Brain Tumor Stem Cell Dependence on Glutaminase Reveals a Metabolic Vulnerability through the Amino Acid Deprivation Response Pathway

In recent years, there has been a resurgence in cancer metabolism research, driven in part by the discovery of an oncometabolite (2-hydroxyglutarate), the observed rewiring of metabolic pathways by oncogenes and tumor suppressors, and the utilization of metabolites other than glucose for energetic and biosynthetic demands (reviewed in ref. 1). A further characterization of the metabolic requirements of proliferating cancer cells has led to the development of novel cancer therapeutics and an improved understanding of the efficacy of some of the earliest cancer therapies, such as antifolates (2, 3). It has long been known that most mammalian cells require glutamine for proliferation (4). In cancer, cells from multiple tumor types have been described as glutamine-addicted (reviewed in ref. 5). Glutaminolysis refers to the hydrolysis of the amino group from glutamine to produce glutamate and ammonium (Fig. 1A). Glutaminolysis is utilized in cancer: as a source of α-ketoglutarate to replenish tricarboxylic acid (TCA) cycle intermediates (anaplerosis) for ATP generation and the synthesis of macromolecules, to generate reactive oxygen species (ROS) scavengers, and to produce the substrates required for multiple epigenetic reactions (1, 6).

Glutaminolysis has also been shown to be required for the maintenance of pluripotency in embryonic stem cells (ESC) and survival of human induced pluripotent stem cells (iPSC; refs. 7, 8). In the context of the normal brain, neural stem cells (NSC) take up and metabolize glutamine to replenish TCA cycle intermediates and macromolecules, whereas differentiated astrocytes take up glutamate then synthesize and secrete glutamine (9). The uptake of glutamate, the major excitatory neurotransmitter, by astrocytes at the neuronal synapse is essential to maintain normal brain function (reviewed in ref. 10). In glioblastoma (GBM), the most common and aggressive primary malignant brain tumor in adults, glutaminolysis is utilized by proliferating GBM cells to support anaplerosis in vitro (11) and GBM tumors have been observed to take up more glutamine than the surrounding normal brain tissue in vivo (12). A population of cells that exhibit cancer stem cell (CSC) properties, termed brain tumor stem cells (BTSC), are hypothesized to be the source of therapeutic resistance, recurrence, and lethality in GBM (13, 14). BTSCs share the biological properties of asymmetric cell division, the ability to undergo quiescence, and gene expression patterns with their normal NSC counterparts (13–15), while also exhibiting the genetic alterations of GBM in common signaling pathways that are known to reprogram cell metabolism (1, 6). We hypothesized that the shared biological properties between NSCs and BTSCs extend to the requirement of glutaminolysis for intermediary metabolism and that inhibiting glutaminolysis would allow us to identify metabolic dependencies in BTSCs.

Here, we show that BTSC lines with low expression of the astrocytic glutamate transporters, termed excitatory amino acid transporters (EAAT), EAAT1 and EAAT2 are dependent on glutaminolysis for cell growth and viability. Targeting glutaminase (GLS) in these sensitive BTSC lines depletes intracellular glutamate, triggering the amino acid deprivation response (AADR) pathway and cell death. The AADR pathway is a stress response pathway triggered by the depletion of intracellular amino acids, resulting in a compensatory increase in the expression of genes involved in amino acid synthesis (16). BTSCs with high astrocytic gene expression also have high EAAT1/EAAT2 expression and these resistant BTSC lines can be sensitized to GLS inhibition by blocking glutamate transport, also triggering the AADR and decreasing cell growth. Overall, we reveal that BTSCs have a unique metabolic dependence on intracellular glutamate to maintain amino acid levels in order to fuel BTSC growth.

BTSC lines were established from GBM surgical procedures after obtaining informed written consent from patients and cultured as described previously (17). Previous characterization of BTSC lines has been performed using stem cell biology assays and detailed analysis of genomic, transcriptomic, and epigenomic data (15, 18). BTSC lines were used in the described experiments within 10 passages after being thawed. Mycoplasma testing was performed using the Universal Mycoplasma Detection Kit (ATCC, cat.# 30–1012K) using the manufacturer's protocol, and the latest date for testing was July 14, 2020. BTSC lines were authenticated by performing short tandem repeat profiling and comparing their profile with the original parental tumor tissue. All experimental procedures were performed in accordance the Health Research Ethics Board of Alberta, Cancer Committee. HF-NSCs were established from human fetal tissue and were cultured and passaged under identical conditions and using the same materials as the BTSC lines. Additional information is available in the Supplementary Materials and Methods.

BTSCs were plated in 96-well plates from 500 to 3,000 per well (depending on the BTSC line used) and 24 hours later, the drugs or compounds were added with equal volumes of the vehicle normalized across all conditions. After the desired treatment time, alamarBlue reagent (Thermo Fisher Scientific, cat.# DAL1025) was added according to the manufacturer's protocol for 6 hours and fluorescence was measured (excitation 540 nm and emission 590 nm).

BTSCs were plated in 96-well plates in a 10 × 6 grid with 10 columns of 1:2 serial dilutions from 512 cells to 1 cell per well (6 replicate wells per cell number) in a final media volume of 100 μL. LDAs were plated in the presence of CB-839 or DMSO and three biological replicates were performed. Experiments were scored 21 days after plating by counting the number of wells per condition, out of six, that formed at least one sphere with a minimum diameter of 50 μm. The sphere forming frequency as a percentage [1/(number of cells needed to form a sphere)*100] and associated statistics were calculated using the Extreme Limiting Dilution Analysis webtool (19).

Microscopic images of BTSC spheres in culture were obtained using the IncuCyte ZOOM live-cell imaging instrument (Essen BioScience) and were acquired and exported using the IncuCyte ZOOM controller software (Essen BioScience, ver. 2016B).

Following a 12-day treatment of with CB-839 or DMSO, BTSCs were incubated for 4 hours with 10 nmol/L Cytotox green reagent (Essen BioScience, cat.# 4633). Four images per well in triplicate wells were acquired and analyzed using the IncuCyte ZOOM controller software. Three biological replicates were performed on separate occasions.

BTSCs were cultured for 5 days after plating then treated with CB-839 (1 μmol/L) or an equal volume of DMSO for 48 hours. Untreated cell samples and media only control tubes were also collected. Three technical replicates were collected for each condition. Metabolomics extraction and analysis was performed at the CHUM Research Centre metabolomics core facility, affiliated with the Université de Montréal as described previously (20). Additional information is available in the Supplementary Materials and Methods.

The BTSC lines BT89 and BT67 were treated with CB-839 or DMSO for 48 hours and pellets of three biological replicates were collected, processed, and sequenced as described previously (21). Gene-level expression data were obtained by collapsing transcript-level data as the sum of transcripts using gene set enrichment analysis (GSEA) software (Broad Institute, ver. 3.0). GSEA software was used to determine the effect of CB-839 treatment on the “krige_amino_acid_deprivation” gene set, from the Molecular Signatures Database (Broad Institute, http://software.broadinstitute.org/gsea/msigdb/, version 7.0), in BT89 and BT67 cells. Gene sets used for ATF4 regulated genes and ER stress response genes were derived from the “reactome_activation_of_genes_by_ATF4” and “GO_ER stress” gene sets from the Molecular Signatures Database. RNA sequencing (RNA-seq) analysis of the 57 BTSC lines was performed on previously published RNA-seq data from Shen and colleagues (18). Additional information is available in the Supplementary Materials and Methods.

RNA was extracted and purified from cell pellets using the RNeasy Kit (Qiagen, cat.# 74104) and reverse transcribed to cDNA using the qScript cDNA Synthesis Kit (Quantabio, cat.# 95047). The cDNA was added to FastStart Essential DNA Green Master mix (Roche, cat.# 06402712001) with 0.5 μmol/L forward and reverse primers. qPCR was performed in triplicate wells using the LightCycler 96 Instrument I (Roche). Using the LightCycler 96 software (Roche, ver. 1.1.0.1320), the relative expression values were obtained and normalized to two control genes: actin beta (ACTB) and beta-2-microglobulin (B2M). XBP1 primers for both the spliced and total length forms were derived from van Galen and colleagues (22). The primers used for qPCR analysis are summarized in Supplementary Table S1.

Western blotting was performed as described previously (21), with the exception that cell pellets for EAAT1 and EAAT2 blots were lysed with 0% and 5% SDS, respectively. Blots were probed using the following antibodies: CHOP (Abcam, cat.# ab11419, RRID:AB_298023), BIM (Cell Signaling Technology, cat.# 2933, RRID:AB_1030947), cleaved caspase-9 (specific to the cleaved form, Cell Signaling Technology, cat.# 20750, RRID:AB_2798848), ASNS (Abcam, cat.# ab40850, RRID:AB_722917), TRIB3 (Abcam, cat.# ab75846, RRID:AB_1310768), ATF4 (Cell Signaling Technology, cat.# 11815, RRID:AB_2616025), EAAT1 (Santa Cruz Biotechnology, cat.# sc-515839), and EAAT2 (Santa Cruz Biotechnology, cat.# sc-365634, RRID:AB_10844832). β-Tubulin (Cell Signaling Technology, cat.# 2146, RRID:AB_2210545), Nucleolin (Abcam, cat.# ab13541, RRID:AB_300442), or actin (Santa Cruz Biotechnology, cat.# sc-1615, RRID:AB_630835) were used as loading controls depending on individual Western blot conditions.

The astrocyte gene signature was developed using the data from two previously published scRNA-seq studies of normal human brain tissue (23, 24), which identified differentially expressed genes across different cell types in the brain. We combined the gene lists for astrocytic genes by including genes that were identified in both studies as having higher expression (log₂-fold change greater than 1.0) in astrocytes. EAAT1 and EAAT2 were removed from the signature, resulting in a 176 gene astrocyte gene signature (Supplementary Table S2). Similarly, we derived an OPC gene signature (Supplementary Table S3) using genes that were more highly expressed in OPCs in both studies. The early RGC gene signature (Supplementary Table S4) was derived from the Nowakowski and colleagues study, using the top 200 genes more highly expressed in early RGCs (24). Gene signature scores, other than for scRNA-seq data analysis (described below), were calculated by performing Z-score analysis for individual genes across all samples and calculating the sum of all genes in the signature.

Normalized gene expression data for 154 GBM tumor samples from RNA-seq analysis, 513 GBM tumor samples using HG-U133a arrays, and a combined 620 samples from the TCGA-GBM and TCGA-LGG RNA-seq datasets were obtained from the TCGA Research Network (https://portal.gdc.cancer.gov/projects/TCGA-GBM and ∼/TCGA-LGG, data version: Feb. 24, 2015) (25). Visualization and data downloads of brain tumor expression datasets were also performed using the GlioVis data portal (26). All gene signature scores were determined using the summed Z-score analysis of individual genes across all samples.

Single-cell suspensions of cryopreserved samples of BT48, BT67, BT89, and BT94 were thawed, checked by trypan blue exclusion for a minimum of 70% viability, and resuspended at 500 to 780 cells/μL. The single-cell library was prepared using the 10X Genomics Chromium Single-Cell v2 Chemistry Reagent Kit (10X Genomics) according to the manufacturer's protocol. Sample preparation, library preparation and sequencing, and data processing and normalization were performed as described previously (27). The gene signature scores were analyzed using the AddModuleScore() function in the Seurat package. All genes in the dataset being analyzed were divided into 25 bins based on averaged expression and 100 genes were randomly selected from each bin, to be used as a control for gene expression. Signature scores were either plotted as individual cells in a scatterplot or divided into quartiles based on the astrocyte gene signature score. Additional information is available in the Supplementary Materials and Methods.

Dissociated BTSCs were resuspended in PBS and 0.5% MACS BSA stock solution (Miltenyi Biotec, cat.# 120–091–376). Nonpermeabilized cells were incubated with the primary EAAT2 antibody (Abcam, cat.# ab17840) or a no primary control for 30 minutes at 4°C each in the dark, washed with PBS, and then incubated with the secondary Alexa Fluor plus goat anti-rabbit 488 nm antibody (Thermo Fisher Scientific, cat.# A32731, RRID:AB_2633280) for 30 minutes at 4°C in the dark. Cells were then washed with PBS and resuspended in 0.5% MACS BSA and 0.4% paraformaldehyde. Gating was established based on the IgG isotype control and then applied to all remaining samples and the percent of positive cells were calculated using FlowJo software (FlowJo, ver. 10.4.2).

Unless otherwise stated in the figure legends, values are given as the mean ± SD for n = 3 replicates and P values were calculated using a two-tailed unpaired Student t test. Statistical comparisons were performed using Microsoft Excel (version 16) and GraphPad Prism (version 8) software.

The bulk RNA-seq data for CB-839 treated BT67 and BT89 lines was deposited in NCBI's Gene Expression Omnibus (GEO) repository, accession number: GSE155300. The scRNA-seq data were deposited in the European Genome-phenome Archive (EGA) repository, in the form of FASTQ files as part of accession number: EGAS00001004656. Processed scRNA-seq data have been made available through the Broad Institute Single Cell Portal, accession number: SCP1155. Bulk RNA-seq data for the large panel of BTSC lines were previously published and deposited in the EGA, accession number: EGAS00001002709 (18). Further inquiries can be made to the corresponding authors upon reasonable request.

To determine whether BTSC lines are dependent on glutaminolysis for cell growth, we blocked glutaminolysis in a panel of 19 BTSC lines using the GLS inhibitor CB-839. CB-839 is a potent and selective inhibitor of GLS with oral bioavailability (28) that is currently undergoing multiple phase I and phase II clinical trials. We observed that of the 19 BTSC lines tested, three had a greater than 75% decrease and nine had a less than 25% decrease in cell growth after a 14-day treatment with 1 μmol/L CB-839 (Fig. 1B). Henceforth, we describe these BTSC lines as GLS-dependent and GLS-independent, respectively. We further characterized the dose response of two GLS-dependent BTSC lines (BT89, BT94), two GLS-independent BTSC lines (BT48, BT67), and a human fetal-derived neural stem cell (HF-NSC) line, all grown under identical neural stem cell culture conditions. Following a 14-day treatment, we observed that the GLS-dependent BTSC lines were highly sensitive to increasing concentrations of CB-839 (Fig. 1C). The differential response of GLS-dependent and GLS-independent BTSC lines to GLS inhibition was further supported by measuring the constitutive protease activity of viable cells (Supplementary Fig. S1A). By following BT89 cells over time using live-cell imaging techniques, we observed that GLS inhibition led to a collapse of the neurosphere morphology (Supplementary Fig. S1B). We next assessed the effect of GLS inhibition on cell death using a cell-impermeable fluorescent DNA stain and trypan blue staining, we found that CB-839 treatment increased cell death in GLS-dependent but not GLS-independent BTSC lines (Fig. 1D; Supplementary Fig. S1C and S1D). Limiting dilution analysis showed that GLS inhibition abolished the sphere-forming capacity of GLS-dependent but not GLS-independent BTSC lines, likely due to the effects on cell viability (Fig. 1E; Supplementary Fig. S1E). These data demonstrate that the GLS-dependent BTSC lines are reliant on glutaminolysis for cell growth, viability, and sphere formation.

We next investigated the metabolic requirements that might underlie the response to GLS inhibition in BTSCs. A major role for glutaminolysis in cancer is to provide a source of carbon for the replenishment of TCA cycle intermediates (11). Treatment of BTSCs with CB-839 decreased intracellular glutamate levels in BTSCs, as measured by LC/MS-MS, with a greater decrease seen in the GLS-dependent line BT89 than in the GLS-independent line BT67 (Fig. 1F; Supplementary Fig. S1F). These data indicate that although CB-839 inhibits GLS function in both groups, GLS-dependent BTSCs are more reliant on GLS to maintain intracellular glutamate levels. If the divergent response to CB-839 treatment was related to the contribution of glutaminolysis to the TCA cycle, then α-ketoglutarate, a direct product of glutamate deamination, would act as a critical metabolic intermediate. However, we found that the levels of α-ketoglutarate were similarly decreased in both GLS-dependent and GLS-independent BTSC lines (Fig. 1G). Furthermore, supplementation with dimethyl-α-ketoglutarate, a cell-permeable form of α-ketoglutarate previously shown to rescue cell growth following glutamine withdrawal (29), did not rescue the growth of the BT89 or BT94 lines following GLS inhibition (Fig. 1H and I). Thus, the conventional role of glutaminolysis as a carbon source for the TCA cycle through α-ketoglutarate does not account for the differential response of BTSCs to GLS inhibition.

To identify a cellular process or pathway that contributes to the divergent response of BTSC lines to GLS inhibition, we assessed changes in the transcriptome by performing RNA-seq analysis in BTSCs treated with CB-839. We observed that the AADR pathway (16) was upregulated in the GLS-dependent line BT89 but not in the GLS-independent line BT67 (Fig. 2A; Supplementary Fig. S2A). The AADR pathway is induced when intracellular amino acids are depleted to the point that uncharged tRNAs accumulate and activate the general control nonderepressible protein 2 (GCN2, encoded by EIF2AK4) kinase. Activated GCN2 phosphorylates the translation initiation factor eIF-2α, resulting in an increase in activating transcription factor 4 (ATF4) expression, translation, and ATF4-mediated transcription of genes in the AADR pathway (Fig. 2B; reviewed in ref. 30). We validated the RNA-seq data using qPCR and Western blot analysis to show that the mRNA and protein levels of the AADR effector ATF4 and three AADR induced genes, C/EBP-homologous protein (CHOP, encoded by DDIT3), asparagine synthetase (ASNS), and tribbles pseudokinase 3 (TRIB3; refs. 30, 31), were increased in two GLS-dependent but not in two GLS-independent BTSC lines after CB-839 treatment (Fig. 2C–E; Supplementary Fig. S2B–S2F).

We next investigated the functional role of the AADR pathway in BTSCs following GLS inhibition. Elevated levels of CHOP can increase the protein levels of the pro-apoptotic gene BCL2 Like 11 (BCL2L11, encoding the apoptotic effector BIM) (32). We asked whether the elevated levels of CHOP, increased as part of the AADR pathway, corresponds with an upregulation of BIM and apoptosis in BTSCs. We observed an increase in the levels of both BIM and cleaved caspase-9, an effector of the intrinsic apoptosis pathway, in GLS-dependent BTSC lines treated with CB-839 (Fig. 2E). Previous work has shown that BSA can be taken up and metabolized as a source of amino acids, specifically as a source of glutamine, to rescue cancer cell proliferation following glutamine withdrawal (33). We found that supplementing BSA to the culture media rescued the growth of two GLS-dependent BTSC lines treated with CB-839 (Fig. 2F; Supplementary Fig. S2G). Interestingly, the use of 10% FBS, of which, albumin is present at a final concentration of approximately 0.2% in complete media (33), when culturing established GBM cell lines could mask the effects of amino acid depletion following GLS inhibition. This may, in part, explain why previous studies observed that GBM cell lines cultured with serum are resistant to GLS inhibition and glutamine withdrawal (34).

The increase in ATF4 and CHOP expression could also indicate that GLS inhibition initiates a more general ER stress response in GLS-dependent BTSC lines. Thus, we evaluated a panel of ER stress response genes (Supplementary Fig. S2H) and ATF4-regulated genes (Supplementary Fig. S2I) in the BT89 and BT67 lines after CB-839 treatment. We did not observe a broad increase in the ER stress response or in ATF4-regulated gene expression. Furthermore, most of the ER stress response and ATF4-regulated genes that did increase in expression are part of the AADR. An additional readout of ER stress is the splicing of XBP1 mRNA by IRE1 and the increased expression of total XBP1 mRNA as mediated by ATF4 (35). Tunicamycin, an inducer of ER stress, increased both spliced and total XBP1 mRNA levels in all four BTSC lines, whereas CB-839 treatment did not (Fig. 2G; Supplementary Fig. S2J). These results show that treatment of GLS-dependent BTSC lines with CB-839 does not trigger a more general ER stress response. Altogether, these data support that GLS-dependent BTSC lines require GLS to maintain amino acid levels to prevent the induction of the AADR pathway.

We next asked whether exogenous glutamate, the product of glutamine deamination by GLS, can rescue the growth of GLS-dependent BTSCs following CB-839 treatment. Surprisingly, supplementation with a supraphysiologic concentration of exogenous glutamate (1 mmol/L) did not rescue the growth of BTSCs treated with CB-839 (Fig. 3A). To test another potential source of glutamate, we supplied exogenous alanine, which can be metabolized by the glutamic-pyruvic transaminases (GPT, GPT2) to produce glutamate (Supplementary Fig. S3A). Although CB-839 treatment led to a greater decrease in the levels of intracellular alanine in the GLS-dependent line BT89 when compared with the GLS-independent line BT67 (Supplementary Fig. S3B), supplementation with alanine (2 mmol/L) did not rescue BT94 cell growth (Supplementary Fig. S3C). We next asked whether modulating other cellular processes could rescue BTSC growth following GLS inhibition. Glutamate is required for the synthesis of glutathione (GSH), a reactive oxygen species scavenger (36). Supplementation with a cell-permeable form of GSH has been used to overcome glutamine withdrawal in pancreatic cancer cells (37); however, in the GLS-dependent line BT94, cell-permeable GSH did not rescue cell growth following CB-839 treatment (Supplementary Fig. S3D). Because HIF1A activity can alter the dependence on glutaminolysis and glycolysis (38), we supplemented the culture media with the HIF1A stabilizer dimethyloxaloylglycine (DMOG) and found that DMOG did not rescue BT94 cell growth following GLS inhibition (Supplementary Fig. S3E). We therefore hypothesized that an impairment in glutamate transport may be preventing the rescue of BTSC growth using exogenous glutamate.

Glutamate is the most abundant excitatory neurotransmitter in the adult mammalian brain and glutamate transport is highly regulated to allow for efficient neurotransmission (10). There are five excitatory amino acid transporters (EAAT) that take up extracellular glutamate. EAAT2 (encoded by SLC1A2, a.k.a. EAAT2 or GLT-1) and EAAT1 (encoded by SLC1A3, a.k.a. EAAT1 or GLAST) are unique in that they are highly expressed in astrocytes (39–41). Notably, EAAT2 has been shown to be responsible for the majority of glutamate uptake in the mammalian brain (42). We next assessed EAAT1 and EAAT2 mRNA expression and protein levels in four BTSC lines and observed that EAAT1 expression level is lowest in the BT89 line and EAAT2 expression is low in both GLS-dependent BTSCs (Fig. 3B and C; Supplementary Fig. S3F and S3G). We then extended this analysis to the panel of BTSC lines treated with CB-839 (Fig. 1B) by analyzing RNA-seq data for the expression of EAAT1 and EAAT2 in the 19 untreated BTSC lines (18). We observed that the combined expression of EAAT1/EAAT2 correlates with resistance to CB-839 treatment (Fig. 3D). These findings suggest that higher expression of EAAT1/EAAT2 allows for the uptake of extracellular glutamate, thereby decreasing the dependence on GLS as a source of intracellular glutamate.

In the nondiseased brain, the expression of EAAT1/EAAT2 is highest in astrocytes, as described above (39–41). We examined whether there is a relationship between astrocytic gene expression and EAAT1/EAAT2 expression in GBM tumors and BTSCs. To assess astrocytic gene expression, we defined an astrocyte gene signature by combining data from two single-cell RNA-seq (scRNA-seq) studies of normal human brain tissue (23, 24). To determine whether the relationship between the astrocyte gene signature and EAAT1/EAAT2 expression persists in GBM tumors, we analyzed a GBM gene expression dataset from The Cancer Genome Atlas (TCGA; ref. 25). We observed a positive correlation, demonstrating that GBM tumors with higher astrocytic gene expression have higher EAAT1/EAAT2 expression (Fig. 4A; Supplementary Fig. S4A). Of note, gene signatures for oligodendrocyte precursor cells (OPC) and early radial glial cells (RGC), less differentiated cell types in the brain, did not correlate with EAAT1/EAAT2 expression as strongly as the astrocyte gene signature in GBM tumors (Fig. 4B; Supplementary Fig. S4B–S4D). Interestingly, we observed in the TCGA dataset that the expression of EAAT1/EAAT2 was lower in GBM (grade 4 glioma) when compared with low-grade gliomas, whether assessed by grade or tumor histology (Fig. 4C; Supplementary Fig. 4E). Within GBM, EAAT1/EAAT2 expression was lower in the more aggressive recurrent disease when compared with primary GBM (Fig. 4D), suggesting that these aggressive tumors are more dependent on GLS to maintain intracellular glutamate. We next asked whether a relationship between astrocytic gene expression and EAAT1/EAAT2 expression exists across BTSC lines by analyzing RNA-seq data across a panel of 57 BTSC lines (18) and we found a positive correlation between the astrocyte gene signature and EAAT1/EAAT2 expression (Fig. 4E). Therefore, a positive correlation between astrocytic gene expression and EAAT1/EAAT2 expression exists across GBM tumor tissue and BTSC lines.

We then sought to determine whether the relationship between EAAT1/EAAT2 expression and astrocytic gene expression observed in bulk RNA-seq analysis of BTSC lines is maintained within individual cells. We thus performed scRNA-seq analysis of 4,785 cells from four BTSC lines (BT48, BT67, BT89, and BT94). When single cells from all four BTSC lines were pooled together, the astrocyte gene signature positively correlated with the EAAT1/EAAT2 expression score (Fig. 4F). To further assess the distribution of EAAT1/EAAT2 expression in BTSCs with varying astrocytic gene expression, we segregated cells into quartiles based on the astrocyte gene signature score. We found that the most astrocytic BTSCs had higher EAAT1/EAAT2 expression compared to the least astrocytic BTSCs (Supplementary Fig. S4F). Subsequently, we analyzed BTSC lines individually and found that, within each BTSC line, BTSCs with higher astrocyte gene signature scores had higher EAAT1/EAAT2 expression scores (Fig. 4G and H; Supplementary Fig. S4G–S4L). Overall, these data demonstrate that the relationship between astrocytic gene expression and EAAT1/EAAT2 expression, that exists in the nondiseased brain, is maintained in GBM tumors and BTSCs. Furthermore, these differences in EAAT1/EAAT2 expression may modulate the metabolic requirements of BTSCs.

We next asked whether EAAT2 function could be augmented to modulate the dependence of BTSCs on GLS. The compound LDN-0212320 (LDN) increases EAAT2 protein levels by increasing the amount of EAAT2 mRNA being translated (43). Flow cytometric analysis of EAAT2, using an antibody targeting an extracellular peptide sequence in nonpermeabilized cells, showed that LDN treatment increased the number of cells with EAAT2 protein expressed on the surface of BT94 and BT89 cells (Fig. 5A; Supplementary Fig. S5A and S5B). Importantly, the addition of LDN completely rescued GLS-dependent BTSC growth following CB-839 treatment (Fig. 5B; Supplementary Fig. S5C). We then asked whether blocking EAAT transporters with L-trans-2,4-PDC (PDC), an inhibitor of EAAT1–5 (44), could sensitize GLS-independent BTSC lines to GLS inhibition. When EAAT transporters and GLS were inhibited, BT67 and BT48 cell growth decreased (Fig. 5C; Supplementary Fig. S5D). Furthermore, this decrease in BTSC growth was accompanied by a further decrease of intracellular glutamate levels (Fig. 5D), and an increase in CHOP protein levels (Fig. 5E) and AADR gene expression (Fig. 5F; Supplementary Fig. S5E–S5G). Collectively, these experiments demonstrate that EAAT1/EAAT2 transporter function in BTSCs can modulate their dependence on GLS.

We next sought to further examine the importance of maintaining intracellular glutamate levels to prevent the AADR in BTSCs. Similar to glutamate (Fig. 1F), we observed that intracellular aspartate levels decreased in both BT89 and BT67 lines following CB-839 treatment, with a greater decrease in GLS-dependent BT89 cells (Fig. 6A). ASNS converts aspartate and glutamine into asparagine and glutamate via a transamination reaction (Supplementary Fig. S6A). However, unlike glutamate (Fig. 5D) and alanine (Supplementary Fig. S6B), aspartate levels did not decrease further when both glutamate transport and GLS were inhibited compared with CB-839 treatment alone (Supplementary Fig. S6C). We found that supplementing CB-839 treated GLS-dependent BTSC lines with exogenous aspartate (0.5 mmol/L) rescued the decrease in cell growth in BT94 but not BT89 (Fig. 6B). Previous work has shown that supplementation with asparagine can rescue cell viability following glutamine deprivation (45). Here, we found that exogenous asparagine (2 mmol/L) did not rescue the decrease in BT94 growth following GLS inhibition (Fig. 6C), suggesting that intracellular glutamate is the product of aspartate metabolism by ASNS that rescues BTSC growth. Supplementation with 0.5 mmol/L exogenous glutamate rescued the decrease in cell growth in the GLS-independent line BT67 after treatment with PDC and CB-839, most likely by competing with PDC (a competitive and transportable EAAT inhibitor) binding to EAAT1/EAAT2 (Fig. 6D). Similarly, supplementation with 0.5 mmol/L exogenous aspartate also rescued BT67 cell growth when both EAATs and GLS are inhibited (Fig. 6E). To determine whether glutamate or aspartate supplementation can maintain amino acid levels, we supplied exogenous glutamate or aspartate and observed that either amino acid can prevent the induction of AADR genes, with a stronger response using glutamate, in BT67 cells following treatment with CB-839 and PDC (Fig. 6F and G; Supplementary Fig. S6D and S6E). Altogether, these results support that maintaining intracellular glutamate levels is required to prevent the AADR in BTSCs.

Glutamine metabolism supports the uncontrolled proliferation of cancer cells by sustaining the energetic and biosynthetic demands of cell proliferation (1). In the nondiseased human brain, glutamine–glutamate homeostasis is maintained by astrocytes, with high levels of EAAT1/EAAT2 (39–41), that take up glutamate to synthesize and secrete glutamine (9). Following reports of glutamine metabolism regulating the cell fate of normal human stem cells (7, 8) and the previously described utilization of glutaminolysis in NSCs (9), we sought to elucidate the role of glutaminolysis in BTSCs. Here, we present a previously unrecognized dependence of BTSCs on glutaminolysis to maintain amino acid levels to prevent the AADR pathway and cell death.

Our finding that the BTSC lines BT89, BT94, and BT248 are highly sensitive to GLS inhibition (Fig. 1B) is in contrast to previous studies on GBM in which glutamine depletion or inhibition of GLS had either no effect or a modest decrease in cell growth or viability (34, 46–48). Interestingly, in the Tardito and colleagues (34) study, which did not observe a decrease of BTSC growth when cultured in the absence of glutamine, it was reported that glutamate was taken up from the media in all three of the BTSC lines tested. This is in line with our finding that when glutamate is taken up at sufficient levels, presumably by the central nervous system specific EAATs (EAAT1/EAAT2), BTSCs are not dependent on glutaminolysis for cell growth. Consistent with the role of glutamate transporters in normal astrocytes, bulk RNA-seq analysis and single-cell RNA-seq analysis of BTSCs showed that a higher astrocyte gene signature score positively correlates with higher expression of EAAT1/EAAT2. We therefore describe a relationship between the expression of astrocytic genes and glutamate transporters, and the dependence on glutaminolysis in BTSCs. This reveals that BTSCs, which meet the criteria of CSCs and have shared biological properties with NSCs (13–15), can express astrocytic genes that, although insufficient for terminal differentiation into mature astrocytes, can dramatically alter BTSC metabolic dependencies.

An alternate interpretation of our results we considered is that the critical metabolite transported by EAATs might be aspartate, which EAATs also transport, and not glutamate. Of note, it has been reported that in colon cancer cells the transport of aspartate by EAAT1 was required for nucleotide synthesis during glutamine starvation (49). Here, supplementation with aspartate rescues cell growth of the GLS-dependent line BT94, but not BT89, treated with CB-839 (Fig. 6B) and of the GLS-independent line BT67 when combining PDC with CB-839 (Fig. 6E). However, several observations suggest that it is the depletion of intracellular glutamate that underlies the induction of the AADR and decrease in BTSC growth following GLS inhibition. First, under conditions where both EAATs and GLS were inhibited, which decreased growth of the GLS-independent line BT67, there was a further depletion of intracellular glutamate but not aspartate when compared to GLS inhibition alone (Fig. 5D; Supplementary Fig. S6C). Second, glutamate, the direct product of the GLS reaction, can rescue BT67 cell growth caused by inhibition of both glutamate transport and GLS (Fig. 6D). Third, under these same conditions, where the BT67 line is treated with PDC and CB-839, glutamate is more effective than aspartate at preventing the induction of the AADR genes CHOP, ATF4, and TRIB3 (Fig. 6G; Supplementary Fig. S6D and S6E). Interestingly, our data suggest a compensatory feedback loop where GLS inhibition leads to the induction of the AADR and increased expression of ASNS, thereby increasing the ability to metabolize aspartate as a source of intracellular glutamate.

In the context of resistant BTSC lines with higher EAAT1/EAAT2 expression, targeting glutamate transport in combination with GLS inhibition decreased intracellular glutamate levels and cell proliferation more than either inhibitor alone (Fig. 5C and D). This result suggests that the maintenance of sufficient intracellular glutamate levels represents a novel metabolic vulnerability that can be targeted more broadly across BTSC lines and in bulk GBM non-CSCs with higher EAAT1/EAAT2 expression. We do note that inhibiting glutamate transport in the brain using PDC may be compromised by a narrow therapeutic window because high levels of extracellular glutamate in the human brain are associated with the induction of seizures and excitotoxic damage (50). However, when administered into the rat brain in vivo, acute exposure to PDC does not induce seizures whereas the EAAT1–5 inhibitor DL-threo-β-Benzyloxyaspartate (DL-TBOA) does (51). Careful future studies will be required to evaluate the potential for the inhibition of glutamate transporters to exacerbate seizures in patients with GBM. An alternate strategy to further deplete intracellular glutamate levels could involve targeting other glutamate-producing enzymes in combination with GLS inhibition. For example, the glutamic-pyruvic transaminases (GPT and GPT2) are involved in the synthesis of alanine (Supplementary Fig. S3A) and under conditions where intracellular glutamate is limited, alanine may be metabolized as a source of glutamate. Concordantly, we observed that when GLS is inhibited alanine levels are lower in the GLS-dependent line BT89 than in the GLS-independent line BT67 (Supplementary Fig. S3B). Future genetic studies will be needed to determine the most effective metabolic enzymes to target in combination with GLS inhibition.

Our finding that GLS inhibition triggers the AADR pathway and cell death demonstrates that inducing the AADR or enhancing its downstream effectors are promising therapeutic strategies to target BTSCs. Depleting amino acid levels to induce the AADR in cancer is an emerging strategy to limit cancer cell proliferation and induce cell death. The depletion of asparagine using l-asparaginase can trigger the AADR (52) and has been used in the clinic to treat acute lymphocytic leukemia (53). In GBM, initial testing of l-asparaginase in mice xenografted subcutaneously with serum-cultured GBM cell lines has shown either limited or no reduction in tumor growth (54, 55). However, neither of these two studies in GBM cells assessed whether l-asparaginase treatment induced the AADR pathway. We suggest that investigating strategies to trigger the AADR pathway or enhance its downstream effectors in GBM, especially in combination with GLS inhibition, warrants further exploration.

In summary, our findings reveal a previously unrecognized dependence on intracellular glutamate for the maintenance of amino acids to prevent the AADR and cell death in GBM-derived BTSCs. As BTSCs express higher levels of the astrocytic glutamate transporters EAAT1/EAAT2, the dependence on glutaminolysis to generate glutamate is reduced. However, inhibiting glutamate transport in GLS-independent BTSC lines increases the dependence on glutaminolysis for intracellular glutamate and cell survival. Our observations indicate that methods to limit intracellular glutamate and enhance the proapoptotic effects of the AADR in BTSCs warrant further investigation. Overall, these data improve our understanding of the underlying metabolic dependencies of BTSCs and suggest novel therapeutic strategies for this aggressive and difficult to treat disease.

No disclosures were reported.

I.J. Restall: Conceptualization, data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. O. Cseh: Data curation, validation, investigation, methodology. L.M. Richards: Data curation, software, formal analysis, validation, investigation, methodology. T.J. Pugh: Resources, data curation, software, supervision, validation, methodology. H.A. Luchman: Conceptualization, data curation, supervision, funding acquisition, validation, methodology, project administration, writing-review and editing. S. Weiss: Conceptualization, resources, data curation, supervision, funding acquisition, validation, methodology, project administration, writing-review and editing.

We thank Rozina Hassam (University of Calgary) for technical support, Paul Gordon (University of Calgary) for help with RNA-seq data processing, and Julien Lamontagne and Erik Joly (Centre hospitalier de l'Université de Montréal) for help with metabolomics acquisition and data processing. A subset of results published here are in part based upon data generated by the TCGA Research Network, https://www.cancer.gov/tcga. This study was supported by a grant from the Canadian Institutes for Health Research (153246 to H.A. Luchman, S. Weiss). S. Weiss and T.J. Pugh were also supported by a Stand Up to Cancer Canada Cancer Stem Cell Dream Team grant (SU2CAACR-DT-19-15) with funding provided by the Government of Canada through Genome Canada and the Canadian Institutes of Health Research. Stand Up To Cancer Canada is a Canadian Registered Charity (Reg. # 80550 6730 RR0001). Research Funding is administered by the American Association for Cancer Research International – Canada, the Scientific Partner of SU2C Canada. I.J. Restall was the recipient of a Clark H. Smith Fellowship award.

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.