A driver role for GABA metabolism in controlling stem and proliferative cell state through GHB production in glioma

Glioblastoma samples from adult patients were obtained from surgical resections. For metabolite measurements and immunohistochemical analysis, glioblastoma fragments were subjected to multisampling.

GBM stem-like cells TG1, TG16, GBM-M, R633 cells were isolated from neurosurgical biopsy samples of human GBM (Table S1), and their stem-like and tumor-initiating properties characterized as previously reported [4, 42, 49, 53]. TG1-miR was derived from TG1 as described [15]. GBM stem-like cells 6240**, stably expressing a luciferase construct, GBM stem-like cells 5706**, and JolMa cells were characterized as described [7, 46]. No mutation in IDH1 or IDH2 coding regions was found (Table S1). TP54, TP80, TP83, TP84 stem-like cells with a K27M H3F3A mutation [58], were isolated from pediatric DIPG and characterized as previously described [52]. Molecular profiles were obtained with transcriptome analysis using Affymetrix Exon 1.0S array (3 independent biological replicates), and proneural, classical or mesenchymal subtype determined with respect to the classification of the TCGA established with a 840 genes list [55]. UT7 leukemia cell line was transduced with lentiviral vector encoding doxycycline-inducible human TET2-GFP cDNA (Fig. S6E). TG1 stem-like cells were transduced with lentiviral vectors encoding doxycycline-inducible human wild-type or catalytically deficient form of TET2-GFP cDNA (Fig. S6F). TG1, 6240**, 5706** and TP54 stem-like cells were transduced with lentiviral vectors encoding a control or an ALDH5A1 shRNA construct (GeneCopeia, Tebu, France). In relevant experiments, cells were treated with GHB or valproate (both from Sigma) or their vehicles (cell medium).

Cells and media were harvested 96 h post-seeding (cell half-doubling time = 4.5, TG1, and 8 days, TG1-miR). Cell pellets were washed in PBS before freezing. Media and cell samples (n = 6) were extracted and analyzed on the GC/MS and LC/MS/MS platforms of Metabolon, Inc. (Durham, NC, USA) as previously described [45]. Following normalization to total protein values for cell data (Bradford assay), log transformation, and imputation with minimum observed values for each compound, Welch’s two-sample t tests were used to identify metabolites that differed significantly between experimental groups. The level of significance was set at p < 0.05. Each metabolite was mapped to pathways based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) (http://​www.​genome.​jp/​kegg/​ ‎), the human metabolome database (http://​www.​hmdb.​ca), and literature mining. Targeted analyses of GHB (4-hydroxybutyrate) and 2-HG (2-hydroxyglutarate) were performed with GC–MS/MS (300MS, Brüker) in the clinical chemistry laboratory at Necker Enfants Malades Hospital (Paris, France). For metabolite measurements in tissues, glioblastoma fragments from adult patients were subjected to multisampling. Each sample was divided into two mirror pieces: one snap-frozen, the second reserved for immunohistological characterization. Tissue punches from weakly proliferative/differentiated (P^LOW/D⁺) and proliferative/non-differentiated (P^HIGH/D⁻) glioblastoma territories were obtained from the snap-frozen pieces.

Cell death was evaluated using the propidium iodide (PI) exclusion test. The cells were incubated with PI (10 µg/10⁶ cells) for 10 min at 4 °C, and the percentage of cells containing PI was measured using FACS (ARIA II, BD Biosciences, France).

Cells were plated at one cell/well (Nunc, 96-deep well plate, non-treated), and treated with 10 mM of GHB each 48 h over four weeks. The cells were then dissociated, and seeded at one cell/well. The percentage of wells containing spheres was scored. At least 500 cells were analyzed for each culture. For extreme limiting dilution assays (ELDA), cells were plated in 96-well plates at 1, 5, 10, 20, 50, and 100 cells/well/100 μl as previously described [4]. The percentage of wells with neurospheres was determined after 10 and 21 days. The analysis of the frequency of sphere-forming cells, a surrogate property of brain cancer stem-like cells [18] was performed with software available at http://​bioinf.​wehi.​edu.​au/​software/​elda/​ [25].

Cells were incubated with BrdU (5-Bromo-2′-deoxyuridine, 10 µM, Invitrogen) for 3 h, and analyzed after DAPI staining using an ARIA II (BD Biosciences, France). Analysis was performed on 10,000-gated events. Violet laser (405 nm) and Pacific Blue filter were used for DAPI detection, and Red laser (640 nm) and APC filter for BrdU detection. Data analysis and figure generation were performed using the FACS Diva version 6.1.2 program (BD Biosciences, France).

Cells were plated in 96-well plates (BD Biosciences, BD BioCoat Poly-d-Lysine) at a density of 5 × 10³ cells/well. The number of adherent cells was counted after 24 h of incubation and expressed in percentage of the total cell numbers. Images were acquired on a digital camera (DXM 1200, Nikon, USA) using AxioVision 4.6 Software (Laboratory Imaging, Ltd).

Cell transfection was achieved with the Amaxa Nucleofector Electroporator using the Nucleofector program A-020 (Amaxa Biosystems, Gaithersburg, MD, USA). Cells were transfected by electroporation with 1 μM of control siRNA (Ambion^® Silencer Negative Control, Cat#AM4611), or anti-ALDH5A1 siRNAs (Ambion^® Cat#16,708, ID si15460, Cat#16,708 ID si15462), or anti-TET2 siRNAs (Ambion^® Cat#4392420, ID si29443). The transfection was performed using the L transfection solution (AMAXA). The cells were chocked twice (at day 0 and day 3) and collected at day 6.

Cells were transfected with Renilla Luciferase mRNA and Firefly luciferase mRNA containing either the wild-type form of ALDH5A1-3′UTR or a mutant form of ALDH5A1-3′UTR with a deletion of the miR-302 putative target sequence (ALDH5A1-3′UTR-DEL). The quantification of Renilla and Firefly luciferase activity was performed using the Dual-Luciferase Reporter Assay System (Promega, France), according to manufacturer’s instructions. Renilla luciferase was used for internal normalization of Firefly activity values.

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). The following antibodies were used for immunoblotting: anti-Nanog (Cell Signaling, 1:1000), anti-p21 (Santa Cruz Biotechnology, 1:200), anti-Actin (Millipore Chemicon, 1:10,000), anti-SSAR (Novus biological, 1:2000), anti-ABAT1 (GABA-T) (Sigma Aldrich, 1:2500), anti-GAD65 (Abcam, 1:500), anti-GAD67 (Abcam, 1:500), anti-HOT (Sigma, 1:2000), anti-GLUD1 (Sigma, 1:2000), and anti-SSADH (Novus Biological, 1:2000). The secondary antibodies were anti-mouse IgG (Santa Cruz Biotechnology, 1:10,000), anti-rabbit IgG (GE Healthcare, 1:10,000) and anti-goat IgG (Santa Cruz Biotechnology, 1:10,000). Signal detection was performed with the ECL + chemiluminescence detection system (PerkinElmer, France). Densitometric analysis was achieved using ImageJ software.

Cells were harvested, PBS washed, smeared on SuperFrost slides (Fischer Scientific, France), and fixed in ice cold methanol for 20 min at −20 °C. Following fixation, cells were washed with PBS, and incubated for 30 min. at room temperature in PBS containing 0.3% Triton X-100 and 5% BSA (Sigma). The following primary antibodies were incubated overnight at 4 °C: Nanog (1:200, R&D), Olig2 (1:200, R&D), GFAP (1:5000, Dako), GLT-1/EEAT2 (1:200, Santa Cruz), ß3-Tubulin (1:3000, Covance), Map-2 (1:1000, Millipore). Secondary antibodies were incubated at room temperature for 1 h (Alexa Fluor^® 488, Alexa Fluor^® 555, Molecular Probes, 1:2000). Immunostaining was analyzed with a fluorescent microscope equipped with an ApoTome module (Axioplan 2, Zeiss). Images were acquired on a digital camera using AxioVision 4.6 Software (Laboratory Imaging, Ltd) and prepared using Adobe Photoshop software (Adobe Systems, San Jose, CA). Immunofluorescent signals were analyzed with Volocity 3D Image Analysis software (PerkinElmer, France) using single optical 240 nm sections.

Morphologic examination was performed on Hematoxylin- and Eosin-stained sections (3–4 μm). Immunolabeling was performed using an automated system (Autostainer Dako, Glostrup Denmark) with the following primary antibodies: anti-Ki67 (MIB-1, Dako, prediluted), anti-Olig2 (goat polyclonal, R&D Systems, 1:500) and anti-GFAP (Dako, 1:4000). Deparaffinization, rehydration and antigen retrieval were performed using the pretreatment module PTlink (Dako). SSADH immunostaining was achieved following overnight incubation with anti-SSADH (Novus Biological, 1:100) at 4 °C. Immunostaining was scored by a pathologist (FBV).

RNA was extracted from the cells with RNeasy Mini Kit (Qiagen, Hilden, Germany, http://​www.​qiagen.​com), submitted to an on-column DNase digestion (RNase-Free DNase Set, Qiagen) and retrotranscribed (QuantiTect Reverse Transcription Kit, Qiagen). PCR was performed using Platinum^® Taq DNA Polymerase High Fidelity (Invitrogen). The PCR products were purified and sequenced (Biofidal, Vaulx en Velin, France, http://​biofidal.​com). Q-PCR assays were performed using an ABI Prism 7700 Sequence Detection system (Perkin-Elmer Applied Biosystems), and the SYBR Green PCR Core Reagents kit (Perkin-Elmer Applied Biosystems). Transcripts of the TBP gene encoding the TATA box-binding protein were used for normalization.

5-hmC and 5-mC were quantified using dot blot assays with anti-5-hmC antibody (Active Motif, Carlsbad, CA, 1:10,000) and anti-5-mC antibody (Millipore Calbiochem, Darmstadt, Germany, 1:2000). DNA was spotted onto nylon Hybond N+ 53 membranes (Amersham) and fixed by U.V. cross-linking. Membranes were air-dried, blocked and incubated with anti-5-hmC or anti-5-mC antibodies and horseradish peroxidase-conjugated secondary antibody (anti-rabbit IgG 1:10,000, GE Healthcare, anti-mouse IgG 1:10,000, Santa Cruz Biotechnology). Signal detection was performed with the ECL + chemiluminescence detection system (PerkinElmer, France). Densitometric analysis was achieved using ImageJ software.

6240** GBM stem-like cells transduced with a luciferase-encoding lentivirus, 5706** GBM stem-like cells and TP54 DIPG stem-like cells stably expressing GFP were used. 100,000 (6240**), 50,000 (5706**) and 25,000 (TP54) cells were injected stereotaxically into the striatum of anesthetized 8- to 9-week-old Nude mice (Harlan Laboratories), 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. 16 mice were grafted with 6240** cells transduced with a shControl construct and 15 mice with a shALDH5A1 construct. Luminescent imaging was performed on an IVIS Spectrum (Perkin-Elmer), after intra-peritoneal injection of luciferin. Total flux (photons per second) values were obtained by imaging mice 14 and 49 days after stereotaxic cell injection and quantified with Live Image 4.0 software. Xenografts of GFP-expressing 5706** and TP54 transduced with a shControl construct or a shALDH5A1 construct were each performed into 3 (5706**) or 4 (TP54) mice per group. Mice were sacrificed at 64 (5706**) or 71 (TP54) days post-graft, and the numbers of GFP-expressing cells determined. The animal maintenance, handling, surveillance, and experimentation were performed in accordance with and approval from the Comité d’éthique en expérimentation animale Charles Darwin N°5 (Protocol #3113).

Statistical analyses were done with Prism 6.0 software (GraphPad) using unpaired t test with Welch’s correction, or one-sample t test when appropriate unless otherwise indicated. Significance threshold was set at p < 0.05. Mean ± SD are shown. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Metabolic rearrangements in differentiated GBM stem-like cells were investigated using unbiased global metabolomic profiling of the TG1 cell line, which was isolated from an IDH1 and 2 wild-type primary GBM (Table S1). We compared naïve cells and cells stably expressing miR-302-367 (referred to as TG1-miR) that are deprived of stem and tumorigenic properties [15], and enriched in differentiation markers (see [15] and Fig. S1). Gas chromatography/mass spectrometry (GC/MS) and liquid chromatography/MS/MS analysis of whole cell extracts and secreted culture media showed that all identified metabolic intermediates and endpoint products of energy metabolic pathways, i.e., glycolysis, tricarboxylic acid (TCA) cycle, and anaplerotic glutaminolysis were significantly reduced in TG1-miR, as exemplified by α-KG a key metabolite of the TCA cycle that can be replenished through anaplerotic reactions (Table S2).

This overall reduction in TG1 energy metabolism upon loss of their stem and tumorigenic properties was accompanied by a broad deregulation of GABA neurotransmitter metabolism (Fig. 1a, b). Decreased GABA levels were associated with increased levels of its metabolic by-products GHB, 2-hydroxyglutarate (2-HG), and 4-guanidinobutanoate (4-GDB) (Table S2). As a result, GABA by-products to α-KG ratios were increased in TG1-miR (Fig. 1a). Since GHB levels were increased in both intra- and extra-cellular compartments, we further focused on understanding the cause for the elevated production of GHB. As depicted in Fig. 1b, glutamate is the entry point of GABA synthesis pathway. It can either be converted into α-KG by glutamate dehydrogenase (or aminotransferases), or into GABA by glutamate decarboxylases (GAD67 and GAD65). GABA transaminase (GABA-T) catalyzes GABA conversion into succinic semialdehyde. The succinic semialdehyde reductase (SSA reductase/SSAR) is responsible for the synthesis of GHB from succinic semialdehyde, whereas hydroxyacid-oxoacid transhydrogenase (HOT) is responsible for GHB degradation. HOT catalyzes a concomitant reaction that additionally produces succinic semialdehyde and 2-HG from α-KG. In silico analysis identified putative miR-302-367 recognition sites in the mRNA of three enzymes of this pathway, i.e., GABA transaminase/GABA-T, hydroxy-oxoacid transhydrogenase/HOT, succinic semialdehyde dehydrogenase/SSADH (Fig. S2A). Only SSADH protein levels were markedly reduced in TG1-miR compared to TG1 (Fig. 1c, Fig. S2B). The transcript levels of ALDH5A1, which encodes SSADH, were also strongly reduced (Fig. 1d). To confirm the direct targeting of the 3′ UTR region of ALDH5A1 mRNA by miR-302, we employed TG1-miR and TG1 cells expressing a Renilla luciferase expression construct containing the ALDH5A1-3′UTR. Luciferase activity was strongly reduced in TG1-miR compared to TG1 (Fig. 1e). Deletion of miR-302 putative target sequence in the 3′UTR of ALDH5A1 mRNA (ALDH5A1-3′UTR-DEL) prevented the binding of the miR, and rescued luciferase activity (Fig. 1e). SSADH converts succinic semialdehyde into succinate, fuelling thus the TCA cycle and thereby limiting the amounts of substrate for SSAR, the enzyme responsible for GHB synthesis (Fig. 1b). Since SSAR levels were unchanged in TG1-miR (Fig. S2B), we verified whether reduced SSADH expression was sufficient to increase GHB levels. SSADH downregulation by siRNA targeting of ALDH5A1 mRNA (Fig. 1f) caused accumulation of GHB not only in stem-like cells of adult GBM but also in stem-like cells of infant DIPG, a form of high-grade glioma with a poor prognosis [58] (Fig. 1g). No change was observed in the levels of 2-HG, the other GABA by-product (Fig. S3). To further test the link between ALDH5A1 downregulation and enhanced GHB production, we used valproate to inhibit SSADH activity. Valproate has long been known to inhibit SSADH activity [54, 57], before being found to also inhibit histone deacetylases [19]. The adult GBM 6240** and DIPG TP54 cell lines were treated with 5 mM valproate. Mass spectrometry measurements showed a five- to sixfold increase in GHB intra-cellular levels in valproate-treated cells compared to control cells (Fig. S4). Altogether, these results show that decreased SSADH expression is sufficient to increase GHB levels.

Using neurosurgical samples of patient tumors, we then compared SSADH expression and GHB/α-KG ratios in GBM territories characterized by low or high proliferative and differentiation features. Coherently to our in vitro observations, we found that SSADH staining prevailed in proliferative/non-differentiated GBM territories (P^HIGH/D⁻) containing mitotic and Olig2 + tumor cells (Fig. 1h, upper panels and Fig. S5A). On the opposite, SSADH expression was scant in non-proliferative/differentiated GBM territories (P^LOW/D⁺) characterized by rare mitotic cells and lack of Olig2 expression (Fig. 1h, lower panels and Fig. S5A). Interestingly, SSADH was also detected in normal cells albeit at a lower intensity than in tumor cells of proliferative/non-differentiated GBM territories (Fig. S5B). Of note, destruction of the normal underlying axonal network distinguishing the solid tumor tissue from infiltrative peri-tumoral tissue was confirmed using neurofilament staining (Fig. S5C). In addition, GC–MS/MS analysis showed that GHB/α-KG ratios tended to be higher in these GBM territories characterized by rare mitotic cells (Fig. 1i). Altogether, these results support the pathological relevance of variations in GHB levels in the context of human GBM.

To further probe the coherence between the in vitro findings and the observations in the patient tumors, we assayed GHB effects on GBM and DIPG cells focusing first on the expression of the transcription factor Olig2 and proliferation. These assays were performed while maintaining the cells in stem-media containing the EGF and bFGF mitogens. GHB inhibited the expression of Olig2 (Fig. 2a, b). Olig2 is required for the proliferation and tumorigenicity of adult GBM cells through repression of the inhibitor of cell cycle progression p21/CDKN1A [35]. Accordingly, Western blotting showed up-regulated p21 levels in GHB-treated cells (Fig. 2c), and upon siRNA-mediated ALDH5A1 downregulation (Fig. S6A).

Inhibition of the self-renewal properties of adult GBM cells (TG1, 6240**, R633) and infant DIPG cells (TP54, TP83) accompanied these changes, as determined through the generation of secondary and tertiary spheres from single cells derived from primary spheres (Fig. 2d) or extreme limiting dilution assays (Fig. S6B). In presence of GHB, the percentage of cells yielding secondary and tertiary spheres dropped from 75–90 to 3–7, and 0–3%, respectively (Fig. 2d). In coherence with these findings, GHB inhibited proliferation of all 7 adult GBM and 4 infant DIPG cells tested. Cell numbers were reduced by 40–70% (Fig. 2e), while cell survival was unaffected (Fig. S6C and D). Cell cycle analysis of TG1 and TP54 cells showed a two- to threefold reduction in the number of cells in S phase (Fig. 2f, g), further confirming GHB inhibitory effects on cancer stem-like cell proliferation. Consistently, enhanced GHB production due to siRNA-mediated ALDH5A1 downregulation in adult GBM (TG1) and infant DIPG (TP54) stem-like cells was accompanied by reduced cell proliferation similar to GHB treatment alone (Fig. 2h). Reduced proliferation rates were also observed in cells transduced with shALDH5A1 compared to shControl (Fig. S6E). Furthermore, enhanced GHB production resulting from valproate inhibition of SSADH activity also resulted in robust inhibition of cell proliferation (Fig. S6F).

We further evaluated the relevance of our findings for the patient tumors by using independent dataset to probe links between SSADH expression and stem cell-like features that have been associated with glioma cell tumorigenicity [9]. The analysis of the TCGA transcriptome dataset of 484 untreated primary GBM was performed using the R2 Genomics Analysis and Visualization Platform (http://​r2.​amc.​nl). The analysis disclosed a correlation positive for ALDH5A1, and negative for AKR7A2 (encoding SSAR), with a majority of genes belonging to the KEGG pathway «Signaling pathways regulating pluripotency of stem cells» (Fig. 3a). 122 of the 142 genes listed in this category were detected in the dataset. 61 were significantly correlated with ALDH5A1 (48 positive/13 negative) and 38 with AKR7A2 (3 positive/35 negative). The same correlations were also found with the R2 platform’ gene category named «cancer gene census only» that contains 487 cancer-related genes (Fig S7A). Of the 371 cancer-related genes detected in the TCGA dataset, 190 were significantly correlated with ALDH5A1 (130 positive/60 negative) and 178 with AKR7A2 (23 positive/155 negative). In addition, the analysis of published transcriptome dataset of early-passage (P3) GBM cells either devoid of or endowed with self-renewing and tumor-initiating properties [32], revealed downregulated ALDH5A1, and conversely up-regulated AKR7A2 expression in non-tumorigenic cells compared to tumorigenic cells (Fig. 3b). Using serum addition as well as growth factor removal as another means to promote cell exit from the stem state, we also observed downregulated ALDH5A1 (Fig. S7B and C). These results led us to determine GHB effects on the expression of the stem transcription factor Nanog, which controls the stem-like and tumorigenic properties of GBM cells [34, 39, 59]. Immunofluorescent imaging showed a robust decrease in nuclear Nanog signal in GHB-treated GBM cells with either mesenchymal (TG1, TG16), or classical profiles (JolMa, 6240**) as well as in infant DIPG cells with pro-neural profiles (TP54, TP80) (Fig. 3c, d). Similar to GHB treatment, siRNA-mediated ALDH5A1 downregulation in TG1 (adult GBM) and TP54 (infant DIPG) cells decreased Nanog nuclear localization (Fig. 3e). Decreased Nanog expression was confirmed by Western blotting (Fig. 3f, TG1), and FACS analysis (Fig. 3g, h, TG1, R633, TP83). We further found that GHB stimulated GBM stem-like cell adhesion on permissive substrates with formation of membrane extensions (Fig. 3i), and increased numbers of cells expressing the differentiation markers ß3-Tubulin and MAP2, or GFAP and EAAT2/GLT1 upon transfer in pro-neuronal or pro-astroglial differentiation media, respectively (Fig. S7D). In a coherent manner, ALDH5A1 downregulation with shRNA, or valproate inhibition of SSADH activity were also accompanied with increased expression of differentiation markers (Fig. S7E and F).

Altogether, these results showed that SSADH-driven GHB accumulation repressed proliferation, clonality and stem-like features of malignant glioma cells. We therefore determined whether these repressive effects affected in vivo tumor development using orthotopic xenografts of GBM (6240**, 5706**) and DIPG (TP54) stem-like cells stably expressing either shControl or shALDH5A1 (Fig. S8). GBM stem-like cells 6240**, which stably express luciferase, were used for bioluminescent imaging. Results showed reduced tumor burden in mice grafted with 6240**-shALDH5A1 compared to mice grafted with 6240**-shControl cells (Fig. 3j). In addition, a significant improvement in survival of the mice grafted with 6240**-shALDH5A1 cells was observed (Kaplan–Meier analysis, Fig. S9A). Decreased tumor burden was also observed with grafts of 5706**-shALDH5A1 and TP54-shALDH5A1, as shown by cell counting (Figs. 3k, S9B).

GHB is naturally present in the brain at low concentrations and acts as an inhibitory neuromodulator through the GABA receptor GABA_BR [12], and a GABA_AR sub-class [1]. Using TG1 cells, we found that agonists of GABA_AR (Isoguvacine hydrochloride) or GABA_BR (Baclofen) did not modify GBM cell proliferation whereas GABA_AR or GABA_BR antagonists (Gabazine and Hydroxysaclofen) did not prevent GHB inhibition of cell proliferation (Fig. S10). GHB action being independent from GABA receptors, we investigated an intracellular mode of action.

GC–MS/MS analysis showed that GHB treatment led to increase in its intra-cellular levels (Fig. 4a). GHB penetration inside the cells, and the similarity of its chemical structure to that of α-KG (Fig. 4b), suggested that GHB could target α-KG-dependent enzymes such as Ten–eleven Translocations (TETs). These nuclear enzymes are needed for ESC maintenance in the stem state, and for reprogramming somatic cells into induced pluripotent stem cell (iPSC) [11, 14, 17]. They notably act by oxidizing 5-methylcytosine (5-mC) into 5-hmC [17], which has been shown recently to recruit a methylosome complex needed for glioblastomagenesis [51].

The evaluation of mRNA levels showed TET2 to be the most abundant of the three TET isoforms in the cells studied here (Fig. S11A). Dot immunoblotting of DNA extracts showed a marked reduction in 5-hmC levels in GHB-treated GBM (TG1) as in DIPG stem-like cells (TP54) compared to control conditions, whereas 5-mC levels were unchanged (Fig. 4c, d). Similar results were obtained using siALDH5A1-transfected TG1 and TP54 cells (Fig. 4e). Compared to TG1, we also found reduced 5-hmC levels in TG1-miR (Fig. S11B) and in TG1 treated with serum (Fig. S11C), a well-known inducer of GBM cell differentiation into less tumorigenic cells. Notably, GHB did not alter 5-hmC levels in human fetal neural stem cells (NSC) (Fig. S11D).

Since GHB did not inhibit expression of TET2, or of the other TET isoforms (Fig. 5a), we pursued the study of GHB effect on TET2 enzymatic activity. To further confirm GHB inhibition of TET2 activity, we first used a UT7 leukemic cell line stably overexpressing TET2 in a doxycycline-dependent manner (Fig. 5b). In UT7 cells, 5-hmC was only detected upon induction of TET2 expression, and GHB strongly reduced 5-hmC levels (Fig. 5c). The same results were obtained with TG1 cells engineered to express either a doxycycline-inducible wild-type or a catalytically defective TET2 (Fig. 5d). Enhanced 5-hmC formation was only observed upon induction of wild-type TET2 expression, and was counteracted in presence of GHB (Fig. 5e). The analysis of TET2 structure confirmed the presence of a GHB binding pocket within TET2. As previously mentioned, α-KG is a co-factor required for TET2 oxidative activity. The crystal structure of TET2 has shown localization of α-KG binding site in the C-terminal domain of the protein [24]. To identify the site of GHB binding pocket in TET2, we performed in silico docking analysis on the GalaxyWEB platform (http://​galaxy.​seoklab.​org/​softwares/​galaxydock.​html [22]). We first confirmed the existence of a unique binding pocket for α-KG, formed by 11 amino acids located as expected in TET2 C-terminal domain (Fig. 5f). A unique binding pocket was also found for GHB (Fig. 5g), almost identical to the α-KG binding pocket: of the 11 amino acids forming the α-KG binding pocket, only 153A was not included in the GHB binding pocket (Fig. 5h). Calculated docking energies were similar for GHB and α-KG (−6.228 and −8.761 kcal/mole, respectively).

Finally, to ascertain further the link between TETs and GHB, we used siRNA to downregulate TET2 (Fig. 6a). TG1 and TP54 cells were used as examples of GBM and DIPG cells, since GHB had the same effects on all the cells we had examined. TET2 downregulation had effects comparable to GHB treatment or ALDH5A1 downregulation, i.e., reduced 5-hmC levels without significant change in 5-mC levels (Fig. 6b), decreased Nanog nuclear signals (Fig. 6c) and reduced cell proliferation (Fig. 6d).

Altogether, these results showed that GHB decreases 5-hmC levels by antagonizing TET activity, and that TET2 participates in the mechanisms by which GHB induces GBM and DIPG cell differentiation into less aggressive cells (Fig. 6e).