Cystathionine-γ-lyase drives antioxidant defense in cysteine-restricted IDH1-mutant astrocytomas

Abstract Background Mutations in isocitrate dehydrogenase 1 or 2 (IDH1/2) define glioma subtypes and are considered primary events in gliomagenesis, impacting tumor epigenetics and metabolism. IDH enzyme activity is crucial for the generation of reducing potential in normal cells, yet the impact of the mutation on the cellular antioxidant system in glioma is not understood. The aim of this study was to determine how glutathione (GSH), the main antioxidant in the brain, is maintained in IDH1-mutant gliomas, despite an altered NADPH/NADP balance. Methods Proteomics, metabolomics, metabolic tracer studies, genetic silencing, and drug targeting approaches in vitro and in vivo were applied. Analyses were done in clinical specimen of different glioma subtypes, in glioma patient-derived cell lines carrying the endogenous IDH1 mutation and corresponding orthotopic xenografts in mice. Results We find that cystathionine-γ-lyase (CSE), the enzyme responsible for cysteine production upstream of GSH biosynthesis, is specifically upregulated in IDH1-mutant astrocytomas. CSE inhibition sensitized these cells to cysteine depletion, an effect not observed in IDH1 wild-type gliomas. This correlated with an increase in reactive oxygen species and reduced GSH synthesis. Propargylglycine (PAG), a brain-penetrant drug specifically targeting CSE, led to delayed tumor growth in mice. Conclusions We show that IDH1-mutant astrocytic gliomas critically rely on NADPH-independent de novo GSH synthesis via CSE to maintain the antioxidant defense, which highlights a novel metabolic vulnerability that may be therapeutically exploited.


INTRODUCTION
The identification of mutations in isocitrate dehydrogenase 1 or 2 (IDH1, IDH2) has dramatically improved our understanding of glioma genesis 1 and led to a better delineation of glioma subtypes. Whereas most primary glioblastomas (GBM) harbour the wild-type enzyme (IDHwt), mutated IDH (IDHm) largely defines lower grade gliomas and secondary GBM.
LGGs are further subdivided based on 1p19q codeletion for oligodendrogliomas (OD) and frequent mutations in TP53 and ATRX for astrocytomas (AS) 2 . Gliomas most often display the R132H mutation in IDH1. The wildtype enzyme produces α-ketoglutarate (αKG) from isocitrate thereby generating NADPH and CO 2 , while the mutant converts αKG into D-2-hydroxyglutarate (D2HG) and oxidizes NADPH 3 . As a main contributor of cytosolic NADPH, IDH1wt is crucial for redox homeostasis via recycling of glutathione (GSH), the main antioxidant in the brain 4 . Up to 65% of total NADPH may be generated from this reaction in IDH1wt GBM, whereas this contribution is decreased in IDH1m gliomas 5 .
Moreover, several studies suggest that IDH1m enhances chemo-radiosensitivity through GSH depletion and ROS generation [6][7][8][9] , yet it is currently not understood how GSH levels are maintained in these tumors.
We previously identified metabolic aberrations in phospholipid, energy and oxidative stress regulation in IDHm gliomas 10 . Notably, despite a drop in the NADPH/NADP+ ratio, GSH levels were barely affected in IDH1m tumors, while enzymes related to cysteine metabolism and de novo GSH production such as cystathionine-β-synthase (CBS) and Glutamatecysteine ligase catalytic subunit (GCLC) showed increased gene expression, suggesting that de novo GSH synthesis might be active in these tumors 10 .
Based on clinical glioma specimen and unique patient-derived cell lines carrying the endogenous IDH1 mutation, we report that IDH1m astrocytomas specifically rely on cystathionine-γ-lyase (CSE, also known as cystathionase, CTH gene) to increase their cysteine pool for de novo GSH synthesis. Furthermore, we show that the CSE inhibitor propargylglycine (PAG) leads to increased cytotoxicity at low cysteine levels in vitro, and A c c e p t e d M a n u s c r i p t 6 affects tumor growth in vivo. Our data warrant further investigation on the therapeutic potential of CSE in IDH1m astrocytoma patients.

Clinical samples
Clinical glioma samples were obtained from 22 patients (

Patient-derived glioma cell lines
Patient-derived cell lines were grown as 3D spheres in defined supplemented DMEM-F12 medium (Supplementary Methods and Figure S1a) [11][12][13][14] . NCH644, NCH601 and NCH421k correspond to IDH1wt GBM, NCH1681 and NCH551b to IDH1m AS, NCH612 to IDH1m OD, based on molecular classification 2 . Cysteine was titrated by serial BIT dilutions (Supplementary Table 2). To confirm that the toxic effect of PAG in low BIT/cysteine medium was due to the lack of cysteine, the cells were rescued with extracellular cysteine ( Figure S4c) 15 .

Proteomics analysis
Protein extracts from all 6 glioma cell lines were digested and peptides were analyzed on a Q-Exactive HF mass spectrometer (Thermo Scientific) coupled with a Dionex Ultimate 3000

In vitro assays
Western blot, qPCR, gene knockdown, sphere size measurement, cytotoxicity assay and ROS measurements are described in Supplementary Methods.

Metabolite and flux analysis in cell lines
For metabolite analysis cells were grown in 30 µM cysteine. 400 µM of L-Serine (U-13 C 3 , 99%) (CLM-1574-H-0.1 Eurisotop) were added to the culture medium obtaining a 1:1 ratio of 13 C 3 -Serine/natural Serine. Metabolites were isolated after 4 days. LC-MS analysis was performed as described 16,17 . Metabolites were identified using an in-house library of exact mass and known retention time generated using commercial standards on the same LC-MS system. A pilot GC-MS-based approach to determine steady state levels of 13 C 3 -serine in different mice tissues ( Figure S4b) was performed as described in 18 .

In vivo experiments
NCH1681 cells were stereotactically injected into the frontal cortex of NSG mice (NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ) (n=20) and tumor growth was monitored by MRI (FSE-T2 sequence, 3T MRI system, MR Solutions) as described previously 19 . Two months after implantation, mice with tumors (size range 2-10 mm 3 ) were selected and randomized into 2 groups (8 mice per group). PAG (35mg/kg) or saline was given ip (5 times/week during 3 months). For in vivo flux analysis, 130 mg/kg/h U-13 C 3 serine was infused in the tail vein (2 mice/group, 4hrs), plus 2 healthy mice without tumors. Brain, liver, kidney and heart were dissected and for LC-MS analysis. All procedures were approved by the animal welfare structure of LIH

IDH1m astrocytomas upregulate cystathionine-γ-lyase (CSE)
We previously reported the upregulation of CBS and GCLC in IDHm gliomas, suggesting an activation of the transsulfuration pathway (Figure 1a) in these tumors 10 , an important source of cysteine for GSH synthesis in astrocytes 20,21 . To corroborate this at the metabolite level, we analyzed specimen from 22 glioma patients including IDHwt GBM (n=10), IDH1m OD (n=4) and IDH1m AS (n=8) (Supplementary Table 1). As expected, IDHm gliomas showed high D2HG and reduced αKG levels compared to IDHwt GBM (Figure 1b). Interestingly, while GSH levels were similar, we observed a decrease in glutamate and an increase in glucose and cystathionine in IDHm gliomas, supporting activation of the transsulfuration pathway (Figure 1b).
To address this in a model amenable to experimental manipulation, we turned to patientderived glioma stem-like cells (GSCs) carrying the endogenous IDH1 mutation (NCH1681, NCH551b, NCH612) and control cells of IDHwt GBM (NCH644, NCH601, NCH421k) ( Figure S1a). IDH1m GSCs exhibited D2HG levels comparable to patients and display low proliferation rates 11-13 . We confirmed the presence of the mutation at the genetic ( Figure   S1b) and protein level ( Figure 1c) and a more than 20 fold increase in D2HG ( Figure S1c).
IDH1m allele frequency reached almost 100% in astrocytic cell lines (Figure S1b), suggesting a loss of heterozygosity over time. The tendency to develop an IDH1m homozygosity in vitro has already been reported by others 22 . Next we investigated the expression of the main mediators of GSH production, CBS and cystathionine-γ-lyase (CSE) at the protein level. CSE is central to this pathway and the only known enzyme capable to synthesize cysteine, the limiting metabolite for GSH production 23 . CSE generates cysteine, α-ketobutyrate and ammonia through the breakdown of cystathionine which is provided by CBS in the transsulfuration pathway (Figure 1a). While CBS showed similar expression in all tumor types, we found that CSE was upregulated in IDH1m cells both at the protein and mRNA level, (Figure 1c-e). Among glioma subtypes, the upregulation of CSE was mainly seen in the IDH1m AS cell lines (NCH1681, NCH551b) (Figure 1c-e). This was confirmed in A c c e p t e d M a n u s c r i p t 9 vivo in a panel of patient-derived orthotopic xenografts, while there was variable expression in IDHwt xenografts (Figure S1d-f). We further validated these data interrogating our clinical material, which showed increased CSE protein in IDH1m tumors compared to IDHwt, with highest expression in AS (Figure S1g,h). Gene expression data from public databases using the GlioVis portal, further confirmed these observations 15 (Figure S1i).
To obtain a more comprehensive view of protein deregulation in IDH1m gliomas, we  (Figure 1g). In summary, we found that cystathionine accumulates in IDHm gliomas and identify its converting enzyme CSE and related partners to be specifically upregulated in IDH1m AS, suggesting an increased activity of the transsulfuration pathway in this glioma type.

Loss of CSE reduces viability in IDH1m astrocytoma cells under cysteine depletion
To investigate the role of CSE in IDH1m AS, we established stable CSE knockdown (KD) lines in NCH1681 cells (IDH1m AS) using two different shRNAs. Both KD clones (shCSE1 and shCSE2) showed a strong drop in CSE protein (Figure 2a,b), yet we did not observe a significant difference in sphere size over time (Figure 2c) compared to control (shCTR), indicating that the proliferation capacity of these cells was unaltered under standard culture conditions. We argued that CSE may not be required if sufficient cysteine is provided from extracellular sources. In standard culture medium the concentration of cysteine (360 µM combined cysteine and its oxidized dimer cystine; see Supplementary Methods) is higher than in human blood (80-180 µM) 24 (source: HMDB). In contrast, in cerebrospinal fluid (CSF), a reference for nutrient values in the brain, cysteine is one of the lowest amino acids with values below 1 µM 25 (source: HMDB). To reach a concentration more relevant to neural tissue, cysteine was serially diluted in cysteine-free DMEM. This led to a gradual loss of sphere viability (based on GFP fluorescence) in CSE KD cells (Figure 2d, e). Control cells only were affected upon complete cysteine withdrawal. The selective impact on CSE KD cells was most prominent at 30 µM cysteine, where the decrease in viability was already observed after 2 days of culture further increasing over time (Figure 2f). Taken together, these data provide evidence that under physiological cysteine concentrations, CSE is required to maintain viability of IDH1m AS cells.

CSE directs GSH biosynthesis upon cysteine starvation
We next asked whether CSE was required to maintain the cysteine pool for GSH production and whether the observed loss of viability was caused by a decrease in GSH levels. We determined the contribution of CSE to the overall GSH pool, by quantifying metabolites in cultures with 30 µM cysteine at different time points. After a 3 day incubation, no effects of metabolites were observed (Figure S2a), in line with the cell viability results (Figure 2f).
However, at 4 days, we found a decrease of total GSH level in the CSE KD cells (Figure   3a). Presumably all remaining cysteine had been used up which was also reflected in cell A c c e p t e d M a n u s c r i p t 11 viability. This was confirmed in a 2 nd experiment ( Figure S2b). Unfortunately because of the gradual exhaustion of cysteine, many cells were vulnerable at day 4 and not all datapoints could be recovered in these experiments. Interestingly, there was a tendency for decreased serine, glycine, glutamate and glucose in CSE KD cells, along with a slight accumulation in cystathionine (Figure 3a, S2b), suggesting an attempt of the cells to compensate for the loss of GSH. Taken together, these data support the premise that GSH biosynthesis depends at least in part on CSE activity upon cysteine depletion.
To confirm that the drop in GSH in CSE knockdown cells was due to a decrease of de novo GSH synthesis, we traced 13 C 3 -serine in cells cultured with 30 µM cysteine for 4 days (ratio 13 C 3 -serine/natural serine = 1:1) (Figure 3b). As aforementioned, the limiting culture conditions prevented to recover sufficient biological replicates to perform statistical analysis (Figure 3c and Figure S2c respectively). Comparable levels of 13 C 3 -serine were observed in all cells (Figure 3c, S2c), whereas the natural serine isotopologue was reduced in the KDs ( Figure S2d). This drop in natural serine together with the decrease in total glucose ( Figure   3a) suggests a preference for glucose-derived serine when there is a high demand for this metabolite. In fact, the enrichment of heavy carbon labelled serine did not exceed 10% of the total pool (Figure S2e). Despite the low serine tracer incorporation, there was a tendency for lower 13 C 2 -glycine in CSE KD cells (Figure 3c, S2c) suggesting reduced serine-to-glycine conversion. In addition, shCSE cells tended towards increased levels of 13 C 3 -cystathionine and a consistent decrease in 13 C 3 -GSH (Figure 3c, S2c). The ratios GSH m+3/Serine m+3 ( Figure S2f) was significantly lower in the CSE KDs. These data support the notion that under cysteine deprivation CSE is essential to generate cysteine, preferentially via the transsulfuration pathway using glucose-derived serine to sustain GSH biosynthesis.

IDH1m astrocytoma cells are selectively sensitive to CSE inhibition
Next we wished to evaluate the therapeutic potential of CSE inhibition in glioma using available chemical inhibitors of CSE. Propargylglycine (PAG) (Figure S3a) is a nonproteinogenic amino acid that irreversibly blocks CSE through the double interaction with  (Figure S3b). Interestingly, the cytotoxic effect of PAG was specific to IDHm AS cells, and no significant toxicity was observed in PAG-treated IDHwt GBM cells (NCH644 and NCH421k) (Figure 4a). Again the effect was limited to low cysteine (≤ 60 µM) (Figure 4b).
To test if the dependency of IDH1m AS cells on CSE activity was related to their antioxidant capacity, we measured ROS levels in IDH1m and IDHwt cells. In line with the cytotoxicity, ROS increased in one of the PAG treated IDH1m AS cells, although this was not robust in the 2 nd cell line (Figure S3d). Taken together these data provide evidence that CSE inhibition under limited cysteine selectively affects IDH1m AS, possibly mediated by increased oxidative stress.

CSE inhibition delays tumor growth in vivo
To evaluate the efficacy of CSE inhibition on tumor growth in vivo, we implanted IDH1m AS cells (NCH1681) into the brain of mice. After two months, tumors were detectable by MRI (around 5 mm 3 ), mice were randomized into saline and PAG (35mg/kg, 5 times/ week) treated groups (Figure 5a). Treatment continued for 3 months and tumor size was monitored by bi-weekly MRIs. Within the first two months of treatment we found a significant decrease in tumor growth rate in PAG treated mice (n=8) (Figure 5b). A relative difference in tumor size was observed between both groups over time, which became pronounced, though not significant at the end of the 3 rd month (Figure 5c d). The experiment was stopped when two mice of the control group reached clinical endpoint (Figure S4a). To determine the effect of PAG at the metabolite level, labelled 13 C 3 -serine (130 mg/kg) was terminally infused in the tail vein (n=2/group) (Figure 5a), reaching steady state at 4-6 hours in the circulation and in the brain (Figure S4b). Consistent with the CSE KD experiment, PAG treated brain tumors showed a slight decrease in serine and glycine when compared to the control group (Figure 5e). Moreover the total level of cystathionine was considerably increased in PAG-treated tumors (Figure 5e), demonstrating target engagement in vivo.
This was accompanied by some decrease of methionine and a significant decrease of homocysteine (Figure 5e), both precursors of cystathionine. In line with the KD results, GSH and glucose levels were reduced in PAG-treated tumors (Figure 5e). A clear accumulation of cystathionine in PAG treated mice was also observed in the liver, where de novo GSH biosynthesis normally occurs 28 , while most other metabolic differences appeared to be tumor-specific (Figure S4c). The isotopologue distribution revealed that unlabeled serine decreased in treated tumors, whereas 13 C 3 -serine (M+3) showed a significant increase ( Figure S4d). Again, labeled serine only accounted for ~10% of the total pool, reflected in the small amount of labeled cystathionine (M+3) in contrast to the strong accumulation of unlabeled one (M+0) (Figure S4d), suggesting preferential use of compensatory glucosederived serine of IDH1m AS tumors also in vivo. In summary, we show that PAG inhibits CSE activity in IDH1m AS brain tumors which causes a significant delay in tumor growth, providing a rationale for a therapeutic potential of CSE inhibitors in IDH1m astrocytomas.

DISCUSSION
Due to the central role of IDH1 in NADPH recycling 4,29 , several groups including ours suggested that IDH1m tumors may display vulnerabilities in redox metabolism and GSH production 5,6,10,30,31 . Nevertheless, to date there is limited data in part due to the difficulties of establishing IDHm cellular and animal models relevant to human disease. Our data, based on patient specimen and patient-derived cell lines with the endogenous IDH mutation indicates that CSE, the only known enzyme capable of synthesizing cysteine, is specifically upregulated in IDH1m AS and is essential to maintain GSH production under limited cysteine  (Figure 6). Inhibition of CSE reduced tumor growth in vivo, thus uncovering a novel druggable metabolic vulnerability in this aggressive glioma subtype.
Our proteomics analysis identified CSE as the most upregulated protein in IDH1m tumors.
The importance of CSE as a source of cysteine for GSH production is well established in astrocytes 20,21 and an early study in C6 glioma cells showed that CSE expression increases upon GSH depletion 32 . However in IDHwt GBM the expression of CSE is variable, suggesting a more plastic response to cysteine depletion, which is consistent with the specific sensitivity to CSE inhibition that we found in IDHm vs IDHwt cells. The limited expression of CSE in ODs can be explained by the location of its gene CTH on chromosomal arm 1p. In line with this, cystathionine, the substrate of CSE, was found to accumulate in IDH1m ODs 33 , suggesting an alternative path to GSH production in these tumors.
We show that in addition to IDH status, the availability of extracellular cysteine determines the susceptibility to CSE inhibition, in line with recent data where CBS was shown to support cell growth during cysteine depletion 34 . In the brain, where the uptake of amino acids is regulated by the blood brain barrier, the cysteine concentration is reportedly very low 25   Despite the modest effect of PAG in vivo, it should be noted that a minor inhibition of growth in a slow growing tumor might well result in a valuable clinical outcome. Moreover, our data provide a rationale for combinatorial approaches, e.g. the addition of PAG may increase the vulnerability to oxidative damage caused by radiation 7 . Recently the combination of radiotherapy with a glutaminase inhibitor was shown to increase mouse survival 45 . The antitumor effects of CSE inhibition may be potentiated by a dietary restriction of cysteine to simultaneously interfere with exogenous and endogenous cysteine supply. This approach was well tolerated and more effective in depleting GSH in the brain of healthy rats compared to PAG alone 46 . In other cancer models the enzyme cyst(e)inase was found to efficiently degrade circulating cysteine, thereby increasing oxidative damage and decreasing tumor load 47 . In conclusion, we here identified a specific dependency of IDH1m astrocytomas on CSE to maintain antioxidant homeostasis suggesting pharmacological inhibition of CSE as a potential strategy that warrants further investigation.