Epithelial-to-mesenchymal transition (EMT) is a fundamental developmental process with strong implications in cancer progression. Understanding the metabolic alterations associated with EMT may open new avenues of treatment and prevention. Here we used 13C carbon analogs of glucose and glutamine to examine differences in their utilization within central carbon and lipid metabolism following EMT in breast epithelial cell lines. We found that there are inherent differences in metabolic profiles before and after EMT. We observed EMT-dependent re-routing of the TCA-cycle, characterized by increased mitochondrial IDH2-mediated reductive carboxylation of glutamine to lipid biosynthesis with a concomitant lowering of glycolytic rates and glutamine-dependent glutathione (GSH) generation. Using weighted correlation network analysis, we identified cancer drugs whose efficacy against the NCI-60 Human Tumor Cell Line panel is significantly associated with GSH abundance and confirmed these in vitro. We report that EMT-linked alterations in GSH synthesis modulate the sensitivity of breast epithelial cells to mTOR inhibitors.

Implications:

EMT in breast cells causes an increased demand for glutamine for fatty acid biosynthesis, altering its contribution to glutathione biosynthesis, which sensitizes the cells to mTOR inhibitors.

This article is featured in Highlights of This Issue, p. 1439

Epithelial-to-mesenchymal transition (EMT) is a fundamental developmental process where tightly bound epithelial cells differentiate into migratory mesenchymal cells that can relocate into adjacent or distant tissues. This process is vital for tissue restructuring during embryonic development and is also necessary for proper wound healing in adult tissue. EMT has strong implications in cancer progression and metastasis where primary tumor cells of epithelial origin can take on a motile phenotype with the ability to migrate through the body and establish secondary tumors at distant locations (1).

Metabolic reprogramming is recognized as one of the 10 cancer hallmarks as proposed by Hanahan and Weinberg (2). In contrast to rapidly dividing cancer cells, a mesenchymal phenotype faces a different set of metabolic requirements whose relation to malignant transformation has been intensely studied and associated with enhanced glycolysis, increased glutaminolysis, nucleotide metabolism, and abnormal choline metabolism (3–5). Quantitative understanding of the metabolic requirements of mesenchymal cells is however lacking, particularly the changes to the turnover and quantity of metabolites involved in xenobiotic clearance, that is, the drug response of cells. Cancer cells that undergo EMT have increased resistance to various drugs (6–8), which indicates that the xenobiotic clearance of the cells is altered. There are three phases involved in the metabolism of xenobiotics: (i) modification, (ii) conjugation, and (iii) excretion. Conjugation involves the binding of particular metabolites (e.g., glutathione (GSH), UDP-glucuronate, PAPS, S-adenosylmethionine) to a xenobiotic compound (9), which leads to the assumption that the availability of these metabolites within cells influences the activity of the drugs. Therefore, accurate metabolic measurements of EMT may contribute to better understanding of the drug resistance of cancer cells and lead to novel therapeutic approaches aimed at eliminating metastatic cancer cells.

We have previously used both ultra-performance liquid chromatography coupled mass spectrometry (UPLC-MS) and NMR to study EMT and cancer metabolism (10–12). Integrated analyses of these metabolomics data with transcriptomic and proteomic data within genome-scale metabolic models predicted metabolic differences that occur following EMT in breast epithelium (12). These included alterations to glycolysis, the pentose phosphate pathway, TCA cycle, and fatty acid synthesis. Although these models provided useful insights into metabolic alterations associated with EMT, they lacked accuracy in predicting internal fluxes in a quantitative manner in the compartmentalized central carbon metabolism.

To better understand the metabolic changes that accompany EMT, we characterized the internal flow of metabolites in D492 breast epithelial cells and their mesenchymal variant, D492M, to determine metabolic changes within central carbon metabolism following EMT in breast epithelial cells. We performed stable isotope tracing of 13C labeled glucose and two separate 13C labeled glutamine analogs. UPLC-MS and NMR were used to measure label incorporation into metabolites associated with central carbon metabolism and lipid biosynthesis. We subsequently performed shRNA lentiviral silencing of key genes to further elucidate their role in EMT metabolic re-programming. Finally, using an integrated network analysis of the NCI-60 Human Tumor Cell Line panel and an untargeted metabolomic analysis, we investigate how the EMT-dependent re-routing of central carbon metabolism affects drug responsiveness in D492 and D492M cells.

Cell culture

D492 and D492M cells were kindly provided by the Stem Cell Research Unit, University of Iceland, and were cultured in DMEM/F12-based medium H14 at 37°C in 5% CO2 as described previously (13). All experiments were performed within four passages from thawing, within the range of 30 to 40 passages in total. For the labeling experiments, the cells were fed with medium containing 100% 13C-labeled glutamine at the one or five position (Cambridge Isotope Laboratories, Inc.) or 13C-labeled glucose at the one and two positions (Cambridge Isotope Laboratories, Inc.). Cells were screened for Mycoplasma infections every month using PCR-based tests at the Biomedical Center, University of Iceland.

Lentiviral shRNA production and transduction

HEK293T cells were transfected using TurboFect transfection reagent (Thermo Fisher Scientific) at 80% confluency in T25 cell culture vessels. The cells were then incubated at 37°C and 5% CO2. Viral supernatant was collected at two timepoints, the first being after 48 hours in culture, and the second 72 hours after changing medium at the first timepoint. The viral supernatant was filtered through a 0.45 μm filter using a syringe and stored at −20°C until usage. The lentiviral vectors were acquired from GeneCopoeia. They contained an shRNA construct for the selective targeting of IDH2. The construct was based on a psi-LVRH1MH vector with an mCherry fluorescent reporter, resistance against hygromycin B, and the identical hairpin sequence TCAAGAG. The target sequence was IDH2 5′GTACAAGGCCACAGACTTTGT-3′. The D492 and D492M cell lines were transduced using 1 mL of filtered viral supernatant at 70% confluency and incubated at 37°C and 5% CO2 for 24 hours, at which timepoint the medium was changed to fresh H14 medium. After further 48 hours, the cells were grown in medium containing hygromycin B (200 μg/mL) for 3 weeks to selectively grow cells containing the shRNA construct.

Real-time PCR

Whole-cell RNA was extracted using Tri-Reagent (Thermo Fisher Scientific, AM9738). Reverse transcription was performed using High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific, 4368814). The expression of the genes IDH1 and IDH2 was quantified, where ACTB (Beta-actin) and was used as an endogenous reference gene. The primers for IDH1, IDH2, and ACTB were designed using the Primer3 software in the Benchling website (https://benchling.com). Primers were designed to span exon junctions and have a melting temperature above 55°C. The expression of IDH2 and IDH1 was assessed using real-time PCR (qPCR). Real-time quantitative PCR reactions were carried out using Luna Universal qPCR Master Mix (New England Biolabs) according to manufacturer's instructions on a Bio-Rad CFX384 Touch Real Time System (Bio-Rad). Gene expression levels were determined using CFX Manager Software (Bio-Rad) and differences in relative expression were estimated with the 2ΔΔCt method. The primer sequences used for quantifying the gene expression were: IDH1-forward 5′-CGACATGGTGGCCCAAGCTATG-3′, IDH1-reverse 5′-TCATGCCGAGAGAGCCATACCC-3′, IDH2-forward 5′-ATGAGGCCCGTGTGGAAGAGTT-3′, IDH2-reverse 5′-CAGATGATGGGCTCCCGGAAGA-3′, ACTB-forward 5′-CTTCCTGGGTGAGTGGAGACTG-3′, and ACTB-reverse 5′-GAGGGAAATGAGGGCAGGACTT-3′.

Proliferation assay

Cells were seeded in quadruplicates in 48-well plates (10,000 cells/well). They were grown in a large chamber incubation system (PeCon GmbH) at 37°C in 5% CO2 and imaged for 12 to 72 hours using Leica DMI6000B. Images of cells were opened with Fiji (14), where the cells were counted with the help of an in-house script.

Detection of intracellular NADP+ and NADPH

NADP+ and NADPH were measured using NADP/NADPH-Glo Assay (G9081; Promega). Cells were seeded in triplicates in opaque 96-well plates (10,000 cells/well) and incubated at 37°C in 5% CO2. After 24 hours, the medium was removed, cells were washed with cold PBS, and then supplemented with 50 μL PBS and 50 μL 1% DTAB in 0.2N NaOH solution to induce cell lysis. Next steps were according to manufacturer's protocol. The luminescence was measured 50 minutes after addition of the NADP-NADPH-Glo Detection reagent in SpectraMaxM3 Microplate Reader (Molecular Devices).

Nuclear magnetic resonance (NMR)

For NMR analysis, D492 and D492M cells were cultured in T225 flasks in supplemented DMEM/F12 until they reached approximately 70% confluency. Cells were then fed with either 1,2–13C glucose or 1–13C glutamine for 6 hours. Parallels without 13C tracers were also cultured. Culture medium was collected after incubation. Methanol extracts from glucose- and glutamine-labeled cells were prepared as described previously (15). The cell extracts were freeze dried prior to NMR analysis. For NMR, freeze-dried cell extracts were dissolved in 600 μL D2O in PBS whereas culture medium (500 μL) was diluted with in D2O-based PBS (100 μL). NMR analysis was performed using a 600 MHz Bruker Avance III NMR spectrometer (Bruker Biospin GmbH), equipped with a 5 mm QCI Cryoprobe with integrated, cooled preamplifiers for 1H, 2H, and 13C. Proton spectra were acquired at 300 K using 1D NOESY (Bruker: noesygppr1d) with presaturation and spoiler gradients as described previously (16). The spectra were collected with 32 scans and 4 dummy scans. The acquisition time was 2.73 seconds and relaxation delay 4 seconds, measuring the FID via collection of 64 K complex data points. The 1H spectra were Fourier transformed with a 0.3 Hz exponential line broadening and the chemical shift was calibrated to alanine at 1.48 ppm. 1H spectra from D492 (n = 5) and D492M (n = 6) cells were transferred to MATLAB R2017a for multivariate data analysis. The spectra were baseline corrected using asymmetric least squares method (17) and peak aligned using icoshift (18). The water peak and areas in the spectra with contamination and noise only were removed. All spectra were mean normalized and mean centered. Principal component analysis (PCA) was performed using PLS toolbox v8.2.1 (Eigenvector Research). Proton decoupled 13C spectra (Bruker: zgpg30) were acquired using a power gated decoupling sequence with a 30° pulse angle as described in Bettum and colleagues (19). The spectra were collected with either 4 K (for 1,2–13C-glucose) or 16 K (for 1–13C-glutamine) scans and 16 dummy scans. The acquisition time was 1.65 seconds, relaxation delay 0.5 seconds, measuring the FID via collection of 96 K complex data points over a sweep width of 197.175 ppm. The 13C spectra were Fourier transformed with a 3.0 Hz exponential line broadening and the chemical shift was calibrated to the 3–13C-alanine peak at 19.0 ppm or 1–13C-glutamine peak at 176.4 ppm. 13C-labeled metabolites downstream from the tracers were identified by comparing 13C spectra with natural abundance spectra acquired under the same conditions. Levels of selected metabolites in the extracts were semiquantitatively assessed by integration of resonance signals using TopSpin 4.0.8 (Bruker Biospin GmbH) after correcting for natural abundance levels. The 13C spectra were normalized to the total AUC in the 1H spectra acquired from the same sample.

Metabolomics

Sample extraction

Polar and nonpolar metabolites were extracted from cell cultures by methanol/chloroform/water extraction. Cells were harvested in ice-cold methanol, vortexed vigorously, and let stand on ice for 10 minutes. Equal amounts of water and chloroform were added to a final composition of 1:1:1 (CH3OH:H2O:CHCl3), vortexed, and left to stand overnight at 4°C. The organic phase (lipids) was collected into a glass vial and dried in a stream of N2 and stored under N2 at −80°C until analysis. The aqueous phase (polar metabolites) was stored at −80°C and evaporated in a miVac concentrator (SP Scientific) before analysis.

UPLC-MS

Before UPLC-MS analysis, the organic phase was reconstituted in MTBE before a methanol solution containing 1M NaOH was added (10:1 v/v, respectively). This was incubated for 3.5 hours at 37°C, when 1 μL formic acid was added (to neutralize the solution), the samples were dried in a stream of N2 and then resuspended in isopropanol:ACN:H2O (6:9:1, v/v/v). The aqueous phase (metabolites) were reconstituted in isopropanol:ACN:H2O (2:1:1, v/v/v). UPLC (Acquity) was coupled with a quadrupole-time of flight mass spectrometer (Synapt G2; Waters). For the lipid samples, chromatographic separation was achieved as described previously (20). For the metabolomic samples, chromatographic separation was achieved by hydrophilic interaction liquid chromatography using an Acquity amide column, 1.7 μm (2.1 × 150 mm; Waters). All samples were analyzed in positive ionization and negative ionization mode using acidic and basic chromatographic conditions. In positive mode and in negative acidic conditions, mobile phase A was 100% ACN and B was 100% H2O both containing 0.1% formic acid. The following elution gradient was used: 0 minutes 99% A; 7 minutes 30% A; 7.1 minutes 99% A; 10 minutes 99% A. In negative mode basic conditions, mobile phase A contained ACN:sodium bicarbonate 10 mmol/L (95:5) and mobile phase B contained ACN:sodium bicarbonate 10 mmol/L (5:95). The following elution gradient was used: 0 minute 99% A; 6 minutes 30% A; 6.5 minutes 99% A; 10 minutes 99% A. In all conditions, the flow rate was 0.4 mL/min, the column temperature was 45°C, and the injection volume was 3.5 μL. The mass spectrometer was operated using a capillary voltage of 1.5 kV, the sampling cone and the extraction cone were of 30 and 5 V. The cone and the desolvation gas flow were 50 and 800 L/h, whereas the source and desolvation gas temperature were 120°C and 500°C. MS spectra were acquired in centroid mode from m/z 50 to 1,000 using scan time of 0.3 seconds. Leucine enkephalin (2 ng/μL) was used as lock mass (m/z 556.2771 and 554.2615 in positive and negative experiments, respectively). A typical analytical block consisted of: (i) pooled QC samples to equilibrate the system, (ii) calibrators, (iii) samples and spiked pooled QC samples, and (iv) calibrators.

Data analysis

TargetLynx (v4.1; Waters) was used to integrate chromatograms of all isotopologues of the metabolites of interest. Ion chromatograms were extracted using a window of 0.02 mDa, which was centered on the expected m/z for each targeted isotopologue. The output was a mass distribution vector (MDV) describing the relative amount of each detected isotopologue of the metabolite. Ion chromatograms of isotopologues of interest extracted and corrected for abundance of naturally occurring isotopes using the IsoCor software (21). When calculating the total contribution (TC) of carbon sources to metabolites, we used the following equation (22):

formula

where n is the number of C atoms in the metabolite, i represents the isotopologues, and m is the relative fraction of the isotopologues.

To evaluate the percentage of glucose that enters the oxidative part of the pentose phoshate pathway, and re-enters glycolysis, we utilized a formula from Lee and colleagues (23):

formula

In Equation (2), m1 and m2 are the fractional abundances of M+1 and M+2 lactate isotopologues, respectively (e.g., from Supplementary Fig. S2).

RNA sequencing

Quantified transcript pseudocounts from kallisto (24) were obtained for D492 and D492M in triplicates from Briem and colleagues (from the authors; ref. 25). These data were imported into R and simultaneously log2-transformed and variance-stabilized using DESeq2′s rlog function (26).

Proteomic analysis

A proteomic dataset for the D492 and D492M cells was obtained from the ProteomeXchange Consortium via the PRIDE (27) partner repository with the dataset identifier PXD024164. The raw data were processed using MaxQuant (28) for both the protein identification and quantification.

Western blot analysis

D492 and D492M cells were grown to 80% to 90% confluent as described above followed by lysis in RIPA buffer. The lysates were subjected to five freeze–thaw cycles, centrifuged at 14,000 RCF for 20 minutes at 4°C. Protein concentration was quantified using a Pierce BCA Protein Assay Kit (Thermo Fisher Scientific).

10 to 20 μg of protein were loaded onto precast 4% to 12% NuPAGE Bis-Tris gels (Thermo Fisher Scientific) and transferred to a 0.45 μm nitrocellulose membrane (Thermo Fisher Scientific). The membrane was blocked in 5% BSA (Thermo Fisher Scientific) for 60 minutes, followed by overnight primary antibody incubation at 4°C. The primary antibody was anti-IDH2 monoclonal rabbit (12652; Cell Signaling Technologies) in 1:1,000 dilution. IDH2 levels were normalized against β-actin (MA5–15739; Thermo Fisher Scientific) in 1:10,000 dilution.

Bands were detected by secondary antibody incubation for 2 hours at room temperature using anti-rabbit IgG (H+L) DyLight 800 4× PEG Conjugate and anti-mouse IgG (H+L) DyLight 680 Conjugate (Cell Signaling Technologies) in 1:15,000 dilution. Imaging was performed using Odyssey CLx (LI-COR Biosciences) to scan the films at 169 μm resolution. The results were analyzed in the Image Lab software (Bio-Rad).

Measurement of metabolite exchange rates

Glucose and lactate measurements were performed in an ABL 90 blood gas analyzer (Radiometer). Glutamine uptake was measured using L-Glutamine/Ammonia Assay Kit (K-GLNAM; Megazyme). The following formula was used to quantify the exchange rates of metabolites in the cells:

formula

where |{v_k}$| is the exchange rate for metabolite k, |{[ {{M_k}} ]_{48}}$| and |{[ {{M_k}} ]_0}$| are the concentrations of metabolite k in the culture media after 48 and 0 hours, respectively, and A is the area under the growth curve.

Weighted correlation network analysis

Drug sensitivity data were gathered from the NCI-60 Human Tumor Cell Lines (29) using the rcellminer R-package (30). We focused specifically on FDA-approved drugs, so after filtering the data (using WGCNA‘s goodSamplesGenes function; ref. 31), the final size of the drug sensitivity matrix was 214 drugs × 59 cell lines (one cell line, MDA-N, was removed due to missing data). To construct a network of drug-sensitivities, a weighted correlation network analysis was employed using the WGCNA R-package (31). A soft threshold value of 9 was used to obtain a scale-free network topology (R2 = 0.85). Highly correlated modules that had an average distance <0.65 were merged. To associate the modules to a specific mechanism of action (MOA), the MOA terms for each drug were obtained (from rcellminer) and tested for overrepresentation in the modules. MOA with Bonferroni-adjusted P-value <0.05 were labeled as overrepresented and used for the functional annotation of the drug modules.

The eigenvalues (first principal components) for each module were identified and used to test association of drug-modules with metabolite levels. The correlation of the drug modules to metabolite levels of the cells within the NCI-60 (from ref. 32) was calculated. The R-package igraph (33) was used to visualize the drug correlation network using the Fruchterman-Reingold (34) force-directed layout. An R-script for the network analysis is in Supplementary File S1.

Drug treatment assays

Cells were seeded (3,000 cells/well) in white Costar 96-well plates (Corning) and maintained at 37°C in 5% CO2. After 24 hours, drugs were added with or without buthionine sulphoximine (BSO; B2515; Sigma Aldrich) where DMSO was used as a control. Then, 72 hours later cell viability was evaluated using the CellTiter-Glo (CTG) assay (Promega), by adding CTG assay mix directly into the wells in a 1:1 ratio. After 10 minutes, luminescence was measured by Victor X3 Multiplate reader (Perkin Elmer). The drugs tested were mTOR inhibitor everolimus (Sigma-Aldrich) and taxane drug paclitaxel (Fresenius Kabi).

Detection of intracellular GSH abundance

Cells were seeded (2,000 cells/well) in white 384-well plates (Greiner Bio-One) and maintained at 37°C in 5% CO2. After 24 hours, BSO (treated) or medium (nontreated) was added. After additional 24 (and 48 for nontreated) hours, GSH measurement was performed by using GSH-Glo Glutathione Assay Kit (V6911; Promega) in accordance with the manufacturer's protocol.

Statistical analysis

Student t test was employed for comparison of two treatments. Benjamini–Hochberg adjustment for multiple comparisons was performed when appropriate. ANOVA was used to compare data from three or more treatments or the simultaneous evaluation of the effect of two grouping variables. The asterisks in each figure represent the P values (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; ns, not significant). Data were assumed to be normally distributed. Statistical analysis and image generation was carried out in the R environment (35) using the ggplot2 (36) and ggpubr (37) packages. In our graphs, all data points are plotted and summarized using mean and SE.

Glycolysis rates determine the pentose phosphate shunt in the D492 EMT cell model

A clear difference in the overall metabolic profiles of D492 and D492M cells was confirmed by PCA of their 1H NMR spectra. The score and loading plots from PCA (Supplementary Figs. S1A and S1B) indicated that D492 cells had more intracellular isoleucine, leucine, valine, alanine, arginine, GSH, myo-inositol, asparagine, proline, AMP, ADP, ATP, tyrosine, phenylalanine, and NAD+, and less glutamine, glutamate, phosphocholine, glycine, threonine, glucose, fumarate, NADP, and NADH. The rate of glucose uptake and lactate secretion was also higher in the epithelial phenotype of D492 cells compared with the mesenchymal phenotype (Fig. 1B), indicative of enhanced glycolysis in D492. To determine pentose phosphate pathway (PPP) split ratios in the cell lines, we used 1,2–13C glucose (Fig. 1A), as described previously (38). Label contribution of 1,2–13C-glucose to lactate after 6 hours was higher in D492 epithelial cells than in D492M mesenchymal cells (Fig. 1C), confirming enhanced glycolysis. To determine the differences in PPP activity, we calculated the percentage of glucose diverted into the PPP using measured lactate isotopologue abundances (Supplementary Fig. S2) and Equation (2). Roughly 2% of glucose was found to enter the PPP cycle in both cell lines. Therefore, because of the overall higher glucose uptake, the flux into oxidative PPP is higher in D492. As a result, reduction of NADP to NADPH via the oxidative phase of the PPP in D492M cells is dampened.

Figure 1.

Glucose metabolism of D492 and D492M. A, A schematic overview of label distribution from 1,2–13C glucose into central carbon metabolism. An atom transition map of glucose metabolism showing a part of the metabolic fates of 1,2–13C glucose within mammalian cells, where the 13C-isotopes are shown in black. Dashed lines indicate more than one reaction between metabolites. B, Measured glucose uptake and lactate secretion in D492 and D492M. Total contribution of glucose to (C) lactate, (D) citrate, and (E) palmitate was measured after culturing of D492 and D492M with 1,2–13C-glucose for 6 hours. Metabolites: f6p, fructose 6-phosphate; g3p, glyceraldehyde 3-phosphate; pyr, pyruvate; co2, carbon dioxide; accoa, acetyl CoA; oac, oxaloacetate; αkg, α-ketoglutarate; suc, succinate; fum, fumarate; mal, malate; asp, aspartate. Enzymes: CS, citrate synthase; ACLY, ATP-citrate lyase. Pathways: PPP, pentose phosphate pathway.

Figure 1.

Glucose metabolism of D492 and D492M. A, A schematic overview of label distribution from 1,2–13C glucose into central carbon metabolism. An atom transition map of glucose metabolism showing a part of the metabolic fates of 1,2–13C glucose within mammalian cells, where the 13C-isotopes are shown in black. Dashed lines indicate more than one reaction between metabolites. B, Measured glucose uptake and lactate secretion in D492 and D492M. Total contribution of glucose to (C) lactate, (D) citrate, and (E) palmitate was measured after culturing of D492 and D492M with 1,2–13C-glucose for 6 hours. Metabolites: f6p, fructose 6-phosphate; g3p, glyceraldehyde 3-phosphate; pyr, pyruvate; co2, carbon dioxide; accoa, acetyl CoA; oac, oxaloacetate; αkg, α-ketoglutarate; suc, succinate; fum, fumarate; mal, malate; asp, aspartate. Enzymes: CS, citrate synthase; ACLY, ATP-citrate lyase. Pathways: PPP, pentose phosphate pathway.

Close modal

To examine the contribution of glucose to the TCA cycle, we measured the contribution of the labeled 1,2–13C-glucose to citrate. Citrate is either oxidized for energy production in the TCA cycle or used as a precursor for lipid biosynthesis (39) through ATP-citrate lyase (ACLY) in the cytosol (Fig. 1A). No difference in glucose-dependent citrate generation (Fig. 1D) was observed, which was reflected by the contribution of glucose to palmitate (Fig. 1E). Furthermore, the contribution of glucose to other TCA cycle components and downstream metabolites (malate, aspartate, and glutamate) after 6 hours in culture was minimal (Supplementary Fig. S2). Thus, the only difference in glucose utilization within central carbon metabolism between D492 and D492M was the increased glycolytic activity of the former.

Glutamine fuels citrate and lipogenic acetyl-CoA production via reductive carboxylation following EMT in D492 cells

Glutamine is a major contributor of carbons into the TCA cycle through anaplerosis (i.e., glutaminolysis), particularly in cancer cells (40–42). Glutamine is the second-highest consumed carbon source in D492 and D492M (after glucose; ref. 12) with an average uptake rate of around 60 fmoles/cell/hour in both cell lines (Fig. 2B). We found that both glucose and glutamine are essential for the proliferation of D492 and D492M (Supplementary Fig. S3). Glutamine can replenish the TCA cycle via glutamate and α-ketoglutarate, which can be metabolized within the TCA cycle both oxidatively and reductively (Fig. 2A). To discriminate oxidative and reductive TCA cycle carbon flow, D492 and D492M cells were cultured with isotopic glutamine analogs labeled at either the one or five positions. These isotopomers of glutamine allow the specific quantification of glutamine to citrate either reductively (1–13C-glutamine) or oxidatively (5–13C-glutamine; Fig. 2A). The 5–13C-glutamine analog can additionally quantify the contribution of glutamine to fatty acids solely through reductive carboxylation (Fig. 2A). Importantly, intracellular glutamate label from a labeled glutamine source is a direct measurement of the contribution of glutamine to the glutamate pool. Compared with D492, the glutamate pool in the D492M cells was not as dependent on glutamine as observed from lower labeling incorporation from both 1- and 5–13C glutamine (Supplementary Figs. S4A and S4B). This was supported by NMR measurements (Supplementary Fig. S1C) and is likely due to higher amounts of glutamate being derived from elsewhere (e.g., transamination reactions and protein catabolism) in D492M. As we were specifically interested in the contribution of glutamine to metabolites beyond glutamate (e.g., citrate), we accounted for the differences in glutamine-to-glutamate labeling by dividing the total contribution from glutamine to the target metabolites with the total contribution of glutamine to glutamate in each cell line. The results therefore represent the relative contribution of glutamate to metabolites.

Figure 2.

Glutamine metabolism of D492 and D492M. A, Atom transition map of glutamine metabolism showing the different metabolic fates of 1–13C glutamine (grey) and 5–13C glutamine (black) within mammalian cells. Dashed lines indicate more than one reaction between metabolites. B, Measured glutamine uptake in D492 and D492M. C, Contribution of 1- and 5–13C-glutamine to citrate in D492 and D492M cells after 6 hours in culture. The total contribution of the glutamine analogs to citrate was normalized to the different origins of glutamate in the cells (total contribution of glutamine to glutamate). D, Total contribution of 5–13C-glutamine to palmitate after 6 hours in culture, normalized to glutamate origin. Metabolites: co2, carbon dioxide; accoa, acetyl CoA; oac, oxaloacetate; αkg, α-ketoglutarate; suc, succinate; mal, malate; asp, aspartate. Enzymes: ME, malic enzyme.

Figure 2.

Glutamine metabolism of D492 and D492M. A, Atom transition map of glutamine metabolism showing the different metabolic fates of 1–13C glutamine (grey) and 5–13C glutamine (black) within mammalian cells. Dashed lines indicate more than one reaction between metabolites. B, Measured glutamine uptake in D492 and D492M. C, Contribution of 1- and 5–13C-glutamine to citrate in D492 and D492M cells after 6 hours in culture. The total contribution of the glutamine analogs to citrate was normalized to the different origins of glutamate in the cells (total contribution of glutamine to glutamate). D, Total contribution of 5–13C-glutamine to palmitate after 6 hours in culture, normalized to glutamate origin. Metabolites: co2, carbon dioxide; accoa, acetyl CoA; oac, oxaloacetate; αkg, α-ketoglutarate; suc, succinate; mal, malate; asp, aspartate. Enzymes: ME, malic enzyme.

Close modal

Approximately 60% of the citrate pool was derived from glutamate in both D492 and D492M as shown by the isotope enrichment from 5–13C glutamine (Fig. 2C). Enrichment of 13C in citrate derived from 1–13C glutamine was however higher in D492M (Fig. 2C), supporting increased reductive carboxylation. These results were mirrored in the 5–13C-glutamine-dependent labeling profiles of palmitate, where there was a six-fold increase in reductive contribution of glutamine in D492M cells (Fig. 2D).

Metabolic re-routing following EMT affects redox metabolism in D492 cells

The reductive pathway of glutamine to citrate is typically activated in hypoxia or following changes in electron transport chain activity (41, 43, 44). Reductive glutaminolysis affects the redox status of cells through NADPH which serves as a cofactor for the reversible isocitrate dehydrogenase enzymes, IDH1 and IDH2 (45). Interestingly, we observed an increased NADPH/NADP ratio in D492M over D492 (Fig. 3A), which is not compatible with the PPP fluxes in the cells. However, even though reductive carboxylation is a NADPH-requiring process (Fig. 3B), studies have shown that its heightened activity may lead to increased cytosolic or mitochondrial NADPH levels based on the coordination of different isoforms of isocitrate dehydrogenase (45, 46). NADPH is known for its role in defenses against reactive oxygen species (ROS), where it is used to reduce the oxidized form of glutathione (GSSG) to generate the reduced form of GSH. Due to the observed alterations in glycolysis, glutamine metabolism, and NADPH/NADP+ ratios, we hypothesized that this would be reflected in ROS generation within the cells. This in turn would translate to alterations in GSH metabolism.

Figure 3.

Redox metabolism is altered following EMT of D492. A, NADPH-to-NADP ratio in D492 and D492M cells. B, A schematic showing the connection of mitochondrial and cytosolic reductive carboxylation, and NADP+/NADPH balance. C, Total contribution of glutamate to oxidized glutathione (GSSG) in D492 and D492M cells after 6 hours in culture. D, Measured abundance of GSH in D492 and D492M after 24 and 48 hours in culture. A two-way ANOVA test revealed a significant difference in GSH levels between the cells, independent of time. E, The two reactions needed to convert glutamate into proline. Both reactions oxidize NADPH. P5CS: Delta-1-pyrroline-5-carboxylate synthase, PYCR, pyrroline-5-carboxylate reductase. F, Total contribution of glutamate to proline in D492 and D492M cells after 6 hours in culture. The results from B and D are from the combined analysis of 1- and 5–13C-glutamine results, since both 13C carbons are detected in the proline and GSSG carbon skeletons.

Figure 3.

Redox metabolism is altered following EMT of D492. A, NADPH-to-NADP ratio in D492 and D492M cells. B, A schematic showing the connection of mitochondrial and cytosolic reductive carboxylation, and NADP+/NADPH balance. C, Total contribution of glutamate to oxidized glutathione (GSSG) in D492 and D492M cells after 6 hours in culture. D, Measured abundance of GSH in D492 and D492M after 24 and 48 hours in culture. A two-way ANOVA test revealed a significant difference in GSH levels between the cells, independent of time. E, The two reactions needed to convert glutamate into proline. Both reactions oxidize NADPH. P5CS: Delta-1-pyrroline-5-carboxylate synthase, PYCR, pyrroline-5-carboxylate reductase. F, Total contribution of glutamate to proline in D492 and D492M cells after 6 hours in culture. The results from B and D are from the combined analysis of 1- and 5–13C-glutamine results, since both 13C carbons are detected in the proline and GSSG carbon skeletons.

Close modal

Label incorporation from glutamine to GSH was a significantly lower following EMT in D492 cells as measured by UPLC-MS (Fig. 3C) and NMR (Supplementary Fig. S1C), further supporting EMT-associated alterations in redox metabolism. In addition, intracellular GSH concentration was 33% lower in D492M cells on average (Fig. 3D).

Proline has been shown to be important to maintain redox homeostasis, by recycling NADPH to NADP+ (47), coupling it to NADPH-generating pathways. Proline is synthesized from glutamate in two reactions, both of which oxidize NADPH (Fig. 3E). We traced the proline synthesis from glutamate in D492 and D492M cells and observed a 2-fold increase in proline synthesis from glutamate following EMT (Fig. 3F). In summary, differences in the redox state of the D492 EMT model can be related to decreased glycolytic/PPP flux and concomitant changes to glutaminolysis, which is defined by decreased GSH synthesis, increased proline synthesis, and increased reliance on reductive carboxylation for citrate synthesis.

Mitochondrial isocitrate dehydrogenase 2 (IDH2) is essential for EMT-linked reductive glutamine metabolism

On the basis of the difference in reductive carboxylation and NADPH/NADP+ ratio (Figs. 2C and 3A, respectively) and recent literature (45, 48), we hypothesized that IDH would contribute to the discrimination between the D492 and D492M metabolic phenotypes (i.e., metabotypes) through glutamine consumption and influence redox balance.

Two isoforms of IDH are known to use NADP+ as a cofactor: cytosolic IDH1 and mitochondrial IDH2. Quantification of IDH mRNA (by RNA sequencing) and protein (by MS proteomics) in D492 and D492M cells revealed increased IDH2 levels in D492M as compared with D492 (Fig. 4A). These findings were confirmed by Western blot analysis (Supplementary Fig. S5). No difference was observed in IDH1 levels (Fig. 4A). Using shRNA lentiviral transduction, we knocked down IDH2 expression in both cell lines and investigated the metabolic and morphologic effects.

Figure 4.

IDH2 mediates reductive carboxylation activity and is coupled to redox metabolism of D492M. A, Relative differences in gene expression and protein translation of IDH1 and IDH2 in D492 and D492M. The results are displayed as log-fold D492M/D492 ratio of abundance of IDH1/2 transcripts from RNA sequencing, and IDH1/2 protein from a proteomic analysis of D492 and D492M. Results are depicted as mean + standard error (n = 3). B, qPCR from D492, D492M and an IDH2 -silenced D492M cell line showing the gene expression levels of IDH2 and IDH1. C, Phase-contrast images of D492M-WT and D492M-IDH2 cells. D, Proliferation of D492M-WT and D492M-IDH2 cell lines (mean + SE, n = 8). E, Effect of IDH2 silencing on the contribution of 1–13C-glutamine, 5–13C-glutamine and 1,2–13C-glucose to citrate in D492M, where the former two were normalized to glutamate origin. F, Effect of IDH2 silencing on the contribution of 5–13C-glutamine (normalized to glutamate origin) and 1,2–13C-glucose to palmitate in D492M. G, Effect of IDH2 silencing on the NADPH/NADP+ ratio in D492M. H, Effect of IDH2 silencing on the contribution of 1–13C-glutamine to proline (normalized to glutamate origin). I, Effect of IDH2 silencing on the contribution of 1–13C-glutamine and 5–13C-glutamine to oxidized GSH (normalized to glutamate origin). E,F,H, and I are from cells cultured with 13C-labeled carbon sources for 6 hours. Student two-tailed t test (with Benjamini–Hochberg adjustment for multiple comparisons) was used to estimate significance of results.

Figure 4.

IDH2 mediates reductive carboxylation activity and is coupled to redox metabolism of D492M. A, Relative differences in gene expression and protein translation of IDH1 and IDH2 in D492 and D492M. The results are displayed as log-fold D492M/D492 ratio of abundance of IDH1/2 transcripts from RNA sequencing, and IDH1/2 protein from a proteomic analysis of D492 and D492M. Results are depicted as mean + standard error (n = 3). B, qPCR from D492, D492M and an IDH2 -silenced D492M cell line showing the gene expression levels of IDH2 and IDH1. C, Phase-contrast images of D492M-WT and D492M-IDH2 cells. D, Proliferation of D492M-WT and D492M-IDH2 cell lines (mean + SE, n = 8). E, Effect of IDH2 silencing on the contribution of 1–13C-glutamine, 5–13C-glutamine and 1,2–13C-glucose to citrate in D492M, where the former two were normalized to glutamate origin. F, Effect of IDH2 silencing on the contribution of 5–13C-glutamine (normalized to glutamate origin) and 1,2–13C-glucose to palmitate in D492M. G, Effect of IDH2 silencing on the NADPH/NADP+ ratio in D492M. H, Effect of IDH2 silencing on the contribution of 1–13C-glutamine to proline (normalized to glutamate origin). I, Effect of IDH2 silencing on the contribution of 1–13C-glutamine and 5–13C-glutamine to oxidized GSH (normalized to glutamate origin). E,F,H, and I are from cells cultured with 13C-labeled carbon sources for 6 hours. Student two-tailed t test (with Benjamini–Hochberg adjustment for multiple comparisons) was used to estimate significance of results.

Close modal

Following a significant reduction in IDH2 expression levels, the mRNA levels of the cytosolic isoform IDH1 did not change in the D492M-IDH2 cell line (Fig. 4B). There were no differences observed in neither morphology nor growth rate upon IDH2 knockdown in D492M (Fig. 4C and D). A significant reduction in reductive carboxylation activity was observed, as indicated by the 1–13C-glutamine contribution to citrate and 5–13C-glutamine contribution to palmitate (Fig. 4E and F, respectively). No significant difference was observed in the overall contribution of glutamine to citrate or glucose to citrate (Fig. 4E). The NADPH/NADP+ ratio was lowered upon IDH2 knockdown and proline synthesis from glutamate was reduced (Fig. 4G and H, respectively), suggesting that the redox homeostasis of these cells is coupled to IDH2-mediated reductive carboxylation. The glutamine-dependent synthesis of GSH, however, was not affected by the IDH2 knockdown (Fig. 4I). In contrast to D492M cells, D492 cells increased the expression of the cytosolic isoform IDH1 upon IDH2 knockdown (Supplementary Fig. S6A). The upregulation of IDH1 was accompanied by increased glutamate-to-lipid contribution via reductive carboxylation (Supplementary Fig. S6B) and decreased growth rate (Supplementary Fig. S6C). In addition, we observed a shift in NADPH-to-NADP+ ratio, increased proline synthesis from glutamate, and reduced GSH synthesis in the D492-IDH2 knockdown cells compared with the wild-type D492 cells (Supplementary Figs. S6D–S6F). These results indicate that the epithelial phenotype, but not the mesenchymal phenotype of D492 cells, can compensate for the IDH2 knockdown by increasing IDH1 expression.

Alteration in GSH biosynthesis drives sensitivity to mTOR inhibition

The different metabotypes of the cell lines, characterized by different glycolytic rates, altered carbon source preference for TCA cycle activity and changes to the synthesis of proline and GSH are reminiscent of cancer stem cell metabotypes (49, 50). Because of GSH's role in drug resistance in various cancer cell types (51–53), we focused our attention on the differences in GSH metabolism between D492 and D492M cells (Fig. 3). We hypothesized that metabolic rerouting of glutamine-derived glutamate for GSH synthesis would affect drug sensitivity in these cells.

To identify drugs that are selectively affected by GSH concentrations within cells, we performed an integrated network analysis of (i) drug sensitivity profiles within the NCI-60 Human Tumor Cell Line database (29) and (ii) untargeted metabolomic analysis of NCI-60 cell lines from Ortmayr and colleagues (32). The network analysis revealed that 214 FDA-approved drugs in the NCI-60 database were grouped into eight intracorrelated drug modules (Fig. 5A), whose functional annotation showed that drugs grouped together according to their mechanism of action. The modules were comprised of DNA-damaging agents and cell-cycle arresting compounds (e.g., alkylating agents, nucleotide analogs, and paclitaxel, n = 122), hormones (n = 7), tyrosine kinase inhibitors targeting EGFR and ERBB2 (e.g., erlotinib and lapatinib, n = 21), mTOR and serine/threonine kinase inhibitors (e.g., everolimus, rapamycin, temsirolimus, n = 8), ALK/CDK inhibitors (n = 17), tyrosine kinases targeting VEGFR, PDGFR, and FGFR (n = 7), MAP kinase inhibitors n = 10), and nonspecific drugs (n = 6).

Figure 5.

GSH levels regulate sensitivity to mTOR inhibitors. A, Network analysis of NCI-60 cell lines treated with various FDA-approved drugs (n = 214) suggests the presence of 8 modules of intra-correlated drugs. The efficacy of each individual drug was correlated with GSH levels, represented by node size. B, The correlation of drug modules' eigenvalues to reduced GSH, oxidized glutathione (GSSG), UDP-glucuronate and S-adenosylmethionine (SAM). Upper numbers in table represent Pearson correlation coefficient and the lower numbers (in brackets) represent the correlation p-value. C, D492 and D492M cells treated with increasing concentrations of the mTOR inhibitor everolimus and taxane drug paclitaxel. Results are presented as mean + SE (n ≥ 4). D, Mechanism of GSH synthesis inhibition by BSO. GGC, gamma-L-glutamyl-L-cysteine; GCL, glutamate-cysteine ligase; GSS, GSH synthetase. E, Effect of 24-hour BSO-treatment on GSH concentration in D492 and D492M cells. F, D492 and D492M cells treated with everolimus (0.005 μmol/L) with and without BSO (50 μmol/L). G, D492 and D492M cells treated with paclitaxel (0.005 μmol/L) with and without BSO (50 μmol/L). For F and G, the Y-axis represents percentage of viability compared with nontreated cells after 72 hours of treatment (mean + standard error, n ≥ 4). Two-way ANOVA and Student two-tailed t test (with Benjamini–Hochberg adjustment for multiple comparisons) were used to measure significance of results in C and E–G, respectively.

Figure 5.

GSH levels regulate sensitivity to mTOR inhibitors. A, Network analysis of NCI-60 cell lines treated with various FDA-approved drugs (n = 214) suggests the presence of 8 modules of intra-correlated drugs. The efficacy of each individual drug was correlated with GSH levels, represented by node size. B, The correlation of drug modules' eigenvalues to reduced GSH, oxidized glutathione (GSSG), UDP-glucuronate and S-adenosylmethionine (SAM). Upper numbers in table represent Pearson correlation coefficient and the lower numbers (in brackets) represent the correlation p-value. C, D492 and D492M cells treated with increasing concentrations of the mTOR inhibitor everolimus and taxane drug paclitaxel. Results are presented as mean + SE (n ≥ 4). D, Mechanism of GSH synthesis inhibition by BSO. GGC, gamma-L-glutamyl-L-cysteine; GCL, glutamate-cysteine ligase; GSS, GSH synthetase. E, Effect of 24-hour BSO-treatment on GSH concentration in D492 and D492M cells. F, D492 and D492M cells treated with everolimus (0.005 μmol/L) with and without BSO (50 μmol/L). G, D492 and D492M cells treated with paclitaxel (0.005 μmol/L) with and without BSO (50 μmol/L). For F and G, the Y-axis represents percentage of viability compared with nontreated cells after 72 hours of treatment (mean + standard error, n ≥ 4). Two-way ANOVA and Student two-tailed t test (with Benjamini–Hochberg adjustment for multiple comparisons) were used to measure significance of results in C and E–G, respectively.

Close modal

Interestingly, GSH levels were negatively correlated with the mTOR inhibitor module, represented in Fig. 5A as the size of the nodes. When the modules were represented collectively as a single unit using singular value decomposition, mTOR inhibitors were significantly correlated with both reduced and oxidized GSH (Fig. 5B). This was further supported by the Spearman correlation P value for mTOR inhibitors and intracellular GSH and GSSG levels (Supplementary Fig. S7). These results imply that high GSH availability is associated with low sensitivity to mTOR inhibitors and vice versa. Other conjugation metabolites (i.e., UDP-glucuronate and S-adenosylmethionine) did not display this type of relationship with mTOR inhibitors.

Following this we examined the effects of the mTOR inhibitor everolimus (strong negative correlation with GSH levels) and paclitaxel (no correlation with GSH levels) on D492 and D492M cells. D492M cells were more sensitive to both everolimus and paclitaxel than D492 (Fig. 5C). To establish a functional link between GSH abundance and mTOR inhibitors, we cotreated D492 and D492M cells with BSO, an inhibitor of the rate-limiting enzyme glutamate-cysteine ligase (GCL) in GSH synthesis (Fig. 5D and E), and either everolimus (Fig. 5F) or paclitaxel (Fig. 5G). The sensitivity of both D492 and D492M cells to everolimus was enhanced by cotreatment with BSO, whereas these effects were not observed when the cells were cotreated with paclitaxel and BSO. Together, these data suggest that GSH availability directly affects sensitivity to drugs that specifically affect the mTOR pathway.

D492 and D492M cells represent only two of the numerous phenotypes within the spectrum of EMT (54). Herein, we have thoroughly characterized the central carbon metabolic activity of these cell types using 13C isotope tracers, specialized metabolic assays, and shRNA-mediated knockdown of gene expression. Furthermore, we have evaluated the functional consequences of EMT-mediated differences in redox metabolism using both in silico and in vitro drug-sensitivity analyses.

IDH2 plays a key role in EMT in breast epithelium

The data presented here support previous findings where genome-scale metabolic models (12) and signal-network models constrained with cell type-specific transcriptomic data (53) predicted lower glycolytic activity of D492M cells compared with D492 cells. In addition, the data indicate that D492M cells increasingly rely on reductive carboxylation of glutamine to citrate via isocitrate dehydrogenase activity. Transcriptomic and proteomic data from D492 and D492M cells confirm that the predominant EMT-associated form of isocitrate dehydrogenase is the mitochondrial NADP+-dependent IDH2, with D492M cells showing significantly higher levels on both the transcript and protein levels (Fig. 4A). Knockdown of IDH2 by lentiviral shRNA induction caused a marked reduction of reductive carboxylation of glutamine to citrate in D492M cells but did not affect the overall contribution of either glucose or glutamine to citrate (Fig. 4E). These results were reflected in the label incorporation of 5–13C-glutamine and 1,2–13C-glucose to palmitate (Fig. 4F). The decrease in citrate labeling from 1–13C-glutamine indicates that in these cells, reductive carboxylation primarily takes place within the mitochondria, as opposed to in the cytosol via IDH1 activity.

The knockdown of IDH2 and the subsequent re-routing of glutamine metabolism significantly diminished the ratio of NADPH to NADP+ and the synthesis of proline from glutamate (Fig. 4G and H), consistent with the relationship of reductive carboxylation and proline synthesis to redox homeostasis (45, 55). However, there was no clear connection of IDH2 to GSH synthesis (Fig. 4I).

Our findings highlight the importance of IDH2 in the increased reductive carboxylation following EMT in breast epithelium. However, we cannot exclude the importance of IDH1 in this context. It is reasonable to assume that when reductive carboxylation takes place in the cytosol via IDH1, the resulting citrate is transported into the mitochondria, where IDH2 takes part in its ultimate oxidation as proposed previously (45). When IDH2 levels are diminished, the activity of this pathway would inevitably be halted. Nevertheless, our results demonstrate that IDH2 knockdown significantly affects the reductive carboxylation of glutamine to citrate and ultimately fatty acids which establishes a functional role of IDH2 in this process. Thus, not only are the D492 and D492M cells different in their the overall lipid composition (56), but also the origin of lipid carbons.

Alterations in reductive carboxylation and redox metabolism follow EMT in breast

Our results show that glutamine-derived citrate is utilized for fatty acid synthesis in the D492 cell model, but the reliance on this pathway is enhanced following EMT. We show that there is a concurrent increase in NADPH/NADP+ ratio and proline synthesis along with a decrease in GSH synthesis (Fig. 3C–F). It has previously been shown that anchorage-independent growth relies on increased reductive carboxylation and subsequent mitigation of mitochondrial ROS (45). The role of proline in anchorage-independent growth has also been demonstrated by showing that its degradation and cycling is higher in breast cancer cells grown in 3D culture than in 2D (55), which ultimately altered the NADPH/NADP+ balance. Phang (57) hypothesized that in cancer cells, proline is directed away from protein synthesis and towards redox regulation, a pathway that proline has previously been shown to participate in within mammalian cells (58). More recently, it was shown that NADPH-dependent proline synthesis and reductive carboxylation act as alternative bins for electrons under hypoxic conditions when the electron transport chain is disabled. This way, electron transfer may continue functioning in cancer cells to maintain their viability (50). These observations fit well with the decreased oxygen consumption rate in D492M mesenchymal cells (12), their shift towards higher overall NADPH (cf. Fig. 3B), increased reductive carboxylation and higher proline synthesis. Furthermore, D492M cells display a concomitant decrease in GSH synthesis and intracellular abundance. Interestingly, Snail mediated EMT induction in MCF7 breast cancer cells has previously been shown to result in intracellular GSH reduction and elevation of ROS (59).

Diminished GSH abundance potentiates sensitivity to PI3K/akt/mTOR inhibitors

We have previously reported a reduction in oxidative phosphorylation following EMT in D492 cells (12). This causes a metabolic shift towards anaplerosis, an upregulation of pathways receiving otherwise ETC-directed electrons (i.e., proline synthesis and reductive carboxylation; ref. 50) and decreased glutamine-derived GSH synthesis. GSH is the most abundant nonprotein thiol in animal cells and it plays a crucial role in the conjugation phase of xenobiotic metabolism. This leads to increased water-solubility of foreign compounds (e.g., drugs or toxins) and their reduced efficacy (60, 61).

Because of the reduction in GSH synthesis and overall abundance in the mesenchymal D492M cells compared with epithelial D492 cells (Fig. 3C and D), we hypothesized that this would result in altered drug sensitivity of the mesenchymal phenotype. Integrated network analysis suggested that D492M cells are more sensitive to drugs that specifically target mTOR (Fig. 5A and B). Furthermore, the lack of a significant relationship between UDP-glucuronate and S-adenosylmethionine, compounds known to partake in the conjugation to xenobiotic compounds (9), suggests that mTOR inhibitors are specifically affected by GSH availability. We tested the efficacy of everolimus, a well-known mTOR inhibitor, and found that D492M cells were more sensitive than D492 cells (Fig. 5C). Furthermore, we showed that depleting intracellular GSH levels via BSO treatment (Fig. 5E) increased the sensitivity to everolimus (Fig. 5F). In contrast, BSO treatment did not affect sensitivity to paclitaxel, a microtubule stabilizer and mitotic inhibitor (Fig. 5G). These results indicate that GSH availability primarily affects the sensitivity to drugs that target the mTOR pathway.

In recent years, studies have shown a direct relationship between the mTOR signaling pathway and oxidative stress response (62, 63). Furthermore, mTOR signaling has been shown to be highly relevant in the EMT process and chemoresistance of ovarian cancer cells and melanoma (64, 65). Collectively, our results introduce a valuable mechanistic insight into the altered drug sensitivity following EMT in breast epithelium and support previous findings that GSH depletion in combination with mTOR inhibitors may specifically target the metastatic potential and/or stemness of cancer cells.

Conclusions

In summary, we have defined alterations in central carbon metabolism of a breast epithelial cell model of EMT using 13C carbon tracing. We show that glutamine metabolism is re-routed towards reductive carboxylation to fuel fatty acid synthesis following EMT due to activity of the mitochondrial NADP+-dependent isocitrate dehydrogenase (IDH2). This leads to decreased GSH production and disrupted redox homeostasis within the cells. Integrated network analysis of the NCI-60 Human Tumor Cell Line database revealed a negative correlation between intracellular GSH levels and sensitivity to mTOR inhibitors, and by depleting intracellular GSH levels in D492 and D492M cells, we sensitized them to mTOR pathway inhibitors. Our results highlight a potential metabolic weakness of low-GSH, EMT-derived cells that may be exploited in antimetastatic treatment.

S.T. Karvelsson reports grants from Göngum Saman and Icelandic Research Fund during the conduct of the study. A. Sigurdsson reports grants from Göngum Saman during the conduct of the study. No disclosures were reported by the other authors.

Sigurdur Trausti Karvelsson: Conceptualization, data curation, software, formal analysis, visualization, methodology, writing–original draft, writing–review and editing. Arnar Sigurdsson: Data curation, formal analysis, investigation, writing–review and editing. Kotryna Seip: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. Maria Tunset Grinde: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. Qiong Wang: Formal analysis, investigation, writing–review and editing. Freyr Johannsson: Formal analysis, investigation, methodology, writing–review and editing. Gunhild Mari Mælandsmo: Supervision, funding acquisition, writing–review and editing. Siver Andreas Moestue: Supervision, funding acquisition, investigation, writing–review and editing. Ottar Rolfsson: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing. Skarphedinn Halldorsson: Conceptualization, data curation, formal analysis, supervision, methodology, writing–original draft, writing–review and editing.

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.

1.
Chaffer
CL
,
San Juan
BP
,
Lim
E
,
Weinberg
RA
. 
EMT, cell plasticity and metastasis
.
Cancer Metastasis Rev
2016
;
35
:
645
54
.
2.
Hanahan
D
,
Weinberg
RA
. 
Hallmarks of cancer: the next generation
.
Cell
2011
;
144
:
646
74
.
3.
Lee
SY
,
Jeon
HM
,
Ju
MK
,
Jeong
EK
,
Kim
CH
,
Park
HG
, et al
Dlx-2 and glutaminase upregulate epithelial-mesenchymal transition and glycolytic switch
.
Oncotarget
2016
;
7
:
7925
39
.
4.
Shaul
YD
,
Freinkman
E
,
Comb
WC
,
Cantor
JR
,
Tam
WL
,
Thiru
P
, et al
Dihydropyrimidine accumulation is required for the epithelial-mesenchymal transition
.
Cell
2014
;
158
:
1094
109
.
5.
Koch
K
,
Hartmann
R
,
Schröter
F
,
Suwala
AK
,
Maciaczyk
D
,
Krüger
AC
, et al
Reciprocal regulation of the cholinic phenotype and epithelialmesenchymal transition in glioblastoma cells
.
Oncotarget
2016
;
7
:
73414
31
.
6.
Fischer
KR
,
Durrans
A
,
Lee
S
,
Sheng
J
,
Li
F
,
Wong
STC
, et al
Epithelial-to-mesenchymal transition is not required for lung metastasis but contributes to chemoresistance
.
Nature
2015
;
527
:
472
6
.
7.
Zheng
X
,
Carstens
JL
,
Kim
J
,
Scheible
M
,
Kaye
J
,
Sugimoto
H
, et al
Epithelial-to-mesenchymal transition is dispensable for metastasis but induces chemoresistance in pancreatic cancer
.
Nature
2015
;
527
:
525
30
.
8.
Williams
ED
,
Gao
D
,
Redfern
A
,
Thompson
EW
. 
Controversies around epithelial–mesenchymal plasticity in cancer metastasis
.
Nat Rev Cancer
2019
;
19
:
716
32
.
9.
Jancova
P
,
Anzenbacher
P
,
Anzenbacherova
E
. 
Phase II drug metabolizing enzymes
.
Biomed Pap
2010
;
154
:
103
16
.
10.
Jóhannsson
F
,
Guðmundsson
S
,
Paglia
G
,
Guðmundsson
S
,
Palsson
B
,
Sigurjónsson
ÓE
, et al
Systems analysis of metabolism in platelet concentrates during storage in platelet additive solution
.
Biochem J
2018
;
475
:
2225
40
.
11.
Grinde
MT
, et al
Biomarker discovery using NMR-based metabolomics of tissue
.
New York, NY
:
Humana
; 
2019
. p.
243
62
.
doi:10.1007/978-1-4939-9690-2_15
.
12.
Halldorsson
S
,
Rohatgi
N
,
Magnusdottir
M
,
Choudhary
KS
,
Gudjonsson
T
,
Knutsen
E
, et al
Metabolic re-wiring of isogenic breast epithelial cell lines following epithelial to mesenchymal transition
.
Cancer Lett
2017
;
396
:
117
29
.
13.
Sigurdsson
V
,
Hilmarsdottir
B
,
Sigmundsdottir
H
,
Fridriksdottir
AJR
,
Ringnér
M
,
Villadsen
R
, et al
Endothelial induced EMT in breast epithelial cells with stem cell properties
.
PLoS One
2011
;
6
:
e23833
.
14.
Schindelin
J
,
Arganda-Carreras
I
,
Frise
E
,
Kaynig
V
,
Longair
M
,
Pietzsch
T
, et al
Fiji: an open-source platform for biological-image analysis
.
Nat Methods
2012
;
9
:
676
82
.
15.
Itkonen
HM
,
Gorad
SS
,
Duveau
DY
,
Martin
SES
,
Barkovskaya
A
,
Bathen
TF
, et al
Inhibition of O-GlcNAc transferase activity reprograms prostate cancer cell metabolism
.
Oncotarget
2016
;
7
:
12464
76
.
16.
Austdal
M
,
Skråstad
RB
,
Gundersen
AS
,
Austgulen
R
,
Iversen
A-C
,
Bathen
TF
. 
Metabolomic biomarkers in serum and urine in women with preeclampsia
.
PLoS One
2014
;
9
:
e91923
.
17.
Eilers
PHC
. 
Parametric time warping
.
Anal Chem
2004
;
76
:
404
11
.
18.
Savorani
F
,
Tomasi
G
,
Engelsen
SB
. 
icoshift: a versatile tool for the rapid alignment of 1D NMR spectra
.
J Magn Reson
2010
;
202
:
190
202
.
19.
Bettum
IJ
,
Gorad
SS
,
Barkovskaya
A
,
Pettersen
S
,
Moestue
SA
,
Vasiliauskaite
K
, et al
Metabolic reprogramming supports the invasive phenotype in malignant melanoma
.
Cancer Lett
2015
;
366
:
71
83
.
20.
Kotronoulas
A
,
Jónasdóttir
HS
,
Sigurðardóttir
RS
,
Halldórsson
S
,
Haraldsson
GG
,
Rolfsson
Ó
. 
Wound healing grafts: omega-3 fatty acid lipid content differentiates the lipid profiles of acellular Atlantic cod skin from traditional dermal substitutes
.
J Tissue Eng Regen Med
2020
;
14
:
441
51
.
21.
Millard
P
,
Letisse
F
,
Sokol
S
,
Portais
JC
. 
IsoCor: correcting MS data in isotope labeling experiments
.
Bioinformatics
2012
;
28
:
1294
6
.
22.
Buescher
JM
,
Antoniewicz
MR
,
Boros
LG
,
Burgess
SC
,
Brunengraber
H
,
Clish
CB
, et al
A roadmap for interpreting 13C metabolite labeling patterns from cells
.
Curr Opin Biotechnol
2015
;
34
:
189
201
.
23.
Lee
WNP
,
Boros
LG
,
Puigjaner
J
,
Bassilian
S
,
Lim
S
,
Cascante
M
, et al
Mass isotopomer study of the nonoxidative pathways of the pentose cycle with [1,2-13C2]glucose
.
Am J Physiol
1998
;
274
:
E843
51
.
24.
Bray
NL
,
Pimentel
H
,
Melsted
P
,
Pachter
L
. 
Near-optimal probabilistic RNA-seq quantification
.
Nat Biotechnol
2016
;
34
:
525
7
.
25.
Briem
E
,
Budkova
Z
,
Sigurdardottir
AK
,
Hilmarsdottir
B
,
Kricker
J
,
Timp
W
, et al
MiR-203a is differentially expressed during branching morphogenesis and EMT in breast progenitor cells and is a repressor of peroxidasin
.
Mech Dev
2019
;
155
:
34
47
.
26.
Love
MI
,
Huber
W
,
Anders
S
. 
Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2
.
Genome Biol
2014
;
15
:
550
.
27.
Perez-Riverol
Y
,
Csordas
A
,
Bai
J
,
Bernal-Llinares
M
,
Hewapathirana
S
,
Kundu
DJ
, et al
The PRIDE database and related tools and resources in 2019: Improving support for quantification data
.
Nucleic Acids Res
2019
;
47
:
D442
50
.
28.
Cox
J
,
Mann
M
. 
MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification
.
Nat Biotechnol
2008
;
26
:
1367
72
.
29.
Shoemaker
RH
. 
The NCI60 human tumour cell line anticancer drug screen
.
Nat Rev Cancer
2006
;
6
:
813
23
.
30.
Luna
A
,
Rajapakse
VN
,
Sousa
FG
,
Gao
J
,
Schultz
N
,
Varma
S
, et al
Rcellminer: exploring molecular profiles and drug response of the NCI-60 cell lines in R
.
Bioinformatics
2016
;
32
:
1272
4
.
31.
Langfelder
P
,
Horvath
S
. 
WGCNA: an R package for weighted correlation network analysis
.
BMC Bioinformatics
2008
;
9
:
559
.
32.
Ortmayr
K
,
Dubuis
S
,
Zampieri
M
. 
Metabolic profiling of cancer cells reveals genome-wide crosstalk between transcriptional regulators and metabolism
.
Nat Commun
2019
;
10
:
1
13
.
33.
Csardi
G
,
Nepusz
T
. 
The igraph software package for complex network research
.
InterJournal Complex Sy
2006
;
1695
.
34.
Fruchterman
TMJ
,
Reingold
EM
. 
Graph drawing by force-directed placement
.
Softw Pract Exp
1991
;
21
:
1129
64
.
35.
R Core Team
.
R: A language and environment for statistical computing
; 
2020
.
36.
Wickham
H
.
ggplot2: Elegant Graphics for Data Analysis
.
New York
:
Springer-Verlag
; 
2016
.
37.
Kassambara
A
.
ggpubr: ‘ggplot2’ based publication ready plots
; 
2020
.
38.
Metallo
CM
,
Walther
JL
,
Stephanopoulos
G
. 
Evaluation of 13C isotopic tracers for metabolic flux analysis in mammalian cells
.
J Biotechnol
2009
;
144
:
167
74
.
39.
Watson
JA
,
Lowenstein
JM
. 
Citrate and the conversion of carbohydrate into fat. Fatty acid synthesis by a combination of cytoplasm and mitochondria
.
J Biol Chem
1970
;
245
:
5993
6002
.
40.
Gaglio
D
,
Metallo
CM
,
Gameiro
PA
,
Hiller
K
,
Danna
LS
,
Balestrieri
C
, et al
Oncogenic K-Ras decouples glucose and glutamine metabolism to support cancer cell growth
.
Mol Syst Biol
2011
;
7
:
523
.
41.
Yang
C
,
Ko
B
,
Hensley
CT
,
Jiang
L
,
Wasti
AT
,
Kim
J
, et al
Glutamine oxidation maintains the TCA cycle and cell survival during impaired mitochondrial pyruvate transport
.
Mol Cell
2014
;
56
:
414
24
.
42.
Yoo
HC
,
Yu
YC
,
Sung
Y
,
Han
JM
. 
Glutamine reliance in cell metabolism
.
Exp Mol Med
2020
;
52
:
1496
516
.
43.
Mullen
AR
,
Wheaton
WW
,
Jin
ES
,
Chen
P-H
,
Sullivan
LB
,
Cheng
T
, et al
Reductive carboxylation supports growth in tumour cells with defective mitochondria
.
Nature
2012
;
481
:
385
8
.
44.
Metallo
CM
,
Gameiro
PA
,
Bell
EL
,
Mattaini
KR
,
Yang
J
,
Hiller
K
, et al
Reductive glutamine metabolism by IDH1 mediates lipogenesis under hypoxia
.
Nature
2012
;
481
:
380
4
.
45.
Jiang
L
,
Shestov
AA
,
Swain
P
,
Yang
C
,
Parker
SJ
,
Wang
QA
, et al
Reductive carboxylation supports redox homeostasis during anchorage-independent growth
.
Nature
2016
;
532
:
255
8
.
46.
Du
J
,
Yanagida
A
,
Knight
K
,
Engel
AL
,
Vo
AH
,
Jankowski
C
, et al
Reductive carboxylation is a major metabolic pathway in the retinal pigment epithelium
.
Proc Natl Acad Sci U S A
2016
;
113
:
14710
5
.
47.
Liu
W
,
Hancock
CN
,
Fischer
JW
,
Harman
M
,
Phang
JM
. 
Proline biosynthesis augments tumor cell growth and aerobic glycolysis: Involvement of pyridine nucleotides
.
Sci Rep
2015
;
5
:
17206
.
48.
Smolková
K
,
Dvořák
A
,
Zelenka
J
,
Vítek
L
,
Ježek
P
. 
Reductive carboxylation and 2-hydroxyglutarate formation by wild-type IDH2 in breast carcinoma cells
.
Int J Biochem Cell Biol
2015
;
65
:
125
33
.
49.
Aguilar
E
,
Marin de Mas
I
,
Zodda
E
,
Marin
S
,
Morrish
F
,
Selivanov
V
, et al
Metabolic reprogramming and dependencies associated with epithelial cancer stem cells independent of the epithelial-mesenchymal transition program
.
Stem Cells
2016
;
34
:
1163
76
.
50.
Liu
M
,
Wang
Y
,
Yang
C
,
Ruan
Y
,
Bai
C
,
Chu
Q
, et al
Inhibiting both proline biosynthesis and lipogenesis synergistically suppresses tumor growth
.
J Exp Med
2020
;
217
:
e20191226
.
51.
Yang
CH
,
Kuo
ML
,
Chen
JC
,
Chen
YC
. 
Arsenic trioxide sensitivity is associated with low level of glutathione in cancer cells
.
Br J Cancer
1999
;
81
:
796
9
.
52.
Rocha
CRR
,
Garcia
CC
,
Vieira
DB
,
Quinet
A
,
de Andrade-Lima
LC
,
Munford
V
, et al
Glutathione depletion sensitizes cisplatin- and temozolomide-resistant glioma cells in vitro and in vivo
.
Cell Death Dis
2014
;
5
:
e1505
.
53.
Nunes
SC
,
Serpa
J
. 
Glutathione in ovarian cancer: a double-edged sword
.
Int J Mol Sci
2018
;
19
:
1882
.
54.
Zhang
Y
,
Weinberg
RA
. 
Epithelial-to-mesenchymal transition in cancer: complexity and opportunities
.
Front Med
2018
;
12
:
361
73
.
55.
Elia
I
,
Broekaert
D
,
Christen
S
,
Boon
R
,
Radaelli
E
,
Orth
MF
, et al
Proline metabolism supports metastasis formation and could be inhibited to selectively target metastasizing cancer cells
.
Nat Commun
2017
;
8
:
15267
.
56.
Eiriksson
FF
,
Rolfsson
O
,
Ogmundsdottir
HM
,
Haraldsson
GG
,
Thorsteinsdottir
M
,
Halldorsson
S
. 
Altered plasmalogen content and fatty acid saturation following epithelial to mesenchymal transition in breast epithelial cell lines
.
Int J Biochem Cell Biol
2018
;
103
:
99
104
.
57.
Phang
JM
. 
Proline metabolism in cell regulation and cancer biology: recent advances and hypotheses
.
Antioxid Redox Signal
2019
;
30
:
635
49
.
58.
Krishnan
N
,
Dickman
MB
,
Becker
DF
. 
Proline modulates the intracellular redox environment and protects mammalian cells against oxidative stress
.
Free Radic Biol Med
2008
;
44
:
671
81
.
59.
Mezencev
R
,
Matyunina
LV
,
Jabbari
N
,
McDonald
JF
. 
Snail-induced epithelial-to-mesenchymal transition of MCF-7 breast cancer cells: systems analysis of molecular changes and their effect on radiation and drug sensitivity
.
BMC Cancer
2016
;
16
:
236
.
60.
Zhang
K
,
Mack
P
,
Wong
KP
. 
Glutathione-related mechanisms in cellular resistance to anticancer drugs (Review)
.
Int J Oncol
1998
;
12
:
871
82
.
61.
Forman
HJ
,
Zhang
H
,
Rinna
AG
.: 
Overview of its protective roles, measurement, and biosynthesis
.
Mol Aspects Med
2009
;
30
:
1
12
.
62.
Li
J
,
Shin
S
,
Sun
Y
,
Yoon
S-O
,
Li
C
,
Zhang
E
, et al
MTORC1-driven tumor cells are highly sensitive to therapeutic targeting by antagonists of oxidative stress
.
Cancer Res
2016
;
76
:
4816
27
.
63.
Hambright
HG
,
Meng
P
,
Kumar
AP
,
Ghosh
R
. 
Inhibition of PI3K/AKT/mTOR axis disrupts oxidative stress-mediated survival of melanoma cells
.
Oncotarget
2015
;
6
:
7195
208
.
64.
Deng
J
,
Bai
X
,
Feng
X
,
Ni
J
,
Beretov
J
,
Graham
P
, et al
Inhibition of PI3K/Akt/mTOR signaling pathway alleviates ovarian cancer chemoresistance through reversing epithelial-mesenchymal transition and decreasing cancer stem cell marker expression
.
BMC Cancer
2019
;
19
:
618
.
65.
Seip
K
,
Fleten
KG
,
Barkovskaya
A
,
Nygaard
V
,
Haugen
MH
,
Engesæter
, et al
Fibroblast-induced switching to the mesenchymal-like phenotype and PI3K/mTOR signaling protects melanoma cells from BRAF inhibitors
.
Oncotarget
2016
;
7
:
19997
20015
.

Supplementary data