Metabolic reprogramming and dependencies associated with epithelial cancer stem cells independent of the epithelial-mesenchymal transition program

Abstract

In solid tumors, cancer stem cells (CSCs) can arise independently of epithelial-mesenchymal transition (EMT). In spite of recent efforts, the metabolic reprogramming associated with CSC phenotypes uncoupled from EMT is poorly understood. Here, by using metabolomic and fluxomic approaches, we identify major metabolic profiles that differentiate metastatic prostate epithelial CSCs (e-CSCs) from non-CSCs expressing a stable EMT. We have found that the e-CSC program in our cellular model is characterized by a high plasticity in energy substrate metabolism, including an enhanced Warburg effect, a greater carbon and energy source flexibility driven by fatty acids and amino acid metabolism and an essential reliance on the proton buffering capacity conferred by glutamine metabolism. An analysis of transcriptomic data yielded a metabolic gene signature for our e-CSCs consistent with the metabolomics and fluxomics analysis that correlated with tumor progression and metastasis in prostate cancer and in 11 additional cancer types. Interestingly, an integrated metabolomics, fluxomics and transcriptomics analysis allowed us to identify key metabolic players regulated at the post-transcriptional level, suggesting potential biomarkers and therapeutic targets to effectively forestall metastasis.

Introduction

Tumors represent heterogeneous collections of neoplastic and non-neoplastic cells with divergent phenotypes engaged in complex interactions [1]. Two major operative categories of cancer cells are cancer stem cells (CSCs), endowed with self-renewal and tumor-initiating potentials, and non-CSCs [2]. These two broad functional classes of tumor cells do not necessarily reside in mutually exclusive subpopulations as cell plasticity allows phenotypic switching between CSC and non-CSC states [3].

It has been shown that the epithelial-mesenchymal transition (EMT) can endow cancer cells with stem cell properties, switching from an epithelial program to a motile mesenchymal phenotype [4]. However, increasing evidence suggests that CSC populations in solid tumors may be heterogeneous, including a substantial proportion of epithelial stem cells in which stemness is uncoupled from EMT, suggesting new mechanism driving metastatic colonization [5]. In addition, recent studies have demonstrated cooperative interactions between cell subpopulations displaying epithelial CSCs (e-CSCs) and non-CSCs that potentiate the metastatic behavior of the combined tumor cell populations [6–9]. The existence of intrinsic tumor cell heterogeneity has important implications in cancer drug resistance and therefore the identification of the most salient signaling and biochemical networks that characterize CSC and EMT phenotypes, including their prevalent metabolic states, is essential in order to design tumor subpopulation-specific therapies.

Although several common metabolic features of cancer cells have been extensively described, such as energy production via enhanced glycolysis (Warburg effect) [10, 11], few studies have been reported on metabolic states specific to CSCs and even fewer to metastatic CSCs. For example, CSCs have been reported either to display enhanced aerobic glycolysis with a concomitant reduction in mitochondrial activity [12] or to preferentially maintain oxidative phosphorylation (OXPHOS) and β-oxidation [13]. On the other hand, the induction of an epithelial-mesenchymal transition (EMT), which can lead to the acquisition of CSC properties in cancer cells [4], has been associated with reduced mitochondrial metabolism and enhanced glycolysis through the suppression of fructose-1,6-bisphosphatase [14] or the inhibition of cytochrome C oxidase [15]. Moreover, inhibition of the enzyme citrate synthase [16] or succinate dehydrogenase subunit B [17] results in a bioenergetic disorder that contributes to the acquisition of an EMT phenotype. There are also evidences that EMT-driven CSCs can metabolize alternative high-energy metabolites, displaying the phenomenon known as “reverse Warburg effect” [18]. In contrast, it has also been shown that mitochondrial biogenesis sustains stem-like properties [19] or support propagation and motility of circulating cancer cells expressing EMT traits [20]. All these observations lead to a complex scenario in which it is difficult to clearly associate a specific metabolic reprogramming associated with CSCs independent of EMT that supports the processes of invasion and metastasis. Hence, a consensus is still lacking on major characteristics that distinguish CSCs from non-CSCs.

To address this deficit, we have performed a comprehensive metabolic characterization of a dual cell model derived from the PC-3 cell line [6] consisting in one highly metastatic subpopulation (PC-3M) enriched in e-CSC features and a second non-metastatic and highly invasive subpopulation (PC-3S) lacking features of CSC (non-CSCs) and displaying a stable EMT. Regardless of tissue of origin of these cells, this model is unique in that CSC and EMT properties are fully uncoupled and displayed by distinct cell subpopulations and thus it offers an ideal cell model to uncover molecular mechanisms and pathways, including metabolic reprogramming, that can be specifically ascribed to either process.

From the integrated analysis of their main bioenergetic pathways and carbon sources, studied by a combination of metabolomics and fluxomics approaches, we have found that our e-CSC subpopulation, in contrast to the mesenchymal-like non-CSC subpopulation, exhibits a strong Warburg effect and a high potential to use alternative mitochondrial substrates, both metabolic features being essential to sustain its stem cell phenotype. Our results also highlight the contribution of amino acids metabolism and specially point out the importance of glutamine to compensate the acidic stress derived from the Warburg effect in the e-CSCs. A computational analysis based on transcriptomic data has identified a metabolic gene signature significantly associated with e-CSC that supports our metabolic analysis and correlates with tumor progression and metastasis in prostate cancer and other 11 tumor types. We thus provide an integrated view and selective metabolic dependencies and vulnerabilities of e-CSCs that may open new avenues to design therapeutic strategies aimed at targeting specific CSC and non-CSC subpopulations.

Materials and Methods

(detailed descriptions of all Materials and Methods are provided as online Supplemental Materials)

Cells and cell culture

PC-3M and PC-3S, clonally derived from the human cell line PC-3, and PC-3M cell variants were obtained as described [6]. Cells were cultured at 37°C in a 5% CO₂ atmosphere in RPMI 1640 media (Sigma-Aldrich or Biowest) supplemented with 10 mM glucose and 2 mM glutamine (unless otherwise indicated), 10% Fetal Bovine Serum (FBS) (PAA Laboratories), 1% pyruvate (1 mM) (Biological Industries), 1% streptomycin (100 µg/mL) / penicillin (100 units/mL) (Gibco) and 1% nonessential amino acids (Biological Industries).

Cell proliferation and viability assay

Proliferation and viability were assessed by Hoechst staining measuring emitted fluorescence.

Spheroid formation assay

Cells (6×10³/well) were seeded on 24-well Ultra Low Attachement culture plates (Corning) in media containing EGF, BFGF, heparine, B27, insulin and hydrocortisone and allowed to grow for 7 days: At the end of the experiment, spheroids were stained, incubating cells with 0.5 mg/mL MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide) for 2–4 h and then each well was scanned for posterior spheroid quantification, using ImageJ software (free software that can be downloaded from http://imagej.nih.gov/ij/download.html).

Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR)

These parameters were determined with a XF24 Extracellular Flux Analyzer (Seahorse Bioscience).

Concentration of media metabolites

Glucose, lactate, glutamate and glutamine media concentrations were determined by NADH-coupled reactions. All other amino acids were determined by ion-exchange chromatography.

Intracellular ATP, ROS and glutathione levels

ATP levels were quantified with the CellTiter-Glo Luminescent Cell Viability Assay (Promega, Madison, WI). Total intracellular ROS levels were determined using the cell-permeant probe 2',7'-dichlorodihydrofluorescein diacetate (H₂DCFDA) (Invitrogen) and flow cytometry. Total glutathione content was determined by the glutathione reductase enzymatic method.

Enzyme activities

LDH, TKT, G6PDH and GPT specific activities were determined by using NAD(P)H-coupled reactions.

¹³C-tracer-based metabolomics: GC/MS analysis

Cells were cultured in media containing either 100% or 50% [1,2-¹³C₂]-glucose, 100% [U-¹³C₆]-glucose or 100% [U-¹³C₅]-glutamine. ¹³C-isotopologue distributions of metabolites were determined by mass spectrometry-coupled gas chromatography (GC/MS). GC/MS spectral data were corrected by regression analysis to eliminate contribution from natural enriched isotopes in derivatization reagents.

Quantification of metabolic fluxes

Metabolic flux distribution was evaluated through computer simulations of the measured distribution of ¹³C isotopologues of metabolites implemented in Isodyn [21].

Statistical analysis

Two-tailed Student’s t-test for independent samples was applied where appropriate. In figures, bars represent mean ± standard deviation (SD), * indicates significance at p-value < 0.05, **; p < 0.01 and ***; p < 0.001; ns, not significant. All experiments shown were performed at least in triplicate

Results

Glycolysis is essential to support cell growth and stemness features of e-CSCs

As a cellular model to help elucidate major bioenergetic pathways of cells that display CSC properties uncoupled from EMT, we resorted to a dual cell model derived from the PC-3 cell line [6] consisting in one highly metastatic subpopulation (PC-3M) enriched in e-CSC features and a second non-metastatic and highly invasive subpopulation (PC-3S) lacking features of CSC (non-CSCs) and displaying a stable EMT. The CSC features of PC-3M cells have been thoroughly characterized by us [6] and supported by their expression of markers characteristic of stem cells such as KLF4, MYC, SOX2 or LIN28, strong enrichment in an embryonic cell (ES)-like gene module [22] and a MYC-centered gene module [23]. Regardless of tissue of origin of these cells, this model is unique in that CSC and EMT properties are fully uncoupled and displayed by distinct cell subpopulations and thus it offers an ideal cell model to uncover molecular mechanisms and pathways, including metabolic reprogramming, that can be specifically ascribed to either process.

We first studied the state of glycolysis in our PC-3M (e-CSCs) and PC-3S (non-CSCs) dual-cell model. The extracellular acidification rate (ECAR), a surrogate for lactic acid derived from glycolysis, was significantly higher in PC-3M cells than in PC-3S cells (Fig. 1A). Consistently, PC-3M cells consumed more glucose and produced more lactate (Fig. 1B) and exhibited a significantly higher lactate dehydrogenase (LDH) activity than PC-3S cells (Fig. 1C). Studies with incorporation of [1,2-¹³C₂]-glucose indicated that both glycolysis and the pentose phosphate pathway (PPP) contribute to the increased lactate production in PC-3M cells (Fig. 1D and Supplemental Table 1). These data suggest a more robust Warburg effect in PC-3M cells as compared to PC-3S cells.

Figure 1.

Higher glycolytic flux and dependence of PC-3M cells. (A): Comparison of PC-3M and PC-3S cells for normalized basal ECAR, (B): glucose consumption and lactate production rates and (C): LDH activity. (D): Relative contributions of glycolysis, PPP and other sources (OS) to lactate production. (E): Assessment of the Crabtree effect on OCR after exposure of cells to 18.75 mM glucose. (F): Effect of glucose deprivation (−Glc) and 5 mM 2-DG treatment (48 h) on cell proliferation. Shown are percentages of proliferation relative to cells cultured in full media and without 2-DG (Control). (G): Intracellular ATP levels analyzed after 24 h incubation with 5 mM 2-DG, normalized to DNA content. (H): Spheroid growth of PC-3M and PC-3S cells (x200). (I): Effect of 2-DG on the spheroid growth capacity of PC-3M cells (x200). (J): Glucose consumption and lactate production rates. (K): Effect of 5 mM 2-DG treatment (48 h) on cell proliferation. Percentages of proliferation relative to untreated control PC-3M, PC-3M/Snai1 or PC-3M/SKMkd cells (100% proliferation). Two-tailed Student t-test was used for statistical analyses. In panels A-E, H, significance was determined for PC-3M vs. PC-3S; F-G, I, treatment vs. control; J-K, variant cell lines vs. PC-3M. *p < 0.05, **p < 0.01 and ***p < 0.001. Abbreviations: 2-DG, 2-deoxyglucose; DCA, dichloroacetate; ECAR, extracellular acidification rate; LDH, lactate dehydrogenase; OCR, oxygen consumption rate; PPP, pentose phosphate pathway.

To further analyze the preference of PC-3M cells for glycolysis over oxidative phosphorylation, we evaluated the level of suppression of mitochondrial respiration after treatment with high glucose concentrations, or Crabtree effect [24]. Glucose treatment elicited a significantly greater reduction of mitochondrial respiration in PC-3M cells than PC-3S cells (Fig. 1E), illustrating a preference of the e-CSC subpopulation for metabolizing glucose through glycolysis. Glucose deprivation or treatment with the glycolytic inhibitor 2-deoxyglucose (2-DG) decreased the proliferation of PC-3M cells more than PC-3S cells (Fig. 1F), partly explained by cell death (Supplemental Fig. 1A) and an accumulation in the G1 phase of the cell cycle (Supplemental Fig. 1B).Treatment with 2-DG decreased cellular ATP levels in both cell subpopulations but more so in PC-3M cells (Fig. 1G). Cell growth in anchorage-independent conditions is a functional assay that correlates with stemness. PC-3M cells have a better ability than PC-3S cells to form spheroids under such conditions (Fig. 1H). Furthermore, we found that disruption of glycolysis by 2-DG treatment significantly affected the capacity of PC-3M cells to grow in suspension (Fig. 1I), highlighting the importance of glycolysis for capacity of PC-3M cells to grow in suspension as spheroids.

To determine the involvement of EMT and self-renewal gene networks for the above glycolytic phenotype, PC-3M cells were induced to acquire an EMT through forced overexpression of Snai1 (PC-3M/Snai1), or their self-renewal properties inhibited through knockdown of the pluripotency factors SOX2, KLF4 and MYC (PC-3M/SKMkd). Both genetic manipulations caused a reduced glycolytic phenotype (Fig. 1J) and diminished the sensitivity of PC-3M cells to 2-DG (Fig. 1K). Therefore, the marked glycolytic phenotype and glucose dependence observed in PC-3M cells requires the maintenance of robust epithelial and self-renewal gene programs.

Mesenchymal-like non-CSCs are more dependent on mitochondrial function than e-CSCs

The above results demonstrate a marked Warburg effect and a higher dependence on glycolysis in PC-3M cells than PC-3S cells. We further investigated the reliance of these cells on mitochondrial respiration using four mitochondrial drugs (Fig. 2A). Oligomycin, an ATP synthase inhibitor, reduced more oxygen consumption rate (OCR) (Fig. 2B) and had a more significant dose-dependent proliferation inhibitory effect (Supplemental Fig. 2) in PC-3S cells than PC-3M cells, suggesting a greater dependence on mitochondrial metabolism for cell growth of the non-CSC subpopulation. However, this drug did not significantly affect ATP levels in either subpopulation (Fig. 2C). Exposure of cells to trifluorocarbonylcyanide phenylhydrazone (FCCP), a mitochondrial uncoupling agent, boosted OCR more in PC-3S cells than PC-3M cells (Fig. 2B), indicative of a significantly higher mitochondrial respiratory capacity of the non-CSC subpopulation. Moreover, exposure to the mitochondrial complex I and III inhibitors rotenone and antimycin reduced OCR in both cell subpopulations, although significantly more in PC-3S cells (Fig. 2B). Consistent with mitochondrial metabolic activity being the main source of reactive oxygen species (ROS) [25], PC-3S cells exhibited higher basal levels of ROS than PC-3M cells (Fig. 2D). In support of the importance of the epithelial and pluripotency gene programs for these metabolic readouts, PC-3M/Snai1 and PC-3M/SKMkd variants were more sensitive than parental PC-3M cells to all mitochondrial drugs (Fig. 2E).

Figure 2.

Enhanced mitochondrial function of PC-3S cells relative to PC-3M cells. (A): OCR profiles generated after exposure to oligomycin, FCCP + Pyr and rotenone + antimycin. (B): OCR fold change (Log₂) after oligomycin, FCCP + Pyr or rotenone + antimycin injections. (C): Intracellular ATP levels analyzed after incubation (24 h) with 2 µM oligomycin. (D): Intracellular ROS levels. (E): OCR fold change (Log₂) after oligomycin, FCCP + Pyr and rotenone + antimycin injections. (F): Glutamine consumption in response to disruption of glycolysis, determined after 48 h incubation (48 h) in standard media conditions (Control), without glucose or with 5 mM 2-DG. (G): Dependence of PC-3M and PC-3S cells on glucose and glutamine availability for mitochondrial respiration. Normalized basal OCR levels determined in full and restricted media conditions. (H): Ketogenic amino acids consumption rates. (I): Effect of BPTES, etomoxir and oligomycin on spheroid growth capacity of PC-3M cells (x200). Significant differences were assessed by two-tailed Student’s t-test. In panels B, D, G-H, significance was determined for PC-3M vs. PC-3S; E, variant cell lines vs. PC-3M; I, treatment vs. control. *p < 0.05, **p < 0.01 and ***p < 0.001. Abbreviations: 2-DG, 2-deoxyglucose; FCCP, trifluorocarbonylcyanide phenylhydrazone; OCR, oxygen consumption rate; Pyr, pyruvate.

We next stained cells with MitoTracker to investigate the possible association of mitochondrial morphology with the above bioenergetics phenotypes (Supplemental Figure 2). PC-3M cells exhibited fragmented mitochondrial networks consistent with extensive mitochondrial fission, which has been associated with less efficient OXPHOS and a higher reliance on anaerobic glycolysis for bioenergetics [26]. In contrast, the mitochondrial morphologies in PC-3S cells were more compact, consistent with active fusion events [27], a process coupled to a robust mitochondrial function and bioenergetics [28]. Thus, the non-CSC subpopulation displays a mitochondrial organization predicted to support a better coupled mitochondrial respiratory chain for the generation of ATP.

A higher flexibility of mitochondrial metabolism is required for spheroid growth of e-CSCs

To investigate the degree of metabolic flexibility of these cells in the use of alternative carbon substrates, we evaluated the consumption of glutamine under glucose deprivation and with 2-DG treatment. These conditions caused enhanced glutamine consumption in PC-3M cells but not PC-3S cells (Fig. 2F), suggesting that PC-3M cells are more adept at using alternative carbon substrates. This was further explored by evaluating baseline OCR in full media (with glucose and glutamine), in restricted media (lacking glucose or glutamine) or minimal media (lacking glucose and glutamine). PC-3M cells showed higher baseline OCR values than PC-3S cells in all restricted media conditions, while no differences were detected in full media (Fig. 2G). This illustrates a robust switching to alternative carbon sources in PC-3M cells and also indicates that, although PC-3M cells preferentially metabolize glucose through glycolysis rather than oxidative metabolism, their mitochondrial respiration is functional and can be boosted when prompted. Interestingly, baseline OCR values for PC-3S cells in the absence of glutamine were lower than those in full media conditions (p = 0.004) (Fig. 2G), suggesting that glutamine is essential for the mitochondrial respiration of PC-3S cells but not PC-3M cells.

To test the ability of PC-3M cells to use alternative mitochondrial substrates, we analyzed the consumption of ketogenic amino acids, which can be degraded into acetyl-CoA. PC-3M cells consumed more glutamine (Fig. 2F) and the ketogenic amino acids leucine, isoleucine, lysine, threonine, tyrosine, tryptophan and phenylalanine than PC-3S cells (Fig. 2H). We examined the contribution of fatty acid β-oxidation by exposing cells to the carnitine palmitoyltransferase 1 (CPT1) inhibitor etomoxir, which impairs fatty acid β-oxidation, thus reducing OCR [29]. Etomoxir treatment decreased OCR in PC-3M cells more than in PC-3S cells (Supplemental Fig. 2). Consistently, PC-3M cells expressed higher CPT1 protein levels than PC-3S cells (Supplemental Fig. 2). To test if oxidation of fatty acids and glutamine as well as global synthesis of mitochondrial ATP are essential for the ability of PC-3M cells to grow in suspension, we treated these cells with the drugs bis-2-(5-phenylacetamido-1,3,4-thiadiazol-2-yl)ethyl sulfide (BPTES) (glutaminase inhibitor), etomoxir and oligomycin (Fig. 2I). Spheroid growth was significantly inhibited after these treatments, demonstrating that the alternative metabolic activities regarding mitochondrial metabolism are essential to sustain the stemness state of e-CSCs in our cell model.

The tricarboxylic acid (TCA) cycle in e-CSCs is mainly fueled by glutamine due to post-transcriptional inactivation of pyruvate dehydrogenase (PDH)

To investigate the contribution of glucose to TCA cycle intermediates, cells were cultured with [1,2-¹³C₂]-glucose and the isotopologue distribution for citrate, glutamate, fumarate, malate and aspartate determined by GC/MS (Supplemental Table 2). We observed that PC-3S cells incorporated label into TCA cycle intermediates more efficiently than PC-3M cells (Supplemental Table 2, 24 h). The labeling of pyruvate was similar in both cell types, suggesting that the e-CSC subpopulation diverts more pyruvate derived from glycolysis away from mitochondria than the non-CSC subpopulation, consistent with a marked Warburg effect. The relative rates of pyruvate oxidation in the TCA cycle were investigated by calculating the ratio of the m2 isotopologues of the above mentioned metabolites (m2TCA) to the m2 pyruvate isotopologue (m2Pyr) (Fig. 3A). We found lower ratios of m2TCA to m2Pyr in PC-3M cells than in PC-3S cells (Fig. 3B), indicative of decreased incorporation of labeled pyruvate into the TCA cycle.

Figure 3.

Glucose and glutamine contribute differentially to the synthesis of TCA cycle intermediates in PC-3M and PC-3S cells. (A): Labeling distribution of TCA cycle intermediates from [1,2-¹³C₂]-glucose. In red, labeling patterns obtained from PDH; in blue, patterns corresponding to ME and PC. (B): Ratios of m2 citrate, glutamate, fumarate, malate and aspartate normalized to m2 pyruvate labeling. (C): Labeling distribution of TCA cycle intermediates from [U-¹³C₅]-glutamine, considering oxidative metabolism (red) or reductive carboxylation (blue) of glutamine. (D): Left, m4 (m5 for glutamate) labeling of TCA cycle intermediates. Right, m3 (m5 for citrate) labeling of aspartate, malate and fumarate after incubation (24 h) with 100% [U-¹³C₅]-glutamine. (E,F): Sequential DCA and oxamate injections (1^st, 2^nd, 3^rd and 4^th) performed at 10, 30, 50 and 70 mM oxamate, and 10, 20, 30 and 40 mM DCA. (G): Gene expression levels of PDHK1 and PDP2, determined by real-time RT-PCR. Inset: PDH and PDH phosphorylated (PDH-P) protein levels, determined by Western blotting. (H): Isodyn predictions of increased TCA cycle fluxes and mitochondrial respiration in PC-3S cells. Left, schematic representation of reactions participating in the TCA cycle. Right, estimated metabolic fluxes in PC-3M and PC-3S cells. Metabolites: α-KG, α-ketoglutarate; AcCoA, acetyl-CoA; Cit, citrate; Glu, glutamate; Fum, fumarate; Mal, malate; OAA, oxalacetate; Pyr, pyruvate. Subscripts: m, mitochondrial; c, cytosolic. Fluxes: akgfum, α-ketoglutarate to fumarate; citdmc, citrate from mitochondria to cytosol; citakg, citrate to α-ketoglutarate CS, citrate synthase; ME, malic enzyme; PC, pyruvate carboxylase; PDH, pyruvate dehydrogenase; resp, mitochondrial respiration (NADH-linked respiratory consumption rate). In panels B, D, G, significance was determined for PC-3M vs. PC-3S; E-F, oxamate vs. DCA treatments. Differences were considered significant when *p < 0.05, **p < 0.01 and ***p < 0.001 (Student’s t-test). Abbreviations: DCA, dichloroacetate; PDHK1, pyruvate dehydrogenase kinase 1; PDP2, pyruvate dehydrogenase phosphatase 2; TCA, tricarboxylic acid.

Next, to investigate the contribution of glutamine to the TCA cycle in our cell model, we analyzed the incorporation of ¹³C in the TCA cycle intermediates after incubation with [U-¹³C₅]-glutamine (Supplemental Table 3). We compared m4 isotopologues (fumarate, malate, citrate, aspartate) and m5 isotopologues (glutamate) to evaluate the glutamine oxidative pathway (Fig. 3C). We observed that the contribution of glutamine to these labeling patterns was more pronounced in PC-3M cells than in PC-3S cells (Fig. 3D, left). Analysis of m5 citrate and m3 aspartate, malate and fumarate indicated that the metabolism of glutamine through reductive carboxylation (Fig. 3C) is more active in PC-3M cells than in PC-3S cells (Fig. 3D, right). These results are consistent with a higher ability of PC-3M cells for using alternative metabolic pathways to sustain energy production and to rapidly adapt to changes in nutrient availability.

PDH is a key enzyme that mediates the entry of pyruvate to mitochondria whose activity is regulated by its phosphorylation status through the activity of pyruvate dehydrogenase kinase (PDHK) [30]. Treatment of cells with the PDHK inhibitor dichloroacetic acid (DCA) or the LDH inhibitor oxamate resulted in a more pronounced increase in mitochondrial respiration in PC-3M cells than in PC-3S cells, particularly in response to DCA (Fig. 3E–F). This suggests that PDH has a lower baseline activity in PC-3M cells than in PC-3S cells. Indeed, PDH was markedly more phosphorylated (i.e. less active) in PC-3M cells than in PC-3S cells (Fig. 3G, inset) and transcript levels for PDHK1 were higher and the PDH phosphatase PDP2 lower in PC-3M cells than in PC-3S cells (Fig. 3G).

We next used Isodyn [21] to integrate ¹³C-based isotopologue distributions (Supplemental Tables 2 and 3) and biochemical data (Supplemental Table 4) to estimate comparative metabolic flux distributions relative to glucose uptake in central carbon metabolic networks in both cell types (Supplemental Fig. 3). The resulting metabolic flux simulations involving TCA cycle-related reactions lend further support to the occurrence of an enhanced entry of pyruvate via PDH into the TCA cycle, increased TCA cycle fluxes (CS, citakg and akgfum) and a higher flux associated with mitochondrial respiration (resp) in PC-3S cells than in PC-3M cells, consistent with a higher diversion of glucose into mitochondrial metabolism (Fig. 3H).

Glutaminolysis protects cells from acidity generated from a marked Warburg effect: an Achilles heel of e-CSCs

To explore the role of an enhanced glutamine metabolism in PC-3M cells, we first determined the expression levels of glutaminase (GLS1), a direct MYC target frequently expressed at high levels in rapidly proliferating cells and CSCs [31, 32]. GLS1 was expressed at significantly higher levels in PC-3M cells than PC-3S cells, in particular the glutaminase C (GAC) isoform (Fig. 4A–B). The higher expression levels of GAC vs. the kidney-type glutaminase (KGA) in PC-3M cells were partly dependent on the maintenance of an epithelial gene program, as shown by a diminished GAC/KGA ratio in PC-3M/Snai1 cells as compared to control PC-3M cells (Fig. 4B). Exposure to the glutaminase inhibitor BPTES inhibited the proliferation of PC-3M cells more significantly than that of PC-3S cells (Fig. 4C), associated with a decreased proportion of cells in the S phase of the cell cycle (Supplemental Fig. 4). The sensitivity of PC-3M cells to growth inhibition by BPTES was attenuated by overexpression of Snai1 (Fig. 4C), illustrating that the strong dependence on glutaminolysis of PC-3M cells relies on the maintenance of an epithelial gene program.

Figure 4.

PC-3M cells are more dependent than PC-3S cells on non-anaplerotic metabolic contributions of glutaminolysis. (A): Protein levels for total GLS1 and its GAC or KGA isoforms as determined by Western blotting. (B): GAC/KGA transcript levels were determined by isoform-specific qPCR and expressed as fold change relative to PC-3M cells. (C): Sensitivity to glutaminase inhibition by BPTES. Cells were incubated with 10 µM BPTES (48 h). Shown are percentages of proliferation relative to control cells. (D): Lactate production after incubation (48 h) without (Control) or with 10 µM BPTES (E): Additive effects of BPTES and 2-DG on PC-3M cell death. after culturing cells in control media or media containing 5 mM 2-DG, 10 µM BPTES or 5mM 2-DG + 10 µM BPTES (48 h). Cell death was assessed by the annexin V-propidium iodide assay. Plots depict percentages of dead cells (apoptotic + necrotic). (F): Failure of DMK to rescue growth of BPTES-treated PC-3M or PC-3S cells. Cells were incubated (48 h) with 10 µM BPTES in the presence or absence of 2 mM DMK. (G): Levels of total intracellular glutathione. (H): Intracellular ROS levels after incubation (48 h) without (Control) and with 10 µM BPTES. (I): Intracellular glutathione levels in cells untreated or treated (48 h) with 10 µM BPTES. (J): Effect on cell proliferation of acidic media, and the combination of low pH with 10 µM BPTES (pH 7.1 + BPTES) or without bicarbonate buffering (pH 7.1 – NaHCO₃) after 48 h of incubation. Percentage of proliferation relative to control conditions (pH 7.9). (K): Effect of acidic media (pH 7.0) and acidic media + 10 µM BPTES on cell proliferation. Shown are percentages of proliferation relative to standard conditions (pH 7.9). In panels B-C, G, K, significance was determined vs. PC-3M; D-F, I-J, treatment vs. control. Significant differences were assessed by Student’s t-test. *p < 0.05, **p < 0.01 and ***p < 0.001. Abbreviations: 2-DG, 2-deoxyglucose; BPTES, bis-2-(5-phenylacetamido-1,3,4-thiadiazol-2-yl)ethyl sulfide; DMK, dimethyl α-ketoglutarate; GAC, glutaminase C; GLS1, glutaminase 1; KGA, kidney-type glutaminase.

Both PC-3M and PC-3S cells increased lactate production upon treatment with BPTES (Fig. 4D), suggesting the participation of metabolic compensatory mechanisms, and thus we tested the effects on viability of a combined inhibition of glutaminolysis and glycolysis. We observed a marked additive effect on cell death by combining BPTES and 2-DG treatments in PC-3M cells but not in PC-3S cells (Fig. 4E), indicating that glycolysis and glutaminolysis are essential for the optimal growth and survival of PC-3M cells.

Glutamine metabolism has been described to be important for several cancer cell types, but its differential essentiality for heterogeneous tumor cell subpopulations has not been addressed thus far [33, 34]. Because glutamine is an anaplerotic TCA cycle substrate [35], growth inhibition by BPTES is expected to be rescued by the cell-permeable α-ketoglutarate (α-KG) analog dimethyl α-ketoglutarate (DMK). Remarkably, cell growth was not rescued by DMK in BPTES-treated PC-3M or PC-3S cells (Fig. 4F), suggesting that glutamine exerts a survival function independent of its anaplerotic function. Glutamine contributes to other metabolic pathways, including synthesis of glutathione, NADPH production and pH homeostasis [36, 37]. In their basal state, PC-3S cells contained higher levels of ROS (Fig. 2D) and total intracellular glutathione (Fig. 4G) than PC-3M cells. Thus, the non-CSC subpopulation potentially counteracts higher levels of ROS with an accumulation of glutathione. Although ROS levels did not increase significantly in either cell subpopulation after BPTES treatment (Fig. 4H), glutathione content was significantly reduced in PC-3S cells (Fig. 4I), demonstrating the importance of glutamine metabolism for glutathione synthesis in these cells.

We next investigated the existence of an alternative mechanism underlying the differential sensitivity of PC-3M cells to glutaminase inhibition. An additional function of glutamine is to provide resistance to acidic conditions through the release of ammonia [36]. Compared to cells grown in standard pH 7.9 media, incubation in more acidic media (pH 7.1) had a more significant impact on the proliferation of PC-3M cells than PC-3S cells (Fig. 4J). Addition of BPTES elicited an additive growth inhibitory effect that was more pronounced for PC-3M cells (Fig. 4J), as did depletion of sodium bicarbonate (NaHCO₃) (Fig. 4J). PC-3M/Snai1 and PC-3M/SKMkd cells showed enhanced resistance to BPTES treatment at low pH compared to control PC-3M cells (Fig. 4K), confirming the importance of the epithelial and pluripotency gene programs in conferring sensitivity to glutaminase inhibition. These observations suggest that the glutaminase reaction contributes to the buffering of surplus protons produced by PC-3M cells through their marked Warburg effect and consequent lactate production.

e-CSCs are characterized by an enhanced serine, glycine and one-carbon (SGOC) metabolism

Metabolic reprogramming during tumor progression induces alterations in biochemical states and preferences for amino acids that are closely linked to the Warburg effect [38, 39]. Glucose metabolism can be diverted from the glycolytic intermediate 3-phosphoglycerate towards the synthesis of serine, which in turn may be converted to glycine and participate in one-carbon metabolism [40]. Both amino acids provide essential precursors for the synthesis of proteins, nucleic acids and lipids [41] and the importance of the SGOC network in cancer has been analyzed in depth [42]. We analyzed the consumption and production profiles of serine and glycine and their isotopologue distributions after incubating cells with a ¹³C- enriched glucose tracer (Supplemental Table 5). PC-3M cells consumed more extracellular serine than PC-3S cells, while extracellular glycine was consumed by PC-3M cells and produced by PC-3S cells (Fig. 5A). After incubation of cells with [U-¹³C₆]-glucose, m3 serine was predominant in PC-3M cells, whereas more evenly distributed representations of m1, m2 and m3 serine isotopologues were found in PC-3S cells (Fig. 5B). The largely homogeneous m3 labeling of serine in PC-3M cells suggests that serine does not accumulate in these cells and is efficiently used as a building block for protein synthesis and/or funneled to catabolic reactions that rapidly transform serine to glycine (Supplemental Fig. 5). In line with this, failure to secrete glycine and relatively low levels of m2 glycine in PC-3M cells (Fig. 5C) suggest that glycine derived from serine is also rapidly used in these cells for biosynthetic purposes or cleaved to further contribute to one-carbon metabolic reactions. In support of this proposal, PC-3M cells expressed 5-fold higher levels than PC-3S cells of glycine decarboxylase (GLDC), the key enzyme catalyzing glycine cleavage (Fig. 5D). Moreover, PC-3M cells also expressed higher levels of many other enzymes involved in serine and one-carbon metabolism (Fig.5D) [40, 43, 44]. Thus, these results suggest that PC-3M cells have a more active SGOC metabolism than PC-3S cells.

Figure 5.

PC-3M cells are characterized by an active serine, glycine, one-carbon metabolism. (A): Serine and glycine consumption/production profiles in PC-3M and PC-3S cells after 96 h of culture. Negative values: consumption; positive values: production. (B): Extracellular serine isotopologue distribution after incubation (48 h) with 100% [U-¹³C₆]-glucose. (C): Extracellular glycine isotopologue distribution after incubation (48 h) with 100% [U-¹³C₆]-glucose. (D): Expression profile of genes involved in the glycine cleavage system, serine and one-carbon metabolism. Relative transcripts levels determined by qPCR represented as Log₂ of ratios between PC-3M and PC-3S cells. Asterisk, genes differentially expressed at p < 0.01 and Log₂ fold-change > 0.5 or < −0.5. In panels A-D, significance was determined for PC-3M vs. PC-3S. ***p < 0.001 (Student’s t-test).

Higher NADPH-generating reactions fuel an enhanced fatty acid synthesis and face oxidative stress in non-CSCs

The oxidative branch of the PPP uses glucose-6-phosphate as a substrate to generate NADPH, providing reducing power for other biosynthetic pathways (e.g. fatty acid synthesis) and to counter free radicals and oxidative stress. The non-oxidative branch of the PPP recycles pentose phosphates to glycolytic intermediates. Importantly, the PPP generates ribose-5-phosphate for nucleotide synthesis. Two key PPP enzymes are glucose-6-phosphate dehydrogenase (G6PDH) in the oxidative branch and transketolase (TKT) in the non-oxidative branch. Label incorporation to ribose from [1,2-¹³C₂]-glucose was significantly greater in PC-3M cells than in PC-3S cells (Fig. 6A and Supplemental Table 2B), suggestive of a larger demand for nucleotide biosynthesis to sustain their higher proliferation rate. Analysis of ribose isotopologue distribution revealed an increase in m1 (oxidative PPP) and m2 (non-oxidative PPP) isotopologues in PC-3M cells (Fig. 6B). However, the analysis of the m1/m2 ribose ratio also indicated a differential contribution of the oxidative and non-oxidative branches of the PPP in these cells (Fig. 6C), suggesting that the highly glycolytic PC-3M cells redirect back part of the glucose-based PPP intermediates to glycolysis through the non-oxidative branch, thus resulting in a higher production of lactate through this pathway (Fig. 1D). The differential use of the PPP branches was further supported by the observation of significantly higher enzymatic activity and expression levels of G6PDH in PC-3S cells and TKT in PC-3M cells (Fig. 6D–F).

Figure 6.

Enhanced NADPH-producing fluxes and fatty acid synthetic reactions in PC-3S cells. (A): Total ¹³C-ribose labeling represented as Σm (m1+m2+m3+m4+m5 labeled ribose). (B): Ribose isotopologue distribution. (C): Oxidative vs. non-oxidative branches of the PPP represented as m1/m2 ribose ratio. (D): G6PDH activity. (E): TKT activity. (F): G6PDH and TKT gene expression levels determined by qRT-PCR. (G): Synthesis of palmitate and (H) stearate from glucose in PC-3S cells and PC-3M cells. Isotopologue distribution of palmitate and stearate after incubation (24 h) with 50% [1,2-¹³C₂]-glucose. (I): ATP citrate lyase (ACLY) protein levels, assessed by Western blotting. Actin was used as a protein loading and transfer control. Two-tailed Student t-test was used for statistical analyses. In panels A-H, significance was determined for PC-3M vs. PC-3S. *p < 0.05, **p < 0.01 and ***p < 0.001. Abreviations: G6PDH, glucose-6-phosphate dehydrogenase; PPP, pentose phosphate pathway; TKT, transketolase.

The above results suggest a higher demand of NADPH in PC-3S cells, likely in order to face higher levels of oxidative stress resulting from accumulation of ROS. Isodyn analysis indicated that the fluxes for NADPH-producing reactions, such as cytosolic malic enzyme (ME) and isocitrate dehydrogenase (participating in the citakg flux), are higher in PC-3S cells than in PC-3M cells (Fig. 3I). Isodyn also predicted an enhanced efflux of citrate from the mitochondrial to the cytosol compartment (citdmc flux) in these cells (Fig. 3I). NADPH reducing equivalents actively participate in many biosynthetic reactions, such as fatty acid synthesis. After incubation of cells with [1,2-¹³C₂]-glucose, greater yields of m2 and m4 labeled palmitate (Fig. 6G) and stearate (Fig. 6H) were detected in PC-3S cells than in PC-3M cells, suggesting that glucose-derived carbons are more efficiently routed towards fatty acid synthesis in PC-3S cells. Protein expression levels of ATP citrate lyase (ACLY), the primary enzyme responsible for cytosolic fatty acid biosynthesis, were also higher in PC-3S cells than PC-3M cells (Fig. 6I), confirming a greater coupling between glycolysis and fatty acid synthesis.

Trancriptome analysis reveals significant associations between the metabolic fingerprint of e-CSCs and malignant progression of several cancers

To determine the potential significance to tumor progression of those metabolic pathways differentially expressed in the e-CSC vs. non-CSC prostate model, we performed a Rank Product analysis [45] of transcriptomic data generated for the PC-3M and PC-3S subpopulations [6]. The resulting PC-3M-overexpresed metabolic gene signature included enzymes involved in the metabolism of serine and glycine (GLDC, CBS), branched-chain amino acids (BCAT1), and glutamate and proline (ASS1, AGMAT), which support the metabolic flexibility accompanying the CSC phenotype. It also included genes that participate in the synthesis of purine nucleotides (GUCY1A3 and PDE3B) as part of reactions that converge on substrates that are actively metabolized in PC-3M cells, such as glycine and one-carbon units from tetrahydrofolate. This analysis allows to examine the metabolic underpinnings of the cellular programs that increase cancer cell aggressiveness and to understand how cellular metabolism contributes to these and other proliferation-independent features of cancer.

Next, by applying Gene Set Enrichment Analysis (GSEA) (Subramanian et al 2005), we found that the PC-3M-overexpressed metabolic gene set was significantly enriched concomitant with malignant progression in prostate cancer (Fig. 7) and in 11 other types of human tumors, including ovarian, bladder, adrenocortical, head and neck, stomach or rectal carcinoma, or melanoma (Supplemental Table 7). These observations link the metabolic pathways and reactions differentially expressed in our CSC vs. non-CSC cell model to human tumor progression.

Figure 7.

A metabolic gene set differentially expressed in PC-3M vs. PC-3S cells associated with cancer progression. Left, Gene Set Enrichment Analysis of a prostate cancer expression dataset [62] showing a significant enrichment of the PC-3M metabolic gene set in metastases (M) relative to primary tumors, and in T3 and T4 stage primary tumors relative to T1 and T2 stage primary tumors. A Pearson correlation was applied to determine linear relationships between gene profiles and four phenotypes (class 1: metastatic; class 2: T4 stage primary; class 3: T3 stage primary; class 4: T1 and T2 stage primary) taken as continuous variables. Right, heatmap illustrating the relative expression levels of the 21 genes of the PC-3M metabolic gene set in 150 prostate cancer samples of the same dataset, highlighting selected genes.

Discussion

The cancer stem cell hypothesis implies intratumoral phenotypic heterogeneity. Neoplastic cells can reversibly switch between CSC and non-CSC states [3], presenting a hurdle to efforts at targeting specific tumor subpopulations that may escape from therapeutic pressure [46]. A successful strategy to forestall these resistance mechanisms should incorporate the characterization of biochemical and gene networks, including those related with cell metabolism, to uncover endogenous vulnerabilities pertaining to both CSCs and non-CSCs.

Previous studies have failed to discriminate whether the metabolic reprogramming observed in tumor cells with CSC features were explained by EMT [14, 16, 20, 47] or CSC states, which recent studies have found that may occur in cancer cells unlinked to EMT [6, 48]. In order to address this issue, we have undertaken a systematic comparative metabolic characterization of a dual-cell tumor cell model in which CSC and EMT programs are clearly uncoupled. Our comparative analysis sheds light on major metabolic pathways and vulnerabilities accompanying self-renewal and stemness states uncoupled from EMT as schematically summarized in Supplemental Figure 6. Thus, the e-CSC PC-3M subpopulation is highly dependent on aerobic glycolysis for cell proliferation and bioenergetics, largely uncoupled from the TCA cycle and OXPHOS. In contrast, the mesenchymal-like non-CSC PC-3S subpopulation relies more on mitochondrial respiration for the generation of energy and metabolizes significant amounts of glucose through the TCA cycle. However, in nutrient-restricted media, PC-3M cells showed their versatile metabolic activity resulting from their greater ability to use alternative substrates to feed into the TCA cycle. Based on these metabolic characteristics, a promising strategy to target CSC subpopulations should consider the interplay between the various metabolic pathways (e.g. glycolysis, glutaminolysis and fatty acid oxidation) that supports their viability and stemness properties [49].

Published evidences suggest a predominance of glycolysis and a Warburg effect in association with breast and lung cancers and glioblastoma CSCs [12, 50–52], which may also display an EMT. Together with such data, our observations lead us to conclude that CSCs engage a Warburg effect, independently of whether their phenotype is associated with an epithelial (our model) or a mesenchymal (other models) gene program. In seeming contradiction, other authors have found that pancreatic CSCs and/or drug-resistant tumor populations preferentially use mitochondrial metabolism over glycolysis for cellular energetics [13, 53]. However, in at least one of these studies it was found that the emergence of CSCs resistant to a specific drug treatment (metformin) was associated with a significantly enhanced glycolytic capacity and lactate production [53], an observation that suggests that enhanced glycolysis may be a feature of the more aggressive CSC subpopulations. In this regard, we note that the PC-3M cells used in our study as our model CSCs are a highly metastatic subpopulation isolated through positive selection in vivo. Importantly, our data demonstrate that the expression of EMT features does not necessary correlate with a higher glycolytic phenotype. High glycolytic fluxes likely benefit cells with elevated proliferation rates through the production of glycolytic intermediates for de novo biosynthesis of nucleotides, lipids and proteins [54] and previous studies have suggested that the glycolytic metabolism may be a broadly conserved stem cell property [55] and associated with undifferentiated states [56, 57]. The situation described here, in which one tumor cell subpopulation displays high glycolytic activity and marked Warburg effect, while a second subpopulation relies more strongly on mitochondrial metabolism for energy supply, is akin to the so-called “metabolic asymmetry” of tumors [58], in which different cell populations within a tumor, including neoplastic vs. non-neoplastic cells, express very different metabolic profiles [59] that contribute to potential co-operation between such subpopulations to foster more aggressive overall phenotypes [60, 61]. Our observations suggest that neoplastic cell subpopulations expressing uncoupled CSC or EMT features may cooperate not only through cell-cell contacts or diffusible protein factors [60], but also through metabolic co-operation.

Amino acid metabolism has been far less explored than glucose metabolism but its biological relevance for the study of tumor progression is gaining ground. Our stable isotope labeling studies also revealed that PC-3M cells actively metabolized the amino acids serine and glycine, while PC-3S cells presented a more limited use of these amino acids as shown by an accumulation of m1, m2 and m3 serine isotopologues and excretion of surplus glycine. Therefore, differences in the expression levels of genes involved in the SGOC metabolism could be good biomarkers of tumor progression to metastatic disease [42].

Glutaminolysis normally serves to anaplerotically fill the TCA cycle, in particular in cells that shut down the entry of pyruvate into mitochondria as a consequence of PDH inhibition [35]. Interestingly, our analysis showed that, of the different functions of glutamine investigated, its protective role against acidic conditions better explained the growth and survival properties of PC-3M cells. Moreover, several metabolic pathways use glutamate-derived glutamine, including reactions involved in serine metabolism (Supplemental Fig. 5b). Targeting these metabolic pathways could additionally compromise the proliferation and viability of PC-3M cells and, by extension, CSCs, not only because of their specific role in amino acid metabolic networks but also as a consequence of compromised cell pH buffering capacity. In contrast, glutamine metabolism in PC-3S cells contributed substantially to their mitochondrial respiration and was more closely linked to glutathione synthesis which confers protection from oxidative stress. Thus, reasonable strategies to target non-CSCs that constitute part of the tumor bulk and with similar metabolic phenotypes might include interventions to limit antioxidant adaptive responses.

Finally, PC-3M cells displayed a more active non-oxidative branch of the PPP, which may be advantageous for their high glycolytic phenotype, allowing a reversible and dynamic connection between glycolysis and the PPP. In contrast, PC-3S cells showed a preference for the oxidative branch of the PPP, a key source of NADPH to support fatty acid synthesis and redox homeostasis maintenance. PC-3S cells also had a more pronounced fatty acid synthetic activity than PC-3M cells, in which fatty acid oxidation was more prevalent. From these observations, we predict a higher sensitivity of the non-CSC PC-3S cell subpopulation to perturbations in lipogenesis than the e-CSC PC-3M subpopulation, which in turn should be more susceptible to inhibitors of fatty acid β-oxidation.

Additionally, our computational analysis has permitted to further analyze the correlation between the metabolic adaptations and tumor malignancy in prostate cancer as well as in other tumor types. The metabolic pathways associated with the metabolic gene signature characteristic of PC-3M cells are consistent with our metabolomics and fluxomics analysis, which indicates the reliability of the metabolic gene signature inferred from our transcriptomic analysis. Nevertheless, this analysis failed to predict specific reactions that we have found, through our experimental approaches, to be relevant in the metabolic reprogramming of CSCs in our cell model. Thus, unlike the transcriptomic-based computational analysis, metabolomics and fluxomics approaches revealed that PDH is more active in PC-3S than in PC-3M cells, whereas the activities of GLS and CPT1, representatives of glutamine and fatty acid metabolism respectively, were increased in PC-3M cells. This is likely due to the fact that transcriptomic analyses do not consider the functional consequences of post-transcriptional modifications, such as the regulation of PDH activity by phosphorylation, the effects of substrate availability or the need for buffering capacity due to excess of protons derived from a marked Warburg effect.

Conclusion

In summary, we conclude that metabolic programs underlying CSC and EMT phenotypes can be mutually uncoupled and that each can be associated with preferred metabolic dependencies required for optimal cell proliferation and sustaining specific cell phenotypes. In this work, we have tackled the study of complex cellular phenomena associated with tumor progression and malignancy from a novel perspective that combines metabolomic, fluxomic and transcriptomic approaches. These complementary studies have widened the scope of our results, establishing new grounds for targeting tumor cell metabolism with a subpopulation-specific approach, providing insights into mechanisms supporting tumor progression and with potential impacts on the management of metastatic disease.

Supplementary Material

Supp Fig S1. Supplemental Figure 1.

Effect of the disruption of glycolysis on cell death and cell cycle. (A): Dead cells (apoptotic + necrotic) in PC-3M and PC-3S subpopulations untreated (Control) or treated with 5 mM 2-DG for 48 h. Cells were incubated with Annexin-V and analyzed by flow cytometry. (B): Cell cycle analysis of PC-3M (upper panel) and PC-3S cells (lower panel) untreated (Control) or treated with 5 mM 2-DG for 48 h. The harvested cells were stained with propidium iodide and their DNA content analyzed by flow cytometry. Data are provided as the means ± SD. *p < 0.05 and **p < 0.01 (Student’s t-test). Abbreviations: 2-DG, 2-deoxyglucose.

Supp Fig S2. Supplemental Figure 2.

(A): Higher sensitivity of PC-3S cell proliferation to oligomycin. Cells were incubated with the indicated oligomycin concentrations for 24 h and cell proliferation assessed. Percentages of proliferation relative to cells cultured in the absence of oligomycin (Control). Data are provided as means ± SD. *p < 0.05 (Student’s t-test). (B): Mitochondrial morphologies of PC-3M and PC-3S cells. Fluorescence emission from PC-3M and PC-3S cells after staining with MitoTracker (red) and DAPI (blue), which target mitochondria and nucleus, respectively. These stainings were carried out as described in Supplemental Materials. Scale bar, 10 µm. (C): The contribution of fatty acid metabolism to mitochondrial respiration was measured as the fold change (Log₂) in OCR after injection of 30 μM etomoxir, using a XF24 Extracellular Flux Analyzer, as described in Supplemental Materials. Bars represent mean ± SD. ***p < 0.001 (Student’s t-test). (D): Protein levels of CPT1 determined by Western blotting. Actin was used as a protein loading and transfer control. Abbreviations: CPT1, carnitinepalmitoyltransferase 1; OCR, oxygen consumption rate.

Supp Fig S3. Supplemental Figure 3.

Metabolic network implemented in Isodyn. Representation of the metabolic network considered in Isodyn. The experimental data integrated in Isodyn include the isotopologue distributions shown in Supplemental Tables 2A and 3 and the biochemical data shown in Supplemental Table 4. Green arrows: Glycolysis. Metabolites: F6P, fructose-6-phosphate; FBP, fructose bisphosphate; G6P, glucose-6-phosphate; Glc, glucose; Lact, lactate; PEP, phosphoenolpyruvate; Pyr, pyruvate; T3P, triose phosphates. Reactions are catalyzed by: ALD, aldolase; FBPase, fructose bisphosphatase; HK (hexokinase) combined with glucose transport into cells; PFK, phosphofructokinase; PK, pyruvate kinase. Magenta arrows: Pentose phosphate pathway. Metabolites: E4P, erytrose-4-phosphate; R5P, ribose-5-phosphate; S7P, sedoheptulose-7-phosphate; Xu5P, xylulose-5-phosphate. Reactions are catalyzed by: G6PDH, glucose-6-phosphate dehydrogenase combined with other reactions transforming G6P into R5P; TA, transaldolase; TKT, transketolase. Red arrows: TCA cycle with mitochondrial metabolites: AcCoA, acetyl-CoA; Cit, citrate; α-KG, α-ketoglutarate; Fum, fumarate; Mal, malate; OAA, oxaloacetate. Reactions: citakg, transformation from citrate to α-ketoglutarate; CS, citrate synthase; ME, malic enzyme; PC, pyruvate carboxylase and other anaplerotic/cataplerotic reactions; PDH, pyruvate dehydrogenase; PEPCK, phosphoenolpyruvatecarboxykinase. Dashed arrows indicate exchanges between cytosolic and mitochondrial metabolite pools: citdmc, exchange of citrate, from mitochondria to cytosol. Amino acids: Ala, alanine; Arg, arginine; Asp, aspartate; Cyst, cysteine; Gln, glutamine; Glu, glutamate; Gly, glycine; His, histidine; Ile, isoleucine; Leu, leucine; Lys, lysine; Met, methionine; Phe, phenylalanine; Pro, proline; Ser, serine; Thr, threonine; Tyr, tyrosine; Val, valine. Other: MTHF, 5-methyltetrahydrofolate; THF, tetrahydrofolate.

Supp Fig S4. Supplemental Figure 4.

Effect of BPTES on cell cycle distribution. Cell cycle analysis of PC-3M (upper panel) and PC-3S (lower panel) cells untreated (Control) or treated with 10 μM (BPTES) for 48 h. The harvested cells were stained with propidium iodide and their DNA content analyzed by flow cytometry. Plots depict the variation of the percentage of cells in each phase (G1, S or G2) of the cell cycle. Data are means ± SD. *p < 0.05 and **p < 0.01 (Student’s t-test).

Supp Fig S5. Supplemental Figure 5.

(A): Synthesis of glutamate from [U-¹³C₆]-glucose. Schematic representation of label distribution from [U-¹³C₆]-glucose to glutamate, after metabolization through glycolysis and one or two turns in TCA cycle. The entry of unlabeled acetyl-CoA coming from sources other than glucose is also considered. (B): Serine, glycine and one-carbon metabolism in PC-3M and PC-3S cells. Schematic representation of the reactions involved in the synthesis of m1, m2 and m3 serine from [U-¹³C₆]-glucose in PC-3M (upper panel) and PC-3S (lower panel) cells. PC-3M cells rapidly metabolize serine and glycine for biosynthetic pathways and additionally, glycine can be cleaved by the glycine cleavage system that is more active in these cells. As a result, glycine is not accumulated in these cells and cannot be converted back to serine, so only m3 serine is detected. On the other hand, PC-3S cells accumulate not only m3 serine, but also m1 and m2 serine. In these cells a limited activity of the glycine cleavage system can contribute to the accumulation of glycine and m1 and m2 serine isotopologues can result from the conversion of glycine back to serine, considering the participation of labeled and unlabeled substrates. In the scheme there is no distinction of the cellular compartments where reactions take place. GLDC is shown as a representative enzyme of the glycine cleavage system. Abbreviations: α-KG, α-ketoglutarate; GLDC, glycine decarboxylase; Glu, glutamate; OAA, oxaloacetate; PHGDH, phosphoglycerate dehydrogenase; PSAT, phosphoserine aminotransferase; PSPH, phosphoserine phosphatase; Pyr, pyruvate; SHMT, serine hydroxymethyltransferase; TCA, tricarboxylic acid; THF, tetrahydrofolate.

Supp Fig S6. Supplemental Figure 6.

Metabolic landscape and vulnerabilities of PC-3M and PC-3S cells. PC-3M cells display a high metabolic flexibility and capacity for using different metabolic pathways. The most relevant metabolic pathways and inhibitors in PC-3M cells are colored in red and those most relevant in PC-3S cells in blue. Inhibitors are inside purple ellipses. Abbreviations: 2-DG, 2-deoxyglucose; aa, amino acids; α-KG, α-ketoglutarate; Asp, asparatate; BCAA, branched-chain amino acids; BCKA, branched-chain keto acids; BPTES, bis-2-(5-phenylacetamido-1,3,4-thiadiazol-2-yl)ethyl sulfide; DCA, dichloroacetate; FCCP, carbonilcyanide p-trifluoromethoxyphenylhydrazone; GA3P, glyceraldehyde-3-P; Glu, glutamate; Lys, lysine; OAA, oxaloacetate; OXPHOS, oxidative phosphorylation; PPP, pentose phosphate pathway; THF, tetrahydrofolate.

*These reactions can take place either in the cytosol or the mitochondria.

Graphical Summary caption.

e-CSCs preferentially rely on aerobic glycolysis for bioenergetics, display an active serine-one-carbon-glycine metabolism and show an increased metabolic flexibility to utilize different carbon sources, such as fatty acids and glutamine, that offset the decreased diversion of glucose-derived carbons into the TCA cycle. Glutaminolysis is essential for the proliferation and survival of e-CSCs largely because the ammonia molecules released from glutamine metabolism neutralize the excessive levels of protons (lactic acid) that result from their marked Warburg effect. In contrast, non-CSCs display a strong coupling between aerobic glycolysis and OXPHOS and a strong dependence on the mitochondrial energy metabolism, which results in ROS production. As an adaptive response, non-CSCs engage several antioxidant defense mechanisms, such as enhanced synthesis of reducing equivalents (NADPH), to prevent the harmful effects of excessive ROS accumulation. e-CSCs, epithelial cancer stem cells; non-CSCs, non-cancer stem cells; THF, tetrahydrofolate; TCA, tricarboxylic acid; NADPH, reduced nicotinamide adenine dinucleotide phosphate; ROS, reactive oxygen species, OXPHOS, oxidative phosphorylation.