Glioblastoma (GBM) comprises distinct subtypes characterized by their molecular profile. Mesenchymal identity in GBM has been associated with a comparatively unfavorable prognosis, primarily due to inherent resistance of these tumors to current therapies. The identification of molecular determinants of mesenchymal transformation could potentially allow for the discovery of new therapeutic targets. Zinc Finger and BTB Domain Containing 18 (ZBTB18/ZNF238/RP58) is a zinc finger transcriptional repressor with a crucial role in brain development and neuronal differentiation. Here, ZBTB18 is primarily silenced in the mesenchymal subtype of GBM through aberrant promoter methylation. Loss of ZBTB18 contributes to the aggressive phenotype of glioblastoma through regulation of poor prognosis–associated signatures. Restitution of ZBTB18 expression reverses the phenotype and impairs tumor-forming ability. These results indicate that ZBTB18 functions as a tumor suppressor in GBM through the regulation of genes associated with phenotypically aggressive properties.
Implications: This study characterizes the role of the putative tumor suppressor ZBTB18 and its regulation by promoter hypermethylation, which appears to be a common mechanism to silence ZBTB18 in the mesenchymal subtype of GBM and provides a new mechanistic opportunity to specifically target this tumor subclass. Mol Cancer Res; 15(8); 998–1011. ©2017 AACR.
Glioblastoma (GBM) is the most malignant primary brain tumor, characterized by a highly invasive nature, poor prognosis, and resistance to aggressive therapies (1). Over the past decade, gene expression profiling has contributed to the identification of multiple GBM subclasses with distinct molecular and clinical characteristics (2, 3). In particular, the mesenchymal (MES) and the proneural (PN) groups appear as the most consistent subclasses reported in both studies (2, 3). The MES subtype is characterized by resistance to radiotherapy (4).
Using bioinformatics tools (5, 6), several transcription factors have been identified as master regulators of a “mesenchymal gene expression signature” (MGES) in GBM, including STAT3, CEBPB, and WWTR1 (a.k.a. TAZ; refs. 7, 8). More recently, a role of NF-κB in controlling the expression of the three master regulators and consequent mesenchymal differentiation was reported (4). Notably, the transcriptional repressor Zinc Finger and BTB Domain Containing 18 (ZBTB18; formerly ZNF238) was identified as a potential negative regulator of the MGES in GBM (8). ZBTB18 is a transcription factor that belongs to the Broad complex, Tramtrack, Bric à brac (BTB) or poxvirus and zing finger (POZ)-zinc finger (BTB/POZ-ZF) protein family and plays a crucial role in brain development and neuronal differentiation (9–12). Previous findings revealed that ZBTB18 is downregulated or lost in mouse gliomas and human GBM cell lines and have implicated ZBTB18 as a putative tumor suppressor in the brain (11). However, the mechanism of ZBTB18 downregulation in GBM remains to be defined.
Mounting evidence suggests that alteration of methylation pathways, which can induce silencing of tumor suppressor genes, is one of the earliest events in carcinogenesis that grant a predisposition to mutational changes (13, 14). The DNA repair gene O-6-methylguanine-DNA methyltransferase (MGMT) is one such gene for which promoter methylation has been shown to result in gene silencing in many cancers (15–17).
Here, we have characterized the role of ZBTB18 as a transcriptional repressor of gene signatures associated with aggressive properties and poor prognosis in GBM. Our results indicate that ZBTB18 serves as a tumor suppressor in the brain and is silenced by DNA methylation in mesenchymal GBM.
Materials and Methods
Tumor samples and culture of GBM-derived cells
GBM and cortical samples from epilepsy surgery were collected at the Department of Neurosurgery of the University Medical Center Freiburg (Freiburg, Germany) in accordance with an Institutional Review Board–approved protocol. Informed consent was obtained from all patients, in accordance with the declaration of Helsinki. Patient-derived GBM stem cells (BTSCs) were prepared from tumor specimens as previously described (18). For passaging, neurospheres were incubated in nonenzymatic cell dissociation solution (Sigma) and mechanically dissociated by pipetting. Proneural cells 3047, 3082, and 3111 were generated at the University of Uppsala (19). GBM cell lines (SNB19 and LN229) and HEK 293T cells were routinely grown in DMEM with 10% FBS. SNB19 and LN229 cells have been authenticated on 3/2/2017 by PCR-single-locus-technology (Eurofins Medigenomix). All cells were mycoplasma-free.
Classification of brain tumor stem cells
The classification of brain tumor stem cells (BTSC) was performed using 510 genes of the 840 classifier genes used by Verhaak and colleagues to classify 260 GBM samples (3) and 529 GBM tissue samples from The Cancer Genome Atlas (TCGA) with assigned subtypes for reference (Cancer Genome Atlas Research Network, 2008). The 510 genes were selected such that they reliably classify the extended set of 529 TCGA samples and were represented on the Illumina HumanHT-12v3 expression BeadChip arrays. The expression levels for these genes on the Illumina arrays and in the TCGA dataset were converted into z-scores, and the combined matrix was used to classify each BTSC sample based on a k-nearest neighbors (k = 10) and voting procedure, in which a subtype was assigned on the basis of the majority subtype among the 10 TCGA samples with highest correlation coefficients for these genes with respect to the BTSC sample. All data manipulations were performed in R (R Core Team, 2012) and MATLAB (The MathWorks, Inc.).
The classification of BTSCs was performed using a method of hierarchical clustering with a Euclidean distance metric to cluster the samples alongside the TCGA samples. Two mRNA datasets with subtype assignment were used as references: an mRNA dataset of 529 patient samples from TCGA assignment and a subset of the 810 signature genes published by Verhaak and colleagues (3). Gliomas (BT) and BTSC classification is reported in Supplementary Table S1.
ZBTB18 vector construction, lentiviral production, and infection
ZBTB18 coding sequence was PCR amplified from normal brain RNA using the following primers: BstXI-flag-hZBTB18v1-sense (TGGCCACAACCATGGACTAC AAGGACGACGATGACAAGTGTCCTAAAGGTTATGAAGACAG) and PmeI-hZBTB18-antisense (GCCTTGGTTTAAACTTATTTCCAAAGTTCTTGAGAG). The PCR product was first cloned into pDrive cloning vector (Qiagen), sequence validated, and subsequently transferred in the pCHMWS-EGFP lentiviral vector (kind gift from V. Baekelandt, University of Leuven, Leuven, Belgium). Lentiviral infections were performed as previously described (18).
For protein analysis of the 30 kDa band detected by the ZBTB18 antibody, SNB19 cells transduced with either control or FLAG-ZBTB18 lentiviral vector were subjected to immunoprecipitation using the M2 FLAG antibody (Sigma). The immunoprecipitated proteins were analyzed by gel electrophoresis and stained by Coomassie blue staining. The 30 kDa band was cut from the gel and analyzed by MS (Agilent 6520 Q-TOF) at the Core Facility Proteomics of the Center for Biological System Analysis (ZBSA) at the University of Freiburg (Freiburg, Germany). The identified peptides were aligned using the mascot software (Matrix Science).
DNA/RNA extraction and quantitative real-time-PCR
DNA and RNA were extracted from human cortex, tumor tissue, or cell culture using the All Prep DNA/RNA Mini Kit or miRNeasy Mini Kit (Qiagen). First-strand cDNA synthesis was generated using the Superscript cDNA synthesis Kit (Invitrogen). Quantitative RT-PCR was performed using SYBR Green (Applied Biosystems) and analyzed relative to 18sRNA (housekeeping) using the ΔCt method. Primer sequences are listed in Supplementary Table S2.
Total protein extracts were prepared in RIPA buffer supplemented with protease inhibitor cocktail (Thermo Scientific), phosphatase inhibitor cocktail (Sigma), and PMSF (Sigma). The following antibodies were used: mouse anti-FLAG M2 (Sigma), rabbit anti-ZBTB18 (Abcam, ab118471), and mouse anti–β-actin (Abcam, ab7291).
For microarray expression profiling, total RNA was prepared using the RNeasy Kit or the all Prep DNA/RNA/Protein mini Kit (Qiagen) and quantified using 2100 Bioanalyzer (Agilent). A total of 1.5 μg of total RNA was processed and analyzed at the Deutsches Krebsforschungszentrum (DKFZ) in Heidelberg (Germany). Hybridization was carried out on Illumina HumanHT-12v4 expression BeadChip. Microarray data were analyzed using the GSEA software (http://www.broadinstitute.org/gsea/index.jsp). Microarray gene accession number: GSE97350 (subseries GSE97347 and GSE97349).
Methylation analysis and pyrosequencing
Genomic DNA from tissues and cell lines was bisulfite modified using the EZ DNA methylation–gold Kit (The Epigenetics Company) according to the manufacturer's instructions. A pool of normal brain tissues or glioma samples were subjected to PCR to amplify ZBTB18 promoter CpG islands using a PyroMark PCR kit (Qiagen) and primers summarized in Supplementary Table S3. Non–CpG-containing primers, methylation-specific primers, and unmethylation-specific primers were designed to cover the entire region. PCR products were cloned into pDrive cloning vector (Qiagen) and submitted for sequencing with SP6 primer and T7 primer (GATC).
Pyrosequencing analysis was performed using a PyroMark Q96 instrument (Qiagen), following the manufacturer's protocol. Primers are summarized in Supplementary Table S3. The results were analyzed by PyroMark CpG software (Qiagen). The methylation index for each sample was calculated as the average value of CpG methylation in the CpG examined.
In vitro methylation and luciferase reporter assay
Different regions of the ZBTB18 promoter were amplified by PCR from brain (cortex)-derived RNA and cloned in the pGL3 vector (Promega). The activity of each promoter was measured by dual-luciferase reporter assay system (Promega) according to the manufacturer's instructions. Cells were seeded at 50% confluence in 6-well plates 24 hours prior to transfection. Cells at 80% to 90% confluence were transiently transfected with 2.5 μg of each pGL3-ZBTB18 promoter vector or pGL3 as a negative control, together with 0.5 μg of pRL-TK (Renilla luciferase reporter, Promega). Transfections were done using Lipofectamine 2000 as directed by the manufacturer (Life Technologies) in serum-free medium. All transfections were done in triplicate. After 48 hours, cells were lysed with passive lysis buffer (Promega). Luciferase activity of each sample was determined in technical triplicate using a Thermo Scientific Appliskan luminometer. All data were reported relative to luciferase activity (firefly/renilla). Twenty micrograms of pGL3-ZBTB18 promoter 4 vector was methylated with SssI methylase or with HpaII methylase (all from New England Biolabs, 2.5 U/μg DNA) in the presence of 160 μmol/L S-Adenosylmethionine (SAM; New England Biolabs), in a manufacturer-supplied buffer at 37°C for 2 hours. The unmethylated DNA was treated as above but without methylases or SAM. The plasmid DNA was extracted by phenol/chloroform, ethanol precipitated, and quantified using a Nanodrop Spectrophotometer (Peqlab). The completion of methylation reaction was controlled by digesting both methylated and unmethylated DNAs using the methylation-sensitive restriction enzyme Hpa II and the methylation-insensitive restriction enzyme McrBC.
Promoter analysis was performed with the MatInspector software (www.Genomatix.de). Primers were designed using the Primer 3 software (http://bioinfo.ut.ee/primer3-0.4.0/) and are listed in Supplementary Table S4. Chromatin immunoprecipitation (ChIP) was performed as previously described (20) with some modifications. SNB19 cells expressing ZBTB18 were first crosslinked with 1% formaldehyde (Polysciences) for 30 minutes at room temperature and quenched by addition of 125 mmol/L Glycine for 15 minutes. Lysates were sonicated for 15 minutes (30 seconds on/30 seconds off) using Branson digital sonifier and centrifuged at 14,000 rpm for 15 minutes at 4°C. Fifty micrograms of sonicated chromatin per IP was precleared and incubated with primary antibody (4 μg) anti-ZBTB18 (Abcam, ab118471). Chromatin–antibody complexes were eluted by two 15-minute incubations at 30°C with 250 μL Elution Buffer (1% SDS, 100 mmol/L NaHCO3). Chromatin was reverse-crosslinked by adding 20 μL of NaCl 5 mol/L and incubated at 65°C for 12 hours. DNA was extracted by phenol-chloroform after RNase and proteinase K digestion. Quantitative RT-PCR was performed using SYBR Green (Applied Biosystems).
For methylation analysis, DNA was prepared using the All Prep DNA/RNA/Protein Mini Kit (Qiagen) and quantified using NanoDrop2000c (Thermo Scientific). A total of 1.5 μg of DNA for each sample was processed at DKFZ by Illumina HumanMethylation450 (Illumina). Data analysis was performed using Integrative Genomics Viewer (IGV) software (Broad Institute; refs. 21, 22).
Migration, invasion, and proliferation assay
Migration and invasion assays were performed as described before (18). Images of migrating cells were taken every 24 hours. For BTSC233 and JX6 cells, laminin-coated (Invitrogen; 4 μg/mL) 60 mm dishes containing a culture insert (Ibidi) were used. Cell migration was calculated using the following formula: (premigration area − migration area)/premigration area × 100).
For invasion assay, 1 × 105 BTSC233, JX6, or SNB19 (both 2.5 × 104) cells were seeded in triplicates in the upper compartment. PDGF-BB (20 ng/mL; R&D Systems) was used as a chemoattractant. Pictures were acquired using an Axioimager 2 Microscope (Zeiss). The assays were validated in two independent experiments.
Cell proliferation was assessed using a commercially available kit for EdU detection (EdU Cell proliferation assay, base click). Cells were plated at a density of 2.0 × 104 per well in a 24-well microplates containing laminin-coated coverslips. After 24 hours of seeding, cells were incubated with EdU solution overnight following the manufacturer's instructions. Upon EdU detection, images were acquired using a fluorescent microscope (Axiovert; Zeiss).
A total of 100 mmol/L stock solutions of 5-AZA-2′-deoxycytidine (5-AZA-dC; Sigma-Aldrich) were prepared by dissolving the substances in DMSO (GIBCO) and stored at −80°C. Immediately before treatment, stock solutions were diluted in cold PBS and added to the cell culture medium. LN229 cells not expressing ZBTB18 were treated with 10 μmol/L 5-aza-dC for 2, 3, 4, 5, and 8 days. JX6 cells and BTSC161s cells were treated with equal amounts of 5-aza-dC for 3 and 6 days. Medium containing fresh drug was changed every 24 hours.
Intracranial injection and immunohistochemistry
Intracranial injections were performed in NOD/SCID mice (Charles River Laboratories) in accordance with the directive 86/609/EEC of the European Parliament, following approval by regional authorities. Experiments were performed as described before (8). Animals were monitored daily until the development of neurologic symptoms by a blinded operator. Histology was performed as previously described (23). For MIB1 staining, primary antibody MIB1 (1:50; Dako; M7240) was used.
Linear regression analyses and graphical model validation were executed using R software. Scatterplots and locally weighted least squares (LOWESS) smooths were used to confirm the suitability of linear regression analyses, and statistical significance of these relationships was assessed according to the P value for the estimated slope of the regression line. A multiple linear regression model was computed based on the ordinary least squares method. Expression analysis of the TCGA data was performed by R software. Publicly available Level 3 TCGA (https://tcga-data.nci.nih.gov/docs/publications/tcga/) data were used for analysis. Data were downloaded at the UCSC Cancer Genome Browser. Only patients with full datasets (expression, methylome, and clinical information) were included. Expression analysis was performed based on Agilent array data (TCGA GBM G4502A) for high-grade glioma and RNA-seq data (TCGA LGG HiSeqV2 PANCAN) for low-grade glioma. Both datasets were normalized and log2 transformed. Methylation data from Infinium HumanMethylation450 BeadChip for lower-grade and high-grade glioma were used for further analysis. For TCGA Expression/Methylation Analysis, normalized expression values were analyzed in tumor subtypes (mesenchymal, proneural, classical, and neural) by a one-way ANOVA model. Survival analysis was performed by Cox proportional hazards model and plotted by Kaplan–Meier survival statistics. Patients without survival data were censored. Robustness was ensured by 10-fold cross-validation. Methylation of the cg23829949 was extracted and analyzed by Wilcox regression model and one-way ANOVA. Significant level was defined as P < 0.001. Analysis was performed by survival package included in R-Software. Different tumor numbers used in the analyses reflects sample availability in different genomic datasets and their overlap.
ZBTB18 is downregulated in high-grade gliomas
To study the role of ZBTB18 in gliomas, we looked at its expression in 1,161 low- and high-grade gliomas from TCGA. ZBTB18 expression was lower in GBM (WHO grade IV) compared with grade II and III gliomas (P < 0.001; Fig. 1A). Interestingly, ZBTB18 expression appeared strongly associated with the proneural subclass when we focused our analysis on high-grade gliomas only (n = 561; Fig. 1B). This finding was confirmed by cluster analysis of ZBTB18 correlating genes in high-grade gliomas (Fig. 1C). ZBTB18 was mostly associated with the proneural subtype and expressed at lower levels in the mesenchymal and classical subtype (Fig. 1C), consistent with our previous identification of ZBTB18 as a putative transcriptional repressor of the MGES described by Phillips and colleagues (2, 8). ZBTB18 protein expression analysis showed low expression of ZBTB18 in GBM-derived cells compared with normal brain samples (cortex; Fig. 1D and E), further reinforcing the notion that ZBTB18 is downregulated in high-grade gliomas. Interestingly, a lower band around 30 kDa was also detected.
ZBTB18 directly regulates MGES genes
Our previous study identified ZBTB18 as a putative negative regulator of the MGES of GBM (8). Moreover, cluster analysis shown in Fig. 1B indicates that ZBTB18 is downregulated in both mesenchymal and classical GBMs, suggesting that low ZBTB18 expression is associated with the more aggressive GBM subtypes. To validate ZBTB18-repressive function, we first analyzed the expression of mesenchymal genes previously predicted to be negatively connected to ZBTB18 in the ARACNE network (8), using SNB19 GBM cell lines transduced with FLAG-ZBTB18 or control vector (Fig. 2A; Supplementary Fig. S3A and S3B). Among 13 putative targets analyzed, 8 genes were clearly downregulated upon ZBTB18 overexpression (Fig. 2A). ChIP revealed ZBTB18 binding at ACTN1, PTRF, SERPINE1, and CD97 promoters (Fig. 2B), indicating that, at least for a subset of targets, ZBTB18 is directly involved in gene repression. To better understand the role of ZBTB18 in high-grade gliomas, we performed overexpression studies and subsequent GSEA (http://software.broadinstitute.org/gsea/index.jsp) on mesenchymal patient-derived BTSC-like cell line (BTSC233; Figs. 1D and E and 2C and D) and a patient-derived GBM xenoline (JX6; ref. 24) which was classified as classical according to the Verhaak study (Figs. 1D and E and 2E and F). Interestingly, we also detected a lower band similar to those observed in GBM-derived cells at around 30 kDa (Figs. 2C and E and 1D). Sequence analysis by MS confirmed that the peptides correspond to the N-terminal portion of ZBTB18 up to around amino acid 270 (Supplementary Fig. S1A and S1B). Western blot analysis of the transduced cells with a polyclonal anti-ZBTB18 antibody showed almost no expression of ZBTB18 in the control cells; the shorter band was weakly recognized because the antibody is directed to central region of ZBTB18 sequence (amino acids 228–498; Supplementary Fig. S1C and S1D). GSEA analysis for the gene signatures described by Phillips and colleagues (2) revealed a strong enrichment for mesenchymal genes in BTSC233 cells expressing a control GFP vector compared with BTSC233 cells expressing ZBTB18 (Fig. 2G; Supplementary Fig. S2A). Surprisingly, the same analysis using gene signatures from the Verhaak classification (3) did not show any specific signature enrichment (Supplementary Fig. S2B). Validation by qRT-PCR in BTSC233 expressing ectopic ZBTB18 confirmed the downregulation of several mesenchymal genes (Fig. 2H). The repressive function of ZBTB18 on a subset of genes was further validated in JX6 cells (Fig. 2I). GSEA showed a strong downregulation of the Phillips proliferative signature in JX6 transduced with a ZBTB18 expressing lentiviral vector compared with the control vector, but again no significant change was induced in the Verhaak signatures (Fig. 2J; Supplementary Fig. S3A). As the classification by Phillips is based on gene expression data of GBMs and grade III gliomas, with the goal of identifying survival-associated genes, we reasoned that the different results might indicate a role of ZBTB18 in the negative regulation of genes associated with poor survival (i.e., proliferative and mesenchymal genes). Interestingly, some of the validated ZBTB18 targets have been previously associated with unfavorable prognosis in glioma (25–27). Further examination of the top downregulated genes by ZBTB18 in JX6 cells by gene expression array highlighted many genes previously reported to play a role in epithelial-to-mesenchymal transition (EMT; Fig. 3A; refs. 28–34). ID1 and ID3 were also downregulated, as consistent with previous findings (Fig. 3A; ref. 35). The repressive role of ZBTB18 was validated by qRT-PCR in JX6 (Fig. 3B) and BTSC233 (Fig. 3C) cells. Many of the validated genes have been reported as part of a multi-cancer gene expression signature associated with prolonged time-to-recurrence in GBM (36, 37). As such, GSEA showed a strong loss of this multi-cancer signature (Anastassiou_cancer_mesenchymal_transition signature) upon ZBTB18 overexpression (Fig. 3D). These data further suggest that ZBTB18 downregulation in high-grade gliomas leads to re-expression of genes associated with malignant features and poor outcome. Interestingly, ZBTB18 re-expression in BTSC233 and JX6 cells led to upregulation of epithelial markers which are often repressed during EMT (refs. 38–43; Fig. 3E and F), further reinforcing the idea that ZBTB18 could play a role in suppressing an EMT-like phenotype in GBM. Consistent with our data and with its previously reported tumor-suppressive role (11), ZBTB18 reexpression in SNB19 affected cell proliferation, migration, and invasion (Supplementary Fig. S4). The same effect on cell proliferation, migration, and invasion was confirmed in BTSC233 cells (Supplementary Fig. S5A, S5C, and S5D). In JX6 cells, ZBTB18 overexpression also reduced cell proliferation and migration (Supplementary Fig. S5B and S5E), although no clear effect on invasion was observed, probably due to the higher invasive properties of those cells (data not shown).
ZBTB18 inhibits brain tumor growth in vivo
We next addressed the role of ZBTB18 in tumor formation in vivo. Immunocompromised (NOD/SCID) mice were intracranially injected with JX6 or BTSC233 cells stably expressing either ZBTB18 or control-GFP vector. Histology analysis of the mouse brains revealed that mice injected with JX6 cells expressing control vector developed bulky tumors. Conversely, only one mouse injected with JX6 cells expressing ZBTB18 formed a very small tumor (Fig. 4A and data not shown). In accord, overall survival was significantly increased in the ZBTB18 group (Fig. 4B). The same experiment in BTSC233 confirmed the effect of ZBTB18 on survival even though the mice still developed tumors (Fig. 4C and D). Ki67/MIB1 staining showed a high level of proliferation in BTSC233 transduced with the control-GFP vector, whereas cells expressing ZBTB18 appeared to be less proliferative or confined in small satellite areas, suggesting that ZBTB18 might somehow impair proliferation or restrict it to specific tumor regions (Fig. 4E). These results suggest that expression of ZBTB18 prolongs animal survival by delaying or inhibiting tumor formation and that the extent of tumor inhibition might depend on the cell background.
Promoter methylation silences ZBTB18 in GBM
To elucidate the mechanism by which ZBTB18 is downregulated in GBM, we examined the ZBTB18 promoter in silico (http://genome.ucsc.edu). The analysis revealed the presence of two CpG islands (Fig. 5A and B), suggesting that DNA hypermethylation could play a role in ZBTB18 transcriptional repression. To verify this hypothesis, we first cloned several promoter regions covering the two CpG islands from a pool of normal brains, GBM samples (BTs), and GBM cell lines after bisulfite modification (Fig. 5A and B). Sequence analysis of the cloned DNA fragments revealed no change in DNA methylation in the more upstream CpG island (CpG1, containing 27 CpGs; Fig. 5B). Conversely, higher methylation in CpG island 2 (CpG2, containing 9 CpGs) was detected in the pool of tumor samples (Fig. 5B). Pyrosequencing of promoter regions containing 6 CpGs located in CpG2 and 2 more downstream CpGs located in the 5′UTR revealed that, although not statistically significant, methylation of ZBTB18 CpG2 tended to be higher in the glioma samples compared with normal brain (Supplementary Fig. S6A). Furthermore, methylation of the 2 CpGs in the 5′UTR (5′UTR CpG -1 and 5′UTR CpG -2) was higher in gliomas compared with normal brain samples (P = 0.0184, unpaired t test; Supplementary Fig. S6B and S6C). The significance was even higher when only GBM samples were included in the analysis (P = 0.0017, unpaired t test; Fig. 5C).
Treatment with the hypomethylating agent 5′-Aza-2′-dC (Decitabine) in GBM cell line LN229 and two patient-derived mesenchymal GBM cell lines (BT161s and JX6), all showing downregulation of ZBTB18 and hypermethylation of the ZBTB18 promoter, resulted in reexpression of ZBTB18 after 72 hours (Fig. 5D). Pyrosequencing analysis confirmed concomitant reduction of 5′UTR CpG-1 and 5′UTR CpG-2 in LN229 cells (Supplementary Fig. S6C). Together, these data support a role of promoter hypermethylation in the silencing of ZBTB18 in GBM.
To further prove that the ZBTB18 region, including 5′UTR CpGs-1 and -2, is important for ZBTB18 promoter activity, we cloned several ZBTB18 promoter regions and analyzed their activity by luciferase reporter assay. As shown in Fig. 6A and B, the ZBTB18 promoter region 3, which does not contain the core promoter and the 5′UTR CpGs-1 and -2, had no promoter activity compared with other cloned promoter regions in which the core promoter and 5′UTR CpGs-1 and -2 were included (Fig. 6A and B). Next, we investigated the effect of DNA methylation on the luciferase reporter activity controlled by the ZBTB18 promoter region with the highest activity. As displayed in Fig. 6C and D, the promoter activity was completely inhibited by the SssI methylase, an enzyme that methylates all CGs, and to a less extent by HpaII methylase, which methylates only CGs in the CCGG context (Fig. 6C and D). These results further indicate that the ZBTB18 promoter region, which includes the core promoter and 5′UTR CpGs-1 and -2, is responsible for promoter activity and sensitive to DNA methylation. This is consistent with the expected role of methylated CpGs close to the TSS in gene expression regulation (44).
ZBTB18 promoter methylation is a hallmark of mesenchymal GBMs
Correlation analysis between ZBTB18 expression and DNA methylation in 251 GBM samples from TCGA (HumanMethylation27 platform, http://cancergenome.nih.gov) revealed a significant inverse correlation between ZBTB18 expression and promoter methylation (P = 1.96 × 10−05, linear regression; Fig. 7A). Interestingly, the examined probe (cg23829949) mapped into the same region analyzed by pyrosequencing (CpG island 2). Consistently, in our set of patient-derived glioma samples, ZBTB18 promoter methylation of 6 CpGs located at the 3′ end of CpG island 2 (Fig. 7B) and of 5′UTR CpGs-1 and -2 (Supplementary Fig. S6B) measured by pyrosequencing was higher in a subset of samples showing low ZBTB18 expression (high-grade gliomas, red line), whereas low promoter methylation was detected in samples with higher ZBTB18 expression (low-grade gliomas, blue line; Fig. 7B). Thus, methylation of CpG island 2 and 5′UTR CpGs-1 and -2 correlates with ZBTB18 expression, at least in a subset of GBM. However, a fraction of glioma samples that did not show promoter hypermethylation still had low ZBTB18 expression, suggesting that DNA hypermethylation is not the only mechanism regulating ZBTB18 expression, or alternatively, that additional methylated regions might be involved.
Promoter methylation analysis of the subset of gliomas with low ZBTB18 expression/high promoter methylation (indicated in red) versus high ZBTB18 expression/low promoter methylation (indicated in blue) previously analyzed using the Infinium HumanMethylation450 BeadChip (ref. 45; Supplementary Fig. S7D) confirmed differential methylation of CpG island 2 between the two tumor groups covering CpGs located in the same region analyzed by pyrosequencing (probes cg19698993 and cg12869659). Intriguingly, tumor samples with high ZBTB18 expression and low ZBTB18 promoter methylation analyzed by DNA methylation array showed high levels of global methylation (45), suggesting that silencing of ZBTB18 by promoter methylation could be a hallmark of non G-CIMP gliomas. Consistent with this hypothesis, ZBTB18 methylation correlated with G-CIMP and IDH1 mutation status so that ZBTB18 was more methylated in IDH1 wild-type compared with IDH1-mutant gliomas (Fig. 7C P = 1.914e−10, Welch two-sample t test). Consistent with this, ZBTB18 expression was higher in G-CIMP GBMs (Fig. 7D). We then investigated the relationship between ZBTB18 methylation (HumanMethylation27 platform) and GBM subclasses in 283 TCGA GBM samples. This analysis revealed a highly significant association between ZBTB18 methylation in a region confirmed by pyrosequencing (probe cg23829949) and mesenchymal GBM subtype (Fig. 7E). Accordingly, regression analysis of ZBTB18 expression on ZBTB18 methylation revealed a strong correlation only in mesenchymal tumors but not in nonmesenchymal tumors (all GBMs: P = 0.000026; mesenchymal tumors: P = 0.0136; nonmesenchymal tumors: P = 0.13, linear regression), further highlighting that methylation-induced silencing of ZBTB18 might be particularly important in the mesenchymal subclass (Supplementary Fig. S7A–S7C). This is consistent with previous analysis indicating that the mesenchymal tumors are usually non–G-CIMP (46, 47). Expression analysis of selected mesenchymal genes (CD97, ACTN1, EMP3, and CHI3L1) revealed an inverse association with ZBTB18 promoter methylation (Fig. 7E and F), further indicating that silencing of ZBTB18 through promoter hypermethylation could play a role in mesenchymal differentiation in GBM.
Survival modeling did not show a statistically significant association with patient survival (data not shown). Instead, we observed a significant link between time-to-tumor progression and ZBTB18 methylation in a two-class model of 109 GBM patients stratified according to lower-than-median versus higher-than-median ZBTB18 methylation (log-rank P = 0.029), such that patients with high methylation demonstrated a comparatively unfavorable outcome (Fig. 7G). This association between ZBTB18 methylation and time-to-tumor progression was also evident in a continuous univariate Cox model [univariate Cox model P = 0.037, HR for tumor progression with methylated ZBTB18: 8.30, 95% confidence interval (CI), 1.13–60.80] and prevailed in a multivariate Cox model that included GBM subclass (classical, mesenchymal, neural, proneural) as a covariate (multivariate Cox model P = 0.048; HR, 7.49; 95% CI, 1.01–65.32), suggesting that ZBTB18 methylation portends a more aggressive tumor phenotype.
Here, we describe a new role for the transcriptional repressor ZBTB18 as a negative regulator of signatures associated with poor survival in GBM, and we propose DNA methylation as a mechanism to silence ZBTB18 in the mesenchymal tumor subtype. Our finding is in accordance with a previous study reporting that ZBTB18 is lost in established human GBM cells and identifying ZBTB18 as a brain tumor-suppressor gene (11). We demonstrate that ZBTB18 is mostly expressed in low-grade gliomas and proneural GBMs but less expressed in mesenchymal GBMs. This is consistent with our previous identification of ZBTB18 as a transcription factor negatively associated with mesenchymal GBMs (8). ZBTB18 reexpression in primary BTSCs dampens the adherence poor-prognosis proliferative and mesenchymal signatures, which were identified as mutually exclusive to proneural in the subtype classification by Phillips and colleagues. Although the association between patient survival and mesenchymal subtype has not been confirmed in the previous TCGA study involving GBMs only (18), mouse models for glioma show that sequential mutations causing a shift from proneural to mesenchymal GBM are also associated with reduced survival (19, 20). Moreover, because tumors often show mixed subtype profiles (20, 21), it is possible that this might mask an association with survival. This was clearly shown by Patel and colleagues who demonstrated that pure IDH1 wild-type proneural tumors were associated with a better survival compared with highly heterogeneous proneural tumors containing cells with other subtypes (21). Similarly, tumor purity was recently shown to be an important parameter to determine a positive association of MGMT methylation with patient survival (17). We show that ZBTB18 attenuates the expression of genes associated with EMT and with time-to-tumor progression in GBM (36, 37). This is in line with our analysis that presented a strong link between ZBTB18 methylation and time-to-tumor progression of GBM but only a trend of association with patient overall survival (data not shown). This finding is in accordance with the prevailing argument that time-to-tumor progression might relate more closely to tumor repopulation, aggressiveness, and therapy resistance, which are biological properties also associated with mesenchymal differentiation of GBM. Given this link to tumor progression and the fact that the mesenchymal phenotype is more prevalent in recurrent GBM (2), ZBTB18 hypermethylation might play a role in both the mesenchymal differentiation characteristics of the de novo tumor and the progression toward a more mesenchymal phenotype. However, as reported in our previous study (8) and also confirmed by others (7), several transcription factors usually cooperate to regulate specific GBM subclasses. So, it would be interesting to study how deregulation of ZBTB18 fits in the previously described regulation of mesenchymal genes by other master regulators (i.e., STAT3, CEBPB, and TAZ). The recent report that, at the single-cell level, a GBM often consists of mosaic of subtype makes the picture even more complicated (48). Still, identifying regulators of different GBM subtype could be important since regulators of different subclasses co-existing in the tumor could be targeted.
The mechanism leading to ZBTB18 downregulation in GBM remained to be defined. Our data reveal a link between ZBTB18 promoter methylation and loss of ZBTB18 expression. However, we observed that, in some cases, low expression of ZBTB18 also occurs in the absence of promoter hypermethylation, suggesting that additional molecular mechanisms downregulating ZBTB18 are potentially operative in GBM. We further demonstrate a strong association between ZBTB18 promoter hypermethylation and the mesenchymal subtype of GBM, implying that silencing of ZBTB18 by promoter methylation is a particular hallmark of this unfavorable GBM subtype. We also report a strong correlation between ZBTB18 methylation and IDH1 wild-type, which is consistent with previous studies indicating that the majority of mesenchymal GBMs are non–G-CIMP (46, 47). Consistently, we show that key mesenchymal genes (2, 8), which are also differentially methylated in G-CIMP versus non–G-CIMP gliomas (46), are highly expressed in GBM exhibiting promoter hypermethylation/low expression of ZBTB18.
Collectively, our data identify ZBTB18 as a candidate tumor suppressor and transcriptional regulator of poor prognosis-associated signatures in GBM. We have identified promoter hypermethylation as a common mechanism to silence ZBTB18 in the mesenchymal subtype of GBM, which provides a new mechanistic opportunity to specifically target this tumor subclass.
Disclosure of Potential Conflicts of Interest
A. Califano is Cofounder and Chief Scientific Advisor at, and has an ownership interest (including patents) in, DarwinHealth, Inc. No potential conflicts of interest were disclosed by the other authors.
Conception and design: V. Fedele, A.P. Masilamani, S. Nelander, A. Iavarone, M. Bredel, M.S. Carro
Development of methodology: V. Fedele, F. Dai, A.P. Masilamani, R. Ferrarese, A. Iavarone, M. Bredel, M.S. Carro
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.P. Masilamani, D.H. Heiland, E. Kling, S. Doostkam, A. Weyerbrock, M. Prinz, M. Bredel, M.S. Carro
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Dai, D.H. Heiland, A.M. Gätjens-Sanchez, R. Ferrarese, H. Kim, S. Nelander, A. Califano, M. Bredel, M.S. Carro
Writing, review, and/or revision of the manuscript: V. Fedele, F. Dai, A.P. Masilamani, R. Ferrarese, S. Nelander, A. Weyerbrock, A. Califano, A. Iavarone, M. Bredel, M.S. Carro
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Prinz, M. Bredel
Study supervision: M.S. Carro
Other [Experimental assay (migration and proliferation assay); analysis of experimental assay; qPCR; ChIP; Cell culture)]: A.M. Gätjens-Sanchez
Other (technical support): L. Platania
The authors thank M. Oberle for mice brain histology, M. Lübbert for providing access to pyrosequencing facility, T. Feuerstein for normal brain tissues, C. Stein for technical assistance, R. Claus for help with methylation data analysis, and D.Ó. hAilín for editing the article (all University of Freiburg), V.D. Marinescu for BTSCs classification (University of Uppsala), S. Nozell for input on the article (UAB, Birmingham); Y. Gillespie (UAB, Birmingham) for providing JX6 cells; and V. Baekelandt (Katholieke Universiteit Leuven) for pLV-eGFP lentiviral vector.
This study was supported by Marie Curie International Reintegration Grant (MC IRG 268303), Deutsche Forschungsgemeinschaft Grant (DFG, CA 1246/2-1; both to M.S. Carro), and German Cancer Aid Grant Award (107714, M. Bredel). M. Prinz was supported by the DFG (SFB 992).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.