Changes in chromatin state reveal ARNT2 at a node of a tumorigenic transcription factor signature driving glioblastoma cell aggressiveness
Although a growing body of evidence indicates that phenotypic plasticity exhibited by glioblastoma cells plays a central role in tumor development and post-therapy recurrence, the master drivers of their aggressiveness remain elusive. Here we mapped the changes in active (H3K4me3) and repressive (H3K27me3) histone modifications accompanying the repression of glioblastoma stem-like cells tumorigenicity. Genes with changing histone marks delineated a network of transcription factors related to cancerous behavior, stem state, and neural development, highlighting a previously unsuspected association between repression of ARNT2 and loss of cell tumorigenicity. Immunohistochemistry confirmed ARNT2 expression in cell sub-populations within proliferative zones of patients’ glioblastoma. Decreased ARNT2 expression was consistently observed in non-tumorigenic glioblastoma cells, compared to tumorigenic cells. Moreover, ARNT2 expression correlated with a tumorigenic molecular signature at both the tissue level within the tumor core and at the single cell level in the patients’ tumors. We found that ARNT2 knockdown decreased the expression of SOX9, POU3F2 and OLIG2, transcription factors implicated in glioblastoma cell tumorigenicity, and repressed glioblastoma stem-like cell tumorigenic properties in vivo. Our results reveal ARNT2 as a pivotal component of the glioblastoma cell tumorigenic signature, located at a node of a transcription factor network controlling glioblastoma cell aggressiveness.
KeywordsBrain cancer Glioma Xenograft ChIP
De novo glioblastoma, the most common and malignant primary brain tumor in adults, is a paradigmatic example of heterogeneous tumors [11, 49, 54, 64]. This heterogeneity stems from clonal selection of genomic and phenotypic variants, which arises not only from the accumulation of mutations but also from dynamic changes in cell states [27, 28]. As a result, cells with different functional properties co-exist such as proliferative versus non-proliferative, migratory versus static, stem-like versus non-stem, pro-angiogenic versus non-pro-angiogenic. Understanding the basis for this heterogeneity is of importance to efficiently target pivotal tumor cells, especially in glioblastoma that exhibits a dismal prognosis despite aggressive therapies.
Studies of glioblastoma cells endowed with stem-like and tumor-initiating properties (GBM stem-like cells) have shown that aside from the heterogeneity linked to distinct mutational loads, cancer cell diversification can be achieved at the functional level within an unchanged genomic background . Glioblastoma cells have been shown to adopt distinct transcriptomic profiles combined with potentially distinct phenotypes and functional behaviors in response to environmental cues, which either favor acquisition of stem-like and tumorigenic properties [3, 24, 52] or in contrast induce their loss [41, 57]. Epigenetic plasticity has been shown to accompany GBM stem-like cell adaptations to their changing microenvironment [21, 22, 52, 71].
An important source of epigenetic plasticity is brought by post-translational histone modifications, such as methylation, acetylation, phosphorylation or ubiquitinylation of histone lysine (K) and arginine (R) residues . These histone modifications alter either the affinity between DNA and histones or create binding sites for chromatin remodeling factors, thereby controlling DNA compaction and accessibility, subsequent transcription and hence ultimately functional outcomes [7, 65]. Pioneer studies in embryonic stem cells (ESC) first revealed the link between histone H3 K4 and K27 trimethylation (H3K4me3 and H3K27me3) with transcriptional expression and repression, respectively [50, 53, 78], the importance of which has been confirmed by large scales epigenomic studies notably in the brain . In addition, these studies reported the existence of bivalent genes bearing both H3K4me3 and H3K27me3 histone marks in ESC [4, 6] as well as in adult multipotent/somatic stem cells [13, 51]. These bivalent genes are associated with RNA polymerase II at their transcription start sites and are thought to be in a “poised” state ready to be fully activated or repressed during differentiation [1, 10, 37, 50].
Materials and methods
GBM stem-like cells with mesenchymal (TG1), and classical transcriptome profiles (6240** and 5706**) were isolated from neurosurgical biopsy samples of human primary glioblastoma affecting 62–68-year-old patients, with a IDH wild-type status, and characterized for their stem-like and tumor-initiating properties as described [2, 25, 56, 62, 63, 67]. TG1-miR was derived from TG1 as described . GBM stem-like cells 6240** and 5706** were stably transduced with a lentiviral construct encoding the firefly luciferase (6240**) or the firefly luciferase and the fluorescent protein GFP (5706**) . All cells were cultured in defined medium containing bFGF and EGF. TG1, 6240**, and 5706** stem-like cells were transduced with lentiviral vectors encoding a control or an ARNT2 shRNA construct (pLKO.1-HPGK-puro-U6-non mammalian shRNA control, and pLKO.1-HPGK-puro-CMV-TGFP-U6-shARNT2, Sigma, France). All non-transduced cells were eliminated following puromycin treatment (2 µg/ml) for 10 days. Lentivirus was produced by the Plateforme vecteurs viraux et transfert de gènes (Necker Federative structure of research, University Paris Descartes, France).
Viable cell counting
Trypan blue exclusion test was used to determine the numbers of viable cells (Trypan blue solution, ThermoFisher, 0.4% v/v, 3 min incubation at room temperature). Blue and white cells (dead and alive, respectively) were counted with the Countess automated cell counter (Thermo Fisher, France).
Extreme limiting dilution assays (ELDA)
Cells were plated in 96-well plates at 1, 5, and 10 cells/well/100 μl as previously described . The percentage of wells with cell spheres was determined after 7 days. The analysis of the frequency of sphere-forming cells, a surrogate property of brain cancer stem-like cells  was performed with software available at http://bioinf.wehi.edu.au/software/elda/ .
ChIP-seq sample preparation and analysis
ChIP assays were performed using ChIP-IT Express Magnetic Chromatin Immunoprecipitation kit following the manufacturer’s protocol (Active motif, France) and 2 × 106 cells per sample and per epitope. Briefly, TG1 and TG1-miR-302–367 cells were cross-linked in 0.5% formaldehyde/PBS for 10 min at room temperature and then treated with 0.125 M glycine in PBS pH 7.4 for 5 min at room temperature. Samples were subsequently washed twice with ice-cold PBS and once with ice-cold PBS supplemented with protease inhibitors cocktail prior to be lysed. Chromatin fragments ranging from 200 to 500 bp were obtained by sonication (10 pulses at 40% of amplitude, 20 s ON, 50 s OFF, Sonics Vibracell VCX 130 sonicator, Sonics and materials, USA). Chromatin was then incubated overnight at 4 °C on a rotor with anti-H3K4me3 (Millipore, 07-473, France) or anti-H3K27me3 (Millipore, 07-449, France). The chromatin–antibody complexes were then washed, eluted and reverse cross-linked at 65 °C for 5 h. The eluted DNA was treated sequentially with Proteinase K and RNase A, and purified with the MinElute Reaction Cleanup Kit (Qiagen, 28204, France). The amount of DNA obtained was measured with a Qubit fluorometer (ThermoFisher, France). Library preparation was performed using the ChIP-Seq Sample Preparation kit (Illumina) on 10 ng of purified ChIP DNA samples. Libraries were sequenced on a Hiseq 2000, 1 library per lane, following standard procedures (Sequencing Platform of Montpellier GenomiX, MGX, France). An input control was sequenced for each cell type, and used for normalization. Alignments of the reads to the hg19 human reference genome were performed with CASAVA (1.8.2 version, Illumina). Alignments with more than two mismatching bases within the 32 first bases of the read were discarded. Visualization was performed with the Integrative Genomics Viewer (www.broadinstitute.org/igv/home). Peak detection was performed using the MACS software version 1.4.2 (http://liulab.dfci.harvard.edu/MACS/)  with input control libraries from the corresponding cell types. Peaks were then annotated using a window of ± 20 kb with respect to the coordinates of the beginning and end of RefSeq transcripts. More than 150 million short reads were obtained for all samples. These short reads were uniquely aligned to the human genome, resulting in a 77 and 76% of the genome covered in TG1 and TG1-miR, respectively. The data have been deposited in NCBI’s Gene Expression Omnibus  and are accessible through GEO Series accession number GSE98330 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE98330). ARNT2 ChIP was performed as described above using anti-ARNT2 antibodies (Santa Cruz, Cliniscience sc-5581X, France) and 100–1000 bp 5706** chromatin fragments. QPCR analysis was performed on total (input) and immunoprecipitated chromatin, and results normalized over the corresponding input signal. Enhanced representation of the regions of interest was compared to TBP promoter negative control. Sequences of all primers used for ChIP-qPCR are listed in Online Resource 1.
Gene expression analysis
Total RNA was prepared using the RNeasy Plus Universal kit (Qiagen, France) according to the manufacturer’s instruction. An on-column DNase digestion was performed during the extraction to yield a pure RNA fraction (RNase-Free DNase Set, Qiagen). cDNA was prepared using the QuantiTect Reverse Transcription Kit (Qiagen) according to manufacturer’s instructions. Expression profiles of TG1 and TG1-miR-302–367 were determined using Affymetrix 1.0 Human Exon ST arrays according to the manufacturer’s instructions in three successive cell passages (Strasbourg France Génomique platform, France). The signals obtained were normalized to a series of housekeeping genes (30 in total), and log2 transformed. RT-QPCR assays were performed using a Quantstudio6 (Applied Biosystems, France). PCR was performed using the SYBR Green PCR Core Reagents kit (Applied Biosystems, France). The thermal cycling conditions comprised an initial denaturation step at 95 °C for 10 min and 45 cycles at 95 °C for 15 s and 60 °C for 1 min. Transcripts of the TBP gene encoding the TATA box-binding protein (a component of the DNA- binding protein complex TFIID) were quantified as an endogenous RNA control. Quantitative values were obtained from the cycle number (Ct value), according to the manufacturer’s manuals (Applied Biosystems). Sequences of all primers used for QPCR are listed in Online Resource 2.
Statistical and graphical analyses of ChIP-seq and microarray data were performed using the R software version 3.2.3 (http://cran.r-project.org/). Gene ontology (GO) analysis was performed with DAVID software (version 6.8, http://david.abcc.ncifcrf.gov/). GO analysis of all genes changing histone marks in TG1-miR compared to TG1 was achieved using all human genes as background (Homo Sapiens from DAVID). GO analyses of genes exchanging an active for a repressive histone mark and vice versa between TG1 and TG1-miR were achieved using as background all the genes with differing histone marks in TG1-miR and TG1. Genes encoding transcription factors were retrieved using the KEGG (http://www.genome.jp/kegg/), and Genomatix databases (Genomatix, Germany). Interactions between the retrieved set of 202 transcription factors were analyzed with the STRING database (version 10.0, http://string-db.org/). Heat maps and z scores were downloaded from the IVY dataset (http://glioblastoma.alleninstitute.org), and analyzed with XLSTAT version 1.2. z-score graphs were generated with Prism 6.0 software (GraphPad). ARNT2 mRNA expression was analyzed using publicly available data using the R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl) (Lee, mixed glioblastoma dataset, GEO ID: GSE4536) and the HGGC website (http://126.96.36.199/hgcc/). TCGA transcriptome dataset of 481 surgical tissue samples of untreated primary glioblastoma (tcga 540 glioblastoma) was analyzed using the R2 Genomics Analysis and Visualization Platform. Single glioblastoma cell transcriptomes were obtained at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE57872, and analyzed with XLSTAT version 1.2.
Cells were harvested, washed with PBS and cell lysis was performed in 50 mM Tris–HCl pH 7.4 buffer containing 1% Triton X-100, 150 mM NaCl, 0.5 mM EGTA, 0,5 mM EDTA and anti-protease cocktail (Complete Protease inhibitor Cocktail Tablets, Roche, France). Protein extracts (30 μg) were separated by SDS-PAGE and transferred to Hybond-C Extra nitrocellulose membranes (GE Healthcare, USA) as described . The following antibodies were used for immunoblotting: anti-actin (Millipore Chemicon, 1:10000), anti-ARNT2 (Santa Cruz, 1:2000), anti-histone H3 (Abcam, 1:50000), anti-trimethyl-histone H3 (Lys 4) (Cosmobio, 1:500), and anti-trimethyl-histone H3 (Lys 27) (Upstate-Millipore, 1:3000). The secondary antibodies were anti-mouse IgG (Santa Cruz Biotechnology, 1:10000) and anti-rabbit IgG (GE Healthcare, 1:10000). Signal detection was performed with the ECL + chemiluminescence detection system (PerkinElmer, France). Densitometric analysis was achieved using ImageJ software.
Morphologic examination of patients’ glioblastoma resections was performed on Hematoxylin and Eosin stained sections (3–4 μm). Immunolabeling was performed using an automated system (Autostainer Dako, Glostrup Denmark). Deparaffinization, rehydration and antigen retrieval were performed using the pretreatment module PTlink (Dako). ARNT2 immunostaining was achieved using anti-ARNT2 (Santa Cruz, 1:50) and anti-Ki67 antibodies (MIB-1, Dako, prediluted). Immunostaining was scored by a pathologist (FBV).
Xenografted mouse brains were dissected after killing of the mice at 45 days post-graft of 6240** or 42 days post-graft of 5706** GBM stem-like cells expressing shControl or shARNT2. The brains were fixed in 4% paraformaldehyde in PBS for 48 h at 4 °C, cryoprotected in 30% sucrose in PBS at 4 °C until the tissue sank, frozen in isopentane at – 40 °C, and stored at – 80 °C. Cryostat sections of 30-µm-thickness were cut in the frontal plane. Thirteen sections, from the olfactory bubs to the posterior end of cerebellum were selected for the analysis. Sections were incubated with DAPI (Sigma, France) for 10 min at room temperature. Sections staining was analyzed with a fluorescent microscope (Axioplan 2, Zeiss). Images were acquired on digital camera (DXM 1200, Nikon, USA) using Zen 2 software (Zeiss) and prepared using Adobe Photoshop software (Adobe Systems, San José, USA).
The animal maintenance, handling, surveillance, and experimentation were performed in accordance with and approval from the Comité d’éthique en expérimentation animale Charles Darwin No. 5 (Protocol #3113). 6240** and 5706** GBM stem-like cells transduced with a luciferase encoding lentivirus and either a shControl or a shARNT2, were used. 140,000 (6240** and 5706**), 40,000 (6240**, 5706**), 20,000 (6240**) or 10,000 (6240**) cells were injected stereotaxically into the striatum of anesthetized 8- to 9-week-old nude mice (Envigo Laboratories, France), using the following coordinates: 0 mm posterior and 2.5 mm lateral to the bregma, and 3 mm deep with respect to the surface of the skull. Luminescent imaging was performed on a photonImager Biospace (Biospace Lab, France), after intra-peritoneal injection of 150 μl luciferin 20 mM (Thermo Fisher, 88293). Tumor formation was monitored by bioluminescence until all mice of the control group showed a signal. Bioluminescent signals were visualized with M3 Vision software (Biospacelab).
R version 3.2.3, XLSTAT version 1.2 or Prism 6.0 software (GraphPad) were used for statistical analyses. The level of significance was set at p < 0.05. The type of statistical test used is provided in the figure legends. All experiments were performed using independent biological samples with the exception of the ChIP-seq. All experiments were repeated at least three times in an independent manner with the exception of the microarray experiment. PCA analysis was performed on XLSTAT version 1.2, based on a Pearson correlation matrix. First and second component (F1 and F2 axis) were used to generate a correlation circle where the variables (genes) were plotted as vectors according to their correlation with F1 and F2 axis.
The figures were prepared using Adobe Illustrator (Adobe Systems).
Repression of GBM stem-like cell properties is accompanied by discrete changes in epigenetic profiles
Lentiviral expression of miR-302–367 in the TG1 human GBM stem-like cell line (referred to as TG1-miR) resulted in loss of their stem-like and tumorigenic properties . H3K4me3 and H3K27me3 profiling of TG1 and TG1-miR was performed by ChIP followed by deep sequencing (data accessible at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE98330). For each cell type analyzed, approximately 16,000 genes (~ 70% of the complete human exome) were found to be associated with the H3K4me3 and/or H3K27me3 mark (Online Resource 3). This analysis revealed a predominance of genes (~ 48%) associated with the active H3K4me3 mark in TG1 and in TG1-miR (Fig. 1b). Only ~ 10% were associated with the repressive H3K27me3 mark. An equivalent proportion (~ 10%) was associated with the bivalent mark (H3K4me3 and H3K27me3) (Fig. 1b). Western blot assays further demonstrated similar H3K27me3 and H3K4me3 protein levels in TG1 and TG1-miR (Fig. 1c). As described in other cell types [5, 78], both marks were enriched in TG1 and TG1-miR at the level of the TSS, with the H3K27me3 mark being in addition spread along the gene bodies (Online Resource 4). Furthermore, as expected, the highest transcript levels were observed in the group of genes associated with the H3K4me3 mark, the lowest transcript levels in the group of genes associated with the H3K27me3 mark, whereas genes associated with the bivalent mark had intermediate expression levels (Fig. 1d). The mean transcript level of the group of genes associated with none of the marks was slightly above the mean expression level of the genes carrying the H3K27me3 mark, suggesting that these genes tended to be repressed. Altogether, these results show that miR-302–367 does not alter global levels of H3K4me3 and H3K27me3, or the proportion of genes associated with either modification or the repartition of the histone marks along the genes.
Changes in histone modifications highlights ARNT2 as a core member of a transcription factor network associated with maintenance of GBM stem-like cell properties
To identify the function of the genes whose chromatin state is modified following repression of the properties of GBM stem-like cells, we performed functional enrichment analysis using DAVID toolbox [14, 15]. In a first step, we performed a gene ontology (GO) analysis using the whole set of the 5151 genes associated with different histone modifications in TG1 and TG1-miR (Fig. 2c, Online Resource 5). Several terms related to the central nervous system were significantly enriched as expected for cells derived from the brain. Consistent with the drastic change in cell functional state induced by the miR-302–367 cluster , we also found terms grouping genes located at the core of cell behavior (such as transcription, metabolism), and related to development and differentiation. We also found categories associated with cell motility (cell adhesion, differentiation, and chemotaxis) consistent with the propensity of TG1-miR to adhere to a permissive plastic support and with the loss of their invasive capacity . Functional enrichment analysis restricted to genes that permute from an active to a repressive histone mark showed enrichments in terms related to the maintenance of the undifferentiated features of the cells (Notch and Wnt signaling pathways, negative regulation of neuron differentiation, Fig. 2d, Online Resource 5). Conversely, genes that changed from the repressive H3K27me3 to the active H3K4me3 mark showed enrichments in terms related to neural cell differentiation (Fig. 2e, Online Resource 5).
ARNT2 is functionally associated with a molecular signature linked to glioblastoma cell tumorigenicity within the patients’ tumors
We first analyzed ARNT2 expression in two published independent transcriptome datasets of glioblastoma cells either devoid of or endowed with tumor-initiating properties [41, 73]. In agreement with our observations, we found that ARNT2 expression was downregulated in non-tumorigenic cells compared to tumorigenic cells in both datasets (Fig. 4d, e). Further, the analysis of the TCGA transcriptome dataset of 481 surgical tissue samples of untreated primary glioblastoma using the GlioVis Platform  showed lower ARNT2 mRNA levels in glioblastoma tissues than in non-tumoral brain tissues (Online Resource 8A). The finding of higher ARNT2 mRNA levels in normal brain tissues than in GBM tissues from which neurons are absent is coherent with the high ARNT2 expression in mature neurons [18, 19, 32]. Analysis of the TCGA glioblastoma dataset and the French glioma dataset gse16011 , showed no variation in ARNT2 expression according to MGMT status, IDH1 mutation or EGFR amplification (not shown). In accordance with the narrow distribution of ARNT2 expression levels across glioblastoma samples (Online Resource 8A), no correlation could be disclosed using the R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl) between variations in ARNT2 mRNA levels and the overall survival of patients (Online Resource 8B).
To further explore the relevance of ARNT2 expression in the context of the human tumors, we compared its expression in the tumor core areas of glioblastomas (IVY dataset) with the expression of genes associated with glioblastoma cells endowed with tumorigenic and stem-like properties. We used a 28 molecules’ signature delineated by the Bernstein laboratory from the combined analysis of single cell transcriptome profiles of cultured GBM stem-like cells and of 254 cells sampled from five different glioblastomas . We retrieved expression data from 27 of the 28 components of the signature from the IVY dataset (Online Resource 10). Principal component analysis (PCA) showed that ARNT2 expression co-varied with the tumorigenic/stem signature along the first principal component axis (F1 axis) (Fig. 5c). We verified whether this co-variation occurred also at the single cell level using the published transcriptome profiles of 254 glioblastoma cells . This dataset contains 25 of the 28 signature’s components (Online Resource 10). Principal component analysis showed that ARNT2 expression co-varied also with the stem signature at the single cell level (Online Resource 11).
Taken together, these results demonstrate that ARNT2 is expressed at the mRNA and protein level within the tumors of patients with glioblastoma. Further, they show that ARNT2 is part of a tumorigenic/stem signature of glioblastoma cells, and regulates the expression of transcription factors previously shown to be involved in the control of glioblastoma cell tumorigenicity.
ARNT2 is essential for the maintenance of glioblastoma cell tumorigenic properties
The above results, associated with our initial finding of ARNT2 down-regulation in glioblastoma cells deprived of tumorigenic properties, led us to evaluate the role of ARNT2 in the control of glioblastoma cell tumorigenicity using orthotopic xenografts of GBM stem-like cells (6240**, 5706**) expressing either a shControl or a shARNT2. ARNT2 knockdown inhibited the proliferation and the clonality of the cells in vitro (Fig. 6c–e, Online Resource 13B and C). Of note, our observation of increased ARNT mRNA levels upon ARNT2 knockdown (Fig. 6a, b) indicates that ARNT cannot compensate for ARNT2 knockdown. Orthotopic xenografts of 1×, 2×, 4× or 14 × 104 6240** or 5706** GBM stem-like cells stably expressing luciferase and either shControl or shARNT2 were used to follow tumor development with bioluminescent imaging. Results showed a striking reduction in tumor incidence in mice grafted with 6240**- and 5706**-shARNT2 compared to mice grafted with 6240**- and 5706**-shCTL cells (Fig. 6f, i, Online Resource 16A–D). Bioluminescence imaging 42 days post-graft revealed tumor formation in six out of seven, and in six out of six mice engrafted with 14 × 104 6240**-shControl and 5706**-shControl, respectively (Fig. 6f, i). In contrast, no bioluminescent signal was detected in the mice grafted with 6240**-shARNT2 or 5706**-shARNT2 (Fig. 6f, i). Similar results were obtained when grafting smaller numbers of cells (Online Resource 13). The reduced tumor development in the mice grafted with 6240** and 5706**-shARNT2 was confirmed by immunohistochemistry (Online Resource 16E-F). Survival assays revealed differing long-term consequences of ARNT2 knockdown according to the GBM stem-like cells grafted. Although we observed a significant improvement in the survival of the mice grafted with either 6240** or 5706** cells expressing shARNT2 (Kaplan–Meier analysis, Fig. 6h, k), only mice grafted with 6240**-shARNT2 eventually developed tumors. Determination of human ARNT2 mRNA levels by QPCR in these tumors showed ARNT2 as well as OLIG2, POU3F2 and SOX9 transcripts levels similar to tumors of the shCTL group (Online Resource 16G). This result indicates that the 6240** cells that formed the tumors escaped ARNT2 inhibition, further pointing to an essential role of this transcription factor for glioblastoma cell aggressiveness. Taken together, these results show that ARNT2 participates in the control of the tumorigenicity of glioblastoma cells.
Understanding the molecular basis of the varying functional cell states that co-exist within glioblastoma and participate in tumor resistance to treatments is of great importance to improve current therapeutic management. Differences in the ability of glioblastoma cells from the same tumor to initiate neoplasms has notably been highlighted by grafting cells sorted from glioblastoma surgical resections in immune-deficient mouse brains [35, 58, 73]. Recent studies have also shown the striking phenotypic plasticity of glioblastoma cells, which can adopt more or less aggressive states during the course of the tumor evolution and treatment [3, 33]. Here, we identified changes in the chromatin state of transcription factors, which accompany the passage of GBM stem-like cells from a highly aggressive to a poorly tumorigenic state. We uncovered a novel transcription factor controlling glioblastoma cell tumorigenicity, which is localized at a node of a transcription factor network controlling glioblastoma cell aggressiveness, and which clusters with a tumorigenic/stem signature of glioblastoma cells at both tissue and single cell levels.
Using the human glioblastoma cell line TG1 expressing or not expressing the micro-RNA cluster miR-302–367 as a model system , we profiled histone modifications. The results of our analyses uncovered a subset of genes showing changes in H3K4me3 or H3K27me3 between TG1 cells and TG1-miR cells in which the stem-like and tumorigenic properties have been repressed by expression of the miR-302–367. In agreement with the previously reported association of miR-302–367 with differentiation of GBM stem-like cells , , ontological pathway analysis of the subset of genes with changes in histone marks showed enrichment in ontological gene groups related to development and engagement in differentiation pathways. Enrichments in terms related to nervous system were also obtained (neuron, neurogenesis, synapse, forebrain), indicating conservation in the tumor cells of an imprint of their tissue of origin. Retrieval of the 202 transcription factors undergoing a change in histone marks further highlighted molecular pathways already identified as important players in the regulation of neural stem/progenitor cell but also of ESC and GBM stem-like cell behaviors, illustrating the pertinence of mapping histone epigenetic marks for identifying regulators of glioblastoma cell properties. For example, we observed an increased H3K27me3 associated with the gene encoding Nanog, a key factor in ESC pluripotency, and which has also been implicated in the maintenance of GBM stem-like cell properties [22, 52, 75]. Similarly, changes were observed for LEF1, an effector of the Wnt signaling pathway known to be involved in neurogenesis  and maintenance of GBM stem-like cell [38, 77, 79], TCF3 and TCF7 that are transcriptional regulators of the Wnt pathway in neural stem cells and ESC [40, 74] and the HES bHLH genes and FOXCs transcription factors implicated in the Notch signaling pathway, which is activated in neural stem cells and GBM stem-like cells [36, 72].
Mapping known and predicted protein–protein interactions between transcription factors exhibiting changes in histone marks in GBM stem-like cell lacking tumorigenic properties generated a network articulated around ARNT2. ARNT2, like its paralog ARNT, is considered to act as a dimerization partner of HIF1/2α, the heterodimers triggering the expression of hypoxia-related genes [46, 61]. ARNT2 expression is especially abundant in the central nervous system and kidney, while that of ARNT is ubiquitous [18, 32]. In the central nervous system, ARNT2 mRNA and protein are enriched in neurons . Although ARNT2/HIFs and ARNT/HIFs heterodimers are equally efficient to ensure neuronal responses to hypoxia , ARNT2 protein levels do not increase under hypoxic conditions unlike those of ARNT and HIF1/2α [42, 47]. The role of ARNT2 in cancer is poorly explored. ARNT2 has been associated with increased as well as decreased growth of non-cerebral cancers [39, 43, 46, 48, 59, 60]. In glioblastoma HIF1 and 2α, but not ARNT2, have been associated to adaptation of cancer cells to hypoxic conditions [30, 42]. We found that the profile of ARNT2 expression in glioblastoma does not correspond with that expected for a hypoxia-related molecule. ARNT2 expression was highest in glioblastoma core zones rather than in the hypoxic necrotic and pseudopalisading zones. This observation favors a hypoxia-independent transcriptional role for ARNT2.
Of note, ARNT2 is one of the few transcription factors switching from the active H3K4me3 to the repressive H3K27me3 mark in GBM stem-like cells expressing miR-302–367. Analysis of publically available data sets indicated that down-regulation of ARNT2 mRNA occurs not only in the TG1-miR cell line, but importantly is also observed in non-tumorigenic glioblastoma cells either directly sorted from patients’ tumors  or following serum-induced differentiation of GBM stem-like cell . This was confirmed at the protein level by immunohistochemistry of ARNT2 expression in sub-populations of cells within proliferative zones of patients’ glioblastoma. Furthermore, we demonstrated that ARNT2 knockdown inhibits tumor-initiating properties in vivo, supporting a role of ARNT2 in the tumorigenicity of glioblastoma cells.
Examination of tumor patients’ transcriptome datasets further associated ARNT2 with a tumorigenic/stem signature of glioblastoma cells  at both the tissue and single cell levels. To ascertain the functional relevance of this association, we focused on three transcription factors members of this stem signature SOX9, POU3F2 and OLIG2, since the knockdown of these factors has previously been reported to inhibit glioblastoma cell tumorigenicity in vivo [31, 44, 68]. We found that ARNT2 knockdown not only impaired the cell tumorigenicity in vivo but also resulted in decreased expression of SOX9, POU3F2 and OLIG2, hence placing ARNT2 at the core of transcriptional regulations of glioblastoma cell tumorigenicity.
In conclusion, our results uncover a novel transcription factor essential for glioblastoma cell tumorigenic properties, show its functional relevance within the context of the patients’ tumor, and shed new lights on the combinatorial organization of the transcription factor networks that regulate glioblastoma cell aggressiveness.
We are obligated to the members of the Cancer stem cell network of Ile-de-France headed by Dr. Christine Chomienne, for their continuous support and helpful discussions. We are grateful to A. Dias-Morais for technical assistance, and Dr. Tomohiro Yamaki (Chiba Ryogo Center, Japan) for helpful discussions. We thank for their precious help A. Borderie, S. Destree and C. Hagnere (Hôpital Pasteur, CHU Nice).
Compliance with ethical standards
All the procedures performed in studies involving human participants were in accordance with the 1964 Helsinki declaration and its later amendments and to the French laws. The institutional review board of the Sainte-Anne Hospital Center—University Paris Descartes (Comité de protection des personnes Ile de France III) approved the study protocol (Protocol Number DC-2008-323). All samples were obtained with informed consent of patients. The animal maintenance, handling, surveillance, and experimentation were performed in accordance with and approval from the Comité d’éthique en expérimentation animale Charles Darwin No. 5 (Protocol #3113).
This work was supported by La Ligue nationale contre le cancer (Equipe Labellisée LIGUE 2013, Equipe Labelisée LIGUE 2016 HC/MPJ), Institut National du Cancer (INCa 2012-1-PLBIO-07-INSERM-1, INCa-AAP Epigénétique et cancer 2014, HC/MPJ), Fondation pour la recherche sur le cerveau (FRC), Agence Nationale de la Recherche (ANR-13-1SV1-0004-03, JH/HC), Cancéropole Région Ile-de-France (EAE and AB fellowships), CAPES/COFECUB (Coordination pour le perfectionnement du personnel de l’enseignement supérieur/Comité français d’evaluation de la coopération universitaire et scientifique avec le Brésil, LGD fellowship), and Fundação Ary Frauzino para o Câncer (LGD fellowship).
Conflict of interest
The authors declare that they have no conflict of interest.
- 2.Assad Kahn S, Costa SL, Gholamin S, Nitta RT, Dubois LG, Feve M et al (2016) The anti-hypertensive drug prazosin inhibits glioblastoma growth via the PKCdelta-dependent inhibition of the AKT pathway. EMBO Mol Med 8:511–526. https://doi.org/10.15252/emmm.201505421 CrossRefPubMedPubMedCentralGoogle Scholar
- 10.Brookes E, de Santiago I, Hebenstreit D, Morris KJ, Carroll T, Xie SQ et al (2012) Polycomb associates genome-wide with a specific RNA polymerase II variant, and regulates metabolic genes in ESCs. Cell Stem Cell 10:157–170. https://doi.org/10.1016/j.stem.2011.12.017 CrossRefPubMedPubMedCentralGoogle Scholar
- 13.Cui K, Zang C, Roh TY, Schones DE, Childs RW, Peng W et al (2009) Chromatin signatures in multipotent human hematopoietic stem cells indicate the fate of bivalent genes during differentiation. Cell Stem Cell 4:80–93. https://doi.org/10.1016/j.stem.2008.11.011 CrossRefPubMedPubMedCentralGoogle Scholar
- 16.Debruyne DN, Turchi L, Burel-Vandenbos F, Fareh M, Almairac F, Virolle V et al (2017) DOCK4 promotes loss of proliferation in glioblastoma progenitor cells through nuclear beta-catenin accumulation and subsequent miR-302–367 cluster expression. Oncogene. https://doi.org/10.1038/onc.2017.323 PubMedGoogle Scholar
- 21.El-Habr EA, Dubois LG, Burel-Vandenbos F, Bogeas A, Lipecka J, Turchi L et al (2017) A driver role for GABA metabolism in controlling stem and proliferative cell state through GHB production in glioma. Acta Neuropathol 133:645–660. https://doi.org/10.1007/s00401-016-1659-5 CrossRefPubMedGoogle Scholar
- 22.Fareh M, Turchi L, Virolle V, Debruyne D, Almairac F, de-la-Forest Divonne S et al (2012) The miR 302–367 cluster drastically affects self-renewal and infiltration properties of glioma-initiating cells through CXCR4 repression and consequent disruption of the SHH-GLI-NANOG network. Cell Death Differ 19:232–244. https://doi.org/10.1038/cdd.2011.89 CrossRefPubMedGoogle Scholar
- 25.Galan-Moya EM, Le Guelte A, Lima Fernandes E, Thirant C, Dwyer J, Bidere N et al (2011) Secreted factors from brain endothelial cells maintain glioblastoma stem-like cell expansion through the mTOR pathway. EMBO Rep 12:470–476. https://doi.org/10.1038/embor.2011.39 CrossRefPubMedPubMedCentralGoogle Scholar
- 32.Hirose K, Morita M, Ema M, Mimura J, Hamada H, Fujii H et al (1996) cDNA cloning and tissue-specific expression of a novel basic helix-loop-helix/PAS factor (Arnt2) with close sequence similarity to the aryl hydrocarbon receptor nuclear translocator (Arnt). Mol Cell Biol 16:1706–1713CrossRefPubMedPubMedCentralGoogle Scholar
- 35.Huang P, Allam A, Taghian A, Freeman J, Duffy M, Suit HD (1995) Growth and metastatic behavior of five human glioblastomas compared with nine other histological types of human tumor xenografts in SCID mice. J Neurosurg 83:308–315. https://doi.org/10.3171/jns.1995.83.2.0308 CrossRefPubMedGoogle Scholar
- 41.Lee J, Kotliarova S, Kotliarov Y, Li A, Su Q, Donin NM et al (2006) Tumor stem cells derived from glioblastomas cultured in bFGF and EGF more closely mirror the phenotype and genotype of primary tumors than do serum-cultured cell lines. Cancer Cell 9:391–403. https://doi.org/10.1016/j.ccr.2006.03.030 CrossRefPubMedGoogle Scholar
- 56.Patru C, Romao L, Varlet P, Coulombel L, Raponi E, Cadusseau J et al (2010) CD133, CD15/SSEA-1, CD34 or side populations do not resume tumor-initiating properties of long-term cultured cancer stem cells from human malignant glio-neuronal tumors. BMC Cancer 10:66. https://doi.org/10.1186/1471-2407-10-66 CrossRefPubMedPubMedCentralGoogle Scholar
- 58.Piccirillo SG, Combi R, Cajola L, Patrizi A, Redaelli S, Bentivegna A et al (2009) Distinct pools of cancer stem-like cells coexist within human glioblastomas and display different tumorigenicity and independent genomic evolution. Oncogene 28:1807–1811. https://doi.org/10.1038/onc.2009.27 CrossRefPubMedGoogle Scholar
- 59.Qin XY, Wei F, Yoshinaga J, Yonemoto J, Tanokura M, Sone H (2011) siRNA-mediated knockdown of aryl hydrocarbon receptor nuclear translocator 2 affects hypoxia-inducible factor-1 regulatory signaling and metabolism in human breast cancer cells. FEBS Lett 585:3310–3315. https://doi.org/10.1016/j.febslet.2011.09.017 CrossRefPubMedGoogle Scholar
- 60.Raitila A, Lehtonen HJ, Arola J, Heliovaara E, Ahlsten M, Georgitsi M et al (2010) Mice with inactivation of aryl hydrocarbon receptor-interacting protein (Aip) display complete penetrance of pituitary adenomas with aberrant ARNT expression. Am J Pathol 177:1969–1976. https://doi.org/10.2353/ajpath.2010.100138 CrossRefPubMedPubMedCentralGoogle Scholar
- 65.Suganuma T, Workman JL (2011) Signals and combinatorial functions of histone modifications. Annu Rev Biochem 80:473–499. https://doi.org/10.1146/annurev-biochem-061809-175347 CrossRefPubMedGoogle Scholar
- 66.Sullivan AE, Raimondo A, Schwab TA, Bruning JB, Froguel P, Farooqi IS et al (2014) Characterization of human variants in obesity-related SIM1 protein identifies a hot-spot for dimerization with the partner protein ARNT2. Biochem J 461:403–412. https://doi.org/10.1042/BJ20131618 CrossRefPubMedGoogle Scholar
- 75.Zbinden M, Duquet A, Lorente-Trigos A, Ngwabyt SN, Borges I, Ruiz I, Altaba A (2010) NANOG regulates glioma stem cells and is essential in vivo acting in a cross-functional network with GLI1 and p53. EMBO J 29:2659–2674. https://doi.org/10.1038/emboj.2010.137 CrossRefPubMedPubMedCentralGoogle Scholar
- 76.Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, Nusbaum C, Myers RM, Brown M, Li W, Liu XS (2008) Model-based analysis of ChIP-Seq (MACS). Genome Biol 9(9):R137. https://doi.org/10.1186/gb-2008-9-9-r137 (Epub 2008 Sep 17. PubMed PMID: 18798982; PubMed Central PMCID: PMC2592715) CrossRefPubMedPubMedCentralGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.