The tricarboxylic acid (TCA) cycle is the central hub of oxidative metabolism, running in the classic forward direction to provide carbon for biosynthesis and reducing agents for generation of ATP. Our metabolic tracer studies in melanoma cells showed that in hypoxic conditions the TCA cycle is largely disconnected from glycolysis. By studying the TCA branch point metabolites, acetyl CoA and citrate, as well as the metabolic endpoint glutamine and fatty acids, we developed a comprehensive picture of the rewiring of the TCA cycle that occurs in hypoxia. Hypoxic tumor cells maintain proliferation by running the TCA cycle in reverse. The source of carbon for acetyl CoA, citrate, and fatty acids switches from glucose in normoxia to glutamine in hypoxia. This hypoxic flux from glutamine into fatty acids is mediated by reductive carboxylation. This reductive carboxylation is catalyzed by two isocitrate dehydrogenases, IDH1 and IDH2. Their combined action is necessary and sufficient to effect the reverse TCA flux and maintain cellular viability.
By profiling different melanoma cell lines, we found that reverse TCA flux by reductive carboxylation is a general feature of melanoma metabolism. Consequently, reductive carboxylation by isocitrate dehydrogenases could be significant for the pathology of melanoma.
Many metabolic routes are rewired in tumors to accommodate the need for biosynthetic precursors. The Warburg effect, the most cited change in tumor metabolism, involves increased consumption of glucose along with a corresponding increase in the production of lactate (Warburg, 1923, 1956). This takes place even in the presence of oxygen, hence much of the energy produced by the tumor cell is derived from glycolysis. Other branches of central carbon metabolism are also altered in tumor cells, including up-regulation of serine and fatty acid biosynthesis (Flavin et al., 2010; Kridel et al., 2004; Locasale et al., 2011; Possemato et al., 2011). Most of these changes have been observed in studies done in normoxia; so far, less is known about transitions in metabolic carbon flux resulting from hypoxia. In a recently published study we found that hypoxia induces an adaptive metabolic transition in melanoma cells (Scott et al., 2011). This adaptation is particularly relevant to fatty acid biosynthesis because the pyruvate dehydrogenase complex, which generates the fatty acid precursor acetyl CoA, is down-regulated by hypoxia inducible factor-1a (Kim et al., 2006). These findings suggested the possibility that metabolic adaptation to hypoxia involves a switch from glucose to the alternative nutrient glutamine as a source of carbon for the synthesis of fatty acids.
Glutamine has been recognized to play a role similar to glucose, as it is used to generate ATP and provide precursors for biosynthetic reactions. In this way, glutamine can be routed through the tricarboxylic acid (TCA) cycle to maintain oxidative phosphorylation, a process called glutaminolysis (Deberardinis et al., 2007; Mckeehan, 1982; Wise et al., 2008). But tumor cells could also route glutamine through another metabolic pathway, in which carbon travels through the TCA cycle in reverse (Dalziel and Londesborough, 1968). In this path, glutamine enters the TCA cycle as oxoglutarate and undergoes reductive carboxylation by isocitrate dehydrogenases (IDHs) to produce isocitrate. Isocitrate is converted to citrate by the reverse reaction of aconitases, making the carbon available for ATP citrate lyase, which produces acetyl CoA (Figure 1). This reverse (reductive) flux is more important than initially anticipated (Koppenol et al., 2011; Scott et al., 2011), especially because carbons from glutamine can be found in fatty acids (Gaglio et al., 2011). Reductive carboxylation has been detected in liver and heart (Comte et al., 2002, 1997; Des Rosiers et al., 1994), and glutamine is a major carbon source for fatty acid synthesis in brown adipocytes (Yoo et al., 2008). Despite the evidence for reverse flux of carbon from glutamine through the TCA cycle, its significance for melanoma remains unclear. In this report, we explore the possibility that reverse (reductive) flux through IDHs is necessary for fatty acid biosynthesis by melanoma cells in hypoxia.
Three genes encode the IDHs (Bell and Baron, 1968); IDH1 is localized in the cytoplasm, and IDH2 and IDH3 are localized in mitochondria (Bell and Baron, 1968). IDH3 differs from the other two IDHs because it functions as a heterotetramer comprising two alpha subunits, one beta subunit, and one gamma subunit. IDH3 is regulated allosterically by ATP, NADH, and NADPH (Bell and Baron, 1968; Bell et al., 1962; Bzymek and Colman, 2007; Reynolds et al., 1978), and prefers NAD as a cofactor. In contrast to IDH3, IDH1, and IDH2 function as homodimers, use NADPH as a cofactor, and lack allosteric effectors. IDH1 and IDH2 catalyze the first step in the reductive carboxylation pathway originating from glutamine. They produce isocitrate from oxoglutarate by reductive carboxylation (Dalziel and Londesborough, 1968). Here we show that, because of this reaction, glutamine becomes an important source of carbon for fatty acids in melanoma cells cultured in hypoxic conditions. Consequently, reductive carboxylation by IDHs could be significant for the pathology of melanoma.
The source of carbon for acetyl CoA differs between normoxia and hypoxia
One objective of this study was to compare glucose and glutamine utilization in normoxic conditions (21% oxygen) versus hypoxic conditions (1% oxygen). We compared the incorporation of carbon from glucose and glutamine into lipogenic metabolites in melanocytes (H3A) versus melanoma cells (WM35 and LU1205). These experiments were performed with isotopically labeled (13C) precursors at pre-steady state conditions to gauge the rate of carbon incorporation into citrate, acetyl CoA, fatty acids, and glycerol (Figure 2). In normoxia, glucose is the main source of carbon for acetyl CoA in all three cell types. However, in hypoxia, flux from glucose to acetyl CoA is dramatically reduced and glutamine becomes the major contributor of carbon (Figure 2A). Flux to citrate follows the same trend (Figure 2B). Whereas flux from glutamine to acetyl CoA is observed in both melanocytes and melanoma cells, melanocytes use substantially less glucose or glutamine than melanoma cells.
Next we measured the incorporation of 13C from glutamine or glucose into saturated fatty acids of the lipid fraction of cells. The H3A melanocytes exhibit no significant fatty acid biosynthesis in normoxia or hypoxia (Figure 2C). In melanoma cells, carbon from both precursors is observed in fatty acids, but in hypoxia the contribution of glucose carbon to fatty acids is decreased by about half, whereas the contribution from glutamine increases two-fold (Figure 2C). As a control, we measured the incorporation of isotope into the polar glycerol headgroup of lipids where the carbons are derived from glycolysis (Figure 2D). Thus, although glutamine becomes the primary source of carbon for fatty acids, it is not metabolized for the synthesis of glycerol headgroups.
Citrate is generated from glutamine by reductive carboxylation in hypoxia
There are two major routes by which carbon from glutamine can enter the acetyl CoA pool: glutaminolysis, wherein glutamine enters and traverses the TCA cycle in the forward direction, or reverse flux through IDHs to citrate. These two possibilities can be distinguished using [U-13C]-glutamine because the two routes generate different isotopic labeling of citrate (Figure 3A). In the case where [U-13C]-glutamine travels via the TCA cycle by glutaminolysis, its carbon backbone is broken into separate pieces. This is evidenced by the domination of the NMR coupling pattern of C2 of citrate by singlet and doublet signals of carbon fragments. However, in the case of direct flux to citrate, the carbon backbone of glutamine remains intact, and the NMR coupling of C2 is a doublet of doublets. In brief, forward, oxidative flux from acetyl CoA or oxaloacetate results in singlets or doublets, but reverse, reductive flux from oxoglutarate generates doublets of doublets (Figure 3B). WM35 and LU1205 melanoma cells showed a predominant doublet of doublets in hypoxia, indicating that the majority of carbon from [U-13C]-glutamine is routed in the direction of reductive carboxylation from oxoglutarate. Whereas some citrate was synthesized via the reductive IDH route in normoxia, the contribution of this route to citrate synthesis (relative to total synthesis) nearly tripled in hypoxia in the melanoma cell lines (Figure 3C).
The metabolic flux from glutamine into fatty acids is routed through IDH1 and IDH2
We sought to determine which of the three human IDHs is responsible for the reductive flux from glutamine into citrate and ultimately into acetyl CoA. This was accomplished by siRNA knockdown of each IDH and a subsequent assessment of flux from glutamine into fatty acid. These experiments were done in WM35 and LU1205 melanoma cells under hypoxic conditions. Knockdown of either IDH1 or IDH2 in LU1205 cells reduced reductive carboxylation and showed a contribution of both enzymes in the reductive direction (Figure 4A), but general metabolic flux into polar metabolites, including organic acids and TCA cycle intermediates oxoglutarate, citrate, and malate, was not significantly changed (Figure 4B). As a next step, we tested the combined role of IDH1 and IDH2 as well as the impact of this route on the metabolic endpoint fatty acids. In WM35 cells, knockdown of either IDH1 or IDH2 individually reduced flux from glutamine into fatty acids by about 50% (Figure 5A) but the combined knockdown reduced flux by more than 80%. Perhaps not surprisingly, knockdown of IDH1 and 2 also reduced cell viability, and the combined knockdown reduced viability by nearly 80% (Figure 5B). The incorporation of stable isotope label into glutamate from glutamine showed that central carbon metabolism was still functioning despite knockdown of IDH isozymes (Figure 5C). Specific knockdown of each IDH was confirmed by Western blot (Figure 5D).
Reductive carboxylation from glutamine is a general feature of melanoma metabolism
To determine whether reductive carboxylation of glutamine and subsequent routing of carbon into fatty acids is a common feature of the metabolism of melanoma cells, we studied four additional melanoma cell lines (Figure 6). Three parameters were determined – uptake of glutamine, the extent of reductive flux, and the relative incorporation of carbon from glutamine into acetyl-CoA in fatty acids. Each of the melanoma lines utilized glutamine, although the level of uptake differed by as much as sixfold. Uptake was slightly less under hypoxia than under normoxia (Figure 6A). Hence glutamine usage alone is not a strong indicator of reverse, reductive TCA flux. The relative amount of citrate synthesized by reductive carboxylation was higher in all cell lines under hypoxia (Figure 6B). In most cases, the proportion of citrate produced by reductive carboxylation increased at least twofold with hypoxia. The labeling of acetyl CoA units in fatty acids from glutamine was elevated in hypoxia for all cell types (Figure 6C). We must point out that the low glutamine uptake caused weak labeling of metabolites of the SBCL2 cell line. However, the general trend that reductive carboxylation is enhanced in SBCL2 is consistent with other cell lines. Together these findings indicate that in hypoxia the compensatory reductive flux from glutamine toward fatty acids is a general feature in melanoma cells.
The results of this study support the following conclusions on glutamine metabolism of melanoma cells in hypoxia: (i) acetyl CoA required for fatty acid biosynthesis is derived from glutamine by reverse flux through the TCA cycle, (ii) IDH1 and IDH2 each contribute to this reverse flux, and (iii) reverse flux through IDH1 and IDH2 is necessary for melanoma cell viability. Our first finding is that the majority of fatty acids synthesized in melanoma cells under hypoxic conditions are derived from glutamine. Carbon from glutamine is routed through reverse flux of the TCA cycle to produce citrate and ultimately acetyl CoA (Figure 7). This metabolic pathway is distinct from glutaminolysis, where glutamine is used for the production of energy rather than for biosynthesis. Our measurements show that despite different levels of glutamine uptake among melanoma cell lines, they all use reverse TCA flux through IDHs when exposed to hypoxic conditions. Increased utilization of glutamine as a carbon source for fatty acid biosynthesis was observed in melanoma lines harboring mutations in B-Raf and N-Ras. As a general property, then, when the supply of oxygen is limiting, the ability of melanoma cells to run one section of the TCA cycle in reverse ensures that cellular fatty acid synthesis is maintained. This is in contrast to melanocytes, where the 13C incorporation into fatty acids was on the order of natural abundance enrichment in normoxia and hypoxia, indicating a lack of de novo fatty acid synthesis. Although the switch of carbon source for acetyl CoA and citrate is similar in melanocytes and melanoma cell lines, this reductive flux only translated into fatty acids in melanoma cells because melanocytes do not synthesize fatty acids. The general validity of this hypoxic switch in combination with the importance of flux into fatty acids, which is elevated in cancer cells (Kridel et al., 2004), may create a therapeutic opportunity. Our second and third conclusions address the contribution of IDH1 and IDH2 to reverse flux. We show that knockdown of either IDH1 or IDH2 lowers reverse TCA cycle flux. Although we cannot exclude the possibility that knockdown of one IDH isoform in some way causes a decrease in reductive flux through the other isoform, the simplest and most direct interpretation of our results is that the enzyme which is knocked down participates in reverse flux. Hence, we conclude that both IDH1 and IDH2 are capable of mediating reverse flux. Importantly, the effects on reverse flux are mirrored in the effects on cell viability. The effects of IDH knockdown on viability do not impact our measurements of flux because this parameter is always normalized to total protein content, the incorporation of stable isotopes into other TCA metabolites was not affected, and measurements of flux were made at time points prior to extensive cell death.
While the present study was underway, three studies were published in high-profile journals that arrive at related conclusions on reductive carboxylation. One study focused on osteosarcoma cells that have an impaired TCA cycle. The authors showed that these cells can partially compensate for the mitochondrial defect by routing glutamine through the reductive flux by IDH1 and IDH2 (Mullen et al., 2012). In melanoma, we observed no impairment of the TCA cycle (Scott et al., 2011), although hypoxia reduced input of carbon from glucose into the TCA cycle. A second report pointed to a critical role for reductive flux through IDH1 in the production of acetyl CoA from glutamine (Metallo et al., 2012). In that study, knockdown of IDH1, but not IDH2, reduced reverse reductive flux and proliferation in a number of different tumor cell lines, although none of these was melanoma, and the knockdowns were performed only in normoxia. Our finding in melanoma cells under hypoxia differs because we observed a clear contribution of both IDH1 and IDH2. Curiously, a third report on human glioblastoma cells suggested that IDH2, but not IDH1, mediates reductive carboxylation of glutamine in hypoxia (Wise et al., 2011). The report did not address a potential role of IDH1, which is an important omission, since in glioma the majority of tumors contain heterozygous mutations in IDH1 but only a few percent contain mutations in IDH2 (Yan et al., 2009). In addition, this third report primarily focused on the generation of 2-hydroxyglutarate, whereas we focused on the generation of fatty acids. Since 2-hydroxyglutarate is a product of mutated IDHs (Ward et al., 2010) and its role in metabolism still needs to be defined, our focus on core TCA metabolites, such as acetyl-CoA and citrate, or fatty acids, the metabolic end-product of reductive carboxylation, allows us to suggest a physiological role for this pathway. The idea that reductive carboxylation mediates reverse TCA flux when the supply of carbon from acetyl CoA is limited is entirely consistent with our data from melanocytes and melanoma cell lines. Therefore we conclude that in the studied cell lines, IDH1 and IDH2 play complementary roles in reductive carboxylation.
In summary, this study shows that in a panel of melanoma cell lines containing either B-Raf or N-Ras mutations, glutamine becomes a major source of carbon for lipogenesis in hypoxia. This metabolic adaptation is distinct from glutaminolysis and is controlled by IDH1 and IDH2. Both enzymes function in the reverse direction to catalyze reductive carboxylation of oxoglutarate to produce isocitrate. Aconitase converts isocitrate to citrate, which is broken down into acetyl CoA units for lipogenesis by ATP-citrate lyase. Together with our previous findings (Scott et al., 2011), our results indicate that IDH1 and IDH2 may be viable drug targets in melanoma.
Tissue cell culture
Human H3A melanocyte, WM35, LU1205, WM793, WM1366, WM3629, SBCL2 melanoma cell lines were maintained in supplemented melanocyte media (M254-500; Invitrogen Life Technologies, Carlsbad, CA, USA) or supplemented minimal essential media (MEM). Supplemented media contained MEM (15010CV; CellGro Corning, Manassas, VA, USA), 10% v/v fetal bovine serum (SH3039603, Lot AUD34653; Hyclone, Logan, UT, USA), 1% v/v antibiotic/antimycotic solution (PS-20; Omega Scientific Inc., Tarzana, CA, USA), 1% v/v MEM non-essential amino acids (SH30238.01; Hyclone, Logan, UT, USA), 1% v/v MEM vitamins (25020CI; CellGro Corning, Manassas, VA, USA), 1 mM l-glutamine (25005CI; CellGro Corning, Manassas, VA, USA). The H3A cell lines were additionally supplemented with 1% v/v human melanocyte growth supplement (HMGS S0025; Invitrogen Life Technologies, Carlsbad, CA, USA). Cells were labeled with [U-13C] glucose (389374; Sigma-Aldrich Isotec, Mamisburg, OH, USA) or [U-13C] glutamine (605166; Sigma-Aldrich) by seeding 4 × 106 cells into 150-mm dishes. At time 0 h, the complete medium was replaced with MEM supplemented with 2 g/l glucose total, of which 50% was [U-13C] glucose or [U-13C] glutamine. After a 24-h stable isotope labeling period, approximately 2 × 107 cells were harvested by incubation with 2.5 g/l trypsin for 5 min (25200-056; Invitrogen Life Technologies, Carlsbad, CA, USA). For experiments in normoxia, cells were maintained, in 21% O2, 74% N2 and 5% CO2; in hypoxia, cells were maintained in 1% O2, 94% N2 and 5% CO2.
Small interfering RNA knockdown experiments
WM35 melanoma cells (1 × 106) were transfected in serum-reduced OptiMEM medium (31985070; Invitrogen) with RNAiMAX Lipofectamine (13778-150; Invitrogen) using 10 nM siRNA (IDH1, J-008294-11; Thermo Scientific Dharmacon, Lafayette, CO, USA; IDH2, J-004013-09; Dharmacon; IDH3A, 4392420 s7122; Ambion Life Technologies, Carlsbad, CA, USA). Forty-eight hours after transfection, the medium was changed and cells were incubated in supplemented MEM with 1 mM [U-13C] glutamine in hypoxia. The efficiency of knockdown was assayed at the protein level 72 h after transfection by Western blotting. The cellular proteins were extracted with CelLytic MT Cell Lysis buffer (C3228; Sigma-Aldrich, Saint Louis, MO, USA), and 50 μg of total proteins were separated in 10% Tris–HCl Criterion SDS-PAGE gel (345-0010; BioRad, Hercules, CA, USA). The protein level of IDH isoforms was determined by monoclonal antibodies produced in rabbit (IDH1-41-55, SAB1100101; Sigma-Aldrich, Steinheim, Germany; IDH2-31-45, SAB1100103; Sigma-Aldrich, Steinheim, Germany; IDH3A, ab58641; Abcam, Cambridge, MA, USA) and with anti-rabbit IgG-horseradish peroxidase conjugate secondary antibody produced in goat (170-6515; BioRad, Hercules, CA, USA). The protein level of tubulin was used as loading control (T6199; Sigma-Aldrich, Steinheim, Germany).
Cellular metabolites were extracted using cold methanol in phosphate buffer and chloroform. Cell pellets were suspended in 20 mM potassium phosphate buffer, pH 6.0. The cells were extracted by adding equal volumes of cold (253 K) absolute methanol (32213; Sigma-Aldrich, Saint Louis, MO, USA) and chloroform (319988; Sigma-Aldrich, Saint Louis, MO, USA). After rapid mixing, the tube was sonicated in an ice bath for 5 min. After centrifugation (18 000 g, 10 min), polar and aqueous fractions were collected. Samples were adjusted for the amount of total cellular protein (5000205; Quick Start Bradford, BioRad, Hercules, CA, USA), although the determination of relative stable isotope enrichment or carbon connectivity does not require this step. The cell extracts were dried by vacuum evaporation and then dissolved in 200 μl of 99.9% D2O with 0.75% 3-(trimethylsilyl)propionic-2,2,3,3-d4 acid (293040; Sigma-Aldrich, Saint Louis, MO, USA) or chloroform-d with 1.0% v/v TMS (151831; Sigma-Aldrich, Saint Louis, MO, USA) for chemical shift referencing. The extract was cleared by centrifugation (18 000 g, 10 min) and transferred into a 3-mm microtube (S3HT7; Norell, Landisville, NJ, USA) for determination of carbon-resolved stable isotope incorporation.
Glutamine concentration was determined in culture medium removed from cells after 18 h labeling in [U-13C] glutamine medium, and uptake was measured by comparison with glutamine content of unused medium (Scott et al., 2011).
Gas chromatography mass spectrometry (GCMS)
GCMS protocols, including cell extraction and metabolite derivatization methods, and methods for calculation of relative reductive flux into citrate and labeling of acetyl units in fatty acids (myristate and palmitate), were as previously described (Scott et al., 2011). Reductive carboxylation was measured as relative intensity of the M + 5 peak in citrate corrected for natural abundance.
Nuclear magnetic resonance (NMR) spectroscopy
NMR experiments were performed on a 600 MHz Bruker Avance spectrometer with a 5-mm TCI cryoprobe (Bruker-BioSpin, Karlsruhe, Germany) at 298.15 K. Positional 13C enrichments were calculated from the ratio of stable isotope incorporation to natural abundance background. Two-dimensional 1H, 13C J-HSQC spectra were recorded with an experimental time of 24 h, 5120 indirect 13C data points at 80 ppm 13C sweep width, 40 ppm carrier position, 4096 direct 1H increments, 2-s recycling delay, and eight scans. All processing and analysis components were scripted in AWK for the NMR software nmrview. Spectra were processed with the nmrpipe and nmrdraw suites. Before Fourier transformation, the data were multiplied by a squared sine-bell window function, phase-corrected, and zero-filled to 8192 data points. Relative labeling incorporation from stable isotope precursor was corrected for natural abundance 13C and averaged over all 1H, 13C detectable carbons of the observed metabolite. Specifically, those were carbon 2 of acetyl CoA, carbons 2 and 4 of citrate, carbons 1–3 of glycerol, and carbons 2–7, ω–2, ω–1, and ω of saturated fatty acids.
We thank Yongmei Feng for assistance with tissue culture. This work was supported by the National Institutes of Health Grants CA-154887 to F.V.F., CA-128814 and CA-140427 to J.W.S.