Cancer Stem Cells: Metabolic Characterization for Targeted Cancer Therapy

The subpopulation of cancer stem cells (CSCs) within tumor bulk are known for tumor recurrence and metastasis. CSCs show intrinsic resistance to conventional therapies and phenotypic plasticity within the tumor, which make these a difficult target for conventional therapies. CSCs have different metabolic phenotypes based on their needs as compared to the bulk cancer cells. CSCs show metabolic plasticity and constantly alter their metabolic state between glycolysis and oxidative metabolism (OXPHOS) to adapt to scarcity of nutrients and therapeutic stress. The metabolic characteristics of CSCs are distinct compared to non-CSCs and thus provide an opportunity to devise more effective strategies to target CSCs. Mechanism for metabolic switch in CSCs is still unravelled, however existing evidence suggests that tumor microenvironment affects the metabolic phenotype of cancer cells. Understanding CSCs metabolism may help in discovering new and effective clinical targets to prevent cancer relapse and metastasis. This review summarises the current knowledge of CSCs metabolism and highlights the potential targeted treatment strategies.


INTRODUCTION
Cancer causes significant deaths worldwide, despite major innovations in treatment therapy strategies, radiation-and chemo-therapy and drug delivery technologies. A major contributor to the cancer treatment-associated toxicities and resistance (1)(2)(3), is their inability to eradicate subset of cancer stem cells (CSCs) which drive tumour growth and heterogeneity. CSCs presence makes tumors resistant to conventional therapies (4). Density of CSCs is a proven prognostic marker in various cancers (5,6), thus targeting CSCs is an effective way for treating cancer.
Till date the origin of CSCs remains elusive. Two models are postulated to explain the genetic and functional heterogeneity of cancer in a single patient: the clonal evolution model and the cancer stem cell (CSC) hypothesis (19). The clonal evolution model suggests that multiple stepwise oncogenic mutations in somatic cells leads to tumor formation and natural selection favors the tumor cells with aggressive phenotype (20,21). The CSC hypothesis suggests that metabolic events occurring in cancer epithelial cells may generate CSCs (Figure 1). Altered metabolic events in cancer cells may affect chromatin organization and activate epigenetic program (22) which may further fuel metabolic-reprogramming of CSCs. Two proposed models explain how metabolic alterations could affect epigenetics (22). In the first model, metabolism reprogramming facilitates differentiation of one cell type to another by altering chromatin modifications without affecting the epigenomic landscape. The second model proposes that altered metabolism induces new potential cell types via creation of novel stable epigenetic states, thus reshaping the entire epigenomic landscape. In this model, altered metabolism remodels chromatin by either inducing gene expression or affecting availability of substrates and cofactors for chromatin-modifying enzymes. In either case, the end-result is a novel cell state that is irreversible as epigenomic landscape has changed.
Metabolic characterization of CSCs has been a challenging task, as CSCs lack a common metabolic phenotype across cancer types. CSCs metabolic pattern differ from adult stem cells (SCs) and use either glycolysis or OXPHOS ( Figure 1) triggering cellular plasticity in CSCs (23). Thus, an understanding of CSCs metabolic features will help target CSCs specifically and prevent cancer progression. Current review summarizes the various metabolic features of CSCs along with therapeutic interventions that can be adopted to target key energy processes in CSCs. Targeting the metabolic flexibility in CSCs can emerge as an effective strategy for preventing or minimizing disease progression and recurrence.

METABOLIC FEATURES OF CSCs
Growth factors, nutrients and oxygen in the tumor microenvironment provide necessary energy sources and growth signals for CSCs generation and proliferation. Recently, metabolism has been identified as a major component in CSCs biology, as oncogenic alterations has been observed to cause metabolite-driven dissemination of CSCs (19). Multipotent SCs use glycolysis (24,25), have fewer mitochondria and produce less reactive oxygen species (ROS) (26,27). Higher ROS levels cause SCs dysfunction (28)(29)(30) and shift to OXPHOS with increased ROS production leads to differentiated SCs progeny.
NANOG-expressing hepatocellular CSCs have higher glycolysis and fatty acid oxidation (FAO) rates, and lower OXPHOS and ROS generation (43). CSCs secretome have enriched levels of glycolytic and antioxidant pathways proteins and secreted high levels of ALDH than differentiated cells from colorectal tumors (24). ALDH detoxifies anticancer drugs such as maphosphamide and CSCs secreting ALDH promoted selfpreservation and protected nearby differentiated mature cancer cells, leading to therapy resistance (24). Ovarian CSCs with glycolysis enrichment, de novo fatty acid synthesis, and decreased mitochondrial respiration and anaplerotic flux, led to aggressive tumors with therapy resistance to cisplatin in comparison to mature cancer cells (34).

Mitochondrial Respiration
As an energy source, OXPHOS is more efficient than glycolysis, but has a slower rate to produce energy. Quiescent or slow-cycling tumor-initiating CSCs prefer OXPHOS metabolism over glycolysis (Figure 1), consume less glucose, have lower lactate and higher ATP levels (55)(56)(57). OXPHOS-dependent CSCs with low glycolytic reserves are shown in acute myeloid leukemia, CD133 + glioblastoma, melanoma, pancreatic and ovarian cancer (58)(59)(60)(61)(62)(63). In breast CSCs, elevated OXPHOS levels trigger chemotherapeutic resistance through synergistic action of MYC and MCL1 (64). CSCs using OXPHOS have higher mitochondrial mass with increase in membrane potential and rates of oxygen consumption (62,65,66). Mitochondrial mass is a vital metabolic biomarker of CSCs (65,67). Tumor cells without mitochondrial DNA (mtDNA) grew slowly and acquisition of mtDNA from host cells led to tumor-initiation and drug resistance in these tumor cells (68), suggesting mitochondrial function as a target for CSCs treatment. Master mitochondrial biogenesis regulator, peroxisome proliferator-activator 1 alpha (PGC1a) maintained stemness characteristics (69) in breast cancer (70) and pancreatic CD133 + CSCs (66) and increased chemoresistance in CSCs (64,(71)(72)(73). NANOG is a pluripotency gene that supports tumorigenesis through OXPHOS and fatty acid metabolism (43). Some breast CSCs show elevated glucose consumption and ATP production, higher mitochondrial activity but lower lactate levels, suggesting that OXPHOS and glycolysis may not be mutually exclusive to CSCs (62).

Glutamine Metabolism
Glycolysis and OXPHOS may not completely support CSCs metabolism, thus glutamine compensates for glucose shortage (74,75). Although a non-essential amino acid, glutamine becomes essential for cancer cells (76) and CSCs from lung, pancreatic and ovarian cancer have shown glutamine dependence (77,78). CSCs rely on glutamine for carbon and amino-nitrogen for protein, nucleotide and lipids biosynthesis (79). Glutamine metabolism is rewired by mutations in mitochondrial DNA (mtDNA) (80) and oncogenic alterations in KRAS (81,82) and c-Myc (83) in tumor cells. Glutamine metabolism in c-Myc-over-expressing cells suggests a pluripotency gene profile dependence on glutamine (84). In pancreatic CSCs, glutamine unavailability reduced stemness characteristics and increased radiation therapy sensitivity (77). L-DON (a glutamine analog) inhibited glucose metabolism and prevented systemic metastasis to liver, lung and kidney in mice (85).

Lipid Metabolism
Cells use an anabolic process of fatty acid synthesis (FAS) to derive energy from fatty acid metabolism for cell growth and proliferation, and a catabolic process of fatty acid oxidation (FAO) for NADH and ATP production (86). CSCs are extremely reliant on de novo lipid biosynthesis, lipid oxidation and lipid metabolizing enzymes (87,88).
CSCs use mitochondrial FAO for ATP and NADPH generation to survive loss of matrix attachment (97,98).
Elevated purine synthesis promoted stemness in brain tumor initiating cells (BTICs) and correlated with significantly poorer overall survival in glioblastoma patients (105). MYC regulates purine synthesis enzymes and its liaison with de novo purine synthesis mediated selective dependence of BTICs on glucosesustained anabolic metabolism. Inhibition of purine synthesis prevented BTICs growth by inhibiting their self-renewal capacity, but differentiated glioma cells remained unaffected (105). Thus frailty of purine synthesis in CSCs makes it a potential therapeutic target, Lysine catabolism promoted self-renewal of CD110 + colorectal cancer tumor-initiating cells (TICs) by generating acetyl-CoA. Acetyl-CoA triggered LDL receptor-related protein 6 (LRP6) acetylation and phosphorylation, and finally activation of WNT signaling (106). Lysine catabolism promoted drugresistance and metastasis to liver in CD110 + TICs by glutamate and glutathione synthesis, which modulated the redox status (106). Collectively, CSCs use an array of metabolism alterations to fuel their self-renewal, thus making these metabolic dependencies open to targeted therapies.

CLINICAL IMPLICATIONS
CSCs have both distinct and flexible metabolic phenotypes between glycolysis and OXPHOS-dependent. Despite limited clinical evidence, targeting CSCs through selective metabolic modulation is an effective and promising avenue for cancer treatment. In our view, synergistic treatments using a standard cytotoxic agent and a metabolic-based therapy will improve eradication of CSCs. Table 1 lists the available metabolic targeting agents undergoing clinical trials in various cancers.

Targeting Glycolysis
Glycolytic CSCs can be targeted for glycolytic enzymes (hexokinase (HK), phosphoglycerate kinase, pyruvate kinase) and glucose transporters (GLUT1-4). Direct inhibition of GLUTs results in a total disruption of glucose uptake and hence energy metabolism, and GLUT inhibitors such as phloretin, fasentin and WZB117 have shown anticancer effects in preclinical models (107)(108)(109)(110). However, ubiquitous expression of GLUTs even in normal cells challenges the explicit inhibition of CSCs glucose uptake and leads to side-effects.
Pyruvate is converted into mitochondrial acetyl-CoA in the cytosol and is negatively regulated by pyruvate dehydrogenase kinase (PDK) enzyme. This shifts cellular metabolism from OXPHOS to glycolysis and thus targeting PDK can inhibit cellular proliferation of CSCs. Dichloroacetate (DCA) activates mitochondrial pyruvate dehydrogenase (PDH) by inhibiting PDK (120), is fairly well-tolerated with fewer side effects and is being tested in several anticancer clinical trials (121,122).
CSCs can oscillate between metabolic phenotypes during oxygen deprivation and glucose starvation, and thus targeting mechanisms underlying these metabolic adaptations can effectively eliminate CSCs. Hypoxia-inducible factors (HIFs) promote tumor progression in response to localized hypoxia by switching to glycolysis from OXPHOS, activating Notch pathway and expression of Oct4 transcription factor (123,124). This suggests HIF-1a's role in self-renewal and multipotency and targeting HIFs can be a prospective treatment for CSCs. Metformin, although an antidiabetic drug, attenuated glycolysis flux in hepatocellular carcinoma cells (125) and improved radiotherapy response in prostate and colon cancer tumor xenograft models (126). Epigallocatechin gallate (EGCG) is an inhibitor of glycolysis and its co-treatment with gemcitabine enhanced pancreatic cancer cell death both in vitro and in xenografts (127).

Targeting Mitochondrial Respiration
Several OXPHOS-targeting pharmacological agents are being explored in clinical trials for cancer treatment ( Table 1) and have potential to target CSCs. OXPHOS inhibition overcame drug resistance in slow-cycling melanoma cells and mitochondria-targeted antibiotics prevented sphere formation and tumorigenesis in CSCs (61,128). Metformin inhibited mitochondrial electron transport chain complex I and diminished OXPHOS (129). Metformin caused energy emergency and hence apoptosis in OXPHOS-dependent pancreatic cancer stem cells (CSCs), but spared their glycolytic differentiated progenies (66). Diabetic patients receiving metformin have a lower mortality rate from cancer and hence a better prognosis (130,131). Phenformin, a biguanide formerly used in diabetes and a mitochondrial inhibitor induced nonsmall cell lung cancer (NSCLC) cells apoptosis (132).
CSCs mitochondrial mass and metabolism can be targeted using approved antibiotics like tetracyclines, salinomycin and erythromycins. Antibiotic salinomycin inhibits OXPHOS (133)   and salinomycin treatment reduced breast CSCs gene expression. Antibiotic tigecycline inhibited mitochondrial translation in mitochondrial associated ribosomes in OXPHOS-dependent leukemia cells (134). CSCs using OXPHOS have a higher mitochondrial membrane potential (Dym) and thus Dym can be explored for selective accumulation of cytotoxic drugs. Triphenylphosphonium (TPP) accumulates in the mitochondrial matrix (135) and conjugation of TPP to doxorubicin prevented drug efflux by enhancing drug selectivity in cancer cells (136). Dual inhibition of glycolysis and OXPHOS in sarcoma cells, using 2-DG and oligomycin/ metformin co-treatment (137), suggests that simultaneous inhibition of glycolytic and mitochondrial respiration is more effective to eradicate CSCs (138,139).

Targeting Glutamine Metabolism
Although a non-essential amino acid, glutamine becomes essential as a favored respiratory fuel for cancer cells and thus depriving glutamine is a potential anti-cancer strategy. Glutamine metabolism can be blocked by inhibiting glutaminase 1 (GLS1), an enzyme that converts glutamine to glutamate. GLS1 inhibition disrupted redox balance in CSCs and sensitized lung and pancreatic cancers to radiotherapy (77,140). GLS1 inhibitors, BPTES (141) and CB-839 reduce intracellular glutamate and 2-hydroxyglutarate (an oncometabolite) levels. Lower glutamate levels inhibited cell growth, induced apoptosis and differentiation in Acute Myeloid Leukemia (AML) cells (142). CB-839 is under clinical trials for various cancers including renal cell carcinoma, hematologic cancer and leukemia ( Table 1).

Targeting Lipid Metabolism
Cancer cells predominately use glycolysis for ATP production instead of oxidizing energy-rich substrates. However, unlike non-cancerous cells dependence on dietary lipids, cancer cells use de novo lipogenesis. Thus targeting fatty acid synthase (FASN), a central enzyme to lipogenesis, is a promising strategy to eliminate CSCs. FASN inhibitor cerulenin reduced de novo lipogenesis and in turn proliferation, migration and stemness of glioma stem cells (GSCs), induced apoptosis in colon cancer cell lines (92,143) and blocked proliferation of pancreatic spheres (93). C75 decreased HER2+ breast cancer cells selfrenewal capacity at non-cytotoxic concentrations (144). However, due to toxicity issues in in-vivo studies owing to high selectivity of FASN inhibitors, only one FASN inhibitor (TVB-2640) is under clinical trials to date ( Table 1).
Studies show that increased fatty acid production in cancer cells raises their dependence on desaturases (enzymes that add double bonds into acyl-CoA chains). Thus targeting desaturase enzyme activity may provide a novel approach to selectively interfere lipid metabolism in CSCs. Several stearoyl-CoA desaturase-1 (SCD-1) inhibitors have effectively targeted stemness in pre-clinical models of cancer. Inhibitors like CAY10556 and SC-26196 reduced stem-ness markers and inhibited in-vitro sphere formation and in-vivo tumorigenicity, by down-regulating Hedgehog and Notch expression in aldehyde dehydrogenase (ALDH)-and CD133-enriched ovarian cells and had no effect on differentiated cells (96). Similarly, SCD-1 inhibitors (SSI-4 or A939572) promoted differentiation in chemo-resistant hepatospheres with little toxicity in vivo (145). MF-438 reduced expression of self-renewal and pluripotency markers in lung ALDH1 + cells (146).
Highly proliferating cells also have a higher demand for components of cell membrane like cholesterol. Cholesterol is either taken up from exogenous sources or synthesized using FASN or mevalonate pathway (154). Statins inhibit cholesterol synthesis through the mevalonate pathway and their target enzyme is 3-hydroxy-3-methyl-glutharyl-coenzyme A reductase (HMGCR). Statins treatment decreased CSCs selfrenewal capacity and number in breast (155), nasopharyngeal (156) carcinomas and CD133 + brain TICs (157). MYC controls over-expression of mevalonate pathway genes and thus anti-CSCs effects of statins could be due to MYC inhibition (157).
Synthesized or accumulated fatty acids are also converted to signaling lipids and energy via FAO, in addition to membrane incorporation or being stored. FAO is an essential energy source in non-glycolytic tumors (158,159), as CSCs show higher FAO in nutrient-deprived conditions (63,86,160,161). FAO promotes pluripotency and chemoresistance (94) by reducing ROS production (162,163) and promoted metastatic capacity in sphere-derived cells (164). Etomoxir, an inhibitor of FAO, inhibited mammosphere formation in hypoxic breast CSCs (165) and eradicated half of quiescent leukemia SCs (99), suggesting that FAO inhibitors hinder CSCs survival. In hepatocellular carcinoma, etomoxir sensitized CSCs to sorafenib treatment (43). Soraphen A, cerulenin and resveratrol inhibited FAO and lowered stemness markers and spheroid formation in CSCs (92,166,167). Lipids also support CSCs functionality by being second messengers in signal transduction pathways. Sphingolipids, eicosanoids (prostaglandin E2) and glycerophospholipids (lysophosphatidic acid (LPA)) boost CSCs number by activation of Notch, AKT and NF-kB pathways in breast, bladder, colorectal (CRC) and ovarian cancer (168)(169)(170)(171). Lipidmediated signaling in CSCs thus can be targeted using inhibitors and dietary supplements. Inhibition of autotoxin (ATX) (a lysophosphatidic acid (LPA)-producing enzyme) with S32826 or PF8380 reduced tumorigenicity and chemoresistance in-vivo (171). Inhibition of LPA production in cancer cells modulated the immune system by inducing monocytes differentiation to macrophages and launching cancer-associated fibroblasts (CAFs) phenotype (172,173). Prostaglandins are major lipid mediator in CSCs and celecoxib treatment of Apc Min /þ mice reduced number of CD133 + CD44 + cells and tumor burden (170). Celecoxib reduced patient-derived CSCs content and liver metastatic tumors number in NOD scid gamma (NSG) mice and weakened chemoresistance in bladder carcinomas, indicating its potential as an adjuvant therapy (169). In contrast, reduction of CD34 + cells in chronic myelogenous leukemia (CML) xenograft model by EP4 receptor (prostaglandin receptor) agonist misoprostol or PGE1 (FDAapproved), suggests a context-dependent role of prostaglandins in stem-ness (174). Further, dietary omega-3 polyunsaturated fatty acids (w-3 PUFA) decreased CRC risk and reduced CD133 + content in CRC cell lines (175,176). Eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA) supplementation decreased breast tumorspheres proliferation (177) and EPA with chemotherapy suppressed tumor growth in mice (178), suggesting an anti-CSCs properties of w-3 PUFAs.

Combination Treatments
CSCs can also attain a combined metabolic phenotype where both glycolysis and OXPHOS are utilized (Figure 1). This phenotype can be attained by direct association of AMP-activated protein kinase (AMPK, master regulator of OXPHOS) and HIF-1 (master regulator of glycolysis) activities (179). High AMPK/HIF-1 activities leads to higher glycolysis and OXPHOS, and provide enhanced proliferation and clonogenicity compared to only glycolytic or OXPHOS phenotype (179). In addition, CSCs metabolize glutamine along with glucose for carbon and aminonitrogen to synthesize amino acids, nucleotides and lipids (79). Additionally, CSCs also use de novo lipogenesis to increase their bioenergetic requirements and are linked in tumor metastasis (88). Also preclinical and clinical setting has shown that targeting a single metabolic pathway like glycolysis has low success rates and enhanced side effects as GLUT transporters are ubiquitous. Also, inhibition of hexokinase II with ionidamine showed no significant improvement in overall survival but led to elevated toxicity (114,(180)(181)(182). Thus combination treatments targeting two or more metabolic pathways will majorly erase CSCs, prevent tumor relapse and prevent side-effects of a single treatment.
Further, combining a standard cytotoxic therapy with a metabolic inhibitor will probably enhance CSCs eradication. Combinations of metformin and JQ-1 (bromodomain and extraterminal motif (BET) inhibitor) in pancreatic cancer (66) or PI3K inhibitor in ovarian cancer (183) blocked both OXPHOS and glycolysis. Apart from direct metabolic inhibition, targeting oncogenes regulating cellular metabolism will also eradicate CSCs effectively. KRAS mutation occurs in about 90% of pancreatic cancer cases (184) and KRAS drives glycolysis and nucleic acids synthesis (185,186). c-MYC is essential for glycolysis in cancer (187,188) and MYC suppression prevents mitochondrial inhibitors resistance (66,75). Thus combination approaches can be extended to target CSCs as an anti-cancer strategy. Table 1 lists the clinical trials using combination treatments for various cancers. Figure 1 summarizes the known CSCs' metabolic phenotypes and how these phenotypes switch with metabolic stressors like nutrient deprivation and hypoxia. However, melanoma cells attain a drug-tolerant "idling state" after enduring MAPK inhibition (MAPKi) and this state has a metabolically Low/ Low (L/L) phenotype, where both AMPK/HIF-1 activity and OXPHOS/glycolysis are minimal (189). L/L phenotype does not favor tumorigenicity but supports cell division. These idle L/L drug-tolerant cells accumulate mutations to promote relapse post MAPKi melanoma treatment (189).

FUTURE CHALLENGES
Further adding to the complexity of CSCs metabolism, Luo et al. (190) showed that breast cancer stem cells (BCSCs) have two states: quiescent mesenchymal-like (M) and proliferative epithelial-like (E). Proliferative E-BCSCs showed higher mitochondrial OXPHOS, whereas M-BCSCs have enrichment of glycolysis and gluconeogenesis pathways and hypoxia promotes M to E transition in BCSCs (190). Thus CSCs' multiple metabolic phenotypes (glycolytic, OXPHOS, combined and L/L) explain the futility of current efforts to eradicate CSCs and a deeper understanding of CSCs metabolic plasticity would translate to better therapeutic strategies.

CONCLUDING REMARKS
CSCs provide treatment resistance and promote metastasis during tumor growth and targeting metabolism holds potential in overcoming cancer recurrence and metastasis by CSCs. Deciphering metabolic reprogramming in cancer showed differences between metabolic phenotypes of CSCs and their differentiated counterparts. CSCs metabolism shuffles between glycolysis and OXPHOS primarily, however the mechanisms of CSCs metabolic heterogeneity are still unknown. Current knowledge suggests that carefully designed metabolic therapies have potential to be more effective against CSCs. Further, co-targeting CSCs using metabolic drugs and traditional anticancer treatments could be more efficient. The ongoing clinical trials targeting CSCs show a promising future for cancer therapy and are worth exploring further. More preclinical and clinical studies are thus required to uncover novel metabolic targets in CSCs.

AUTHOR CONTRIBUTIONS
SB conceptualized the article and reviewed the literature, JK did literature search and drafted the manuscript. Both SB and JK contributed to the final version.