In one of the first assessments of CSC metabolic states, Vlashi et al. studied the oxygen consumption rate and external acidification rate of GSC neurospheres derived from three independent GBM samples, compared with differentiated progenies cultured as monolayers (81). The group found that GSCs exhibited lower glucose uptake rates, lower lactate production, higher ATP levels and were therefore less glycolytic. Moreover, the group proposed that differentiation could induce a switch from a dominant use of OXPHOS for ATP production to glycolytic dependency (81). By contrast, GSCs derived in the same way from U87 tumours by Zhou et al. showed no significant difference in ATP levels in hypoxic versus normoxic conditions (82). When assessed under exclusive normoxia, elevated GSC lactate production was observed as well as a 4-fold increase in glucose uptake, suggestive of predominant glycolysis (82). Rationalisation of this result with findings from other GSC studies include the increased expression of neuronal GLUT3 by these cells for higher affinity glucose uptake, with GLUT3 knockdown in studies by Flavahan et al. conferring a significant decrease in the growth of GSCs (83). In addition, Zhou et al. showed that GSCs derived from two GBM surgical specimens expressed lower levels of voltage-dependent anion channel 2 (VDAC2) than the differentiated phenotype, believed to be essential for preservation of stem cell features, tumorigenicity and phosphofructokinase (PFKP) mediated glycolysis (84).

By collectively interpreting contrasting results from the GSC studies discussed, more detailed analysis reveals additional complexity, suggestive of heterogenous metabolic phenotypes within a single tumour (85). For example, Hoang-Minh et al.’s separation using CellTrace dyes and metabolic characterisation of GBM cell cycling speeds relative to bioenergetic strategy indicated that slow cycling cell (SCC) tumour populations, believed to be enriched in GSCs, were shown to survive predominantly using OXPHOS and lipid metabolism (86). SCCs derived from primary human GBM cell lines also displayed elevated chemoresistance and invasive capacity (86). Furthermore, a multitude of studies have recorded GSC metabolic subtype disparity, in which Gene Set Variation Analysis (GSVA) of TCGA GBM subtypes reveal that MES GBMs predominantly correlate with glycolytic pathways (87). Moreover, the metabolic enzyme analyses of MES GBM have revealed preferential glycolytic enzyme expression such as aldehyde dehydrogenase 1A3 (ALDH1A3) (87). This metabolic heterogeneity has been further highlighted by Duraj et al.’s application of 4 different metabolic drugs (metformin (MF), dichloroacetate, sodium oxamate and diazo-5-oxo-L-norleucine) to three GSC types, shown to exhibit different drug sensitivities due to initial cellular glycolytic/oxidative tendencies (88).

The employment of updated techniques to directly quantify metabolites for metabolic flux analysis in vitro and in vivo has also catalysed an increased acceptance for metabolic plasticity, challenging the categorisation of discrete malignant metabolic phenotypes pioneered by Warburg  (89). In more contemporary studies, Shibao et al. derived isogenic glioma initiating cells (GICs) from neural stem cells (NSCs) expressing the H-Ras^V12 oncoprotein and showed that orthotopic primary tumour initiation was independent of initial cellular metabolic state and glycolytic enzyme expression (90). Furthermore, clonally derived GSCs from the same GIC 14 days post in vivo initiation showed tumour sustenance of metabolic diversity, thought to demonstrate the possibility for coexistence of GSCs with different bioenergetic strategies within the same tumour, relative to their environmental niche (90). Transcriptional up/downregulation of the glycolytic enzymes HK2, PKM2, LDHA and PDK1 were also recorded following the observation that cells were able to glycolytically compensate on exposure to hypoxia, and vice versa using OXPHOS after glycolytic inhibition, indicative of cellular metabolic coping mechanisms for continued biosynthesis (90).

Regardless of the potential that the Warburg hypothesis brought for glycolytic inhibitor targeting, cancer cell survival and resistance remains a barrier to therapeutic success. As time has passed, the elucidation of a tangled cellular metabolic network has precipitated increasing futility of unimodal targeting strategies excluding tumoral heterogeneity and TME remodelling. Therefore, it is important to consider how the glycolytic pathway fits into a larger picture of cellular metabolism ( Figure 3 ) and thus, isolated targeting of this pathway may be increasingly misjudging the tangled network of cell coping mechanisms. DeBerardinis et al. noted that simply observing a heavy cellular reliance on aerobic glycolysis may be accessory to a temporary metabolic strategy, facilitating the production of proliferative precursors as opposed to a complete impairment of the oxidative pathway (91). The group also illustrated that the catabolism of glutamine in addition to other carbon sources such as acetate shown by Mashimo et al. may be an imperative feature of Warburg’s observed cancer cell phenotype, ‘picking up’ the burden of TCA anaplerosis and nicotinamide adenine dinucleotide phosphate (NADPH) synthesis for continued cell maintenance (91, 92).

Glutamine is an abundant plasma nutrient essential for cellular catabolic regulation, flux and processing of carbon, nitrogen and reducing equivalents (91). Integrating results from the study of GBM, other cancers and neurodegenerative disorders, a crucial balance between glutamine synthesis and catabolism has been recognised (93). Glutamine uptake by the alanine/serine/cysteine transporter 2 (ASCT2) transporter is followed by cytoplasmic glutaminase (GLS) conversion to glutamate and subsequent fate processing dependent upon cell requirements. Glutamate can be directly processed to the antioxidant glutathione via glutamate-cysteine ligase (GCL) enzymatic combination with intracellular cysteine or exported though through xCT (cysteine/glutamate transporter) subject to cysteine import (94). The GBM xCT mediated export of glutamate has been shown to confer tumoural invasive expansion of rat striata implanted glioma cell clones with staining showing neuronal degeneration and inflammation of the surrounding TME (95). Moreover, immunoprecipitation analysis of three glioma cell lines and two primary human GBM cells by Tsuchihashi et al., revealed the control of xCT surface expression was subject to direct interaction with the EGFR intracellular domain, a mechanism which has since been further delineated by the group  (96, 97). Glutamate is also instrumental for non-essential amino acid synthesis as well as purine and pyrimidine nucleotide precursors (94). Furthermore, glutaminolysis for the continued TCA flux and maintenance of biosynthetic intermediates has been a prominent feature in the study of cancer transformation.

In the healthy brain glutamine production via the glutamine synthetase (GS) catalytic condensation of glutamate and ammonia exclusively by astrocytes can be cooperatively utilised by neuronal GLS for glutamate hydrolysis and subsequent neurotransmission; termed ‘the glutamine-glutamate cycle’ (98). The ammonia utilised by GS is a product of amino acid breakdown and this enzyme is therefore essential for detoxification as well as uncontrolled neurotransmission through clearing glutamate from the synaptic cleft. However, importantly astrocytic de novo glutamate synthesis for neurotransmitter pool maintenance is dependent upon the enzyme pyruvate carboxylase (PC) and has been used as a specific marker of astrocytes. PC shuttling of glucose-derived pyruvate into the TCA for glutamate synthesis is subsequently converted to glutamine through GS.

An unmistakable metabolic adaptation drawn from glioma studies including GBM is an ‘addiction’ to glutamine for survival and growth, particularly in hostile conditions. In vitro experiments by Wise et al. using the paediatric GBM cell line SF188 revealed glutamine growth dependence, such that removal of glutamine from the culture medium inhibited cell survival despite the presence of glucose (99). This finding is well established for other cells including cancer with many early studies reporting cell survival in glutamine and nucleoside supplemented culture media in the absence of any sugar (100–102). Furthermore, inhibition of the established oncogenic drivers of glucose addiction PI3K and Akt by Wise et al. showed no effect on glutaminolysis mediation, precipitating the finding that elevated MYC expression resulted in upregulation of glutamine dependence genes (99). Myc was shown to stimulate increased SLC1A5 (ASCT2) transcription for higher rates of uptake and glutamine mediated TCA anaplerosis, essential for the replenishment of precursors for growth limiting macromolecule synthesis (99). This data implied the possibility that anaplerotic nutrient use was driven by discrete oncogenic systems. Other studies to further analyse high MYC expression and nutrient dependence in a range of human cell lines including human lung fibroblasts and human lymphoblastoid cells have found that Myc dependent miR-23a and miR-23b transcriptional repression upregulates GLS expression as well as sensitising cells to apoptosis during glutamine deprivation (103, 104).

Besides studies implicating glutamine as an extremely important molecule for TCA anaplerosis and cell survival, it is important to recognise that a balance exists between glucose, glutamine, and other cellular respiratory substrates. As mentioned for astrocytic systems, glucose can be catabolised to pyruvate and diverted away from PDH mediated entrance into the TCA, instead being carboxylated to oxaloacetate by PC. Nonetheless, observations consistent with the Warburg effect have rarely implicated glucose as the predominant precursor for neoplasm anaplerosis, supported by findings from metabolic flux experiments in well established “glutamine addicted” SF188 paediatric GBM cells with negligible PC activity (91). In fact, in vitro SF188 metabolic flux analysis has shown near exclusive nutrient favouring, with DeBerardinis et al. stating that cells derive up to 90% of anaplerotic oxaloacetate production from glutamine (91). Nonetheless, studies in human lung cancer tissue and hepatocellular carcinoma cell lines in vivo and in vitro respectively have evidenced preferential PC anaplerosis mediation, resulting in different sensitivities to experimental nutrient withdrawal (105, 106). As a result, some GBM experimental designs have since focussed on delineating neoplasm anaplerotic carbon source flexibility during nutrient deprivation. This is of clinical importance due to the potential for therapeutic confounding resulting from cellular complementation and survival mechanisms using alternative pathways (105). Accordingly, further analysis using GLS suppression in SF188 and adult LN229 GBM cell lines by Cheng et al. have since revealed that despite limited growth, cells are able to compensate using other GLS independent amidotransferase-catalysed glutamate generating pathways as well as glucose-derived carbon incorporation into TCA intermediates through PC (105). The additionally observed upregulation of PC after glutamine deprivation led the group to conclude: “PC is dispensable for growth of glutamine-replete glioblastoma cells, but required when glutamine supply is limited’’ (105).

Further to studies using single cell lines, the use of patient derived primary GBM samples has since been imperative for the delineation of tumoural metabolic heterogeneity. Metabolic analysis of 14 patient GBMs by Oizel et al. revealed phenotypic disparity and clustering as two distinct groups (107). One group exhibited substrate flexibility including glutamine utilisation for TCA directed nicotinamide adenine dinucleotide (NADH) formation, as well as higher expression of SLC1A5, GLS and Glutamic-Oxaloacetic Transaminase 1 (GOT) (107). The other exhibited glucose dependency for survival and growth. Substrate removal and substitution experiments revealed that the metabolically flexible phenotype was essential for maintaining cell proliferation when glucose was removed (107). In addition, the long-term inhibition of glutamine metabolism using the inhibitor, epigallocatechin gallate allowed cells to use glucose mediated TCA flux (107). Data such as these have been essential for recognising the concurrent existence of cells with different substrate sensitivities within a whole tumour. Later, the group combined metabolic analysis with molecular subtyping data, revealing that cells with higher substrate flexibility and metabolic adaptation potential could be categorised as MES, whilst glucose dependent cells represented one of the other subtypes. Further extension of experiments to the U87 cell line associated with GSC features, conformed to the MES subtype and thus, exhibited glutamine-mediated anaplerotic shuttling (107). The MES subtype has been heavily implicated for hypoxic survival and represents a subset of GBM (9, 10). These findings correlate with experimental data from Flavahan et al. using patient derived GSCs confirmed by assessing self-renewal, proliferation and expression of stem cell markers to show the association of high GLUT3 expression with GSCs in the PN subtype (83). However, earlier isotopic analysis studies using ¹³C mass spectrometry of isolated GSCs in neurospheres cultured from GBM orthotopic mouse models by Marin-Valencia et al. displayed rapid consumption of glucose for PC mediated anaplerosis and mitochondrial oxidation in addition to de novo glutamate production (108). Corresponding transcription data revealed the low expression of GLS in contrast to PC and GS levels, accessory to the observation that glucose-derived glutamate was used to maintain a large intracellular pool of glutamine, mediated by GS (108). Corroborating data from other studies have shown that GBM cells with high GS expression showed little to no reliance on extracellular glutamine uptake for growth preservation via glutaminolysis, but instead a dependence on de novo glutamine synthesis for intracellular pool replenishment (109). Assumptions based on this data have since exclusively defined GSCs as displaying high GS levels, contrary to data from Oizel at al. in which molecular subtyping is believed to be the more dominant driver of metabolic phenotype, with both GS-positive and GS-negative GSCs described (107, 109) Molecular subtyping of the GSCs studied by Marian-Valencia et al. would have revealed whether this data correlated with finding by Oizel et al., such that these cells showed subtyping distinct from the MES phenotype.

Nonetheless, combining these study outcomes, a distinction has been made between GBM cells displaying low GS levels as being mainly reliant on glutamine for TCA anaplerosis, with the ability to switch to largely glucose as a main substrate under glutamine repression (107). In contrast, GBM cells with high GS levels frequently defined as GSCs are believed to rely upon PC-mediated anaplerosis as a major source of glutamate GS mediated glutamine pool replenishment. In this way, these cells show major similarities to untransformed astrocytic systems in which PC and GS activities are required for glucose-derived glutamine pool maintenance, with little reliance on external glutamine uptake for cell proliferation. In studies by Tardito et al. proportionate protein levels were recorded for astrocytes and GS-positive GBM cells (109). The heavy reliance on de novo glutamate production as observed in GSCs with high GS expression, is believed to be an abundant source of nucleotide precursors for sustained purine biosynthesis (49, 108, 109).

Metabolic crosstalk between GBM cells and the stroma in the context of plastic substrate utilisation is essential for transformed cell survival during chaotic TME dynamics. In addition, studies evidencing the ability of subpopulations of tumour cells to adapt during nutrient restriction has led to investigations of cancer-stromal alliances in a range of cancers for survival (110–112). Tardito et al. reported low kinetic uptake of glutamine from the blood by GS-negative GBM cells and the healthy brain (109). Since GS-negative GBM cells have shown high glutamine uptake for anaplerotic pathways, orthotopic mouse models of GS-negative GBM revealed that other GS expressing tumour cells as well as astrocytes were a major source of the amino acid. Moreover, co-culture of astrocytes and GS negative LN18 GBM cells revealed growth sustenance based on astrocytic glutamine production in the absence of media supplementation (109). This data shows that subject to transformation, glutamine excreted into the TME for endogenous neuronal transmission, can instead support the growth of malignant cells that require a supply of metabolic nutrients. Furthermore, Kallenberg et al.’s magnetic resonance spectroscopy data illustrated a higher concentration of glutamine in the hemispheres of GBM patients relative to healthy controls, reported as a marker for early tumour infiltration (113). In addition, cross section staining for astrocytes and GS level of human derived GBM xenografts display a symbiotic positioning of astrocytes surrounding GS-negative GBM cells (109).

The widespread acceptance of the Warburg effect as well as experimental evidence supporting this metabolic shift in GBM tumours has been instrumental for the emergence of glycolytic targeting strategies. Glucose uptake inhibition has been dominated by the use of GLUT1 antagonists such as indinavir and ritonavir, shown in vitro to reduce GBM cell proliferation including in GSC cell lines (114). However, despite the synergistic effect of ritonavir when combined with chemotherapeutics in vitro, both compounds are unable to effectively cross the blood brain barrier (BBB), and therefore not able to reach effective concentrations in the brain (114). Alternatives have included compounds such as 2-fluoro-6-(m-hydroxybenzoyloxy) phenyl m-hydroxybenzoate (WZB117), displaying successful in vitro inhibition of GSC self-renewal ability (115). However, despite the beneficial effects observed for the inhibition of this transporter, the robustness of the targeting strategy has been challenged due to widespread GLUT1 expression in the human brain and therefore potential for off target effects (116). Consequently, GLUT3 has since emerged as a more promising target with protein expression analysis by Flavahan et al. showing elevated levels of the transporter in GSCs (83). Moreover, studies by Cosset et al. have interrogated GLUT3 expression further, showing that targeting with therapeutics may specifically inhibit classical/proneural cells within GBM tumours (117).

Other glycolytic inhibition strategies have included the suppression of HK2 with antifungals, as well as use of the glucose analogue 2-deoxy-D-glucose (2-DG) for competitive inhibition of cellular glucose uptake, and overall reduction in glycolysis (118, 119). The depletion in ATP production resulting from 2-DG treatment has been shown to promote cellular endoplasmic reticulum (ER) stress, followed by the unfolded protein response (UPR), similarly observed as an effect post radiotherapy (119). It is believed that GSCs have a greater ability to escape this stress with superior autophagic pathway promotion and reestablishment of ER homeostasis (120). Therefore, currently, there are efforts to promote tumour apoptosis by preventing autophagic and UPR mediated ER homeostasis reestablishment, with some groups finding that blocking ER stress altogether and the downstream protective mechanisms could enhance the cytotoxic effects of small therapeutic compounds such as shikonin (121). An increasingly established downstream effector of the UPR is 78-kDa glucose-regulated protein (GRP78) - a member of the heat shock protein family, instrumental for downstream activation of ER homeostasis regulators as well as observed cell surface translocation, characteristic of invasive cancers (119, 122). Whilst the translocation of GRP78 from the ER to the cell surface is not well understood, it is believed that GRP78 may also regulate cellular interactions with the surrounding TME (123).

Further to treatment strategies requiring administration of small molecule inhibitors, the modification of patient dietary nutrient intake has become a potentially effective approach for diverting tumour metabolism, such as short-term starvation shown by in vitro colon cancer studies for apoptosis induction (124). The dietary reduction of serum glucose concentrations involves restricting carbohydrate consumption, traditionally implemented by following a 4:1 fat to protein and carbohydrate ratio (commercially available as KetoCal^®) that is well tolerated in patient studies for targeting GBM energy metabolism (125–127). As a result, multiple clinical trials are either complete or ongoing evaluating the safety/tolerance and effectiveness of the ketogenic diet (KD) as an adjuvant to current treatment in GBM: NCT01865162, NCT023939378, NCT04691960, NCT02302235, NCT00575146, NCT03451799, NCT01754350, NCT03075514, NCT03278249, NCT01535911, NCT04730869, NCT02286167. This dietary limitation forces the liver to metabolically adapt, using fats to produce ketone bodies which can be used as an alternative fuel source for cellular energy metabolism; originally pioneered for the beneficial reduction in seizures in epileptic children (128). Furthermore, Otto Warburg’s description of tumoural exploitation of aerobic glycolysis provided initial rationale for the approach in GBM, suggested as an effective strategy to slow tumour growth and reduce ability of cells to exhibit a Warburg phenotype (129). As such, the efficacy of this treatment has been shown by Abdelwahab et al’s. use of a bioluminescent mouse model of malignant glioma to confer increased survival (127). Moreover, the retrospective statistical analysis of patient data suggests that postoperative hyperglycaemia is associated with poor survival; an effect that could be suppressed by lowering glucose levels during treatment (130–132).

Despite studies demonstrating efficacy of the KD for increased GBM survival, Sperry et al. have explored GBM intratumoural heterogeneity to reveal that populations of tumour cells could in fact thrive using ketone bodies and upregulated fatty acid oxidation for survival, therefore mitigating the effectiveness of the KD (133). However, the authors highlight that the KD in combination with additional metabolic inhibition of enzymes such as carnitine palmitoyltransferase 1A (CPT1A) – the rate-limiting enzyme for fatty acid oxidation, or implementation of additional calorie restriction to exacerbate the effect of serum glucose reduction could increase the efficacy of the KD, as evidenced by Shelton et al. in GBM (133, 134). Tumour survival via the exploitation of pathways that circumvent low glucose concentrations such as glutaminolysis have also become a concern for implementation of the KD as an effective treatment (135). Moreover, despite the maintenance of low glutamine levels in the healthy brain via the glutamine-glutamate cycle, studies by Takano et al. have shown that glioma glutamate secretion into the surrounding brain could be readily recycled by tumours for sustained energy metabolism and growth (95). Therefore, strategies implementing the calorie restricted KD in combination with 6-diazo-5-oxo-L-norleucine (DON) – a gluatminolysis antagonist has been tested in vivo GBM mouse models by Mukherjee et al. conferring increased survival and tumour cell apoptosis (135).

Besides a reduction in glycolytic flux resulting from the KD, other downstream cellular responses have been investigated by Ji et al. using GSCs derived from both patients and cell lines (136). Culture medium high in β-hydroxybutyrate for mimicking the effects of the KD resulted in reduced GSC proliferation and tumorigenicity as well as revealing the possibility for the induction of damaging morphological and functional mitochondrial alterations (136). In addition, studies by Seyfried et al. using in vivo astrocytoma models exposed to restricted standard and KDs showed alterations in metabolic modulators such as a reduction of IGF-I (137). Both IGF-I, the IGF1 receptor (IGF-1R) and associated signalling network have been linked with tumour survival in a range of cancers, with experimental data showing the presence of the receptor providing protection from apoptosis following cytotoxic treatment (138, 139). Moreover, the IGFR1 exhibits elevated expression in GBM cells when compared with normal brain cells and is therefore believed to be a marker of reduced patient survival and inhibition of tumour cell apoptosis (140, 141). In fact, Zhang et al. carried out in vitro culture studies comparing control U87 GBM cells to those overexpressing the IGF-R1, revealing that in response to hydrogen peroxide exposure, the cells with higher IGF-R1 could inhibit apoptosis. In addition, increasing concentrations of IGF1 or overexpression of the IGF-1R resulted in Akt phosphorylation, believed to increase PI3K/Akt pathway activation and subsequent apoptosis suppression (140).

The 2010 retrospective review of high-grade glioma patients by Grommes et al. exposed that diabetic patients receiving treatment with PPARγ agonists (Thiazolidinediones) had a median survival increase of 13 months in comparison to those not taking this medication (142). Despite analysis of the data defining this result as not significant due to small sample size, expression data showing elevated levels of PPARγ in malignancies such as colon, duodenal, lung, prostate, thyroid, primary and metastatic breast cancer has provoked an increased interest in the clinical relevance of this receptor family, of which there are 3 mammalian members: PPARα, PPARβ/δ and PPARγ (143–148). The PPARs are part of the nuclear receptor superfamily and function as ligand-inducible transcription factors which under normal conditions bind dietary fats as well as regulating both adipocyte and macrophage biology (149).

PPAR agonists have therefore gained traction with studies by Keshamouni et al. demonstrated that treatment of non-small cell lung cancer (NSCLC) patient tumours with the PPARγ ligand troglitazone conferred anti-proliferative effects due to G0/G1 cell cycle arrest and reduction in cyclin D/E expression (145). The same cell cycle arrest has also been recorded in GBM tissue samples using the PPARγ ligand pioglitazone by Zang et al., with other research groups showing that this ligand could additionally stimulate β-catenin mediated apoptosis of GBM cell lines (150, 151). Interestingly, Chearwae and Bright (152) treated neurospheres generated from commercial GBM cell lines with the PPARγ agonists 15-deoxy-Δ^12,14-prostaglandin J₂ (15d-PGJ2) or all-trans retinoic acid to induce apoptosis and inhibit neurosphere formation and expansion through Tyk2-Stat3 inhibition (152). The notion that these ligands exert effects though the Janus kinase-signal transducer and activator of transcription (JAK-STAT) pathway has been complemented by studies by Mo et al. using PPARγ ligands in mouse embryonic stem cells (153). Since STAT3 is crucial for the self-renewal of GSCs, the potential downstream inhibition of this pathway could be important for sensitisation to chemotherapeutics such as temozolomide (TMZ), as shown by Villalva et al. (154).

Further to the delineation of different populations of GBM cells with distinct cycling kinetics by Hoang-Minh et al., enhanced cytotoxic resistance of slow-cycling populations have been described by studies performed in colon and breast tumour cells by Moore et al. despite being successful against the more proliferative tumour bulk (86, 155). Therefore, the induction of differentiation in stem populations has become a promising therapeutic strategy for cellular sensitisation to standard chemo and radiotherapy for GBM (Stupp protocol) (4). Pestereva et al. used imatinib to inhibit the PDGFR and stem cell factor receptor (c-Kit), with encouraging reductions of stem cells markers and tumorigenicity of tissue derived GSCs (156). There is also promise that PPARγ agonists could be useful for inducing differentiation of GSCs within tumours due to indications that ciglitazone and 15d-PGJ2 caused altered expression of stemness genes such as SOX2 and NANOG (157). This was again tested in mouse derived NSCs cultured as neurospheres by Kanakasabai et al. with stem cell gene expression analysis before and after treatment with iglitazone or 15d-PGJ2 revealing a downregulation of stem and differentiation associated genes (158). However, despite the potential benefits of this approach, studies by Caren et al. have highlighted the importance of ensuring terminal GSC differentiation in a stable manner, due to the variable responses of cells to bone morphogenic protein (BMP) therapies. Therefore, studies are required to investigate the mechanisms used by cells to evade patterns of commitment, making them vulnerable to dedifferentiation (159).

In addition to the potential beneficial effect of PPARγ agonists on GBM, the elevated expression of PPARα recorded in GBM tissue by Haynes et al. has been complemented by data demonstrating tumour growth suppression following the use of agonists such as gemfibrozil (160, 161). This beneficial effect has also been observed in NSCLC cells, where induction of PPARα inhibited growth and angiogenesis, as well as leading to apoptosis in ovarian cancer cells when combined with PPARγ induction (162, 163). However, a compelling aspect of PPARα is that it is imperative for ketogenesis due to transcriptionally regulating 3-hydroxy-3-methylglutaryl-CoA (HMGCS) – the rate-limiting enzyme for the conversion of acetyl-CoA to β-hydroxybutyrate and acetoacetate (164, 165). Therefore, PPARα can also downregulate IGF/Akt signalling discussed for the KD as well as GLUT1 and 4 receptors, essential for supressing the Warburg phenotype (166–168).

Further to studies revealing that GBMs undergo glucose oxidation, the discovery of mitochondrial aberrations including electron transport chain (ETC) components and mitochondrial reserve capacity have provided insight into additional GBM survival mechanisms (169). Mitochondrial DNA (mtDNA) profiling has revealed mutations in complex I, III and IV of the ETC, affecting the balance between OXPHOS and aerobic glycolysis (170, 171). GBM patient tumour tissue analysis by Lloyd et al. revealed that at least one mitochondrial DNA (mtDNA) mutation is present in 43% of patients (171). Further to this, large scale mtDNA alterations, ETC remodelling, and enzyme activity modulation has been shown to be critical for TMZ resistance in both cell lines and human GBM specimens, with chemoresistance largely driven by cytochrome C oxidase (COX) (172). Moreover, subject to research by Oliva et al., COX subunit-IV (COX-IV) was shown mainly to be associated with COX isoform I (COX-IV-I) for cellular nutrient sensing and modulation of energy production in TMZ resistant cells (172). This finding was later extended to GSCs, with COX-IV-I displaying cooperation with downstream targets to enhance tumorigenicity and self-renewal (173). Consequently, ETC pharmacological and genomic intervention has been further investigated, with results suggesting TMZ sensitisation following COX-IV-I targeting (174, 175).

Furthermore, the anti-diabetic biguanide drug, MF, has been analysed as a promising therapeutic strategy for decreasing cancer proliferation and inducing apoptosis subject to nutrient availability. Investigations in colon cancer cells have revealed that MF reversibly inhibits the function of ETC complex I subject to the presence of membrane potential and glucose (176, 177). Other studies have revealed that in the absence of glucose, MF can induce cellular apoptosis as well as a possible selective effect on CSCs (177–181). Further to this, phenformin - a MF analogue with elevated potency, was used by Jiang et al. in vitro and for in vivo tumour mouse models, revealing multiple beneficial effects including a reduction in GSC self-renewal and overall prolonged mouse survival (182). However, since biguanide use is associated with lactic acidosis induction, combining MF with the inhibition of PDK by dichloroacetate (DCA) by Haugrud et al. conferred a cell survival advantage, both reducing lactate production and inhibition of oxidative metabolism in breast cancer cells (182, 183). GBM oxidative stress induction has also been described for other compounds such as the anthelmintic drug Ivermectin, similarly shown to induce apoptosis via ETC complex I inhibition as well as targeting angiogenesis and the Akt/mTOR pathway in GSC cell lines (184). Interestingly, this drug also induces apoptosis in human microvascular endothelial cells (HBMEC), thus breaking an important protective and synergistic relationship believed to be critical for GBM niche maintenance (184).

Further to the identification of isolated metabolic targets, the combination of existing GBM chemoradiation protocols with agonists/antagonists of multiple signalling pathways such as sonic hedgehog (Shh), murine double minute 2 (MDM2), p53, PI3K/mTOR, EGFRvIII, poly(ADP-ribose) polymerase 1 (PARP1), cyclin-dependent kinase 4/6 (CDK4/6) have exhibited cytotoxic sensitisation efficacy of GBM models (185–191). Further experimental combinatorial strategies include the use of impridone compounds for inhibiting both glycolysis and OXPHOS via Akt/ERK dual inhibition, c-Myc degradation and apoptosis in GSCs (192). In addition, Yuan et al. effectively used the glycolytic inhibitor 3-bromo-2-oxopropionate-1-propyl ester (3-BrOP) and the alkylating chemotherapeutic carmustine/BCNU in GSCs that were initially highly glycolytic, causing major ATP depletion and abrogation of DNA repair capacity (193). Furthermore, the development of the arsenic-based mitochondrial toxin, 4-(N-(S-penicillaminylacetyl)amino) phenylarsonous acid (PENAO) by Shen et al. to trigger mitochondrial apoptosis had better efficacy when used with DCA for dual targeting of glucose metabolism (194). Other combinations of drugs undergoing clinical development for GBM have also been tested with existing compounds such as Navitoclax/ABT-263 (a Bcl-2/Bcl-xL inhibitor) with encouraging in vivo low toxicity and suppression of tumour growth (195). Moreover, studies using chemotherapeutics with both MF and phenformin have shown beneficial effects tumour growth inhibition, with Jiang et al. using phenformin and TMZ in vivo for prolonged mouse survival (182).

However, the epidemiological identification of existing drugs with anti-tumour effects is not restricted to the biguanides, many repurposed compounds are currently under scrutiny in clinical trials in combination with existing chemoradiation protocols (see Table 1 ). Moreover, clinical trials to test the coordinated blocking of multiple different cell survival pathways has become an attractive strategy, with efforts to use multiple repurposed compounds that will work synergistically for the greatest therapeutic benefit. In 2016, the CUSP9v3 trial (identifier: NCT027703780) started, combining TMZ with Aprepitant, Minocycline, Disulfiram, Celecoxib, Sertraline, Captopril, Itraconazole, Ritonavir and Auranofin in additive treatment cycles for the treatment of recurrent GBM cases; a strategy originally proposed in 2013 by Kast et al. (216, 217). Although the results have not been released at the time of writing this review, experimental studies to test the robustness of the CUSP9 strategy have been published, including Skaga et al.’s investigations using 15 GSC cultures, derived from 15 patient GBMs including relapsed tumours (18). The group found that the combination of drugs was significantly more effective than when used alone, as well as increasing the therapeutic benefit of TMZ for sphere eradication in most of the cell lines, with the highest resistance in a proneural population (18). In addition to testing compound efficacy with conventional therapy, it has come to light that patient tumour profiling prior to therapeutic application may give a greater therapeutic benefit. Recently, the antiproliferative effects of four different compounds: MF, dichloroacetate (DCA), sodium oxamate (SOD) and diazo-5-oxo-L-norleucine (DON) were tested on 4 GSC subpopulations with different initial oxidative/glycolytic metabolism tendencies by Duraj et al. (88). The group found that GSC inhibitor sensitivity differed relative to Warburg-like/OXPHOS phenotype, and Seahorse XF to determine glycolytic/mitochondrial ATP production shifts following treatment. The authors suggest that predictive metabolic shifts despite initial bioenergetic plasticity following treatment, could be used as “metabolic priming” (88). In this way, the malignant cells could be pushed towards exhibiting a phenotype with enhanced sensitivity to subsequent small molecular inhibitors, chemo- or radiotherapy.