Nitrogen Trapping as a Therapeutic Strategy in Tumors with Mitochondrial Dysfunction

Cellular metabolic reprogramming is an essential step toward tumorigenesis. Cancer metabolism not only has to support the cell's high anabolic needs but also has to respond to various challenges such as low oxygen and nutrient availability pertaining to the tumor environment. Several canonical oncogenes have been shown to regulate cancer cell metabolism (1). The discovery of cancer-associated mutations in the tricarboxylic acid (TCA) cycle enzymes isocitrate dehydrogenase 1 and 2 (IDH1/2), succinate dehydrogenase (SDH), and fumarate hydratase (FH) indicates that significant alterations in metabolic pathways can also drive tumorigenesis (2). Rewiring of metabolism may render cancer cells more dependent than normal cells on specific cellular processes that could be targeted for therapeutic benefit (3, 4).

It is important to note that although all cancer cells utilize glucose and secrete lactate in conditions with ample oxygen, a phenomenon termed aerobic glycolysis or the Warburg effect, many cancers also maintain mitochondrial metabolism and require respiratory competency (5, 6). However, we now know that OXPHOS defects play a crucial role in a subset of cancers. For example, FH- and SDH-mutant cancers manifest pronounced mitochondrial respiration deficiencies (7–9). Furthermore, pathogenic mitochondrial DNA (mtDNA) mutations occur frequently in a broad range of cancer types (10, 11). In addition, cancer cells that have limited access to oxygen in vivo may exhibit OXPHOS defects (12).

Interestingly, a series of reports have recently demonstrated that cancer cells under hypoxic conditions and cancer cells with TCA cycle or electron transport chain (ETC) mutations display very similar metabolic reprogramming phenotypes. To survive the severe truncation of the OXPHOS pathway, these cells undergo multiple metabolic rearrangements, such as increased glycolysis and utilization of glutamine via reductive instead of oxidative carboxylation to replenish TCA cycle metabolites (13, 14). In culture, respiration-incompetent cells are auxotrophic for pyruvate because of its role in maintaining redox balance to support aspartate biosynthesis. Concordantly, aspartate is a common limiting factor for their proliferation (12, 15–18). Moreover, it has been found that cytosolic aspartate synthesis via the glutamic-oxaloacetic transaminase 1 (GOT1) becomes essential when the ETC is inhibited (15).

Clinically, OXPHOS-defective cancers are often difficult to treat. For example, hypoxia enhances cancer virulence and significantly reduces the efficacy of radiotherapy, chemotherapy, and targeted therapy (19). Loss-of-function mutations in FH can cause an aggressive form of kidney cancer called hereditary leiomyomatosis and renal cell carcinoma (HLRCC). HLRCC-associated kidney cancer occurs early in life and can metastasize even when tumors are small (<1 cm; refs. 20, 21). In addition, using mitochondrial transfer and cybrid cells, studies have shown that some mtDNA mutations can enhance tumor progression (22–24). The close resemblance in the metabolic phenotypes despite the varying causes of OXPHOS deficiency suggests that it may be possible to develop a unifying therapeutic approach for such cancers. In the current work, we demonstrate that cell-permeable forms of the TCA cycle metabolite α-ketoglutarate (aKG) lead to potent cytotoxicity specifically in OXPHOS-incompetent cancer cells by targeting their dependence on the aspartate biosynthesis pathway.

Compounds dmaKG (dimethyl aKG, 349631), deaKG (diethyl aKG, CDS008208), etaKG (2-oxo-pentanedioic acid 5-ethyl ester 1-(3-trifluoromethyl-benzyl) ester, SML1743), antimycin A (A8674), rotenone (R8875), aspartate (L-aspartic acid potassium salt, A9381), 2-DG (2-deoxy-D-glucose, 25972), adenine (A2786), ATP (adenosine 5′-triphosphate, A7699), ADP (adenosine 5′-diphosphate, A5285), AMP-PCP (β,γ-methyleneadenosine 5′-triphosphate, M7510), AOA (aminooxyacetic acid hemihydrochloride, C13408), and 3-bromopyruvate (16490) were purchased from Sigma Aldrich. Oligomycin (11341), necrostatin (11658), atpenin (11898), and metformin (13118) were purchased from Cayman Chemical. 3-NPA (3-nitropropionic acid, 4849) was purchased from Tocris. Cyclosporine A (BMLA1950100) was purchased from Enzo Life Sciences. Z-VAD-FMK (A19025) was purchased from Apexbio Technology. MG-132 (S2619) was purchased from Selleck Chemicals. Phenformin (P29690) was purchased from Toronto Research Chemicals.

All of the cell lines were cultured in the standard medium consisting of DMEM (Invitrogen, #11965) supplemented with 10% FBS, 1 mmol/L of sodium pyruvate, and 1% penicillin–streptomycin at 37°C with 10% CO₂ unless specified otherwise. UOK262 and UOK268 cells were gifts from Drs. Y. Yang and W.M. Linehan at the NIH (Bethesda, MD; refs. 25, 26). HLRCC isogenic cell pairs were generated by infection with viral particles containing pBabe-puro (UOK262-ev and UOK268-ev) or pBabe-puro-FH (UOK262-FH and UOK268-FH) and were maintained in 2 μg/mL of puromycin. UOK262-SLC1A3 and 143B-wt cells expressing SLC1A3 were generated by infection with viruses containing pMXs-SLC1A3 and were maintained in 5 and 4 μg/mL of blasticidin, respectively. 143B cybrid pairs (143B-wt and 143B-CytB) were gifts from Dr. M.G. Vander Heiden (Massachusetts Institute of Technology, Cambridge, MA) with the permission of Dr. N.S. Chandel (Northwestern University, Evanston, IL), and they were cultured in the standard medium supplemented with 0.1 mg/mL uridine (17). pMXs-GOT1 and pMXs-IRES-blasticidin empty vector were introduced into UOK262, 143B-wt, and 786-O by viral transduction to generate stable lines. pMXs-SLC1A3 (Addgene, #72873) and pMXs-GOT1 (Addgene, #72872) were gifts from David Sabatini (Massachusetts Institute of Technology, Cambridge, MA; ref. 15). UMRC2-ev and UMRC2-VHL are human clear cell renal cell carcinoma cell lines carrying pBabe-puro and pBabe-puro-VHL, respectively. These cells were provided by O. Iliopoulos (Massachusetts General Hospital (MGH), Charlestown, MA) and were maintained in the presence of 2 μg/mL of puromycin. 786-O and UMRC-3 were provided by O. Iliopoulos. NCI-H522, DU-145, Cal-12T, and HCT116 were gifts from Dr. Cyril Benes (Center for Molecular Therapeutics, MGH). These cells have not been authenticated during this study. Cells were maintained and propagated for two months after thawing. Mycoplasma test was performed routinely using the MycoAlert PLUS Mycoplasma Detection Kit (Lonza, #LT07-701). The FH mutations in HLRCC isogenic cell pairs have been sequence confirmed. Multiple batches of HLRCC isogenic cell pairs were generated and the results were reproducible from different batches.

All of the cytotoxicity results shown as the percentage of dead cells were determined using the trypan blue dye exclusion assay. For this assay, 20,000 cells per well were seeded in 24-well plates, and treatments started the next day. Twenty-four hours after treatment, both floating and adherent cells were harvested to determine the percentage of cell death using trypan blue at 0.2 w/v% final concentration.

Apoptosis was measured using FITC Annexin V Apoptosis Detection Kit with PI (BioLegend, 640906). Cells were treated with dmaKG for indicated duration and were harvested by trypsinization, washed with ice-cold PBS, and finally resuspended in binding buffer. The staining was performed as recommended by the manufacturer. After 20 minutes of incubation in the dark, samples were analyzed on a Fortessa X-20 flow cytometer (BD Biosciences). The apoptotic cell fraction was measured based on the fraction of cells that are FITC⁺ and PI⁻ (early apoptotic) or that are FITC⁺ and PI⁺ (late apoptotic) in a dot plot. At least 10,000 events were assessed per measurement.

Cells/well (10,000) were seeded in 12-well plates, and dmaKG was added the following day. Cells were grown for 4 days under the indicated conditions. Cells were then washed twice with PBS and incubated with 0.2% filtered crystal violet solution for 15 minutes. Cells were washed with water approximately three times until all extracellular crystal violet was removed. Intracellular crystal violet was then dissolved in methanol. Absorbance of the solution was measured at 570 nm using the FLUOstar Omega Microplate Reader (BMG Labtech).

Cells/well (2 × 10⁵) were plated in 12-well plate and were treated with dmaKG (6.6 mmol/L) for 8 hours. Twenty minutes before termination, tetramethylrhodamine, ethyl ester (TMRE) was added to cells at a final concentration of 0.2 μmol/L and incubated at 37°C. Cells were trypsinized, washed, and suspended in 0.5 mL PBS. Trifluoromethoxy carbonylcyanide phenylhydrazone (FCCP; 20 μmol/L) was used as a negative control. The cell fluorescence was measured by flow cytometry with appropriate channels.

The cells were plated in a XF96 cell culture microplate at 25,000 cells/well and maintained in regular cell growth conditions for at least 12 hours prior to the experiments. On the day of the assay, cells were treated with indicated compounds for indicated times. Oxygen consumption rate (OCR), extracellular acidification rate (ECAR), and proton efflux rate (PER) were measured according to the manufacturer's instructions for the Seahorse XF Glycolysis Stress Test, the XF Cell Mito Stress Test, and the XF Real-Time ATP Rate Assay. Data analysis was performed using the Seahorse Wave software.

The ATP content was measured using the CellTiter-Glo Luminescent Cell Viability Assay (Promega, G7570). Briefly, 5,000 cells/well were plated in a 96-well plate. On the next day, cells were treated with compounds as indicated in the figure legends and then lysed using the assay reagent followed by the luminescence measurement using the FLUOstar Omega Microplate Reader (BMG Labtech). Relative ATP levels was calculated as compared with the control of the same cell types.

Cells/well (2 × 10⁵) were plated in a 12-well plate. On the next day, cells were treated as indicated in the figure legends. At the end of the experiments, cells were washed with cold saline followed by ATP extraction with the boiling method and sample processing per the manufacturer's protocol. Cryopreserved mouse organs or xenograft tumors (20 mg) from in vivo efficacy studies were processed through trichloroacetic acid deproteination followed by extraction into 10 mmol/L Tris-EDTA (TE) buffer using ether that is presaturated in TE buffer. The levels of the total adenine nucleotide pool or the levels of individual adenine nucleotides (ATP, ADP, or AMP) were measured using the ATP/ADP/AMP Assay Kit (Biomedical Research Service, Department of Biochemistry, University at Buffalo, Buffalo, NY, A-125). The luminescence was measured using the FLUOstar Omega Microplate Reader (BMG Labtech).

Cell lysates were prepared in Laemmli sample buffer (Bio-Rad, #1610737) supplemented with Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific, #78440). The proteins were separated by 12% SDS-PAGE and transferred onto polyvinylidene fluoride membranes. After blocking with 5% nonfat milk in Tris-buffered saline and 0.1% Tween 20, membranes were incubated with primary antibodies at 4°C overnight, followed by 1:20,000 horseradish peroxidase–conjugated secondary antibody (Cell Signaling Technology) for 2 hours. Bands were visualized using Clarity ECL Western Blotting Substrate (Bio-Rad, #1705060) and ChemiDoc Touch Imaging System (Bio-Rad Laboratories). Antibodies were purchased from Cell Signaling Technology for prolylhydroxy-HIF1α (34345), phospho-p70S6K1 (9234S), p70S6K1 (2708S), phospho-S6 (5364S), S6 (2317S), phospho-4EBP1 (9451S), and 4EBP1 (9644S) and from BD Biosciences for HIF1α (610958).

All animal experiments were performed with approval of the Institutional Animal Care and Use Committee at the MGH.

For UOK262 xenograft tumor model, athymic nude mice were injected subcutaneously in the right flank with 7.5 × 10⁶ UOK262 cells. In approximately two months, palpable tumors were formed, and treatment was started. dmaKG was diluted in phosphate buffered saline (75 mg/mL) daily before the injection. The solution was administered via intraperitoneal injection based on body weight (200 μL/20 g) twice a day with a 2-hour interval, which equals to 750 mg/kg per injection. Blind measurements of tumor mass were performed twice a week using a caliper and volume was calculated using 0.5*(short dimension)²*(long dimension). Body weight was measured once a week.

For 143B-CytB and 143B-wt xenograft tumor models, athymic nude mice were injected subcutaneously on the right side of abdominal region with 5 × 10⁶ 143B-CytB or 143B-wt cells. Tumor growth was monitored at regular intervals and once the tumor volume was above or equal to 200 mm³, mice were randomly assigned into groups and treatment was started. All mice were injected intraperitoneally (100 μL/20 g) twice a day with a 2-hour interval between two injections. The control group received PBS. The dmaKG group received dmaKG dissolved in PBS (1,000 mg/kg per injection). The phenformin group received phenformin dissolved in PBS (25 mg/kg per injection). The cotreatment group receive PBS containing both dmaKG (1,000 mg/kg per injection) and phenformin (25 mg/kg per injection). Tumor dimensions and body weights were measured every third day and volume was measured using the above-mentioned formula.

At the conclusion of the experiments, some tumors and organs were harvested and fixed in 4% paraformaldehyde/PBS, pH 7.4. The tissues were then embedded in paraffin and serially sectioned. For routine histologic screening, tissue sections were stained with hematoxylin and eosin by standard methods. In addition, some tumors and organs were harvested and snap frozen in liquid N2 for additional downstream analyses. Moreover, blood was collected by cardiac puncture after the animal was euthanized and was submitted to MGH CCM Clinical Pathology Lab for chem panel and serum analysis.

All statistical analyses were performed in GraphPad Prism. Statistical methods used for each experiment are as indicated in the figure legends.

Additional information can be found in Supplementary Methods.

To uncover metabolic vulnerabilities unique to OXPHOS-incompetent cancer cells, we created isogenic cell pairs carrying an empty vector (ev) or an expression vector of wild-type FH (FH) from two FH-mutant HLRCC cell lines - UOK268 derived from a primary kidney tumor and UOK262 derived from a metastasis of a kidney tumor (25, 26). It has been shown that FH mutation leads to accumulation of fumarate and that fumarate may act as an oncometabolite by impinging on several pathways associated with cancer (8, 27–29). As expected, we found that repletion of FH into HLRCC cells results in normalization of fumarate levels, increases in OCR, and decreases in ECAR. In addition, the ev cells rely more on glucose, pyruvate and glutamine for proliferation as compared to their FH-replete counterparts (Supplementary Fig. S1A–S1F). These results indicate that reconstitution of FH in HLRCC cells restores mitochondrial respiration at least partially.

Because fumarate can compete with aKG for binding to aKG-dependent dioxygenases, we initially surmised that addition of dimethyl aKG (dmaKG), a cell-permeable form of aKG, might reverse some of the phenotypes observed in FH-mutant cells attributed to fumarate accumulation. Serendipitously, we noticed that dmaKG at low millimolar concentrations caused rapid and massive cell death within 24 hours in both UOK262 and UOK268 ev cells but not in their FH-replete counterparts (Fig. 1A). Even at subtoxic concentrations, dmaKG displayed antiproliferative effects more potently on the ev cells as compared with the FH-replete cells. Of note, we have found that cells become more sensitive or resistant to dmaKG at a lower or higher pH, respectively. Moreover, removing pyruvate or supplementing additional pyruvate in the medium has no significant impact on dmaKG sensitivity (Supplementary Fig. S2A–S2C).

Similar to dmaKG, other cell-permeable forms of aKG were also differentially toxic between the cell pairs, indicating that aKG maybe the active moiety (Fig. 1A). α-Ketoglutaric acid without a cell-permeable moiety was not cytotoxic up to 13.2 mmol/L, indicating that entry of aKG into the cell is critical for toxicity. Furthermore, we did not find any differential toxicity with cell-permeable forms of other TCA cycle metabolites, highlighting the special property of aKG esters in targeting FH-mutant cancer cells (Supplementary Fig. S3A and S3B).

To characterize the mechanisms leading to cell death, we found that dmaKG caused a significant decrease in the mitochondrial membrane potential and an increase in Annexin-V–positive cells within 6 hours in the ev cells (Fig. 1B and C). Accordingly, we found that cyclosporine A (CsA), which prevents the opening of the mitochondrial permeability transition pore, and the pan-caspase inhibitor Z-VAD-FMK suppressed dmaKG-induced cell death, albeit less effectively with higher concentrations of dmaKG (Supplementary Fig. S4A–S4D). These results suggest that cell death in the presence of dmaKG involves a mitochondria-regulated pathway and may occur via both caspase-dependent and caspase-independent mechanisms.

To investigate whether dmaKG exerts cytotoxicity under other OXPHOS-incompetent cancer conditions, we employed an isogenic cybrid pair of human osteosarcoma 143B cells containing either wild-type mtDNA (143B-wt) or mtDNA with a mutation in the ETC complex III cytochrome b gene (143B-CytB; refs. 14, 17). The data showed that various aKG esters induced pronounced death of the respiration-deficient 143B-CytB cells but not the respiration-proficient 143B-wt cells (Fig. 1D). However, chemical inhibition of the ETC complex I, II, III, or V in either UOK262-FH or 143B-wt cells elicited dmaKG sensitivity in these cells (Fig. 1E). Likewise, treatment with phenformin, an ETC complex I inhibitor, induced dmaKG toxicity in a variety of cancer cell types (Supplementary Fig. S5). Changing the culture condition from 21% to 1% O₂ also induced dmaKG cytotoxicity in a spectrum of cancer cell lines, although its potency varied depending on the cancer cell types (Fig. 1F). We additionally tested an isogenic cell pair of UMRC2, which is a VHL-deficient renal cancer cell line that expresses both HIF1α and HIF-2α (30). UMRC2 carrying an empty vector (ev) and UMRC2 VHL-replete (VHL) cells were both insensitive to dmaKG in normoxia but became equally sensitive in hypoxia, suggesting that a strong suppression of oxidative metabolism, not the pseudohypoxia phenotype, is associated with dmaKG sensitivity (Fig. 1F). In summary, the findings strongly suggest that cell-permeable aKG confers potent cytotoxicity selectively towards OXPHOS-incompetent cancer cells.

We initially hypothesized that, upon dmaKG treatment, OXPHOS-deficient cells died because they accumulated a higher level of aKG compared to OXPHOS-proficient cells. Using mass spectrometry, we found that intracellular aKG levels increased 5- to 25-fold within 0.5 hour of treatment. After 5 hours of incubation, aKG levels were higher in OXPHOS-deficient conditions compared with OXPHOS-proficient conditions in both UOK268 and 143B cell pairs, whereas in UOK262 pair the levels were similar (Supplementary Fig. S6A and S6D). Hence, OXPHOS deficiency may increase the amount of aKG accumulating in the cells. Nevertheless, even when the intracellular aKG levels are similar as in the UOK262 pair, the OXPHOS-deficient cells die more than the OXPHOS-proficient cells, implying an inherent difference in their responses to high levels of aKG.

In addition, we considered the possibility that aKG was converted to 2-hydroxyglutarate (2-HG) inside the cell and that 2-HG might cause cell death. 2-HG has two enantiomers. While D-2HG is known as an oncometabolite of the IDH1/2 mutations, L-2HG elevates naturally under hypoxia or acidic conditions (31, 32). 2-HG may also be converted to aKG via the enzymes D-2HGDH and L-2HGDH. We found that dmaKG significantly increased L-2-HG levels in UOK262, UOK268, and 143B cell pairs except 143B-wt. However, dimethyl L-2HG (dmHG) up to 13.2 mmol/L was not toxic to UOK262-ev or UOK268-ev cells, and yet it potentiated the toxicity of dmaKG possibly due to a further increase in the intracellular aKG concentrations (Supplementary Fig. S6A–S6D). These results suggest that aKG but not 2-HG accumulation elicits cytotoxicity in OXPHOS-defective cells.

aKG may inhibit mitochondrial ATP synthase (33), which may decrease ATP level, leading to cell death. Interestingly, we observed that dmaKG decreased ATP levels acutely in less than 4 hours. Moreover, the reduction was significantly greater in the ev cells than in the FH-replete cells (Fig. 2A). However, oligomycin, a well-characterized inhibitor of ATP synthase, did not reduce ATP levels. Rather, inhibition of glycolysis using 2-deoxy-D-glucose (2DG) rapidly decreased ATP levels in the ev cells within the same timeframe and to a similar extent as did dmaKG. Nonetheless, 2DG reduced ATP levels indiscriminately between ev and FH-replete cells. Furthermore, 2DG was not cytotoxic even after 24-hour treatment (Fig. 2B). Although these results do not clarify the mechanism by which dmaKG decreases ATP levels in HLRCC cells, they indicate that lowering of ATP levels is not sufficient to cause HLRCC cell death.

Next, we examined whether dmaKG might inhibit glycolysis using the Seahorse Real-Time ATP Rate Assay. In both UOK262-ev and -FH cells, glycolysis contributed more towards overall ATP production than did mitochondrial respiration (Fig. 2C). Meanwhile, we found that dmaKG reduced glycolytic ATP production in UOK262-ev but not in UOK262-FH. To confirm these findings, we measured the ECAR using the glycolysis stress assay. The results showed that dmaKG reduced both glycolysis and glycolytic capacity preferentially in the ev cells (Fig. 2D). Similarly, dmaKG lowered ECAR in UOK268-ev and 143B-CytB but not in UOK268-FH and 143B-wt. Moreover, we found that glucose uptake in UOK262-ev was unaffected in the presence of dmaKG (Supplementary Fig. S7A–S7C). Although ECAR is an indirect measurement of glycolysis, these results suggest that dmaKG inhibits the glycolytic pathway at the steps downstream of glucose transport and only in OXPHOS-incompetent cells.

It is known that HLRCC cells exhibit a pseudohypoxic phenotype due to normoxic stabilization of hypoxia-inducible factor-1α (HIF1α), a key regulator of glycolysis (28). Because aKG is a cofactor of HIF prolyl hydroxylases (PHD), enhancing PHD activities may lead to HIF1α degradation. Unexpectedly, we observed that, upon dmaKG treatment, HIF1α protein was decreased in UOK262-ev and UOK268-ev, whereas it was increased in UOK262-FH and UOK268-FH (Fig. 2E). The related compound deaKG conferred similar effects as dmaKG (Fig. 2F). Cell-permeable aKG derivatives have been shown to both increase and decrease HIF1α protein stability, suggesting that the effects may depend on cell type, culture conditions, and the aKG derivatives used (34, 35). Hou and colleagues recently reported that dmaKG causes HIF1α stabilization in various cancer cell lines by inhibiting PHD2 (34). In UOK262-FH, our data also suggest that dmaKG stabilizes HIF1α by inhibiting its prolyl hydroxylation (Supplementary Fig. S8A). Therefore, it is puzzling why HIF1α does not accumulate in OXPHOS-incompetent cells as in OXPHOS-competent cells in response to cell-permeable aKG. In UOK262-ev, the decreases in HIF1α protein levels were not due to changes in its mRNA levels or protein degradation (Supplementary Fig. S8A). Results in the 143B cell pair were consistent with those in the HLRCC cell pairs (Fig. 2G; Supplementary Fig. S8B). Collectively, these data indicate that dmaKG may increase the stability of HIF1α in OXPHOS-competent conditions. However, in OXPHOS-incompetent conditions, dmaKG may downregulate HIF1α through translation via mechanisms such as the mTOR pathway (36).

Although the pseudohypoxia phenotype may play some roles in FH-mutant cancers, previous studies have shown that reductions in HIF1α level do not cause HLRCC cell death (37, 38). To explore alternative cytotoxic mechanisms, we performed LC-MS–based metabolic profiling of key and relevant metabolites following 0.5 and 5 hours of dmaKG treatment in the UOK262 cell pair (Fig. 3A). Not surprisingly, we found that dmaKG increased citrate levels implying reductive carboxylation in both ev and FH-replete cells (14). Nonetheless, these cells showed clear differences in their oxidative metabolism in response to dmaKG. In FH-replete cells, dmaKG increased all TCA cycle metabolites downstream of aKG including succinate, fumarate, and malate. In ev cells, although succinate was increased, there was a decrease in fumarate and no increase in malate. We also found that dmaKG significantly reduced aspartate levels within 0.5 hour in both UOK262-ev and UOK262-FH cells. However, although aspartate level gradually recovered in UOK262-FH, it did not recover in UOK262-ev. Meanwhile, dmaKG caused only modest changes in other amino acids and the changes were largely similar between ev and FH-replete cells (Fig. 3B).

Metabolic profiling of the 143B cell pair following 3 and 6 hours of dmaKG treatment showed similar results (Fig. 3C and D). In 143B-CytB, dmaKG increased citrate levels due to reductive metabolism of aKG (14). Moreover, dmaKG caused significant increases in succinate but decreases in fumarate and malate, suggesting that oxidative metabolism downstream of succinate may be inefficient in these cells. In comparison, in 143B-wt cells in which oxidative metabolism is efficient, dmaKG did not increase citrate or succinate. Notably, dmaKG increased malate levels in 143B-wt and UOK262-FH but not in 143B-CytB or UOK262-ev. In both UOK262 and 143B cell pairs, basal aspartate levels were lower in OXPHOS-incompetent conditions compared with OXPHOS-competent conditions. Treatment with dmaKG caused a further 80% to 90% reduction in aspartate levels in UOK262-ev and 143B-CytB, but only 40% to 60% reduction in UOK262-FH and 143B-wt. However, in 143B-wt cells treated with phenformin, dmaKG again reduced aspartate level to only 10% of the control level (Supplementary Fig. S9A and S9B). These results indicate that dmaKG induces severe aspartate deficiency selectively in OXPHOS-incompetent conditions.

It has been previously shown that aspartate is synthesized by cytosolic GOT1 under OXPHOS-incompetent conditions in contrast to mitochondrial GOT2 under OXPHOS-competent conditions (15). GOT1/2 are transaminases that carry out the reversible reactions from glutamate and oxaloacetate to aKG and aspartate (Fig. 3E). Hence, increasing intracellular aKG concentrations may be incompatible with aspartate synthesis. Moreover, our data suggest that dmaKG may preferentially affect GOT1 as compared with GOT2. Concordantly, a potent, nonselective transaminase inhibitor aminooxyacetic acid (AOA) suppressed aspartate levels to similar levels in UOK262-ev and UOK262-FH (Fig. 3F). We noticed that dmaKG reduced aspartate level even more than AOA did in UOK262-ev, suggesting that aKG may deplete aspartate further by trapping its nitrogen through GOT1 and yield glutamate. To test this hypothesis, we investigated the transfer efficiency of the ¹⁵N atom from isotope-labeled aspartate to aKG, which would produce ¹⁵N-labeled glutamate. Because most cells do not take up aspartate efficiently, we expressed the glutamate-aspartate transporter (SLC1A3) in 143B-wt cells to enhance the uptake of labelled aspartate. The data showed that dmaKG significantly elevated the ratio of labeled glutamate to labelled aspartate in the presence of phenformin, but the effect was less in the absence of phenformin (Fig. 3G). Moreover, overexpression of GOT1 enhanced dmaKG toxicity in both UOK262-ev and 143B-CytB cells (Fig. 3H). As compared with dmaKG, AOA exerted a cytostatic effect in UOK262-ev cells and GOT1 overexpression decreased this effect (Fig. 3I). Similarly, overexpression of GOT1 augmented dmaKG toxicity in 143B-wt and 786-O, a renal cancer cell line, under hypoxia or phenformin-treated conditions, whereas it suppressed AOA and phenformin mediated cytotoxic effect (Supplementary Fig. S10A and S10B). These results highlight the differences in the actions of dmaKG and AOA, and they strongly suggest that dmaKG consumes aspartate and causes cell death in OXPHOS-incompetent cells through GOT1 activity.

As previously reported, we found that GOT1 expression became essential under respiration-deficient conditions (15). To determine whether aspartate depletion might be the cause of dmaKG cytotoxicity in OXPHOS-incompetent cells, we supplemented aspartate in the culture media. The results showed that adding 10 mmol/L of aspartate almost completely rescued both HLRCC cell lines and 143B-CytB cells from dmaKG's toxic and anti-proliferative effects (Fig. 4A and B; Supplementary Fig. S11A and S11B). Moreover, in UOK262 cells expressing SLC1A3 ectopically, 150 μmol/L of aspartate effectively prevented dmaKG-induced cell death (Fig. 4C). Thus, the data support that dmaKG mediates cytotoxicity by inducing aspartate deficiency in these cells.

Because a major role of aspartate in proliferating cells is to participate in nucleotide synthesis, we assessed whether the dmaKG-mediated decrease in ATP might be in part due to adenylate depletion. Using a biochemical method, we found that adenylates (ATP/ADP/AMP) exhausted quickly upon dmaKG treatment in UOK262-ev but not in UOK262-FH cells (Fig. 4D). Moreover, aspartate supplementation restored adenylate levels in both UOK262-ev and 143B-CytB treated with dmaKG, indicating that the exhaustion was caused by aspartate deficiency (Fig. 4E). In another experiment using the UOK262 pair, we analyzed ATP, ADP, and AMP levels individually and found that dmaKG reduced all of them in ev but not FH-replete cells. In contrast, the glycolysis inhibitors 2-DG and 3-bromopyrvate (3-BrPy) reduced only ATP but not ADP or AMP in both UOK262-ev and -FH cells (Supplementary Fig. S12). To determine the role of adenylate deficiencies, we added adenine or ATP to the media and observed significant rescue of dmaKG cytotoxicity, except that rescue with ATP was not seen in 143B-CytB (Fig. 4F). Uptake of ATP has been shown to vary among different cell types (39). We tested whether the ATP analogue β,γ-methyleneadenosine 5′-triphosphate (AMP-PCP) might permeate cell membrane and found that it substantially reduced dmaKG sensitivity in 143B-CytB. We confirmed that adenylate levels were restored by ATP addition in UOK262-ev but not in 143B-CytB and by AMP-PCP in 143B-CytB (Supplementary Fig. S13A and S13B). In summary, these findings demonstrate that in OXPHOS-incompetent cells, adenylates are rapidly depleted when aspartate is limited. This depletion may lead to not only cessation of DNA synthesis and cell proliferation but also other downstream consequences that cumulatively contribute to cell death.

We sought to investigate further the downstream consequences of aspartate and adenylate deficiencies and to explore the mechanism by which dmaKG suppressed glycolysis in OXPHOS-incompetent cells. Surprisingly, supplementation of aspartate in UOK262-SLC1A3 as well as ADP or ATP in UOK262-ev nullified the effects of dmaKG on glycolysis and glycolytic capacity (Fig. 5A). Supplementation of aspartate and AMP-PCP also restored glycolysis and glycolytic capacity in dmaKG-treated 143B-CytB cells (Fig. 5B). These data imply that glycolysis might be halted due to the lack of adenylates. In addition to ADP and ATP, NAD⁺ is also an essential cofactor for glycolysis. We measured NAD⁺ and NADH in the UOK262 pair and found that the NAD⁺/NADH ratios were not significantly changed by dmaKG. However, total NAD⁺ and NADH level decreased in the presence of dmaKG, possibly because biosynthesis of NAD⁺ via both de novo and salvage pathways requires ATP (Supplementary Fig. S14A and S14B). Likewise, we observed that NAD⁺ and NADH declined more slowly compared with the adenylates in UOK262-ev. Collectively, these findings suggest that adenylates are essential for glycolysis. Therefore, adenylate depletion mediated by dmaKG and aspartate deficiency can inhibit glycolysis.

The deficiencies in aspartate and adenylates also prompted us to investigate the mTOR pathway, which is a major nutrient-sensing mechanism in the cell (40). Both amino acid and purine nucleotide shortages can abrogate mTOR activity, which normally stimulates protein synthesis via phosphorylation of P70S6K1 and 4E-BP1 (41–43). In HLRCC and 143B cell pairs, we found that dmaKG suppressed phosphorylation of P70S6K1, 4E-BP1, and S6 more strongly in the OXPHOS-deficient cells than in the OXPHOS-proficient cells (Fig. 5C). Supplementation of aspartate and ATP in UOK262-ev or aspartate and AMP-PCP in 143B-CytB reversed these effects, indicating that dmaKG-mediated aspartate and adenylate deficiencies can lead to mTOR inhibition (Fig. 5D). Inhibition of mTOR will lead to growth arrest and may contribute to cell death (40). Meanwhile, because HIF1α is known to be translationally regulated by mTOR, these findings suggest that dmaKG downregulates HIF1α in HLRCC cells but not in their FH-replete cell pairs due to its selective inhibition of mTOR in OXPHOS-incompetent cells (36).

To gain insight into dmaKG's effects on OXPHOS-deficient tumors in vivo, we generated xenograft tumors in mice using the 143B cell pair. In the 143B-CytB tumor model, tumor growth curves began to diverge between the control group and the dmaKG group after treatment started (Fig. 6A, left). By day 6, numerous metastatic tumor nodules appeared in two out of six mice in the control group; hence, their tumor volumes were not counted thereafter. On the other hand, in the dmaKG group, tumor metastasis was not found in any of the six mice (Fig. 6B) and tumor growth was impeded. In contrast, in the 143B-wt xenograft tumor model, treatment with dmaKG or phenformin did not affect tumor growth, whereas dmaKG and phenformin cotreatment effectively abrogated tumor progression (Fig. 6A, right). No tumor metastasis was found in any of the 143B-wt xenografts, consistent with a previous report implicating an inverse relationship between OXPHOS competency and metastatic potential of 143B tumors (23). In addition, both aspartate and adenylate levels were significantly reduced by dmaKG treatment in 143B-CytB tumors and by dmaKG and phenformin cotreatment in 143B-wt tumors (Fig. 6C and D). Accordingly, Western blot analysis showed that mTOR signaling was inhibited by dmaKG treatment in 143B-CytB tumors and by dmaKG and phenformin cotreatment in 143B-wt tumors (Supplementary Fig. S15A and S15B). We also extracted brain, heart, kidney, liver, lung, and spleen from the mice bearing 143B-wt tumors and found that neither aspartate nor adenine nucleotide pool levels were significantly affected in these organs by any of the treatments (Fig. 6E and F). The mouse body weight, a complete blood count analysis and a panel of blood chemistry assays measured at the conclusion of the experiment also did not find any toxicity caused by the treatments (Supplementary Fig. S16A–S16C).

We additionally investigated dmaKG's effects using UOK262 xenograft tumor model. It has been previously shown that while UOK262-ev can form xenograft tumors in mice, UOK262-FH cannot, suggesting that the FH mutation is crucial for the tumorigenicity of HLRCC cells (38). Similar to its effect on 143B-CytB tumors, dmaKG significantly reduced the progression, aspartate level, and mTOR signaling in UOK262 tumors (Supplementary Fig. S17A–S17C). In tumor sections from dmaKG-treated mice, regions of dead cells were clearly visible, whereas such regions were seldom observed in tumor sections from saline-treated animals (Supplementary Fig. S17D and S17E). Meanwhile, no tissue damage was seen in sections of heart, liver, and kidneys after 3 weeks of treatment and mouse weight changes between control and dmaKG-treated groups were similar (Supplementary Fig. S18A and S18B). These data indicate that treatment with dmaKG is well tolerated and depletes aspartate in OXPHOS-impaired tumors selectively, providing important evidence of dmaKG's anticancer efficacy in vivo.

Recent studies have shown that aspartate, especially under the conditions of impaired mitochondrial metabolism, is a limiting factor for tumor growth (12, 18). However, a strategy to target aspartate availability in cancers with mitochondrial dysfunction had yet to be identified. Meanwhile, it was unclear whether one could selectively deplete aspartate in cancer without affecting normal cells in vivo. In this report, we identify that cell-permeable aKG esters, such as dmaKG, are toxic to cancers exhibiting mitochondrial metabolism deficiency by promoting aspartate utilization and depletion via GOT1, therefore rendering aspartate inaccessible to alternative biosynthetic usage (Fig. 7). Moreover, the current study provides encouraging in vivo evidence that indicates anticancer efficacy and specificity of this approach, warranting its future clinical investigations.

Notably, the data suggest that there are several differences between respiration-proficient and respiration-deficient cells that contribute to their differential sensitivities to dmaKG (Fig. 7). First, after entering the cell, dmaKG is converted into aKG by esterases, causing a surge of intracellular aKG level in both conditions. However, competency in mitochondrial metabolism can cause aKG levels to normalize quickly, presumably through its oxidative utilization by the mitochondrial TCA cycle. Thus, OXPHOS-incompetent cells may experience a higher amount of aKG accumulation for a longer period of time compared with OXPHOS-competent cells due to their mitochondrial metabolism defects. Second, the increase in aKG level is likely to be higher in the cytosol than in the mitochondria, causing a stronger effect on GOT1 as compared with GOT2. It is known that GOT1 is the main aspartate producer in OXPHOS-incompetent conditions in contrast to GOT2 in OXPHOS-competent conditions (15). Third, aKG can increase TCA cycle metabolites such as malate inside the mitochondria of OXPHOS-competent cells. Although we cannot reliably detect oxaloacetate levels by mass spectrometry, increases in malate may increase oxaloacetate, thereby boosting aspartate production via GOT2 in OXPHOS-competent conditions. All of these mechanisms combined may account for the selective depletion of aspartate in cancer cells with mitochondrial dysfunction. As compared with the GOT1/2 inhibitor AOA, which inhibits aspartate synthesis and reduces aspartate levels equally between OXPHOS-competent and incompetent conditions, dmaKG converts GOT1 from an aspartate producing enzyme to an aspartate utilizing enzyme in OXPHOS-incompetent conditions (Fig. 7). These findings highlight the unique therapeutic specificity of dmaKG for OXPHOS-incompetent tumors, which will reduce the likelihood of adverse effects on normal cells.

Although dmaKG induces potent cytotoxicity in a variety of cancer cell types under OXPHOS-incompetent conditions, the concentration required for its toxicity varies. This may be due to differences in the endogenous metabolite levels and the rates of metabolism in different cell types. However, even at nontoxic concentrations, dmaKG also inhibits cell proliferation preferentially in OXPHOS-incompetent conditions. Similarly, using GOT1 knockout cells, Birsoy and colleagues showed that the level of aspartate depletion determines whether cells die or become growth arrested (15). In addition to OXPHOS deficiency, it is possible that some genetic mutations that alter the aspartate synthesis pathways, aspartate utilization, or the cell's reliance on redox balance mediated by the aspartate–malate shuttle may also induce dmaKG sensitivities (44, 45). Thus, cell-permeable aKG derivatives may confer anticancer effects in cancer types other than those with TCA cycle or mtDNA mutations.

Aspartate is a substrate for biosynthesis from phosphoribosyl pyrophosphate (PRPP) to inosine monophosphate (IMP) and from IMP to AMP. We show that dmaKG leads to a rapid decline of adenylates in OXPHOS-incompetent cells due to aspartate deficiency. This is consistent with the results of Garcia-Bermudez and colleagues, which show that the rate of aspartate-mediated nucleotide synthesis increases under OXPHOS-incompetent conditions (12). Moreover, dmaKG mediates suppression of glycolysis and mTOR inhibition. Unexpectedly, we found that restoration of adenylate levels restores glycolysis. Interestingly, mature human erythrocytes are devoid of de novo nucleotide biosynthesis. It has been reported that restoration of adenylates in aged erythrocytes also restores glycolytic activities in these cells (46). Taken together, these results demonstrate that adenine nucleotides can play a limiting role in glycolysis. In addition, we show that supplementation of either aspartate or adenylates restore mTOR activity. Although mTOR is known to integrate a plethora of signals ranging from nutrients to growth factors to stress, these data suggest that at least one of the early signals that leads to mTOR inactivation in dmaKG-treated cells is adenylate depletion. Previously, two groups independently reported that mTOR complex 1 (mTORC1) is distinctly sensitive to purine but not pyrimidine nucleotide levels (41, 43). Hoxhaj and colleagues show that inactivation of mTORC1 by purine deficiency is not mediated by the energy sensor AMPK (43). Collectively, the evidence indicates that dmaKG mediates cytotoxicity by depleting aspartate and adenylates in respiration-incompetent cancer cells. This leads to cessation of glycolysis, the major energy producing mechanism in these cells, as well as to inactivation of mTOR, the indispensable cell growth signaling pathway.

Using mouse xenograft tumor models, we found that in vivo administration of dmaKG replicates its effects observed in the cell culture system. It hampers the progression of UOK262 and 143B-CytB tumors that have intrinsic mitochondrial dysfunction, greatly reduces aspartate and adenine nucleotide levels, and inhibits mTOR signaling in them. These effects are not seen in 143B-wt tumors that have functional mitochondrial metabolism, indicating the specificity of this therapeutic strategy. Treatment of 143B-CytB–implanted animals with dmaKG also completely abrogated metastatic nodules. In addition, our cell culture experiments indicate that dmaKG sensitivity can be enhanced by hypoxia or by cotreatment with an ETC inhibitor in cancers that do not have intrinsic mitochondrial defects. Since cancer cells in vivo often reside in a variable degree of hypoxic environment, it will be informative to investigate the potential added benefits of dmaKG in conjunction with other standard therapeutic approaches. We demonstrate that dmaKG and phenformin in combination, but not individually, impair the progression of 143B-wt tumors. While we were pursuing these experiments, Sica and colleagues reported that ETC complex I inhibitor BAY87-2243 elicits synthetic lethality with dmaKG and concluded that cell death is caused by energy failure due to inhibition of OXPHOS and a MDM2-dependent inhibition of glycolysis (47). Although their report did not investigate the roles of GOT1, it independently confirms the therapeutic synergy between dmaKG and OXPHOS inhibition. Interestingly, using a number of analyses in mouse blood and various organs, we have not found any significant toxicity in the mice treated with dmaKG or cotreated with dmaKG and phenformin for a short period (15–21 days). Moreover, aspartate and adenylate levels in various mouse organs are unaffected, in contrast to their significant decreases in the tumors. These data may imply that cancer cells and normal cells are differentially sensitive to ETC inhibitors in vivo, which could be due to some intrinsic differences in their mitochondrial metabolism efficiencies or extrinsic differences in the oxygen levels of their environments. In other studies, dmaKG has been shown to confer cardiac protection in a mouse model of thoracic aortic constriction and to extend lifespan in nematodes (33, 48, 49). Together, these findings suggest that aKG derivatives may have positive safety profiles, although a thorough preclinical evaluation of toxicity for a long-term treatment with dmaKG is needed. In conclusion, this study identifies a new and promising therapeutic strategy, which may form the basis for the development of various broad-spectrum anticancer approaches.

A.M. Metelo reports the PhD student grant SFRH/BD/70002/2010 from Fundação para a Ciência e a Tecnologia, Portugal, funded by POPH/FSE, Programa Operacional e da União Europeia. B. Olenchock reports employment at Regeneron Pharmaceuticals, Inc., a pharmaceutical company that is developing cancer therapeutics (not employed at Regeneron while conducting the research included in this publication). A.M. Intlekofer has previously consulted for Foundation Medicine, Inc. (over three years ago). O. lliopoulos reports a patent for 61/363,192 issued (patent filed by MIT). No potential conflicts of interest were disclosed by the other authors.

H.R. Madala: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. I.T. Helenius: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing-original draft, writing-review and editing. W. Zhou: Investigation, methodology, writing-review and editing. E. Mills: Resources, methodology, writing-review and editing. Y. Zhang: Data curation, validation, investigation, methodology, writing-review and editing. Y. Liu: Data curation, formal analysis, investigation, writing-review and editing. A.M. Metelo: Investigation, methodology, writing-review and editing. M.L. Kelley: Investigation, writing-review and editing. S. Punganuru: Resources, writing-review and editing. K.B. Kim: Resources, writing-review and editing. B. Olenchock: Investigation, methodology, writing-review and editing. E. Rhee: Resources, writing-review and editing. A.M. Intlekofer: Investigation, methodology, writing-review and editing. O. lliopoulos: Resources, funding acquisition, writing-review and editing. E. Chouchani: Funding acquisition, methodology, writing-review and editing. J.-R.J. Yeh: Conceptualization, resources, supervision, funding acquisition, investigation, writing-original draft, project administration, writing-review and editing.

The authors thank Drs. Cyril Benes, Matthew Vander Heiden, Lucas Sullivan, Navdeep Chandel, Youfeng Yang, and Marston Linehan for providing human cancer cell lines and for helpful discussions, Dr. Miguel Rivera for assistance on interpreting tumor sections and Dr. Mohammed Mahamdeh for assistance on microscope image acquisition and processing. This work is supported by the Hassenfeld Scholar Award (to J.-R.J. Yeh), NIH R01 CA215410 (to J.-R.J. Yeh), NIH R01 CA160458 (to O. Iliopoulos), and the Claudia Adams Barr Program (to E. Chouchani). A.M. Intlekofer is supported by the NIH/NCI (R37 CA251543, K08 CA201483, P30 CA008748).

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.