Glutaminase-2 Expression Induces Metabolic Changes and Regulates Pyruvate Dehydrogenase activity in Glioblastoma Cells

1. Introduction

The reprogramming of cellular metabolism is a hallmark of cancer [1]. Tumours present metabolic adaptations that allow meeting the increased need for energy generation, biosynthesis of new cell compounds and maintaining biochemical homeostasis [2]. Although the concept of metabolic reprogramming is built upon the notion that cancer cells possess a plethora of metabolic adaptations that increase their fitness, it may also render them vulnerable to specific targeted therapies against these metabolic traits, which are fundamental for their survival and success [3,4]. Glutamine (Gln) provides a large reservoir for nitrogen in animal cells [5] and is by far the most abundant amino acid in human plasma [6]. The importance of Gln in cancer metabolism has been established in seminal articles [6,7]. In cancer, Gln may constitute an essential nutrient, fuelling bioenergetics and serving to obtain ATP, promoting the biosynthesis of non-essential amino acids [8], nucleotides [9], lipids [10] hexosamines, and also being linked to the synthesis of the critical antioxidant glutathione (GSH) [11]. Therefore, Gln constitutes a source of both reduced nitrogen and carbon and is a main anaplerotic substrate in cancer cells, being the principal supplier for the tricarboxylic acid (TCA) cycle in cultured cells in vitro [12].

Gliomas account for the majority of tumours arising from the central nervous system [13]. Although glycolysis provides energy and carbon skeletons for the synthesis of amino acids, lipids and nucleotides in gliomas, glutaminolysis and oxidative phosphorylation (OXPHOS) are also critical for their metabolism [14]. Among gliomas, glioblastoma multiforme (GBM) is the most common and lethal, showing a very low 5-year survival rate [15]. Current treatment of GBM mainly consists of maximal surgical resection followed by radiotherapy and chemotherapy using the alkylating agent temozolomide; however, the efficacy is limited, as median survival time ranges from 12 to 15 months, so new therapeutic options are needed [16]. Gliomas, including GBM, are classified as isocitrate dehydrogenase (IDH)-mutant or IDH-wild type based on whether they harbour mutations in IDH1 or IDH2 [17]. Common mutations in IDH1 or IDH2 confer a gain of function, resulting in a non-canonical enzymatic activity converting α-ketoglutarate (AKG) into the oncometabolite D-2-hydroxyglutarate (D-2HG), which interferes with dioxygenase function leading to a prooncogenic hypermethylation phenotype [18]. Other common genetic alterations in GBMs include the gain of function of the epidermal growth factor receptor (EGFR) [19], or inactivating mutations in tumour suppressors such as the phosphatase and tensing homolog (PTEN), CDKN2A or TP53 [15]. Gliomas and GBMs usually reprogram Gln metabolism, as tumours become avid consumers of Gln [20]. In this regard, it is common the upregulation of glutaminase [21].

Glutaminase (GA; EC 3.5.1.2) is the enzyme responsible for catalysing the oxidative deamidation of Gln, producing glutamate and free ammonium in the first step in glutaminolysis [11,22]. Although there are many other essential enzymes employing Gln as a substrate and producing glutamate, as Gln amidotransferases, their action implies the transfer of the amide group of Gln to an acceptor molecule [23]. Cancer cells usually show much greater rates of ammonium generation and excretion compared to ammonium incorporation into biomass [24]. As the ammonium incorporation rate may be limiting the production of glutamate from Gln by these enzymes, GAs become principal enzymes for tumour cells, allowing for faster Gln-metabolizing rates [25]. Accordingly, GAs’ expression patterns are commonly altered in cancer [26]. In humans, two genes encode for GAs, GLS and GLS2. GLS encodes two isoforms, the larger kidney-type glutaminase (KGA) and the shorter glutaminase C (GAC), and both can be called indistinctly GLS isoforms. On the other hand, the GLS2 gene encodes for a canonical isoform, glutaminase B (GAB), and a shorter one called liver-type glutaminase (LGA), being both collectively named as GLS2 isoforms [27]. Expression of GLS is favoured by oncogenes such as c-Myc [28], while GLS2 was found to be a direct target of p53 [29]. Many tumours, including GBMs, overexpress the GLS isoenzymes, which correlates with malignancy, and therefore GLS isoforms have been considered prooncogenic factors [30]. Several GLS inhibitors have been investigated for potential application in therapy, including compound CB-839, which reached clinical trials [17,26,31]. On the other hand, GLS2 is silenced in several cancer cell lines compared with the normal tissue of origin, as it happens in hepatocellular carcinoma (HCC), colon carcinomas and GBMs, where GLS2 promoter was found to be hypermethylated [32,33]. GLS inhibition by CB-839 induced cell cycle arrest in GLS-highly expressing GBM stem-like cells [34], while slowed proliferation in T98-G and LN-229 GBM cell lines [35].

In this work, we investigated the metabolic changes upon restoring the expression of GLS2 in a GBM cell model, the human cell line LN-229, characterized by no detectable GLS2 expression but expressing GLS at high levels [36]. We developed cell models with stable overexpression of GLS2 and conducted a broad investigation of cellular metabolism including metabolomics and stable isotope tracing to study how GBM cell metabolism changes before and after GLS2 overexpression. We also included concomitant treatment with the GLS inhibitor CB-839, looking to distinguish isoenzyme-specific metabolic changes. These GLS2- or GLS-specific metabolic changes may be potentially of use in the frame of therapies targeting the metabolic rewiring in cancer, which will depend on specific molecular contexts.

2. Results

This section may be divided by subheadings. It should provide a concise and precise description of the experimental results, their interpretation, as well as the experimental conclusions that can be drawn.

2.1. GLS2 Overexpression Slows Cell Proliferation in Human LN-229 Cell Line and Confers Resistance to CB-839 Treatment

To investigate the effects of GLS2 expression on a human GBM cell model originally lacking GLS2 expression, we developed two independent cell models stably expressing GLS2 at different levels, checked by Western blotting using a specific antibody targeting GLS2 (Figure 1A). LN-229 wild-type cells or cells transfected with the empty vector to be used as a control (LN-EV) showed no detectable expression of GLS2 protein, while the two selected stably expressing models overexpressed GLS2 at a lower (called LN-GLS2(+) LO) or higher (called LN-GLS2(+) HI) levels (Figure 1A). A cell proliferation assay was performed to calculate the doubling time for each cell model (Figure 1B). The mean doubling time for LN-EV control cells was 30.6 h, while it increased to 34.8 h for the LN-GLS2(+) LO model, and to 41.2 h for the LN-GLS2(+) HI model. Thus, GLS2 expression increased the time required for cell duplication by 13.7% and 34.6%, respectively, following a dose-dependent pattern with GLS2 expression. When GLS inhibitor CB-839 was added at 1 µM, LN-EV control cells nearly doubled their doubling time compared to untreated cells (Figure 1C), while a slight increase (6.6%) was observed for LN-GLS2(+) LO, and no change was noted for LN-GLS2(+) HI. Additionally, a cell viability assay was performed employing a wide range of CB-839 concentration treatments for 72 h (Figure 1D). Increasing CB-839 concentration decreased the viability ratio of LN-229 wild-type cells and LN-EV control cells to 50%, with calculated IC50s of 47.4 nM for the wild-type cells and 33.3 nM for the LN-EV model. The viability ratio of GLS2-expressing models remained around 100% and only showed a decrease when CB-839 concentration was orders of magnitude higher (10 µM), but never dropped to 50% in viability.

2.2. GLS2 Alters Both Oxidative Decarboxylation and Reductive Carboxylation of Gln-Derived AKG in the TCA Cycle

Glutaminolysis is a main anaplerotic pathway in cancer cells. GAs convert Gln into glutamate, which may subsequently be deaminated by aminotransferases or by glutamate dehydrogenase (GDH) to generate AKG, an intermediate metabolite of the TCA cycle. Thus, the whole carbon skeleton from Gln may enter the TCA cycle in the form of AKG, which normally follows the oxidative pathway and is decarboxylated by AKG dehydrogenase complex to form succinyl-CoA and, subsequently, succinate, fumarate, malate and oxaloacetate, which may react with acetyl-CoA coming from glycolytic pyruvate to form citrate, and citrate may be isomerized to cis-aconitate and isocitrate, which can be decarboxylated to AKG by IDH, closing the cycle. Alternatively, cancer cells have been proven to commonly activate an alternative pathway for AKG, implying its reductive carboxylation to isocitrate in a non-canonical reaction catalyzed by IDHs. Gln metabolism has been shown to promote lipid biosynthesis by this pathway, since it allows for the generation of isocitrate and, later, citrate, which may furtherly serve to generate oxaloacetate and acetyl-CoA in the cytosol, the latter serving for lipid biosynthesis. To analyse how Gln anaplerotic metabolism is affected by GLS2 expression, cell models were supplied with Gln isotopically labeled in all its carbons with ¹³C ([U-¹³C]Gln). We also included the condition of treating the GLS2-overexpressing cell models with the GLS inhibitor CB-839 at 1 µM for 24 h, to account for an experimental condition in which GLS2 is being expressed and, at the same time, GLS activity is being inhibited, so most GA activity in these conditions would be due to GLS2. Metabolites generated from Gln carbon will inherit part or all of the ¹³C label. Thus, succinate, fumarate, malate and aspartate (which may be synthesized by transamination of oxaloacetate, using glutamate’s amino group) produced from Gln following the oxidative TCA cycle metabolism, will inherit the m+4 label, while citrate generated by reductive carboxylation of Gln-derived AKG will be m+5. Fumarate, malate or aspartate produced from reductive citrate will be m+3.

Figure 2 shows the fractional abundances of m+4 isotopologues from several metabolites related to the TCA cycle in the GLS2-overexpressing models compared to the LN-EV control. GLS2 expression increased the labeling from Gln in oxidatively-generated succinate, fumarate, malate and aspartate (Figure 2A). Surprisingly, the LN-GLS2(+) HI shows a 50% reduction in citrate m+4 compared to the LN-EV control or the lower GLS2-overexpressing model. When LN-GLS2(+) LO was treated with CB-839 at 1 µM (Figure 2B), fractional abundances of m+4 fumarate, malate and aspartate returned to the control values, still showing significantly higher labeling for succinate m+4. However, LN-GLS2(+) HI treated with CB-839 1 µM showed almost no change compared to untreated LN-GLS2(+) HI. Interestingly, m+4 citrate reduction did not change even quantitatively with the CB-839 treatment, showing the same differences against the LN-EV control.

Figure 3 shows fractional abundances for reductive carboxylation-associated labeling. Expression of GLS2 increased in a dose-dependent manner the labeling of m+3 fumarate, malate and aspartate (Figure 3A), and in the case of LN-GLS2(+) HI, a large increase in m+5 citrate (from 0.12 of LN-EV Control to 0.35). These changes in isotopologue abundances only showed a slight reduction in LN-GLS2(+) LO when treated with CB-839 1µM (Figure 3B), while treated LN-GLS2(+) HI showed no difference compared to the untreated model.

2.3. Effect of GLS2 Overexpression on the Metabolome

To further analyse the metabolic changes induced by GLS2 expression and by GLS2 expression and concomitant GLS inhibition by CB-839 (1 µM for 24 h), we carried out a broad LC-MS-based metabolomic analysis for the quantification of the levels of a total of 306 metabolites. Of all the metabolites detected and quantified, more than 100 changed their levels significantly in the comparison between the two GLS2-expressing models and LN-EV. When CB-839 1 µM was added, the treated models showed differential amounts of more than 80 metabolites. A hierarchical clustering analysis was performed using the web tool Metaboanalyst. To provide a general view of the metabolic changes induced by GLS2, Figure 4 shows a heatmap for the change in abundance of the top 50 metabolites in the comparison between LN-GLS2(+) HI and LN-EV, while Figure 5 shows the top 50 changing metabolites in LN-GLS2(+) HI +CB-839 1µM vs LN-EV.

Among changes in the metabolome, GLS2 expression was found to affect the levels of several nucleotides (Supplementary Figure S1). LN-GLS2(+) HI showed an increase of 2.3-fold for adenosine monophosphate (AMP), while guanosine monophosphate (GMP) and uridine monophosphate (UMP) levels dropped to 0.72 and 0.52-fold, respectively (Supplementary Figure S1A). LN-GLS2(+) LO showed an increase of 1.94-fold for AMP but showed no statistically significant change for GMP and UMP, although tending to drop (0.76 and 0.63, respectively) (Supplementary Figure S1B). When LN-GLS2(+) HI was treated with CB-839 1 µM, the increase in AMP was slightly attenuated (1.88-fold vs LN-EV), but GMP and UMP levels were quantitatively lower compared to the untreated model (0.64 and 0.40 vs LN-EV, respectively) (Supplementary Figure S1C). LN-GLS2(+) LO + CB-839 1 µM still showed an accumulation of AMP (1.75-fold), and, interestingly, showed a significant drop in GMP and UMP (0.53 and 0.49-fold, respectively) (Supplementary Figure S1D).

GLS2 expression was also shown to alter the levels of two methylated metabolites, 3-methyl-adenine (3-MA) and N,N,N-trimethyl-lysine (Supplementary Figure S2, panels A and B). LN-GLS2(+) LO caused 3-MA levels to rise to 1.6-fold, while higher GLS2 expression caused a further increment to 1.8-fold. Noticeably, treatment of both GLS2-expressing models with CB-839 caused 3-MA levels to go higher, as LN-GLS2(+) LO + CB-839 1 µM showed a 3.2-fold increase in 3-MA, while it was 3.87-fold for LN-GLS2(+) HI+ CB-839 1 µM (Supplementary Figure S2A). Both GLS2-expression and GLS inhibition by CB-839 made 3-MA levels rise. The pattern was quite similar for N,N,N-trimethyl-lysine (Supplementary Figure S2B), which raised to 1.6 and 1.9-fold for untreated LN-GLS2(+) LO and HI models, respectively. When CB-839 1 µM was added, it further increased to 1.7 and 2.3-fold, for LN-GLS2(+) LO+ CB-839 1 µM and LN-GLS2(+) HI+ CB-839 1 µM, respectively.

Interestingly, the levels of the universal methyl donor, S-adenosyl methionine (SAM) also changed when GLS2 was expressed at high levels (Supplementary Figure S2C). LN-GLS2(+) HI showed an increase in SAM levels of 2.3-fold vs LN-EV control, while CB-839 1 µM treatment did not make a difference compared to the untreated model, still showing a 2.2-fold increase vs LN-EV. LN-GLS2(+) LO either untreated or treated with CB-839 did not show a statistically significant difference in SAM levels vs control, however, a tendency may be observed for SAM levels to rise (1.28 and 1.21-fold vs LN-EV for LN-GLS2(+) LO and LN-GLS2(+) LO+ CB-839 1 µM, respectively).

However, GLS2 expression importantly affected glutaminolysis and the TCA cycle, and the metabolic changes related to these pathways will be presented more profusely in the following section.

2.4. GLS2 Reshapes Key Metabolic Pathways in Human GBM Cells

Levels of metabolites related to the TCA cycle significantly changed when GLS2 was overexpressed, as shown in Figure 6. In this regard, LN-GLS2(+) HI showed significantly lower levels of Gln (0.43-fold), while glutamate was higher (2.5-fold) (Figure 6A). Metabolites downstream of glutamate production by GAs also changed, as shown by the levels of AKG (2-fold), succinate (1.9-fold), fumarate (1.8-fold), malate (1.7-fold) and aspartate (2.4-fold). Noteworthy, citrate (0.18-fold), cis-aconitate (0.18-fold) and isocitrate (0.15-fold) levels were strongly reduced in LN-GLS2(+) HI, exhibiting a drop of more than 80% compared to the control. For the LN-GLS2(+) LO, smaller differences were obtained, however, Gln showed lower levels (0.77-fold) and glutamate (1.6-fold), succinate (1.6-fold) and aspartate (1.9-fold) levels were higher (Figure 6B). Nevertheless, this model also showed decreased levels of citrate (0.65-fold), cis-aconitate (0.65-fold) and isocitrate (0.54-fold), whose reduction was proportional to GLS2 levels, as it was greater with higher GLS2 expression and smaller with lower GLS2 expression.

LN-GLS2(+) HI treated with CB-839 1 µM for 24 h showed no changes compared to the untreated condition, as metabolites changed in the same way and amount compared to the LN-EV control (Figure 6C). When LN-GLS2(+) LO was treated with CB-839, the levels of Gln, glutamate, succinate and aspartate returned to the control ones, while malate levels where significantly lower (Figure 6D). Interestingly, the levels of citrate, cis-aconitate and isocitrate were still significantly lower compared to the control and did not show a quantitative change compared to the untreated LN-GLS2(+) LO model, not being affected by CB-839 treatment.

LN-GLS2(+) HI also showed a significant 1.5-fold increase in pyruvate levels vs control (Figure 7), and also in L-alanine (2.4-fold), which may be synthesized by transamination from pyruvate using glutamate as the amino donor. Interestingly, this pyruvate accumulation did not correlate with an increase in lactate, which showed a slight but significant decrease when GLS2 was expressed at high levels (0.90-fold). Additionally, O-acetyl carnitine was significantly decreased to 0.50-fold.

Besides, the levels of several N-acetyl amino acids were decreased when GLS2 was overexpressed, treated or not with CB-839 (Supplementary Figure S3), as it happened in all cases for N-acetyl aspartate, N-acetyl glycine or N-acetyl serine. LN-GLS2(+) HI showed greater differences compared with the control, with a significant decrease in the levels of N-acetyl alanine (0.59-fold), N-acetyl asparagine (0.47-fold), N-acetyl aspartate (0.15-fold), N-acetyl cysteine (0.46-fold), N-acetyl glycine (0.35-fold), N-acetyl leucine (0.65-fold), N-acetyl methionine (0.68-fold), N-acetyl phenylalanine (0.54-fold) and N-acetyl serine (0.23-fold).

2.5. GLS2 Modulates PDH Activity and Phosphorylation Pattern

Previously shown metabolomics results pointed out that GLS2 was causing a significant drop in the 6-carbon intermediates of the TCA cycle (citrate, cis-aconitate and isocitrate), which were not being affected by concomitant CB-839 treatment. While U-¹³C-Gln isotope-tracing data denoted that GLS2 was altering the way that citrate was being synthesized from Gln, as it showed a 50% reduction in citrate m+4 abundance, while citrate m+5 fraction was around 3-fold higher. We wondered if GLS2 could be negatively affecting PDH complex activity in an isoenzyme-specific way, which may explain lower 6-carbon intermediate levels of the TCA cycle, lower citrate m+4 abundance, since citrate m+4 generation implies Gln-derived oxaloacetate condensation with PDH-produced acetyl-CoA, as well as pyruvate and L-alanine accumulation. We developed new LN-229 overexpression models for GLS isoenzyme to test whether an increase in the levels/activity of any GA may cause these effects, looking to compare them with the GLS2-expressing models for distinguishing isoenzyme-specific activities. LN-GLS (+) models were generated similarly to LN-GLS2 (+) models, and two LN-GLS (+) with different GLS overexpression levels were selected (LN-GLS (+) LO and LN-GLS (+) HI) (Figure 8A). To test whether GLS overexpression was affecting cell proliferation, doubling time was evaluated for both GLS-overexpressing models as previously showed for the GLS2-overexpressing models. GLS-overexpression was found not to be causing a significant change in doubling time (Figure 8B), contrary to what happened with GLS2-overexpression (Figure 1B). To ascertain that GLS2 or GLS overexpression was indeed causing an increase in GA enzymatic activity, total GA activity (i.e., GA activity due to any isoform, endogenous and/or overexpressed) was measured for the GLS2- and GLS-overexpressing models, compared to the LN-EV control (Figure 8C). LN-GLS2(+) LO showed an increased of 2-fold for GA activity, while GA activity rose to 2.76-fold for LN-GLS2(+) HI. Similarly, GA activity increased to 2.31- and 3-fold for LN-GLS(+) LO and LN-GLS(+) HI, respectively. To analyse the potential crosstalk between GAs and PDH complex activity, a PDH enzymatic activity assay was performed for the GLS2 and GLS-overexpressing models (Figure 8D). Results showed that GLS2-expressing models (both LN-GLS2(+) HI and LN-GLS2(+) LO) were causing a significant reduction in PDH activity of 0.42- and 0.45-fold vs LN-EV control, respectively. LN-GLS (+) models showed no reduction in PDH activity, since LN-GLS (+) LO showed no statistically significant change compared to the control, while LN-GLS (+) HI showed a small but significant increase in PDH activity. To further ascertain if the GLS2 effect on the PDH complex’ function implied the modulation of its function through phosphorylation of the catalytic subunit PDHA1, PDHA1 phosphorylation status was studied for three serine residues known to downregulate PDH activity when phosphorylated: S232, S293 and S300. PDHA1 phosphorylation in these residues was analysed through Western blotting by using specific antibodies, as well as total levels of PDHA1 protein. For S232 and S300, no phosphorylation was detected at all in the LN-229 cell line (by comparison with the mitochondrial protein of HEK293-T cells used as a positive control; negative data not shown). However, phosphorylated S293 of PDHA1 showed a marked increase in the LN-GLS2 (+) HI model (Figure 8E) compared to LN-EV control, and also to LN-229 wild-type cells. LN-GLS (+) HI did not show increased amounts of phosphorylated S293 in PDHA1. Levels of total PDHA1 protein were similar for LN-229, LN-EV, LN-GLS2(+) HI and LN-GLS (+) HI (Figure 8E).

3. Discussion

Cell models expressing GLS2 showed an increase in their mean doubling times which was proportional to GLS2 expression levels (Figure 1A). Thus, GLS2-expression was able by itself to reduce the proliferation ability of these GBM cells, as it has been found for several GBM cell lines in previous studies [36,37]. However, when GLS was overexpressed in the same way as GLS2 in LN-229 cells, it did not cause significant changes in the proliferation capacity of these cells (Figure 8B), so we proved that this effect on the attenuation of cell proliferation was specific for GLS2, even though that overexpression of both, GLS2 or GLS, was causing an increase in total GA activity in the employed cell models (Figure 8C), which was quite similar among GLS2- and GLS-overexpressing models. These results, again, point to specific GLS2 differential functions that may or may not imply its GA catalytic activity. In the last years, reports have shown that GLS2 may be carrying out some activities that may not be related to its catalytic activity, being considered as potential secondary functions likely related to its protein interaction domains, as it was found to physically associate with Dicer in HCC cells to mediate stabilization and further maturation of miR-34a, resulting in oncogenic Snail repression and an anti-metastatic effect [38]. Interestingly, GLS2 has been reported to localize into the nucleus in GBM cells [37], which may point to roles in epigenetic regulation [39]. Here we showed that although GLS2 expression reduced the proliferation rate of GBM cells, it also made them insensitive to the anti-proliferative effects of GLS inhibition by CB-839 when GLS2 was expressed at sufficient levels (Figure 1C), rendering cells resistant to CB-839 treatment (Figure 1D), and so, being able to compensate for GLS activity inhibition regarding cell proliferation and viability. It seems clear that GLS2 possesses specific functions differentiating it from GLS isoenzymes and unravelling them may shed light on both its roles in physiological conditions and its impact on cancer.

In that regard, we studied how Gln carbon is handled by employing [U-¹³C]Gln when GLS2 is expressed (along with endogenous GLS expression), and also when GLS2 is expressed while CB-839 is used for GLS inhibition, an experimental condition looking to replace most of the endogenous GA activity due to GLS for GLS2 activity. GLS2-expression caused an increase in fractional abundances of succinate, fumarate, malate and aspartate m+4 (Figure 2A), derived from Gln through oxidative metabolism in the TCA cycle of Gln-derived AKG [35], which was expected since these models have higher global GA activity, increasing the rate of glutamate generation from Gln and so the metabolic flux into the TCA cycle. Accordingly, when CB-839 1 µM was added, fractional abundances of these metabolites tended to return to the control levels in the lower GLS2-expressing model, which did not express GLS2 at enough levels to fully compensate for GLS inhibition (Figure 1B). In contrast, citrate m+4, which is a product of oxidative Gln-derived oxaloacetate condensation with pyruvate-derived acetyl-CoA, was reduced when GLS2 was highly expressed (Figure 1A), as citrate m+4 fractional abundance was reduced by 50%, not being affected by CB-839 treatment (Figure 1B). All of this pointed to GLS2 causing a blockade in citrate synthesis through oxidative TCA cycle metabolism, which was not merely caused by an increase in GA activity, since GLS inhibition by CB-839 did not modify citrate m+4 abundance. On the other hand, GLS2 expression increased m+5 citrate (Figure 3A) from reductive carboxylation of Gln-derived AKG, and also of m+3 aspartate, malate and fumarate, which are derived from citrate m+5 [35]. Treatment with CB-839 (Figure 3B) only slightly attenuated this increase for the high GLS2-expressing model, while causing citrate m+5 fraction to return to control levels in the lower GLS2-expressing model. These results suggest that changes in reductive citrate synthesis from Gln were directly proportional to the total GA activity of the cell.

Metabolomic quantification of the levels of TCA cycle-related metabolites (Figure 6) perfectly complemented isotopic tracing results, showing lower levels of Gln, and higher levels of glutamate, AKG, succinate, fumarate, malate and aspartate when GLS2 was expressed at high levels (Figure 6A), while the 6-carbon intermediates of the TCA cycle, citrate, cis-aconitate and isocitrate, were largely reduced. LN-GLS2(+) LO, showed lower Gln, and higher glutamate, succinate and aspartate, but again lower levels of citrate, cis-aconitate and isocitrate (Figure 6B). Reduction in the levels of the 6-carbon intermediates was much accused with higher GLS2 expression (more than 80% lower), compared to the lower overexpressing model (30-40% lower). Integration of both isotope-tracing and metabolomic quantification information suggest that GLS2 expression is increasing the metabolic flux from Gln to the TCA cycle, rising the levels of most metabolites of the cycle, but at the same time causing a blockade in the synthesis of citrate by the oxidative way, denoted by significant reduction of citrate m+4, and global citrate levels (Figure 2A and Figure 6A,B, respectively), while citrate m+5 increase (Figure 3A) may be a consequence of both the hypothetical blockade in oxidative citrate synthesis and increased influx via glutaminolysis into the TCA cycle, both causing a relative increase in the fractional abundance of citrate m+5 compared to the total pool of citrate, which is largely reduced in these conditions. Besides, treatment with CB-839 caused almost no change in the levels of TCA cycle-associated metabolites in LN-GLS2(+) HI (Figure 6C), which proved to express GLS2 at levels sufficient to fully compensate for GLS inhibition. Interestingly, when the lower GLS2-overexpressing model was treated with CB-839 (Figure 6D), levels of Gln, glutamate, succinate and aspartate returned to the control levels and malate even dropped below control levels, while citrate, cis-aconitate and isocitrate levels were not affected, showing the same differences against the control that the untreated model showed (Figure 6B). Again, all of this suggested that the effect of GLS2 on citrate synthesis, and subsequent synthesis of cis-aconitate and isocitrate, is specific for GLS2 and not due to any change in GA activity.

Additionally, LN-GLS2(+) HI also showed an accumulation of pyruvate (Figure 7), the substrate of PDH for acetyl-CoA generation in the mitochondria. Acetyl-CoA is a substrate needed in the condensation reaction with oxaloacetate to form citrate in the TCA cycle. Oxaloacetate could not be quantified for technical reasons, however, since the immediate previous metabolite in the TCA cycle pathway, malate, is increased in LN-GLS2(+) HI, as well as aspartate, which is generated from oxaloacetate by transamination, it does not seem risky to hypothesize that oxaloacetate levels are probably higher as well when GLS2 is being overexpressed at enough levels. In this regard, it appears logical to consider that the GLS2-induced blockade of citrate synthesis could be mediated by a negative-modulatory effect of GLS2 on PDH enzymatic activity that would impair glycolytic acetyl-CoA production, thus making it limiting for citrate synthesis. Moreover, L-alanine, which may be synthesized by transamination from pyruvate, was also accumulated when GLS2 was expressed, while the increase in pyruvate levels did not correlate with higher levels of lactate, which was slightly reduced in LN-GLS2(+) HI. Acetyl-carnitine was equally reduced with high GLS2 expression (Figure 7). We hypothesize that altered acetyl-carnitine levels in these conditions may be reflecting changes in acetyl-CoA levels, as we have discussed in a previous publication [35]. Acetyl-carnitine takes part in an interchange reaction catalysed by carnitine acetyltransferase with acetyl-CoA, for the transfer of the acetyl group to carnitine to be translocated though cell membranes, including the mitochondrial inner membrane, which acetyl-CoA cannot make by itself [40]. Transfer of the acetyl group from acetyl-CoA to carnitine in the mitochondria is essential in some situations such as when the TCA cycle flux is compromised and acetyl-CoA accumulates and sequesters the mitochondrial pool of HS-CoA, preventing it from participating in other reactions [41]. LN-GLS2(+) HI showed a reduction in acetyl-carnitine levels of 50%, which as previously reasoned, can be hypothetically resembling lower levels of acetyl-CoA when GLS2 is expressed. In addition, the levels of several N-acetyl amino acids significantly changed with GLS2 overexpression (Supplementary Figure S3). Many free N-acetyl amino acids are originated from proteins that were acetylated at its N-terminal end and were subsequently degraded, while others may be synthesized directly in its free form, like N-acetyl aspartate. In any case, acetyl-CoA is the acetyl donor for the reaction [42]. We hypothesize that the generalized diminished levels of N-acetyl amino acids, with larger differences for the LN-GLS2(+) HI model, treated or not with CB-839, may be indicative as well for acetyl-CoA levels, given the fact that the levels of amino acids like glutamate, aspartate (Figure 6) or alanine (Figure 7) are significantly higher in these conditions, while its N-acetyl derivatives are diminished (Supplementary Figure 3), suggesting that acetyl-CoA levels may be limiting for the reaction.

To ascertain if there exists an actual crosstalk between GLS2 and the PDH complex, which may explain the previous metabolic changes, a PDH activity assay was performed looking to analyse endogenous PDH activity when GLS2 was expressed. To further confirm if this potential effect on PDH was due to a specific activity of GLS2, and not by other glutaminases, two cell models were developed to overexpress GLS, at lower and higher levels (Figure 8A), in the same way as the GLS2-overexpressing cell models. GLS2- and GLS-overexpressing models showed an increase in total GA activity which correlated with the expression levels of GLS2 and GLS, and this change in GA activity was similar between the higher and lower GLS2- and GLS-overexpressing models (Figure 8C). PDH activity was measured in these cell models and, as hypothesized, was greatly decreased when GLS2 was overexpressed in both models, with a reduction in PDH activity greater than 55% in both cases (Figure 8D). On the other hand, GLS overexpression did not cause lower PDH activity, since LN-GLS(+) LO showed no significant change while LN-GLS(+) HI exhibited even a slight increase in PDH activity. In any case, GLS2’s negative effect on PDH activity was found to be isoenzyme-specific. To further deepen the mechanism of action by which GLS2 was causing lower PDH activity, the phosphorylation status of three serine residues of PDHA1 was analysed by Western blot. Serine 293, serine 300 and serine 232 are known to be phosphorylated by pyruvate dehydrogenase kinases 1-4 (PDK1-4) to negatively regulate PDH activity [43,44]. Phosphorylated serine 232 and serine 300 of PDHA1 were not detected at all in the LN-229 cell line (negative data not shown), however, phosphorylated PDHA1 at serine 293 was found to be importantly increased when GLS2 was overexpressed (Figure 8E), compared to the LN-EV control and LN-229 wild type cells, while no changes were noted when GLS was being overexpressed. Total PDHA1 levels were parallelly analysed to confirm that variations in PDH activity and/or PDHA1 phosphorylation were not due to changes in protein levels. No significant alterations in PDHA1 levels were noted in any condition (GLS2- or GLS- overexpression) for LN-229 GBM cells (Figure 8E). Considering all of the above, GLS2 was found to be specifically causing an increase in phosphorylation of serine 293 of PDHA1, causing PDH activity to decrease by more than 50%. It can be hypothesized that this effect of GLS2 could be preventing glycolytic carbon from being incorporated into mitochondrial metabolism, minimizing the metabolic flux through the full TCA cycle and likely negatively affecting the flux in the electron transport chain and OXPHOS rate, which may be related to the lower proliferation capacity induced by GLS2-expression (Figure 1B). These metabolic traits could be associated with oncogenic phenotypes, as mitochondrial metabolism has been classically considered to be impaired in cancer [45]. However, it is nowadays evident that the picture is more complex and that most cancer cells and tumours need and rely on mitochondrial metabolism [3]. Noteworthy, the metabolic changes induced by GLS2 did not correlate with higher lactate levels (Figure 7), so although GLS2 could be limiting glycolytic flux into the mitochondria, it did not cause an increase in the Warburg effect (increased aerobic glycolysis and lactate generation).

Further research is needed to investigate the mechanism by which GLS2 induces PDHA1 phosphorylation and, hence, lower PDH complex activity. The PDH complex can be regulated at several levels and by various substrates [46]. However, phosphorylation of PDHA1 at serine residues 232, 293 and 300 by PDKs 1-4 are well-studied and established regulation mechanisms, while PDH phosphatases 1 and 2 mediate in its dephosphorylation and activation [47,48,49]. Phosphorylation of one of the three serine residues is enough to inhibit PDH complex activity [50,51]. It seems reasonable to think that GLS2 could be inducing one or more PDKs to increase the phosphorylation of S293 in PDHA1 or inhibiting a PDH phosphatase. Whether this inhibition of PDH activity by GLS2 is a generalizable characteristic of GLS2 in other tumour types or even in physiological conditions, or if by contrary is a specific activity restricted to GBM cells is another key question to be addressed. GLS2 was originally discovered in the liver [52], and its expression in physiological conditions is restricted to certain tissues/organs, like the brain, pancreas or liver, where it is highly expressed [53]. Given the fact that the liver is a gluconeogenic organ, it is tempting to speculate that PDH activity regulation by GLS2 could be related to GLS2’s homeostatic functions in normal physiological conditions in the liver. In this regard, GLS2 could be a more suitable GA isoenzyme for adapting to a gluconeogenic metabolism, favoring glutaminolysis while preventing pyruvate funneling toward the TCA cycle. In this hypothetical scheme, GLS2 could be providing gluconeogenic substrates by glutaminolysis such as oxaloacetate, which can be substrate of phosphoenolpyruvate carboxykinase (PEPCK) to produce the glycolytic intermediate phosphoenol pyruvate [54], while concomitantly inhibiting PDH function, and thus preventing both pyruvate utilization by PDH and further oxaloacetate condensation with PDH-derived acetyl-CoA. Due to PDH blockade and pyruvate accumulation, conversion of PEPCK-produced phosphoenol pyruvate to pyruvate would not be favored and, hence, GLS2 could then be promoting a net directionality towards gluconeogenesis and glucose synthesis. However, although this hypothetical role for GLS2 isoenzyme fits well with its preponderant expression in the liver, only lies now on the ground of speculation, and further efforts are needed to ascertain the role of GLS2 both in physiological conditions and in cancer.

GLS2 expression also affected the levels of several nucleotides (Supplementary Figure S1). LN-GLS2(+) HI showed an accumulation of AMP, while GMP was lower (Supplementary Figure S1A), LN-GLS2(+) LO also showed increased levels of AMP (Supplementary Figure S1B); AMP could be increasing because of higher AMP synthesis in the de novo purine nucleotide biosynthesis pathway due to higher aspartate availability caused by GLS2 expression (Figure 6). However, GMP, the other final product of the de novo purine biosynthesis pathway is significantly lower. It can be hypothesized that inosine monophosphate (IMP) could be preferentially directed to AMP synthesis in the final steps of the purine de novo biosynthesis pathway, to the detriment of GMP synthesis, since aspartate is needed to form adenylosuccinate from IMP, which will later give AMP. For GMP synthesis, IMP needs to accept the amide group from Gln to form xanthosine monophosphate, later giving GMP. So, increased aspartate levels and lower levels of Gln in GLS2-expressing models may point to this interpretation. However, when CB-839 was added, both models overexpressing GLS2 showed an accumulation of AMP and significantly decreased levels of GMP (Supplementary Figure S1C,D). Interestingly, LN-GLS2(+) LO showed no significant drop in GMP, but when CB-839 was added, GMP levels were reduced by almost 50% vs control. LN-GLS2(+) LO +CB-839 1µM showed that Gln and aspartate levels returned to the control levels (Figure 6D) since the increase in GA activity due to GLS2 expression was compensated by GLS inhibition with CB-839. Thus, total aspartate and Gln levels seem unlikely to solely explain changes in GMP and AMP, which could be due to some specific activity carried out by GLS2, since both GLS2 expression and GLS inhibition by CB-839 favoured lower GMP levels. It can be hypothesized that another possibility for AMP levels to rise is due to the effect of GLS2 on PDH activity, which compromises citrate synthesis from glycolytic acetyl-CoA and glutaminolytic oxaloacetate, and thus would be causing lower flux through the complete TCA cycle, minimizing NADH formation and therefore potentially impairing ATP synthesis by OXPHOS, which could result in a higher AMP/ATP ratio and so AMP accumulation. In this hypothetical interpretation, ATP levels would be lower due to GLS2 overexpression and, since hydrolysis of ATP is needed to synthesize GMP from xanthosine monophosphate, could explain lower GMP levels due to lower GMP synthesis rate. However, further research is needed to ascertain how bioenergetics and nucleotide biosynthesis are affected by GLS2.

On the other hand, LN-GLS2(+) HI showed a 48% reduction in UMP levels (Supplementary Figure S1A), which did not change significantly for LN-GLS2(+) LO (Supplementary Figure S1B). When CB-839 was added, UMP levels decreased significantly for treated GLS2-overexpressing models (Supplementary Figure S1C,D). Again, both GLS2 overexpression and GLS inhibition by CB-839 caused lower levels of UMP, so apparently GLS2 and GLS had opposing effects regarding UMP levels. This result was unanticipated since aspartate levels are higher when GLS2 is overexpressed, and aspartate is essential at the start of the pyrimidine de novo biosynthesis pathway. In a previous publication by authors [35] where we analysed the metabolic changes following CB-839 treatment in three GBM cell lines, including LN-229, CB-839 alone caused a reduction of UMP which perfectly correlated with lower aspartate levels. However, it is hard to anticipate now which changes induced by GLS2 could be causing depletion of UMP, since aspartate levels are not diminished but are quite higher in these conditions.

Lastly, GLS2 expression also caused an increase in the levels of the methylated metabolites 3-methyl-adenine (3-MA) (Supplementary Figure S2A) and N,N,N-trimethyl lysine (3meLys) (Supplementary Figure S2B). 3-MA levels correlated with GLS2 expression, increasing to 1.56-fold and 1.8-fold for the lower and higher GLS2 overexpressing models, respectively. 3meLys showed the same tendency, with increases of 1.63-fold and 1.86-fold for both models. As previously discussed by authors [35], 3-MA may appear as a spontaneous modification of DNA, being considered an aberrant methylation mark to be repaired by alkyl-DNA glycosylases (also called 3-methyladenine DNA glycosylases), which recognize and repair this modification through the base excision repair mechanism, releasing free 3-MA [55].

On the other hand, multiple lysine residues in histones are targets for methylation by lysine methyltransferases, which have essential roles in epigenetic regulation by altering chromatin accessibility [56]. Proteolytic degradation of methylated histones has been identified as the primary source of free 3meLys. 3meLys is implicated in the carnitine biosynthesis pathway, being the starting metabolite for four sequential reactions giving carnitine as the final product. Two of these reactions require AKG as a cosubstrate, which is oxidized to succinate [57]. Since AKG is higher when GLS2 is being overexpressed, it does not suggest that carnitine synthesis could be impaired, which would potentially cause 3meLys accumulation. Thus, free 3-MA and 3meLys could be interpreted as indicators of overall methylation levels in DNA and proteins, including histones, respectively. Concomitant treatment with CB-839 largely increased 3-MA accumulation for both models (Supplementary Figure S2A) and also made 3meLys’ levels go higher (Supplementary Figure S2B). Both GLS2 expression and GLS inhibition caused 3-MA and 3meLys to accumulate, hence pointing to opposing effects for GLS2 and GLS in this regard.

GLS2 expression has been previously reported to downregulate the phosphatidylinositol 3-kinase/protein kinase B (PI3K/AKT) signalling pathway in HCC [58] and GBM [36] cell lines, which has been related to a tumour suppressor function for GLS2 in these contexts, correlating with lower proliferation rates and increased sensitivity to oxidative stress [37,60]. Interestingly, the PI3K-AKT pathway has been shown to regulate oxygen metabolism in human head and neck cancer cells through a mechanism involving phosphorylation of PDH [61]. The authors demonstrated that inhibition of this pathway led to increased PDHA1 phosphorylation thus decreasing its activity and O₂ consumption which allows a reduction in tumor hypoxia [61]. In addition, 3-MA has been used as an autophagy inhibitor, which is achieved through downregulation of the PI3K/AKT pathway [62]. Although further research is needed to confirm if endogenous levels of 3-MA can act in this sense, speculations can be made linking increased 3-MA levels when GLS2 is overexpressed and previously noticed downregulation of the PI3K/AKT pathway related to GLS2 expression, suggesting a mechanistic link through 3-MA for GLS2 effect on the PI3K/AKT pathway. In this regard and considering previous reports on head and neck cancer cells [61], it may be hypothesized that the effect of GLS2 on the PI3K/AKT pathway may be related to our finding linking GLS2 and PDH activity downregulation.

Noticeably, SAM levels were significantly higher in the LN-GLS2(+) HI model (Supplementary Figure S2C), while the lower GLS2-expressing model showed a statistically non-significant tendency of SAM to go higher. However, this effect on SAM levels is interpreted to be specific for GLS2, since concomitant CB-839 treatment caused no change in SAM levels, as treated LN-GLS2(+) HI showed no quantitative differences compared to the untreated model. We have shown previously that CB-839 treatment in several GBM cell lines could cause 3-MA, 3meLys and also 5-methyl cytosine levels to increase [35], which was thought to be a consequence of a relative imbalance in AKG and succinate levels that could compromise α-ketoglutarate-dependent dioxygenase function and thus, demethylation rates. In this case, since GLS2 expression causes higher levels of both AKG and succinate, and no relative imbalance between these metabolites was found, the potential effect of GLS2 on overall methylation levels is thought to be mediated by increased SAM levels. SAM is a critical methyl donor for multiple reactions, being a needed substrate for DNA methyl transferases and for lysine methyl transferases that catalyse DNA and histone lysine methylation, respectively. Higher SAM availability could by itself be causing higher overall methylation of DNA and proteins [63]. However, the mechanism by which GLS2 could be causing an accumulation of SAM remains to be elucidated.

Future research work is needed to ascertain how generalizable are the molecular and metabolic changes induced by GLS2, as well as the mechanisms behind them, which could imply or not the catalytic GA activity of GLS2, its interactome or its subcellular location. Previous publications by our group found that GLS2 could be located at the nucleus of GBM cells [37], as well as linked GLS2 function with oxidative stress handling [58] and microRNA regulation [59], also noted previously in HCC by other groups [38], while also being reported modulation of cell signalling pathways by GLS2 in HCC and GBM [36,57]. For all of the above, it is not to be discarded that the GLS2-PDH crosstalk and the effect on the TCA cycle, nucleotide biosynthesis or methylation modulation may be a consequence of epigenetic and/or protein function regulation at different levels carried out by GLS2. Understanding GLS2 functions will shed light on its roles in physiological conditions and in cancer, where the aim is to identify GLS2 expression as an indicator of certain molecular vulnerabilities to be exploited in future targeted therapies.

4. Materials and Methods

4.1. Chemicals

Dimethyl sulfoxide (DMSO), CB-839 (Telaglenastat), methoxyamine hydrochloride, pyridine, tert-butyldimethylsilyl ether (TBDMS) and G418 were obtained from Sigma-Aldrich Co. (St. Louis, MO, USA). U-¹³C-Gln was purchased from Cambridge Isotope Laboratories (Tewksbury, MA, USA). Hoechst and propidium iodide were from ThermoFisher Scientific (Waltham, MA, USA).

4.2. Cell Lines, Culture Conditions and Stable Transfections

LN-229 human GBM cell line was kindly provided by Dr. Monika Szeliga, Department of Neurotoxicology, Mossakowski Medical Research Centre, Polish Academy of Sciences, Warsaw, Poland. Cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% foetal bovine serum, 100 I.U./mL penicillin, 100 μg/mL streptomycin, and 4 mM L-Gln, as previously described by authors [59]. For stable overexpression experiments, LN-229 cells were seeded in 6-well plates and transfected with a pcDNA3 vector (GenScript Biotech, Piscataway, NJ, USA) containing the Neomycin resistance gene and the full-length cDNA of GLS2 (for the GLS2-overexpressing models), the full-length cDNA of the GAC isoform of GLS (for the GLS overexpressing models), or the empty-vector (for control cells), employing Lipofectamine 3000 (ThermoFisher Scientific). Transfected cell pools were incubated for 48 h and then supplemented with 0.75 mg/mL G418 (Sigma-Aldrich) for the selection of transfected cells. Cell media supplemented with G418 was renewed every two days for 7–10 days to allow for the selection and limited expansion of transfectants (not allowing cells to reach high confluence percentages). After this time, cells were detached, and individual cells were isolated by seeding them in 96-well plates by limiting dilution cloning. Wells were checked subsequently to contain individual cells and wells containing no cells or more than one cell were discarded. Monoclonal populations were expanded for 4–6 weeks using media supplemented with G418. Total protein was extracted from each monoclonal population generated to verify the expression of the protein of interest (GLS2 or GLS) by Western blotting using specific antibodies. Two independent cell models were selected for each condition (overexpression of GLS2 or GLS), having different expression levels of the protein of interest. Cells were transfected with the empty vector following the same procedure as the overexpressing models and were employed as controls. LN229-GLS2(+) (including the lower GLS2-expressing model called LN-GLS2(+) LO and the higher GLS2-expressing model called LN-GLS2(+) HI), empty vector-transfected cells LN229-EV (named LN-EV) and LN229-GLS(+) (including the lower GLS-expressing model called LN-GLS(+) LO and the higher GLS-expressing model called LN-GLS(+) HI) were maintained in DMEM routinely supplemented with 0.75 mg/mL G418 for continuous selection of stable transfectants, preventing spontaneous loss of expression of the protein of interest. Stable overexpression of GLS2 or GLS was routinely checked by Western blot, proving stable expression over time. All cultures were maintained at 37°C in a humidified atmosphere with 95% air and 5% CO₂. Cells were routinely checked to be free of mycoplasma contamination.

4.3. Cell Proliferation and Viability Assays

For studying changes in cell proliferation, 4 × 10⁴ cells of each experimental or control (LN-EV) condition were seeded in triplicate in 12 well plates. Cells were allowed to grow for 96 h and then detached and counted using trypan blue as vital staining for discriminating live or dead cells. 4 × 10⁴ cells of each replicate were independently reseeded and the process was repeated for 5 cycles, so each replicate of each condition grew independently for a total of 20 days. The cell number in each counting was referred to initial seeding number to calculate a mean doubling time for each condition.

To analyse cell viability, 2.5 × 10³ cells for each cell model were seeded in 96 well plates. CB-839 was added in a concentration range from 0.01 nM to 10 µM. DMSO was used as a vehicle-control to compare with, considering it the 100% viability condition. Each condition was assayed in quadruplicate. After 72 h of treatment, the culture medium was carefully replaced with medium supplemented with Hoechst at 1 µg/mL and propidium iodide at 1 µg/mL (ThermoFisher Scientific) for total and dead cell staining, respectively. The plate was covered and incubated at 37 °C for 30 min, and then was analysed in a Celigo Image Cytometer (Nexcelom Bioscience, Lawrence, MA, USA). The viable cell number was calculated for each condition and a viability ratio was calculated by considering the DMSO-treated cells as the 100% viability (viability = 1). Inhibitory concentration 50 (IC50) (viability ratio = 0.5) was calculated, when possible, by referring the viability ratio to the CB-839 concentration.

4.4. Metabolomics

For extraction of the cell metabolome, samples were processed as previously described [35]. Briefly, cells were cultured in 6 cm dishes, and when they reached 80–90% confluence, the cells were washed with ice-cold normal saline solution on ice, and then 1 mL of −20 °C cold 80% v/v methanol was added per dish. The cells were scraped on ice, and the suspension was transferred to microcentrifuge tubes and subjected to three freeze-thaw cycles between liquid nitrogen and a 37°C water bath. Lysates were centrifuged at 4 °C for 15 min at 17,000× g, and the metabolite-containing supernatant was transferred to new microcentrifuge tubes and dried overnight in a Speed-Vac concentrator (Thermo Fisher Scientific, SPD1030), keeping the tubes open and applying breathable membranes to avoid cross-contamination. Metabolite-containing pellets were redissolved and transferred to liquid chromatography-mass spectrometry (LC-MS) autosampler vials and injected on a Quadrupole Time of Flight (QTOF) LC-MS analyser (Agilent Technologies, 6550 iFunnel, Santa Clara, CA, USA). The total ion count for each sample was employed to normalise metabolite abundances, and a fold-change was calculated by referring experimental normalised abundances to the control ones.

4.5. [U-¹³C]Glutamine Tracing Experiments

Cells were cultured with or without CB-839 and when reached 80% confluence, cells were washed with PBS twice, and fresh medium containing labeled Gln with or without CB-839 treatment was added to each dish. Then, cells were traced for 6 h, washed with ice-cold normal saline and metabolites were extracted in methanol and processed as described before. Dried metabolite pellets were incubated with 1% methoxyamine hydrochloride in pyridine for 15 min at 70 °C, following derivatization with TBDMS for 1 h at 70 °C. The volume was transferred to gas chromatography-mass spectrometry (GC-MS) autosampler vials, and 1 µL was injected into an Agilent 7890A gas chromatograph coupled to an Agilent 5975C mass selective detector (Agilent Technologies). The results were manually reviewed, and data were expressed as the isotopologue fractions of the total metabolite.

4.6. Protein Expression and Mitochondria Isolation

GLS2 or GLS (GAC isoform) overexpression was assessed by Western blot (Figure 1A and Figure 8A). Cells were washed in PBS and harvested in lysis buffer (RIPA buffer: 1% NP-40, 0.1% sodium deoxycholate, 0.1% SDS, 150 mM NaCl, 50 mM Tris-HCl pH 8.0, supplemented with EDTA-free Halt™ (100X) Protease Inhibitor Cocktail, (ThermoFisher Scientific). Lysates were vortexed, incubated on ice for 5 min, and then centrifuged at 12,000× g, 15 min at 4 °C. Supernatants were collected and assayed or stored at −80 °C. Protein quantification was performed using Pierce™ BCA Protein Assay Kit (ThermoFisher Scientific). Twenty micrograms of total protein were resolved on 10% polyacrylamide-SDS gels and then electrotransferred to nitrocellulose membranes. Membranes were blocked for 1 h at room temperature with 5% non-fat milk or bovine serum albumin (BSA) in Tris-buffered saline-Tween 20 (TBS-T). All antibodies were diluted in 5% BSA in TBS-T. For GLS detection, a primary antibody targeting specifically the GAC isoform was used. This antibody was generated in rabbit against a peptide containing the 20 C-terminal amino acids of human GAC and has been previously demonstrated to react specifically with the GAC isoform of GLS, and not to react with other glutaminase isoforms [59]. For GLS2 quantification, an antibody against GLS2 (ab113509) was purchased from Abcam (Abcam, Cambridge, UK), and used at 1:2000 dilution. An antibody targeting β-actin (ab8227, Abcam) was used at 1:5000 for employing the β-actin signal as a loading control. To analyse pyruvate dehydrogenase E1 subunit alpha 1 (PDHA1) levels and its phosphorylation status, mitochondria were isolated from cells in culture using the mitochondria isolation kit for cultured cells (89874, ThermoFisher Scientific). Isolated mitochondria were lysed with RIPA buffer supplemented with protease (87785, ThermoFisher Scientific) and phosphatase inhibitors (78420, ThermoFisher Scientific), and mitochondrial protein was quantified using the Pierce™ BCA Protein Assay Kit. 15 µg of mitochondrial proteins were loaded and resolved on 15% polyacrylamide-SDS gels and electrotransferred to PVDF membranes. Membranes were blocked for 1.5 h at room temperature with 5% BSA in TBS-T. PDHA1 phosphorylation was studied at three serine residues (AP1063, Sigma-Aldrich, for p-Ser232, 1:500 dilution; ab177461, Abcam, for p-Ser293, 1:1000 dilution; and 29583-1-AP, ThermoFisher Scientific, for p-Ser300, 1:1000 dilution), as well as total PDHA1 (ab168379, Abcam, 1:1000 dilution). COX IV was used as a loading control for mitochondrial protein, using a specific antibody (ab16056, Abcam, 1:5000 dilution). All incubations with primary antibodies were performed overnight at 4 °C in agitation. After incubation with respective primary antibodies, the membranes were washed in TBS-T and incubated for 1 h at room temperature with a secondary HRP-conjugated Goat anti-rabbit antibody (A0545, Sigma-Aldrich, 1:40,000 dilution) and visualized in a ChemiDoc™ Gel Imaging System (Bio-Rad, Hercules, CA, USA), using an enhanced chemiluminescence detection system (SuperSignal™ West Pico, ThermoFisher Scientific). For data quantification in the case of PDHA1/COXIV Western blots, Image Lab Software (Bio-Rad) was used, following the manufacturer’s instructions. In brief, each band was detected and selected, and the background signal was discarded. After that, net values corresponding to the signal of the protein of interest (phosphorylated or total PDHA1) band were normalized to its respective loading control (COX IV) signal (Supplementary Figure S4).

4.7. Glutaminase Enzymatic Activity

Glutaminase activity was estimated in the experimental models overexpressing GLS2 (LN-GLS2(+) LO and LN-GLS2(+) HI) or GLS (LN-GLS(+) LO and LN-GLS(+) HI) by quantification of the ammonium produced in the GA reaction. The method was adapted from the original method described by McGee & Knox [64] and modified by Heini et al. [65], based on the reaction of ammonium and o-phthalaldehyde (OPA), which may be further quantified by spectrophotometry or fluorescence. In brief, about 4 × 10⁶ cells of each condition were cultured and detached using trypsin when reached 80% confluence. Cells were centrifuged and washed twice with PBS, then cell pellets were resuspended in 300 µL of 150 mM K₂HPO₄, pH 8.6. Samples were subjected to two freeze-thaw cycles and vortexed for 1 minute to completely break cell membranes. Samples were centrifuged at 12,000× g, 5 min at 4 °C and supernatant was collected. Total protein was quantified using the Pierce™ BCA Protein Assay Kit (ThermoFisher Scientific) for further normalization. 40 µL of extracts were loaded in 96-well plates, in triplicate, and 56 µL of reaction buffer (171 mM L-Gln, 150 mM K₂HPO₄, pH 8.6) were added and mixed. Plates were incubated for 1 h at 37 °C and then the reaction was stopped by adding 10 µL of 10% v/v trichloroacetic acid. Blanks were included for each sample by incubating sample extracts and reaction buffer separately and mixing just before trichloroacetic acid addition. For quantification of the ammonia generated in the GA reaction, 30 µL of each reaction were transferred to new wells and 150 µL of the OPA reagent (8.6 mM OPA, 3.33 mM β-mercaptoethanol, 160 mM K₂HPO₄, pH 7.4) was added. A standard curve of NH₄Cl ranging from 2.5 mM to 75 µM was assayed in parallel. The plate was covered and incubated for 45 min, at 37 °C in the dark, and then absorbance was measured at 410 nm in a plate reader. The absorbance of the blanks was subtracted from the samples to account for ammonia generation not due to GA reaction, and net absorbance values were used to calculate GA activity by referring them to the NH₄Cl standard curve. Values were normalized to total protein content and referred to the control to calculate a Fold-Change. Four independent experiments were performed.

4.8. PDH Enzymatic Activity

PDH activity was quantified using the Pyruvate Dehydrogenase (PDH) Enzyme Activity Microplate Assay Kit (ab109902, Abcam). Samples were processed and assayed according to the manufacturer’s instructions. Briefly, cells were cultured in 150 mm dishes, employing at least 2 × 10⁷ cells per condition. When cells reached 80% confluence, the medium was discarded, and dishes were washed twice with ice-cold PBS on ice. Cells were scraped in ice-cold PBS and samples were transferred to microcentrifuge tubes. Cells were centrifuged at 300× g, 5 min at 4 °C, and the cell pellet was resuspended in 500 µL ice-cold PBS supplemented with phosphatase inhibitors. A small aliquot was taken for protein quantification and the cell suspension was accordingly adjusted to the equivalent of 15 mg/mL of protein. After that, detergent was added for cell lysis, and lysates were centrifuged at 1000× g, 10 min a 4 °C, transferring supernatant to clean tubes. The extract was diluted to 2–5 mg/mL of protein and 200 µL were loaded in 96-well plates functionalized with monoclonal antibodies against PDH for intact PDH complex immobilization. Samples were incubated for 3 h at room temperature to allow for binding of the PDH complex. After that, the volume was discarded and wells were washed twice; then, a substrate/chromogen solution was added, and the plate was kinetically read at 450 nm in a plate reader for 30 min. Data was analyzed according to the manufacturer’s instructions, by comparing absorbance increase rate between experimental and control samples, normalized to protein content.

4.9. Statistical Analysis

Unless specifically noted otherwise, at least three independent experiments were performed in duplicate; for tracing experiments three independent experiments were made in triplicate; metabolomics was made twice in triplicate. Heatmaps showing the change in abundance of the top 50 altered metabolites were generated by using the web tool Metaboanalyst, following PLS-DA analysis. Dot and bar graphs and statistical analysis (Student’s t-test for experimental against control samples) were made using GraphPad Prism software. Values were expressed as means ± standard error of the mean (SEM). p < 0.05 was considered statistically significant. Statistical significance was expressed using asterisks, as follows: * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.

5. Conclusions

This study focused on the metabolic changes induced by ectopic GLS2 expression in GBM cells, which are characterized by high levels of GLS but lack appreciable GLS2 expression [25]. We have identified for the first time specific crosstalk between GLS2 and PDH, placing GLS2 as a potential key regulator of mitochondrial metabolism, linking glucose and glutamine metabolism, essential metabolic pathways for cancer cells [66]. This finding may explain some of the tumour suppressor functions associated with GLS2 in several contexts, including GBM models [11,36]. Additionally, GLS2 induced accumulation of AMP, and depletion of GMP and UMP, likely reflecting alterations in nucleotide biosynthesis and bioenergetics, which could not be compensated through inhibition of endogenous GLS by CB-839, even though when aspartate and glutamine overall levels where not altered. Besides, GLS2 induced an accumulation of methylated metabolites, potentially indicating higher overall methylation in DNA and proteins, including histones, which was further increased by concomitant GLS inhibition, reflecting the opposite effects of GLS and GLS2. GLS2 induced an accumulation of SAM that was not affected by GLS inhibition, which appears as the mechanistic reason for increased hypothetical methylation. However, the mechanism by which GLS2 may be increasing SAM levels, downregulating GMP and UMP, and increasing phosphorylation of serine 293 of PDHA1 remains unknown. Our findings paved the way for further research regarding deeper mechanistic studies concerning the implication of GLS2 in PDH modulation, nucleotide metabolism and epigenetic regulation. Whether these activities for GLS2 take place in other tumour types, or even are related to the role of GLS2 in normal physiological conditions remains a key open question to be solved.