Cells were mycoplasma tested and authenticated. Human glioma cell lines, U87 (RRID:CVCL_0022; https://scicrunch.org/resolver/CVCL_0022 and https://web.expasy.org/cellosaurus/CVCL_0022) and U-251MG (RRID:CVCL_2219 https://scicrunch.org/resolver/RRID:CVCL_2219 and https://web.expasy.org/cellosaurus/CVCL_0021) (ATCC), were grown in DMEM (Cat# 21969035, Gibco,) supplemented with 10% FBS. Patient-derived glioblastoma cell lines previously isolated from patient samples, as described in,¹⁶ were cultured in serum-free Neurobasal Medium (Cat# 12349015, Gibco) supplemented with B27 (Cat#17504044, Gibco), N2 (Cat# A1370701, ThermoFisher Scientific), 20 ng/mL EGF (Cat# E9644, Sigma), 20 ng/mL FGF (Cat # PHG0261, Gibco), 2 mM glutamine and Penicillin-Streptomycin (100 U/ml) (Cat# 15070063, Gibco) in flasks precoated with Extracellular Matrix (Cat# E1270, Sigma). For hypoxia ­experiments, cells were incubated in a Tri-Gas incubator (1% O₂, 5% CO₂) or in a hypoxia chamber (0.1% or 1% O_2, 5% CO₂).

Procedures were performed in compliance with licenses issued under the United Kingdom Animals Scientific Procedures Act, 1986 and approved by an ethical review body. Cells were implanted orthotopically¹⁶ in 12-week-old (20 g) BALB/c nude mice and 12-week-old (200 g) RNU rats (Charles River,). Randomization was achieved by implanting the different cell lines in mice or rats coming from the same cage. Tumor growth was monitored using MRI (Supplementary Information). For doxycycline treatment, animals were fed a diet containing 200 ppm doxycycline (Cat# TD.180625, Envigo) for 10 days. At least 3 animals were implanted for each condition and samples were only excluded in those cases where the animals had to be culled due to development of symptoms prior to the endpoint. Investigators were not blinded to group allocations.

Protein extracts were prepared from cells and homogenized tissue using cold Pierce RIPA buffer (Cat#89901, Thermo Fisher Scientific, Waltham,) containing EDTA-free Protease Inhibitor Cocktail (Roche, Basel, Switzerland). Samples were run on 4%–12% Bis-Tris precast polyacrylamide gels (NuPAGE) and transferred onto a nitrocellulose membrane using dry (iBlot 2 Dry Blotting System [Thermofisher]) or wet (NuPAGE) transfer. Primary antibody diluents (Supplementary Information) were prepared in Odyssey Blocking Buffer (Licor) with 0.2% Tween-20 or in 5% Bovine Serum Albumin in Tris-Buffered Saline with 0.1% Tween-20.

Rat brains were dissected and fixed in 10% neutral buffered formalin for 24 hours and transferred to 70% ethanol for processing into paraffin blocks. The antibodies used are shown in Supplementary Information and stained sections were imaged using Leica’s Polymer Refine Detection System on the Bond-III platform.

Cells were lysed in cold lysis buffer (50 mM HEPES, 1 mM EDTA, 0.7 % sodium deoxycholate, 1 % Nonidet P-40, 0.5 M lithium chloride, pH 7.6 with EDTA-free Protease Inhibitor Cocktail (Roche)) and assayed as described in.¹⁷ The reaction (5 mM leucine, 5 mM α-KG, 5 mM ammonium sulphate, 0.05 mM NADH, 0.5 mM GTP, 1 mM DTT, 1.9 U leucine dehydrogenase and 5–20 μL of cell extract in 100 mM potassium phosphate buffer (pH 7.4) in a final volume of 200 μL) was conducted in a UV-transparent 96-well plate (Corning) at 37°C and 340 nm absorbance measured in a Clariostar microplate reader (BMG Labtech). The BCAT1 inhibitor, gabapentin (Cat# PHR1049, Sigma-Aldrich), was used to determine the contributions of BCAT1 and BCAT2 to total activity.

RNA was extracted using a Qiagen RNA isolation kit (Rneasy Mini Kit, Cat#74104) and quantified using a Qubit RNA BR Assay Kit (Cat# Q10210 ThermoFisher Scientific). Reverse transcription was performed using M-MuLV Reverse Transcriptase (NEB) and quantitative PCR was performed using Fast SYBR Green master mix (Cat#4385610, Applied Biosystems). QuantStudio software (Applied Biosystems) was used for analysis.

An shRNA sequence targeting BCAT1 (shBCAT1) or a control sequence (shScr) (Supplementary Information) were cloned into the pLKO.1-Tet-OnTet-pLKO-puro lentiviral plasmid¹⁸ (Addgene plasmid # 21915; http://n2t.net/addgene:21915; RRID:Addgene_21915; https://scicrunch.org/scicrunch/resolver/RRID:Addgene_21915; http://www.addgene.org/21915). Plasmids psPAX2 (7.5 μg) and D2G (2.5 μg) (Addgene plasmid # 12260; https://scicrunch.org/resolver/Addgene_12260/; http://n2t.net/addgene:12260; RRID:Addgene_12260 http://www.addgene.org/12260) and Addgene plasmid # 12259; http://n2t.net/addgene:12259; RRID:Addgene_12259 http://www.addgene.org/12259; https://scicrunch.org/resolver/Addgene_12259/) and 10 μg of the lentiviral plasmid were mixed with 500 μL serum-free DMEM and added to 80 μL Lipofectamine 3000 Transfection Reagent in 500 μL serum-free DMEM and the solution was added to HEK293 cells (ThermoFisher Scientific). Cell medium was filtered through a 0.45 μm filter and 1 mL of this was used to infect glioblastoma cells with 8 μL of the 5 mg/mL polybrene solution. Infected cells were selected with 2 μg/mL puromycin (Gibco). Fifty ng/ml Doxycycline Hyclate (Sigma) was used to induce knockdown.

The coding regions of human BCAT1 or luciferase were subcloned from a GeneArt plasmid (Thermo Fisher Scientific) into a pBOBI plasmid (gift from Verma laboratory, Salk Institute La Jolla), which contains an EF1 promoter used to drive the expression of mStrawberry and BCAT1 or mStrawberry and luciferase, respectively. Lentiviruses were produced following transfection of HEK293 cells with the plasmids as described above using pMDL packaging plasmid (Addgene #12251; RRID:Addgene_12251 (https://scicrunch.org/resolver/Addgene_12251/mentions?q=&i=rrid:addgene_12251 and http://www.addgene.org/12251), pCMV-VSV-G envelope vector (Addgene #8454; RRID:Addgene_8454 https://scicrunch.org/resolver/Addgene_8454 and http://www.addgene.org/8454) and pRSV-Rev (RRID: Addgene #12253; https://scicrunch.org/resolver/Addgene_12253/ and http://www.addgene.org/12253) and were used for infection of A11 and S2 cells. Infected cells were FACS sorted and expanded to form BCAT1 overexpressing cell lines or luciferase-overexpressing control cell lines.

Percentage confluence over time was measured in an Incucyte system (Sartorius, Göttingen, Germany). Cell proliferation was also measured using a luciferase-based assay (Cat#G9711, RealTime Glo, Promega).

Cells were seeded in Ultra Low attachment 96-well plates (Corning) (4000 cells/well) and an invasion assay was executed as described previously.¹⁹ Images were acquired with an Incucyte microscope (Sartorius, Göttingen, Germany) and analyzed using a Python script (Supplementary Information).

Cells (10⁶) were fixed in 4.5 mL ice-cold 70% ethanol, centrifuged (5 minutes, 300 g), and washed with PBS. The pellets were resuspended in 600 μL of Propidium Iodide Solution (0.1% Triton X, 2 mg RNAse, 20 mg/mL Propidium Iodide in PBS) and incubated at 37°C for 15 minutes with gentle shaking and then filtered and analyzed on an LSRII (BD) flow cytometer. FlowJo software was used for data analysis.

Cell oxygen consumption rates were measured using a Seahorse XF96 Extracellular Flux Analyzer (Agilent).

A11 cells expressing doxycycline-inducible shScr and shBCAT1 were treated with doxycycline for 6 days before washing and staining with 5 μg/mL DAPI. Live cells were sorted on an Influx cell sorter (BD Biosciences, New Jersey). 1, 5, 10, or 20 cells were seeded per well in a 96-well plate containing complete Neurobasal Medium and incubated for 3–4 weeks. The number of tumor sphere-containing wells per group was counted and data were analyzed using Extreme Limiting Dilution Analysis software.²⁰

Cells were treated with doxycycline or vehicle for 7 days before RNA extraction using a RNeasy Mini Kit (Cat#74104, Qiagen). For tumor samples, the whole brain was frozen in liquid nitrogen and the tumor was extracted using a cryostat and surgical scalpel cooled with dry ice. The tissue was homogenized and RNA was extracted using a RNeasy Mini Kit (Qiagen). RNA was quantified using an Agilent 4200 TapeStation. For library preparation, the Illumina Truseq stranded mRNA kit was used, and single-read sequencing was performed on a HiSeq 4000 machine (Illumina). Data analysis were performed as described in Supplementary Information.

Cells were washed with ice-cold PBS and incubated on dry ice with 500 μL extraction solution (50% Methanol, 30% Acetonitrile, 20% Ultrapure Water, d8-Valine at a final concentration of 5 μM) per million cells added to each well for 15 minutes. Cells were scraped from the wells, agitated (15 minutes, 4°C), incubated for 1 hour at −20°C, vortexed, and centrifuged (21 000 g, 10 minutes, 4°C). Liquid Chromatography Mass Spectrometry (LCMS) was used to quantify the relative concentrations of each metabolite of interest and the absolute αKG concentration, as described in Supplementary Information.

DNA was extracted using a Purelink Genomic DNA Mini Kit (Invitrogen). DNA (1 μg) was incubated with 5 units DNA Degradase Plus (Cat# E2020, Zymo Research) (4 hours, 37°C) and samples were analyzed by LCMS as described in Supplementary Information.

A11 tumor-bearing mice were infused via a tail vein cannula with a bolus of 300 mg [2-¹³C,¹⁵N]leucine/g body weight and then a continuous infusion of 0.0069 mg/g body weight min⁻¹ for 150 minutes.²¹ At the end of the infusion, mice were sacrificed by cervical dislocation followed by blood collection in EDTA-coated tubes and dissection and flash freezing of the tumor in liquid nitrogen. Blood samples were centrifuged (2000 g, 20 minutes) to collect the plasma. Labeled leucine enrichment was measured in plasma and tumor extracts by LCMS, as described in Supplementary Information.

IDH wild-type patient-derived (A11, S2, SP20, and A25) and established human (U87 and U251) and rat (C6) glioblastoma cell lines showed variable concentrations of BCAT1 and BCAT2 (Figure 1B–G; Supplementary Figure S1 A–D). SP20 had very low concentrations of BCAT1 with S2 cells showing no detectable protein (Figure 1B) on western blot. When grown as neurospheres S2 cells still showed no detectable BCAT1 protein (Figure 1C). The same protein expression patterns were observed in rat orthotopic xenografts derived from A11, SP20, and S2 cells. Immunohistochemistry showed the highest BCAT1 staining in A11 xenografts with lower staining in SP20 and minimal staining in S2 xenografts (Figure 1D). These expression patterns were confirmed by western blot for the A11 and S2 xenografts (Figure 1C). There was an inverse relationship between BCAT1 and BCAT2 protein concentrations (Figure 1E–G), which was also reflected in the relative activities of the enzymes (Supplementary Figure S1A–C). A11 cells had higher BCAT1 than BCAT2 activity whereas in SP20 and S2 cells the predominant BCAT activity was due to BCAT2 (Supplementary Figure S1A–C). BCAT2 transcript levels were much higher than for BCAT1 with BCAT1 mRNA being undetectable in S2 cells (Supplementary Figure S1D). Examination of the TCGA dataset for IDH wild-type glioblastoma also showed a wide variation in the expression levels of BCAT1, with A11 representative of tumors with high expression and S2 tumors with low expression (Figure 1H).

A11, S2, and SP20 patient-derived cells have been described previously¹⁶ (referred to previously as GB4, GB1, and GB2, respectively) and were shown to recapitulate the biology of the patient tumors from which they were derived. Following orthotopic implantation, they all showed high expression of the glial cell marker GFAP. Exome sequencing showed PTEN frameshift mutations in S2 and SP20 tumors, resulting in protein loss and a PIK3R1 (V73fs) frameshift mutation in A11, that could lead to activation of PI3K. All the models showed activation of the PI3K/Akt pathway, as indicated by phosphorylation of Akt.¹⁶ S2 also harbors TP53 and RB1 mutations that are not found in A11 (Supplementary Figure S1E). Epithelial-to-mesenchymal transition has been shown to drive BCAT1 expression²² and RNAseq profiles showed that A11 cells have more mesenchymal features whereas S2 cells appear more neural progenitor cell-like (Supplementary Figure S1F).²³ A11 cells are more glycolytic in vitro than S2 cells (Supplementary Figure S1G–H) and previous ¹³C MRI studies of hyperpolarized [1-¹³C]pyruvate metabolism showed higher lactate labeling in A11 tumors than in S2 tumors.¹⁶ A11 tumors also showed higher levels of expression of the glycolytic enzymes, lactate dehydrogenase A and hexokinase 2 and of the monocarboxylate transporters MCT1 and MCT4.¹⁶

Treatment of A11, SP20, S2, and U87 cells with increasing concentrations of gabapentin, a selective BCAT1 inhibitor,²⁴ reduced proliferation of A11, SP20, and U87 cells, and changed A11 and SP20 cell morphology, but had no effect on S2 cell proliferation or morphology (Figure 2A–C, Supplementary Figure S2A). The effects of gabapentin were phenocopied by shRNA-mediated knockdown of BCAT1 expression. Knockdown of BCAT1 using a doxycycline-inducible shRNA (shBCAT1) (Figure 2D–G; Supplementary Figure S3 A,D) decreased proliferation of A11 cells but had a smaller effect on the proliferation of SP20 cells (Figure 2H). Knockdown of BCAT2 in A11 or S2 cells (Supplementary Figure S3B) had no effect on cell proliferation (Figure 2H). Knockdown of BCAT1 in A11 cells resulted in partial cell cycle arrest in G1 phase (Figure 2I; Supplementary Figure S3E) and a more rounded morphology (Figure 2Jj), similar to that observed with gabapentin (Figure 2C). There was no effect on cell viability in A11 or SP20 cells upon BCAT1 knockdown (Supplementary Figure S3C). Overexpression of BCAT1 (Figure 2K–Ll; Supplementary Figure S3F) increased cell proliferation and invasion in A11 cells but inhibited proliferation in S2 cells and had no effect on invasion (Figure 2M–N; Supplementary Figure S3G–I). Overexpression of c-Myc in S2 cells (S2myc), which increased BCAT1 expression (Figure 3H–I) and increased their proliferation (Supplementary Figure S2B), conferred sensitivity to gabapentin inhibition of proliferation (Figure 2A), which now also changed S2 cell morphology (Supplementary Figure S2A).

BCAT1 expression is driven by hypoxia²⁵ and c-Myc.^26,27 Incubation of cells in 1% or 0.1% O₂ increased BCAT1 protein concentration in A11, SP20 and U251 cells (Figure 3A–E) but not in S2 cells (Figure 3G) and had no effect on BCAT2 expression in A11 and SP20 cells (Figure 3F). Overexpression of c-Myc in S2 cells, which had low levels of c-Myc (Figure 3H–I), markedly increased the concentration of BCAT1 (Figure 3J,K) but had no effect on the concentration of BCAT2 (Figure 3L). Increased expression of BCAT1 in hypoxic A11 cells occurred despite a decrease in c-Myc concentration (Figure 3M). In summary, A11, SP20, and U251 cells show upregulation of BCAT1 expression in hypoxia whereas S2 cells display regulation of BCAT1 expression by c-Myc but not by hypoxia. Whole genome sequencing of A11, SP20, and S2 cells performed previously¹⁶ showed no differences in the coding, promoter, or enhancer regions of the BCAT1 gene with the exception of an intronic variant/mutation in the promoter region in 2 A11 samples.

BCAT1 knockdown in A11 cells changed their transcriptional profile (Supplementary Figure S4), with the expression of cyclins and cell cycle checkpoint proteins being downregulated, consistent with partial cell cycle arrest (Figure 2I). HIF transcriptional targets were also significantly downregulated (Figure 4A–B, Supplementary Figure S4). HIF-1α protein, and the product of a target gene, carbonic anhydrase IX (CAIX),²⁸ were significantly downregulated following BCAT1 knockdown in A11 and SP20 cells (Figure 4C–H). HIF-1α, and the products of its target genes, CAIX and hexokinase II (HKII),²⁹ were also significantly downregulated following doxycycline-induced BCAT1 knockdown in orthotopically implanted A11 xenografts (Figure 4I–J). Conversely, BCAT1 overexpression in A11 and S2 cells increased the concentrations of HIF-1α and CAIX in A11 cells and CAIX in S2 cells (Figure 4K–L). Following BCAT1 knockdown, A11 cells were incubated in 1% O₂ for 8 hours and then extracted either immediately or following 10 minutes of incubation in normoxic conditions. BCAT1 knockdown increased the degradation of HIF-1α, indicating that it reduces HIF-1α stability (Figure 4M). Immunohistochemistry (Supplementary Figure S5) showed that A11 tumor BCAT1 expression was co-localized with the expression of the monocarboxylate transporters MCT1 and MCT4, which are also HIF transcriptional targets,^30,31 and with CAIX.

The effect of increased BCAT1 expression on glioblastoma cell proliferation was suggested to result from increased production of BCKAs, from the corresponding BCAAs, and their oxidation in the TCA cycle.⁷ However, inhibition of A11 cell proliferation by BCAT1 knockdown was not relieved by addition of BCKAs (Figure 5A). Moreover, in BCAT1 knockdown cells the basal oxygen consumption rate (Figure 5B) and the rate following addition of the mitochondrial uncoupler FCCP (Supplementary Figure S6A) were not significantly different from control cells, suggesting that BCAT1 knockdown had not reduced flux through the TCA cycle. Infusion of mice bearing orthotopically implanted A11 tumors with [2-¹³C,¹⁵N]leucine showed minimal incorporation of ¹³C into the TCA cycle intermediates succinate, malate, α-KG, and acetyl-CoA, suggesting that BCAAs are not an important carbon source for the TCA cycle in these tumors (Figure 5C).

Knockdown of BCAT1 in A11 cells resulted in higher concentrations of TCA cycle intermediates, suggesting that BCAT1 knockdown not only does not limit flux into the TCA cycle but actually increases it (Supplementary Figure S6B–E), consistent with downregulation of HIF-1α expression. BCAT1 knockdown also resulted in higher concentrations of glutamate and glutathione (Supplementary Figure S6F–G), contrary to what was observed in an immortalized human astrocyte cell line,³² and again consistent with increased flux in the TCA cycle.³³

BCAT1 expression in CML increases with disease progression and drives the production of BCAAs by aminating the corresponding BCKAs. Knockdown or inhibition of BCAT1 in these cells reduced phosphorylation of S6 kinase, suggesting that the increase in BCAAs, particularly leucine, drives disease progression through activation of mTORC1.¹⁵ However, there was no evidence that this mechanism drives proliferation of A11 cells. Knockdown of BCAT1 in A11 cells (Supplementary Figure S6H) produced no change in BCAA concentrations (Figure 5D) or in the ratio of phosphorylated to non-phosphorylated S6 (Figure 5E). BCAT1 knockdown also had no effect on the ratio of phosphorylated to non-phosphorylated S6 in orthotopic A11 tumors in mice (Figure 5F). S2 cells overexpressing BCAT1 showed no differences in S6 phosphorylation when compared to S2 control cells (Supplementary Figure S6I).

In AML stem cells BCAT1 upregulation lowers α-KG concentration, resulting in reduced activity of α-KG-dependent dioxygenases, including Ten Eleven Translocation (TET) enzymes and Egl-9 Hypoxia Inducible Factor 1 (Egln1), leading to a hypermethylated DNA state, similar to that in IDH mutant cells, and stabilization of HIF-1α, respectively.⁵ Treatment of A11 cells with dimethyl α-KG, a cell-permeable analog of α-KG,^34–36 mimicked the effects of BCAT1 knockdown on cell proliferation (Figure 5G). Conversely, supplementation with dimethyl oxalyl glycine (DMOG), a competitive inhibitor of α-KG-dependent dioxygenases,^37,38 restored the proliferation rate of BCAT1 knockdown cells to control levels (Figure 5H). Supplementation of the growth medium with α-KG reduced HIF-1α levels, confirming the role of α-KG in destabilizing HIF-1α in these cells (Figure 5I,J). BCAT1 knockdown increased the levels of 5-hydroxymethylcytosine in DNA, indicating activation of the TET enzymes and supporting the proposal that BCAT1 knockdown leads to the accumulation of α-KG and increased activity of α-KG-dependent dioxygenases (Figure 5K). There was a small increase in α-KG concentration following doxycycline induction of shBCAT1 expression. However, this may be an underestimate of the effect of BCAT1 knockdown since the addition of doxycycline to cells expressing a control shRNA (shScr) depressed the α-KG concentration (Figure 5L). Moreover, the relative change in the cytosolic concentration could be much larger depending on the contribution of α-KG in the mitochondria.

In summary, the effects of changes in BCAT1 activity on A11 cell proliferation are mediated via changes in α-KG concentration. Decreases in BCAT1 activity raise α-KG concentration, leading to increased prolyl hydroxylase activity and HIF-1α destabilization, resulting in inhibition of cell proliferation and invasion. The increase in α-KG concentration also leads to an increase in TET activity resulting in DNA demethylation, as reflected in the increased levels of 5hmC (Figure 5M).

Expression of Forkhead box protein M1 (FOXM1), a downstream target of HIF-1α³⁹ that has been shown to play multiple roles in glioblastoma,^40–44 was downregulated following BCAT1 knockdown (Figure 6A–B) and this was accompanied by downregulated expression of its target genes (Figure 6C–D). In addition to promoting cell proliferation, FOXM1 has been implicated in maintaining a stem-like phenotype and the tumorigenicity of glioblastoma cells.^40,43,45,46 BCAT1 knockdown in A11 cells reduced their capacity to form tumor spheroids and resulted in a significantly reduced frequency of cells with self-renewal capacity (Figure 6E–G).