Interplay between TRAP1 and Sirtuin-3 Modulates Mitochondrial Respiration and Oxidative Stress to Maintain Stemness of Glioma Stem Cells

Glioblastoma multiforme (GBM), the most common malignant brain tumor, quickly develops resistance to both radiation and cytotoxic chemotherapy, which inevitably leads to poor clinical outcomes and tumor recurrence despite multimodal therapy (1). A small subpopulation of self-renewing tumor cells, that is, cancer stem cells (CSC) or tumor-initiating cells, isolated from patients with GBM is responsible for therapeutic resistance and tumor relapse (2, 3). Thus, targeting glioma stem cells (GSC) is a promising therapeutic approach for aggressive and relapsing tumors (3).

Some cancer cells are more dependent on aerobic glycolysis than on mitochondrial respiration; this phenomenon, called the Warburg effect, redirects metabolism toward biosynthetic pathways, reduces cellular oxidative stress, and supports proliferation and survival of cancer cells (4, 5). However, one caveat with respect to a metabolic shift toward glycolysis is a marked reduction in the ATP yield, which ultimately requires cancer cells to increase uptake of glucose (5, 6). This may limit tumor progression under reduced nutrient conditions. Growing evidence suggests that many cancer cells retain mitochondrial function for energy production (7, 8), and that certain cancer cells and CSCs have a metabolic preference for mitochondrial respiration rather than glycolysis (9, 10). GSCs are extremely dependent on mitochondrial respiration for energy production and survival (9, 11); however, little is known about the molecular mechanisms underlying metabolic reprogramming.

TNF receptor-associated protein 1 (TRAP1) is a mitochondrial paralog of the 90-kDa HSP (Hsp90) and is highly expressed by a variety of cancer cells (12, 13). TRAP1 endows cancer cells with a high cell death threshold and confers drug resistance by maintaining mitochondrial integrity (14–17). Thus, small-molecule inhibitors targeting mitochondrial TRAP1 have been developed to disrupt mitochondrial function and thereby induce cell death (14, 16, 18, 19). Recent reports suggest that metabolic regulation by TRAP1 alters mitochondrial respiration and subsequently reprograms cellular metabolism in some tumors (20–22). However, the underlying molecular mechanisms are unclear (23) and have not yet been examined in highly tumorigenic CSCs.

Sirtuins (SIRT) are NAD⁺-dependent protein deacetylases; among the seven mammalian SIRTs (SIRT1–7) identified to date, SIRT3, SIRT4, and SIRT5 are localized to the mitochondria (24). SIRT3, a major mitochondrial protein deacetylase, senses cellular metabolic stress and modulates energy and ROS metabolism (25–28), both of which are closely related to the pathophysiology of tumorigenesis. However, the contribution of SIRT3 to tumorigenesis is controversial; previous studies have identified both tumor-promoting and -suppressive roles (29, 30). This suggests that the precise role of SIRT3 depends on the cellular context; therefore, the specific type of cancer must be taken into account (30, 31). Studies suggest that SIRT3 has a tumor-promoting effect in the brain (32, 33), but no report to date has described SIRT3-dependent metabolic regulation of GSCs.

Here, we found that interplay between TRAP1 and SIRT3 in GSCs rewired cellular metabolism to generate energy-efficient mitochondrial respiration without overproduction of ROS. These metabolic reprogramming functions were crucial for maintenance of self-renewal capacity, lack of differentiation, and survival of GSCs, particularly under limited nutrient conditions. Inactivating either TRAP1 or SIRT3 compromised the metabolic pathways regulated by them and, subsequently, caused GSCs to lose stemness, leading to suppression of sphere formation in vitro and tumor growth in vivo.

All chemicals were purchased from Sigma. Anti-nestin and anti-Sox2 antibodies were purchased from R&D Systems; anti-TRAP1, anti-SOD2, anti-NDUFA9, anti-UQCRC2, anti-ATPB, anti-GFAP, and anti-SIRT3 antibodies were purchased from Abcam; the anti-Cox IV antibody was purchased from Cell Signaling Technology; anti-acetylated lysine, anti-GDH, anti-Ki-67, and anti-SDHB antibodies were purchased from Santa Cruz Biotechnology; the anti-β-actin antibody was obtained from MP Biomedicals; and the anti-Hsp60 antibody was purchased from BD Biosciences.

A total of 150 patients with glioma were enrolled in this cohort study conducted at Seoul National University Hospital (SNUH, Seoul, Republic of Korea). This included 108 randomly selected consecutive patients with glioma who underwent curative resection at SNUH (performed by the same surgical team) in 2014–2015. Patients in this cohort were followed until June 2017, with a median observation time of 22.5 months (range, 2–47 months). Overall survival was defined as the interval between the date of surgery and date of death or last observation. Data for patients not experiencing relapse or death were censored at the last follow-up. All experiments involving human subjects were approved by the Institutional Review Board (IRB) of SNUH (IRB no. 2014-08-004).

A tissue microarray (TMA) was constructed as described previously (34). All glioma cases were histologically reviewed by hematoxylin and eosin staining. Briefly, two cores were taken from each representative tumor tissue and from brain tissue adjacent to the tumor within a distance of 10 mm. These were used to construct TMA slides. Duplicate 1.0-cm diameter cylinders taken from intratumoral and peritumoral areas (a total of four punches per patient) were obtained and transferred to the recipient paraffin block at defined array positions (Biochip Company Ltd). TMA blocks were cut into 4-μm sections and mounted onto slides coated with 3-aminopropyltrioxysilane. After deparaffinization and antigen retrieval in citrate buffer (pH 6.0) for 20 minutes in a pressure cooker, slides were treated for 10 minutes with 3% hydrogen peroxide to block endogenous peroxidase. Anti-TRAP1 (1:200, monoclonal, Abcam, #ab109323) and anti-Sox2 (1:200, polyclonal; R&D Systems, #AF2018) were the primary antibodies. The TMAs were subsequently incubated with the Polink-2 Plus HRP detection system (GBI Labs) and visualized with 3,3′-diaminobenzidine (DAB). Negative controls were treated the same way, but without the primary antibodies. IHC staining was assessed by two independent pathologists with no prior knowledge of patient characteristics. Discrepancies were resolved by consensus. Extent of staining was scored on a scale from 0 to 4, corresponding to the percentage of immune-reactive tumor cells (0%, 1–5%, 6%–25%, 26%–75%, and 76%–100%, respectively). Staining intensity was scored as negative (score = 0), weak (score = 1), or strong (score = 2). A score ranging from 0 to 8 was calculated by multiplying the score for staining extent with that for intensity. Final grades (negative, 1+, 2+, and 3+) were assigned to each specimen with scores of 0–1, 2–3, 4–5, and 6–8, respectively.

Formalin-fixed, paraffin-embedded tissue samples were sectioned (7-μm thick) using a microtome and transferred to glass slides. After paraffin removal and antigen retrieval, the slides were blocked with 5% FBS, 5% BSA, and 0.3% Triton X-100 in PBS, and then incubated overnight at 4°C with antibodies specific for Sox2 and TRAP1. Next, the slides were incubated for 2 hours with Alexa Fluor 488 (green)- or Alexa Fluor 633 (red)-conjugated secondary antibodies (Invitrogen) and mounted in a mounting medium containing DAPI (Vectashield). Stained slides were analyzed under a FV1000 confocal microscope (Olympus).

Resected brain tumor tissues were collected at SNUH (Seoul, Republic of Korea) with written informed consent in accordance with institutional guidelines, and graded pathologically according to WHO grades (35). GBM tissues were formalin-fixed and paraffin-embedded for IHC or immunofluorescence analysis. All human GSC lines were cultured in DMEM/F-12 supplemented with B27 (Invitrogen), EGF (10 ng/mL; R&D Systems), and basic fibroblast growth factor (bFGF; 5 ng/mL; R&D Systems). Differentiated GSC cells were maintained in DMEM/F-12 supplemented with 10% FBS. Brain cancer cell lines (LN229, T98G, U87, and U251) and human astrocytes were purchased from the ATCC and maintained as recommended by the manufacturer. Cells were cultured in DMEM or RPMI medium (Gibco) containing 5% or 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco) at 37°C in a humidified atmosphere of 5% CO₂. To enrich stem cells, LN229 and T98G cells were cultured for 3 weeks in DMEM/F12 (Gibco) medium supplemented with 10 ng/mL EGF, 5 ng/mL bFGF, 0.04% B27, and 1% penicillin/streptomycin.

Oxygen consumption rate (OCR) was measured using a Mito Stress Test Kit and a XF24 Extracellular Flux Analyzer (Seahorse Bioscience), according to the manufacturer's protocol. Briefly, X01 or LN229 cells (2 × 10⁴) were attached to laminin-coated Seahorse XF24 cell culture microplates (Seahorse Bioscience) for 6 hours and then incubated with drugs for 48 hours at 37°C/5% CO₂. At the time of OCR measurement, the medium was changed to DMEM XF assay medium (Seahorse Bioscience) supplemented with 2.5 mmol/L glutamine, 0.5 mmol/L sodium pyruvate, and 17.5 mmol/L glucose (pH 7.4 at 37°C). After equilibration for 30 minutes at 37°C in the absence of CO₂, OCR was recorded using the Seahorse XF Analyzer (Seahorse Bioscience). Oligomycin (0.4 μmol/L), carbonyl cyanide p-(trifluoromethoxy) phenylhydrazone (FCCP; 0.5 μmol/L), and rotenone/antimycin A (0.5 μmol/L) mixtures were injected sequentially to determine mitochondrial function. Readings were collected from each well every 8 minutes; each 8-minute cycle comprised mixing (3 minutes), waiting (2 minutes), and recording (3 minutes). After completion of OCR measurement, the amount of protein in each well was quantified to normalize the results.

Cellular glucose uptake and lactate concentration in the culture medium were measured using the Glucose Uptake Colorimetric Assay Kit (BioVision) and the Lactate Colorimetric Assay Kit (BioVision), respectively, according to the manufacturer's instructions. Briefly, to measure cellular uptake of glucose, X01 or LN229 cells (1 × 10⁴) were plated in 96-well plates. Each well was washed three times with PBS and then starvation was induced by incubating the cells for 40 minutes with 100 μL of Krebs-Ringer-Phosphate-HEPES (KRPH) buffer. Cells were treated for 20 minutes with 1 mmol/L 2-deoxyglucose and lysed with 90 μL of Extraction Buffer. Cell lysates were neutralized by addition of 10 μL of Neutralization Buffer. Fluorescence (Ex/Em = 535/587 nm) was measured in a microplate reader (Synergy NEO, BioTek). To measure lactate production, culture medium was collected at the indicated times and mixed with Lactate Assay Buffer (50 μL/well) in a 96-well plate. Then, 50 μL of Reaction Buffer was added to each well for 30 minutes at room temperature. Absorbance was measured at 570 nm in a microplate reader (Synergy NEO, BioTek).

To measure the NAD⁺/NADH ratio in mitochondria, X01 or 528NS cells (3 × 10⁶) were seeded in 75T flasks and cultured in 5 mmol/L or 25 mmol/L glucose medium for 24 hours. Mitochondria were isolated using a Mitochondria Isolation Kit (Thermo Fisher Scientific), and the NAD⁺/NADH ratio was measured using the NAD⁺/NADH Quantification Colorimetric kit (BioVision), according to the manufacturer's instructions. Briefly, isolated mitochondria were lysed by two freeze/thaw cycles in NAD/NADH Extraction Buffer and vortexed for 10 seconds. After centrifugation, half of the supernatant was used for measurement of NADt (total amount of NAD⁺ and NADH), and the other half was used for measurement of NADH (after decomposition of NAD⁺ at 60°C for 30 minutes). Samples were incubated for 5 minutes with NAD Cycling Mix, followed by NADH Developer solution for 2 hours. Absorbance at 450 nm was then measured. The NAD⁺/NADH ratio was calculated as follows: (NADt−NADH)/NADH.

siRNAs specific for TRAP1 and SIRT3 were synthesized by Genolution using the following sequences: TRAP1-#1 5′-AAACATGAGTTCCAGGCCGAG-3′ and TRAP1-#2 5′-CCCGGTCCCTGTACTCAGAAA-3′; SIRT3-#1 5′-GTGGGTGCTTCAAGTGTTGTT-3′ and SIRT3-#2 5′-GCCCAACGTCACTCACTAC-3′; control 5′-ACUCUAUCUGCACGCUGAC-3′. Cells grown to 50%–75% confluence on 6-well plates were transfected with 50–100 nmol/L siRNA for 48 hours using G-Fectin (Genolution) and then analyzed by Western blotting.

To examine mitochondrial superoxide, cells were incubated for 20 minutes with 200 nmol/L Mito-SOX red (Invitrogen), washed with PBS, and then analyzed immediately using a FACSCalibur System (BD Biosciences). Data were processed using FlowJo software (TreeStar). To measure induction of apoptosis, the DNA content (assessed using propidium iodide) and phosphatidylserine (PS) externalization were analyzed using the Annexin V/Dead Cell Apoptosis Kit (Invitrogen). Labeled cells were quantified using the FACSCalibur system (BD Biosciences) and data were analyzed using FlowJo software (TreeStar).

Total RNA was prepared from cultured cells using RNeasy Mini Kits (QIAGEN) according to the manufacturer's instructions, and cDNA was synthesized using the ProtoScript First Strand cDNA Synthesis Kit (New England Biolabs) with an oligo(dT) primer. PCR reactions were performed in a Mastercycler PCR machine (Eppendorf) with the following sets of forward and reverse oligonucleotide primers: GAPDH, 5′-CGGGAAGCTTGTCATCAATGG-3′ and 5′-GGCAGTGATGGCATGGACTG-3′; TRAP1, 5′- AGCGCACTCATCAGGAAACT-3′ and 5′-TCAAACTCACGAAGGTGCAG-3′; Nestin, 5′- AACAGCGACGGAGGTCTCTA-3′ and 5′-CCTTTCCCAGGTTCTCTTCC-3′; Sox2, 5′-AACCAAGACGCTCATGAAGAAG-3′ and 5′-GCGAGTAGGACATGCTGTAGGT-3′; GFAP, 5′-GCTTCCTGGAACAGCAAAAC-3′ and 5′-CCTCCAGCGACTCAATCTTC-3′; SIRT3, 5′-CATGAGCTGCAGTGACTGGT-3′ and 5′-GAGCTTGCCGTTCAACTAGG-3′; NDUFA9, 5′-CAATTGCTCAACTGTCCAAG-3′ and 5′-AAGACTCTTGCTACCCATCG-3′; SDHB, 5′- TAAATGTGGCCCCATGGTAT-3′ and 5′-GATACTGCTGCTTGCCTTCC-3′; UQCRC2, 5′-CAAAGTTGCCCCCAAAGTTA-3′ and 5′-CATGAGTCTGCGGATTCTGA-3′; COX IV, 5′-CGAGCAATTTCCACCTCTGT-3′ and 5′-GGGCCGTACACATAGTGCTT-3′; and ATPB, 5′-AGCCCATGGTGGTTACTCTG-3′ and 5′-GGCAGGGTCAGTCAAGTCAT-3′.

For immunoprecipitation, 100–200 μg of protein lysate derived from GSCs incubated in immunoprecipitation (IP) lysis buffer (135 mmol/L NaCl, 1% glycerol, 1% NP-40, 20 mmol/L Tris-HCl, pH 8.0, and protease/phosphatase inhibitors) was precleared using 1 μg of rabbit or mouse IgG and 20 μL of Protein A/G PLUS agarose (Santa Cruz Biotechnology) to remove nonspecific proteins. The precleared lysates were then incubated at 4°C for 1 hour with 1 μg of anti-TRAP1 antibody and then mixed with 20 μL of Protein A/G PLUS agarose at 4°C overnight. The immunocomplexes were then precipitated by centrifugation at 1,000 × g for 1 minute at 4°C and washed three times with IP lysis buffer. Bound proteins were analyzed by Western blotting. To perform the pull-down experiments, GST and GST-TRAP1 proteins were induced in E. coli BL21 by overnight incubation with 0.2 mmol/L IPTG at 18°C. Cells were harvested by centrifugation and lysed by sonication in lysis buffer (50 mmol/L Tris-HCl, pH 7.4, 150 mmol/L NaCl, 5 mmol/L MgCl₂, and 1 mmol/L dithiothreitol). After centrifugation at 15,000 × g for 30 minutes, the soluble fractions were mixed with glutathione-Sepharose beads (GE Healthcare). Protein-bound beads were washed three times with wash buffer (50 mmol/L Tris-HCl, pH 7.4, 500 mmol/L NaCl, 5 mmol/L MgCl₂, and 1 mmol/L dithiothreitol). Mitochondria isolated from X01 cells were lysed for 1 hour at 4°C in RIPA buffer containing 50 mmol/L Tris-HCl, pH 7.4, 150 mmol/L NaCl, 1% NP-40, 0.25% sodium deoxycholate, and proteinase inhibitor (Invitrogen) under constant agitation. After centrifugation at 13,000 × g for 5 minutes at 4°C, the supernatant was precleared for 1 hour at 4°C with glutathione beads (GE Healthcare). Then, 100–200 μg of precleared protein extract was incubated overnight at 4°C with GST and GST-TRAP1 beads. Bead-bound proteins were collected by centrifugation at 1,000 × g for 1 minute at 4°C, washed three times with lysis buffer, and analyzed by Western blotting.

For the in vitro limiting dilution assays, cells were plated in decreasing numbers (200, 100, 50, 25, 12.5, and 6.25) in 96-well plates containing DMEM/F-12 supplemented with B27, 10 ng/mL EGF, and 5 ng/mL bFGF. Extreme limiting dilution analysis was performed using software available at http://bioinf.wehi.edu.au/software/elda/. Images of spheres formed by each group were taken on day 5 after plating.

The enzyme activity of SIRT3 in vitro or in vivo was determined using the SIRT3 Activity Assay Kit, according to the manufacturer's instructions (Abcam). SOD activity was determined in vitro using the SOD Activity Assay Kit, according to the manufacturer's instructions (BioVision). The activity of Complexes I (ab109721, Abcam), II (K660-100, BioVision), and V (ab109907, Abcam) was measured using respective activity assay kits according to the manufacturer's instructions. Mitochondria were lysed in the provided buffers to obtain a protein solution (5 μg/μL) and loaded into the provided plates prior to immunocapture. After washing away the unbound proteins, the provided substrate and dye were added to each well and absorbance was measured as described in the instructions.

All animal experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee at the National Cancer Center, Republic of Korea (NCC-15-268). Orthotopic GBM xenografts were produced as described previously (36). Briefly, cells were transplanted following resuspension in DMEM/F-12 medium supplemented with B27, 10 ng/mL EGF, and 5 ng/mL bFGF. Cells were injected stereotactically into the left striatum of 5-week-old female BALB/c nude mice. The injection coordinates were 2.2 mm to the left of the midline and 0.2 mm posterior to bregma, at a depth of 3.5 mm. The brain of each mouse was harvested and fixed in 4% paraformaldehyde. For the subcutaneous mouse model, cells were injected into the hip area on both sides. Tumor growth was measured twice per week using electronic caliper. Tumor volume was calculated using the following formula: V = 1/2 × (width)² × length. The mean tumor volume at the start of drug injection was approximately 50 mm³. Gamitrinib (10 mg/kg) or vehicle (DMSO) dissolved in 20% Cremophor EL (Sigma) in PBS was administered intraperitoneally every day. At the end of the experiment, animals were euthanized and tumors were collected for histology and Western blotting.

To examine histologic characteristics, brains were removed, fixed for 24 hours at 4°C with 4% paraformaldehyde, and stained with hematoxylin (DAKO) and 0.25% eosin (Merck). For IHC staining of CSC markers and TRAP1-associated genes [Sox2 (R&D Systems), Nestin (Abcam), SIRT3 (Cell Signaling Technology), Ki-67 (Abcam), and cleaved caspase-3 (Cell Signaling Technology)], sections were subjected to antigen retrieval in citrate buffer (pH 6.0) and endogenous peroxidase was blocked with 3% hydrogen peroxide. Tissue sections were then incubated overnight at 4°C in a humidified chamber with appropriate primary antibodies in antibody diluent buffer (IHC World). Staining results were visualized by sequential incubation of TMAs with components of the Polink-2 Plus HRP detection system (GBI); DAB (Vector Laboratories) was used as the chromogen. The number of cells positive for Ki-67 and cleaved caspase-3 per hematoxylin counterstained area was counted (200 high-power fields were scanned).

All statistical analyses were performed using GraphPad Prism 7. All experiments were repeated at least twice, and data are presented as the mean ± SEM. Statistical significance of data derived from the following experiments was determined using Student t test: OCR, glucose uptake, lactate production, ATP production, SOD activity, SIRT3 activity, mitochondrial respiratory complex activity, tumor weight measurement, and quantification of IHC staining. A value of P < 0.05 was considered significant. Pearson correlation coefficient (r) was used to determine the correlation between TRAP1 and Sox2 expression in GBM cells and GSCs (using Prism 7). Survival was assessed using the Kaplan–Meier method and the log-rank test (using GraphPad Prism 7).

To understand the relationship between TRAP1 and brain tumor malignancy, we examined expression of TRAP1 in brain tumor tissues derived from human patients. TRAP1 expression correlated significantly with brain tumor malignancy; in particular, TRAP1 expression in grade IV tumors (i.e., GBM) was higher than that in lower grade tumors (Fig. 1A and B; Supplementary Table S1). Similarly, expression of Sox2, a marker of stemness (37), was elevated in higher grade tumors (Fig. 1C; Supplementary Table S2). Interestingly, we found that expression of both TRAP1 and Sox2 at the single-cell level was elevated in samples from patients with glioma (Fig. 1D and E). Consistent with this, flow cytometry analysis of single cells from a GBM cell line (LN229) revealed a significant positive correlation between expression of TRAP1 and Sox2 (Pearson correlation coefficient r = 0.6964; Supplementary Fig. S1A). The correlation was stronger in CSC-enriched LN229 cells (r = 0.8568) and the GSC line X01 (r = 0.9284; Supplementary Fig. S1A and S1B). Therefore, to better understand the clinical implications of TRAP1 and stemness marker protein levels in glioma patients, we constructed Kaplan–Meier curves using clinical information collected at Seoul National University Hospital (SNUH, Seoul, Republic of Korea). Analysis of these data indicated that poor survival of patients with glioma correlated significantly with increased expression of TRAP1 and Sox2 (Fig. 1F and G; Supplementary Fig. S1C and S1D). Survival of patients with high expression of both TRAP1 and Sox2 was much worse than that of patients with low expression of both proteins (Fig. 1H; Supplementary Fig. S1E). The positive correlation between TRAP1 and Sox2 expression was significant for datasets derived from all patients with glioma and GBM (Supplementary Fig. S1F and S1G).

To understand the function of TRAP1 in GSCs, we examined its expression in GSC lines X01 and 528NS. We found that TRAP1 expression at both the protein and mRNA levels was higher in GSCs than in differentiated counterparts (Fig. 2A and B). Furthermore, all ten GSC lines derived from patients with glioma showed consistently higher expression of TRAP1 than normal astrocytes and GBM cell lines (Fig. 2C). Because TRAP1 regulates both mitochondrial respiration and ROS production by several cancers (14, 23, 38), we examined whether TRAP1 regulates mitochondrial metabolism in GSCs. To examine TRAP1 function in mitochondria, we used the mitochondria-accumulating drug gamitrinib, a conjugate of triphenylphosphonium (for mitochondrial delivery) and geldanamycin (for TRAP1 inhibition; refs. 14, 39). Inactivation of TRAP1 by gamitrinib or siRNA (14, 19) led to a marked reduction in mitochondrial respiration in X01 cells, but had no significant effect on the differentiated counterpart (Fig. 2D–G); the basal OCR of X01 cells fell by 33%–58%, whereas spare respiratory capacity (SRC) fell by 73%–92% (Fig. 2E and G). Consistent with this, inhibiting TRAP1 in X01 cells reduced the mitochondrial membrane potential (Supplementary Fig. S2A) and increased lactate production (Supplementary Fig. S2B). In addition to mitochondrial respiration, the mitochondrial ROS concentration in GSCs increased after inactivating TRAP1 with either gamitrinib or siRNA (Fig. 2H and I). Similarly, when CSCs were enriched from GBM cell lines T98G and LN229, the enriched stem cells expressed higher levels of TRAP1 than the original cells (Supplementary Fig. S2C). Consequently, inhibiting TRAP1 expression by enriched stem cells reduced respiration, increased lactate production, and increased ROS production (Supplementary Fig. S2D–S2F). Collectively, these data suggest that TRAP1 plays a critical role in maintaining mitochondrial respiration and reducing ROS concentrations in GSCs.

The pro-respiratory function of TRAP1 appears to be the antithesis of the Warburg effect, an aerobic glycolytic metabolic characteristic of cancer cells (5). However, growing evidence suggests that tumor metabolism is heterogeneous, and that certain tumors (or subpopulations of cells within tumors, including CSCs within GBM) depend primarily on mitochondrial respiration rather than on glycolysis (9, 40). Consistent with this, we found that X01 consumed less glucose and produced less lactate, and displayed slightly higher cellular ATP concentrations, than differentiated X01 cells (Supplementary Fig. S3A–S3D). Furthermore, X01 cells had higher basal OCR and SRC than their differentiated counterpart (Supplementary Fig. S3E). Collectively, these data indicate that GSCs have a preference for using mitochondrial respiration for efficient ATP production, whereas the differentiated counterparts derive energy primarily from glycolysis. However, although respiration in GSCs was elevated, mitochondrial ROS production was significantly lower than that in the differentiated counterparts (Supplementary Fig. S3F). To examine the functional implications of mitochondrial respiration and ROS production with respect to maintaining GSC stemness, we exposed GSCs to the mitochondrial respiratory chain inhibitors antimycin A (Complex III) and oligomycin (Complex V), and to an exogenous ROS (hydrogen peroxide). The respiratory chain inhibitors and hydrogen peroxide not only reduced expression of stemness markers (nestin and Sox2; Supplementary Fig. S3G), but also inhibited sphere-forming activity by X01 cells (Supplementary Fig. S3H–S3J). Collectively, these data suggest that mitochondrial respiration and antioxidant systems within the mitochondria are active in GSCs, and that they are necessary to maintain stemness.

To investigate whether the metabolic regulatory mechanisms associated with TRAP1 affect the stem cell properties of GSCs, we examined expression of stem cell markers after TRAP1 inactivation. Treatment of X01 and 528NS cells with gamitrinib or TRAP1 siRNA reduced expression of Sox2 and nestin (Fig. 3A and B). Consistent with this, inactivation of TRAP1 by gamitrinib or TRAP1 knockdown reduced the sphere- and colony-forming activity of X01 cells (Fig. 3C and D; Supplementary Fig. S4A–S4C). In addition to maintaining stemness, gamitrinib reduced the viability of GSCs (X01 and 528NS) by inducing apoptotic cell death (Fig. 3E and F), but did not affect the viability of normal astrocytes and neural stem cells (Fig. 3E).

Similar to the functions of TRAP1 shown in this study, the major mitochondrial deacetylase SIRT3 modulates mitochondrial energy and ROS metabolism (25–27); therefore, we examined the mechanisms between these two proteins. To identify the relationship between TRAP1 and SIRT3, we first examined their expression in human patients with glioma. We found a positive correlation between expression of TRAP1 and SIRT3 proteins in individual cancer cells (r = 0.7392, P < 0.0001; Fig. 4A and B). A pull-down experiment revealed that TRAP1 interacts directly with SIRT3 (Fig. 4C). Consistent with this, endogenous TRAP1 coimmunoprecipitated with SIRT3 (Fig. 4D). Differentiation of GSCs reduced expression of both SIRT3 and TRAP1 at the protein level, but did not alter expression of SIRT3 mRNA (Fig. 4E), suggesting that SIRT3 protein is destabilized during differentiation of GSCs. Likewise, loss of TRAP1 function after siRNA or gamitrinib treatment led to reduced expression of SIRT3 (Fig. 4F), most likely due to the reduced half-life of the SIRT3 protein (Supplementary Fig. S5A); this suggests that TRAP1 acts as a chaperone that stabilizes SIRT3 protein. This is supported by the finding that SIRT3 enzyme activity in the mitochondrial extract fell markedly after TRAP1 inactivation (Fig. 4G). In addition to stabilizing SIRT3 in vivo, recombinant TRAP1 increased the deacetylase activity of purified SIRT3 in vitro (by more than 2-fold), a phenomenon reversed by exposure to the TRAP1 inhibitor gamitrinib (Fig. 4H). After normalizing enzyme activity to the SIRT3 protein level, we found that gamitrinib still reduced SIRT3 activity (Supplementary Fig. S5B). In addition, short-term (0.5 hour) exposure of X01 cells to gamitrinib (which led to no detectable degradation of SIRT3), reduced the enzymatic activity of SIRT3 (Supplementary Fig. S5C). Collectively, these data suggest that TRAP1 augments the deacetylase activity of SIRT3 by inhibiting degradation of the protein and by increasing its enzyme activity in GSCs.

Knocking down SIRT3 increased mitochondrial ROS production (Supplementary Fig. S5D), supporting previous studies showing that SIRT3 inhibits ROS (41) in GSCs. Here, we found that overexpression of SIRT3 fully inhibited the gamitrinib-triggered increase in cellular ROS production (Fig. 4I), suggesting that TRAP1 acts as an upstream regulator of SIRT3 during ROS metabolism. Likewise, overexpression of SIRT3 led to significant reversal of the cytotoxic activity of gamitrinib (Supplementary Fig. S5E). To better understand the effectors downstream of SIRT3, we focused on superoxide dismutase 2 (SOD2), which is a SIRT3 substrate in mitochondria that regulates ROS metabolism (27, 28). Inactivation of TRAP1 by gamitrinib increased the amount of acetylated mitochondrial proteins in X01 cells (Supplementary Fig. S5F). As a downstream effector of SIRT3, SOD2 was highly acetylated after inhibition of TRAP1 (Fig. 4J; Supplementary Fig. S5G). Treatment with gamitrinib led to a significant reduction in SOD enzyme activity in X01 mitochondria (Fig. 4K) without altering expression of the protein (Supplementary Fig. S5H). Collectively, the data indicate that TRAP1 increases SIRT3 enzyme activity, resulting in reduced ROS production via deacetylation and subsequent activation of SOD2.

Proteomics analyses have identified a number of acetylation sites within TRAP1 (PhosphoSitePlus database, http://www.phosphosite.org). Therefore, we assumed that these sites could be affected by the major mitochondrial deacetylase SIRT3 (25). As expected, nicotinamide, a SIRT inhibitor, increased acetylation of TRAP1 (Fig. 5A). Consistent with this, overexpression of SIRT3 reduced acetylation of TRAP1 (Fig. 5B), whereas knocking down SIRT3 increased acetylation of TRAP1 (Fig. 5C). Taken together, these data suggest that SIRT3 regulates acetylation of TRAP1.

Because TRAP1 (20–23) and SIRT3 (25, 26) interact with and regulate electron transport chain complexes (ETC), we next asked whether interplay between TRAP1 and SIRT3 regulates mitochondrial respiration. Pull-down (Fig. 5D) and coimmunoprecipitation experiments (Fig. 5E) showed that TRAP1 and SIRT3 were associated with Complexes I, II, IV, and V. Pharmacologic and genetic inhibition of TRAP1 led to a marked reduction in expression of NDUFA9 (a Complex I subunit) and SDHB (a Complex II subunit) proteins without altering expression of the respective mRNAs (Fig. 5F–H), indicating increased degradation of ETC subunits upon TRAP1 inhibition. SIRT3 knockdown also reduced expression of NDUFA9 and SDHB, but did not reduce expression of the SIRT3 substrate proteins SOD2 and glutamate dehydrogenase (GDH; Fig. 5I; ref. 42), suggesting that SIRT3-mediated increases in TRAP1 chaperone activity stabilize Complexes I and II. Gamitrinib increased acetylation of the SIRT3 substrate NDUFA9 (Fig. 5J); overexpression of SIRT3 increased basal OCR and SRC (Supplementary Fig. S6A) as previously reported (25, 26). Overexpression of SIRT3, however, did not reverse gamitrinib-induced dysfunction of ETCs (Supplementary Fig. S6A and S6B) due to disruption of Complex I and II after TRAP1 inactivation (Supplementary Fig. S6C). Consistent with this, biochemical analyses showed that gamitrinib reduced the enzymatic activity of Complexes I and II in the mitochondrial fraction of X01 cells without affecting activity of Complex V (Fig. 5K–M). Consequently, SIRT3 knockdown reduced expression of nestin and Sox2 (Fig. 5I) and reduced the sphere- and colony-forming activity of X01 (Fig. 5N; Supplementary Fig. S7A and S7B). Taken together, the data suggest that chaperone activity is required for formation of functional ETC complexes, and that SIRT3 activates ETC functions via deacetylation. Therefore, the positive interplay between TRAP1 and SIRT3 efficiently increases mitochondrial respiration by both quantitatively and qualitatively regulating ETCs in GSCs.

To examine the role of SIRT3 during metabolic adaptation of GSCs, we incubated X01 and 528NS cells in low (5 mmol/L) and high (25 mmol/L) glucose DMEM medium. Low glucose conditions led to increased expression of SIRT3 and Complex I/II subunit proteins, resulting in increased expression of stemness markers (Fig. 6A and B). Accordingly, basal OCR and SRC in low glucose medium increased by 1.7-fold and 2.3-fold, respectively, compared with that in high glucose medium (Fig. 6C). The increase of both ETC expression and OCR was suppressed by treatment with SIRT3 siRNA (Fig. 6D and E), suggesting TRAP1/SIRT3-dependent stabilization of respiratory complexes. Consistent with this, inhibition of TRAP1 by gamitrinib or siRNA reduced expression of Complex I/II subunits and stemness markers (Fig. 6F and G), and fully suppressed the basal OCR and SRC (Fig 6H). Increased interplay between TRAP1 and SIRT3 could be triggered by increased levels of mitochondrial NAD⁺ upon exposure to low glucose medium (Fig. 6I and J). Consequently, GSCs were much more sensitive to TRAP1 inhibition under low glucose conditions than under high glucose conditions (Fig. 6K). These data suggest that interplay between TRAP1 and SIRT3 has a crucial role in metabolic adaptation under reduced nutrient conditions.

The interaction between TRAP1 and SIRT3 means that inactivating either of them effectively impairs crucial metabolic characteristics of GSCs coordinated by these proteins. To investigate the antitumor efficacy of targeting the interdependency between the proteins in vivo, X01 cells infected with a shTRAP1 lentivirus were transplanted orthotopically into the brains of nude mice. Lentiviral-mediated inhibition of TRAP1 led to a significant reduction in tumor growth (Supplementary Fig. S8A) and subsequently prolonged the life span of the mice (Fig. 7A). Consistent with this, pharmacologic inhibition of TRAP1 by gamitrinib reduced the growth of subcutaneously transplanted 528NS and X01 cells (Fig. 7B and C; Supplementary Fig. S8B). Immunohistochemical and Western blot analyses revealed that inhibition of TRAP1 reduced expression of Sox2, nestin, and SIRT3 in vivo (Fig. 7D; Supplementary Fig. S8C and S8D). Furthermore, gamitrinib reduced the number of Ki67-positive cells, but increased the number of cells positive for cleaved caspase-3 (Fig. 7E and F), suggesting reduced cell proliferation and increased cell death in vivo.

Here, we show that SIRT3 and TRAP1 are overexpressed in glioma stem cells (GSC), and that interplay between these two molecules increases activity of ETCs without increasing the ROS concentration in GSC mitochondria. Nutrient conditions regulated interplay between these proteins and subsequently triggered reprograming of cellular metabolism. Inhibiting either TRAP1 or SIRT3 in GSCs compromised this interplay and disrupted the function of both proteins, leading to metabolic dysregulation, overproduction of ROS, loss of stemness properties, and cell death both in vitro and in vivo. Thus, the results clearly indicate that TRAP1 and SIRT3 play an important role by maintaining stem cell populations during development of malignant glioma.

A large proportion of cellular ROS is produced within mitochondrial respiratory complexes (43). TRAP1 counters excess ROS production in many types of cancer cell (38). The molecular mechanisms underlying redox regulation have been elusive; however, we show here that activation of SIRT3 by TRAP1 leads to activation of SOD2. ROS are not only toxic to cells, but also act as a signaling molecule that triggers differentiation of both normal and cancer stem cells (44, 45). Similarly, we found that excess ROS in GSCs led to loss of stemness, indicating the importance of ROS scavenging systems for maintaining GSCs. Furthermore, because active scavenging of ROS is associated with GSC resistance to radiation and chemotherapy (45, 46), the interplay between TRAP1 and SIRT3 may contribute to the therapeutic resistance of CSCs in vivo.

Fast and continuously growing tumor cells suffer local nutritional deprivation and need to generate adaptive responses to thrive under such stressful conditions (5, 47). To meet nutrient demand under conditions of metabolic stresses, some cancer cells such as GSCs secrete angiogenic factors to induce angiogenesis in the tumor microenvironment, thereby allowing greater uptake of nutrients and oxygen (48, 49). Furthermore, GSCs develop an internal strategy to adapt and survive under nutrient-poor conditions; they switch their metabolism to favor energy-efficient mitochondrial respiration (9, 11). Here, we showed that, under limiting glucose conditions, increased mitochondrial NAD⁺ concentrations in GSCs activate SIRT3. This subsequently activates TRAP1 via a positive feedback loop between TRAP1 and SIRT3, which consequently switches cellular metabolism toward energy-efficient respiration to allow adaptation to the harsh microenvironment. In addition, the reprogrammed mitochondrial metabolism increases the SRC, which avoids an ATP crisis under increased workload and stress conditions (50, 51). Thus, this metabolic characteristic may contribute not only to cellular adaptation of GSCs to stress conditions, but also to development of therapeutic resistance. Because many drug-resistant CSCs depend primarily on mitochondrial respiration (40, 52, 53), other cancer cells could also exploit the TRAP1/SIRT3 interaction to improve metabolic adaptation and flexibility.

Although there is a consensus regarding the metabolic rewiring function of TRAP1, the underlying molecular mechanism remains controversial (23). TRAP1 in some cancer cells (such as human colon carcinoma, osteosarcoma, and cervical carcinoma) suppresses mitochondrial respiration by inhibiting Complex II or IV (20, 21), whereas TRAP1 in human prostate cancer, breast cancer, and glioblastoma increases mitochondrial respiration by increasing Complex II activity (22, 54, 55). Considering that the TRAP1 chaperone functions differently in disparate cancer cells, the cellular context seems to define the metabolic functions of TRAP1 (23). This suggests that unique compositions of TRAP1 clients and regulators (such as the SIRT3–TRAP1 interaction examined in this study) may be the factors that determine TRAP1-driven metabolic regulation. With respect to the role of unique cellular composition, we suggest that TRAP1 interactors such as SIRT3 and the mitochondrial c-Src, which increase respiratory activity of ETC (21, 26), may play a role. Interestingly, TRAP1 regulates activity of these two enzymes in an opposing manner: it supports the enzyme activity of SIRT3 but inhibits that of c-Src. Therefore, we can speculate that TRAP1 inhibits or activates mitochondrial respiration in different cancer cell types depending on its preferential interaction with c-Src or SIRT3. TRAP1-dependent metabolic regulation may be better explained after comprehensive analyses of the TRAP1 networks in individual cancer cells.

Collectively, the results presented herein reveal an important metabolic characteristic of GSCs: high mitochondrial respiration without overproduction of ROS. This is closely associated with positive feedback interplay between TRAP1 and SIRT3. Because GSCs are responsible for failure of anticancer therapy and high recurrence rates, inhibiting TRAP1 and/or SIRT3 may be the Achilles heel of GSCs, leading to development of effective and novel strategies for treating patients with GBM.

No potential conflicts of interest were disclosed.

Conception and design: H.-K. Park, J.-H. Hong, J.H. Kim, J.B. Park, B.H. Kang

Development of methodology: H.-K. Park, Y.T. Oh, S.-H. Park, B.H. Kang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H.-K. Park, Y.T. Oh, A.-J. Lee, S.-H. Park, C.-K. Park, J.B. Park, B.H. Kang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H.-K. Park, J.-H. Hong, Y.T. Oh, S.S. Kim, J. Yin, J.H. Kim, B.H. Kang

Writing, review, and/or revision of the manuscript: H.-K. Park, J.-H. Hong, Y.C. Chae, J.B. Park, B.H. Kang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.T. Oh, S.S. Kim, J. Yin, J.H. Kim, S.-H. Park

Study supervision: M.-J. Park, J.B. Park, B.H. Kang

This work was supported by the National Cancer Center (NCC-1810121-1, Republic of Korea), by the UNIST research fund (1.180018.01, Republic of Korea), by National Research Foundation of Korea (NRF) grants funded by MSIT (NRF-2016R1A2B2012624, NRF-2016R1A6A3A11934753, NRF-2017R1A2B4011741, NRF-2017R1D1A1B03033303, NRF-22A20130012280, NRF-2018R1A5A1024340, NRF-2018R1A4A1025860), and by the Korea Drug Development Fund (KDDF), funded by MSIT, MOTIE, and MOHW (KDDF-201512-02, Republic of Korea).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.