139

The expression level of miR-139 in U251 cells was detected using RNA-tailing and primer-extension real-time PCR according to the instructions of All-in-One miRNA Detection Kit (Genecopoeia, Guangzhou, China). miR-139 forward primer and miRNA universal adaptor PCR primer were provided by Genecopoeia corporation (Genecopoeia, Guangzhou, China).

Luciferase reporter assay

To evaluate the binding of miR-139 to 3'UTR of GFAP, three reporter constructs carrying two copies of rs11558961-C allele, G allele or mutant sequences at the 3'UTR of luciferase gene were created. Firstly, three ~100 bp DNA sequences centered at rs11558961 (C or G allele) or mutant sequence were synthesized; then, two tandem copies of these sequences were cloned into pGL-3 Basic vector (Promega, Madison, WI, USA) using restriction enzyme sites- BamH I and Sal I. The schematic diagram for vector construction was shown in Fig. 2A. Day 1, U251 cells were seeded at 1×10⁴ cells/well in a 24-well plate. Day 2, cells were transfected with 0.8 μg pGL-3 Basic vectors (inserted with rs11558961- C allele, G allele or mutant sequences) and 0.16μg pRL-TK vector (Luciferase Assay System; Promega). Cells were transfected with mimic control, antagomir control, miR-139 mimic or antagomir miR-139 (GenePharma, Shanghai) in different group. Day 3, the luciferase activity was examined on Synergy H1 microplate reader (BioTek Instruments, Winooski, VT, USA) using Dual-luciferase Reporter Assay System (Promega, Madison, WI, USA) according to the instructions of manufactures. Results were represented as relative luciferase activity to pRL-TK.

Imatinib-induced apoptosis

The GBM primary cells with rs11558961 CC (n=5), CG (n=3) or GG (n=3) genotypes (5×10³ per well) were incubated with imatinib at the concentration of 10μg/ml or 50μg/ml for 24 hours. The cells were then fixed with 4% paraform followed by staining with 200 μl Hoechst 33258 reagent (Beyotime, Shanghai, China) in the dark for 30 min. After being washed twice with PBS, cells were immediately photographed under an inversion fluorescence microscope (Olympus IX51, Tokyo, Japan) to determine ratios of apoptotic cells. In CC, CG or GG group, cells were quantified by counting at least 3 samples, 5 independent visual fields/sample, to determine the apoptotic ratio. Data was presented as mean ± SD.

Cell migration assay

Cell migration measurements were performed using 24-well Transwell® units (8.0 μm pore, Costar Corning, NY, USA). In brief, different GBM primary cells (1×10³) in 100 μl were added to each matrigel-coated insert, and 500 μl DMEM supplemented with 10% FBS was added to the lower chamber. The plates were incubated at 37 °C, 5% CO2 for 24 h. The non-migrated cells in the insert were wiped away and the migrated cells were stained with ponceau. Finally, the images were captured with a phasecontrast microscope (Olympus IX51, Tokyo, Japan). There were 11 groups in this experiment, rs11558961 CC (n=5), CG (n=3) or GG (n=3). In each group, the migrated cells were quantified by counting at least 3 different samples, 5 independent visual fields/sample, to determine the average invasion ratio. Each assay was performed in triplicate.

Statistical analysis

The differences of the demographic and clinical features, and frequencies of genotypes in case-control study were calculated by the Chi-square test (for categorical variables) or Student's t-test (for continuous variables). Hardy–Weinberg equilibrium (HWE) was detected by online analytical tools (https://ihg.gsf.de/cgi-bin/hw/hwa1.pl). The linkage disequilibrium (LD) of candidate SNPs was analyzed using HaploView v4.2 (Broad Institute, Cambridge, MA, USA). For the determination of main effect of SNPs, univariate and multivariate logistic regression models were conducted to generate odds ratio (ORs) and corresponding 95% confidence intervals (CIs) with adjustment for possible confounders, i.e. age, gender, drinking status and so on. All statistical tests were two-sided and performed using Statistical Program for Social Sciences (SPSS 16.0, Chicago, IL, USA) and R (http://www.r-project.org/). P<0.05 was considered as statistically significant.