DLST-dependence dictates metabolic heterogeneity in TCA-cycle usage among triple-negative breast cancer

Triple-negative breast cancer (TNBC) is traditionally considered a glycolytic tumor with a poor prognosis while lacking targeted therapies. Here we show that high expression of dihydrolipoamide S-succinyltransferase (DLST), a tricarboxylic acid (TCA) cycle enzyme, predicts poor overall and recurrence-free survival among TNBC patients. DLST depletion suppresses growth and induces death in subsets of human TNBC cell lines, which are capable of utilizing glutamine anaplerosis. Metabolomics profiling reveals significant changes in the TCA cycle and reactive oxygen species (ROS) related pathways for sensitive but not resistant TNBC cells. Consequently, DLST depletion in sensitive TNBC cells increases ROS levels while N-acetyl-L-cysteine partially rescues cell growth. Importantly, suppression of the TCA cycle through DLST depletion or CPI-613, a drug currently in clinical trials for treating other cancers, decreases the burden and invasion of these TNBC. Together, our data demonstrate differential TCA-cycle usage in TNBC and provide therapeutic implications for the DLST-dependent subsets.

T riple-negative breast cancer (TNBC) is an aggressive malignancy associated with high recurrence rates for localized disease and increased risk of visceral metastasis, as well as poor prognosis for advanced disease 1,2 . Due to the absence of hormone receptor overexpression and HER2 amplification, TNBC is not suited for targeted therapy directed against these receptors, with chemotherapy being the main treatment option, especially for those without BRCA1/2 mutations 1,3 . Moreover, lacking predictive prognostic biomarkers, it remains difficult for clinicians to select effective therapeutic approaches for this disease 4 . An improved understanding of disease etiology could reveal novel cellular pathways for therapeutic intervention and better TNBC prognostication.
Although targeting metabolic dependencies emerges as an effective approach for the treatment of aggressive malignancies such as acute leukemia 5 , TNBC cells are highly heterogeneous in terms of their dependence on specific metabolic pathways 6 . Early studies indicated that TNBC cells utilize predominantly aerobic glycolysis instead of oxidative phosphorylation (OXPHOS) for energy production 7 . A recent study, however, shows that TNBC cells shift from glycolysis to mitochondrial OXPHOS during metastatic spread 8 . Similarly, upregulated fatty acid oxidation is detected in metastatic breast cancer 9 . Despite these findings, the dependence of TNBC cells on the tricarboxylic acid (TCA) cycle remains unclear.
Inside the TCA cycle, dihydrolipoamide S-succinyltransferase (DLST) is the E2 transferase of α-ketoglutarate dehydrogenase complex (KGDHC) that regulates the irreversible conversion of α-ketoglutarate to succinyl-CoA 10 . Here we show that high DLST expression predicts poor overall and recurrence-free survival among TNBC patients. DLST depletion impairs the growth, survival, and migration of TNBC cells that have relatively intact TCA-cycle function, while minimally affecting TNBC cells with the defective TCA cycle function. Importantly, DLST-dependent TNBC cells exhibit decreased tumor burden and invasion in zebrafish xenografts when treated with CPI-613, a lipoate derivative that can inhibit KGDHC activity and is currently in clinical trials for treating other cancers (NCT03793140; NCT03699319; NCT04217317; and NCT04593758). Together, our data demonstrate that DLST-dependence in TNBC dictates their usage of the TCA cycle to promote tumor cell growth and invasion. Our studies provide mechanistic insights for metabolic heterogeneity of TNBC and therapeutic implications for the DLST-dependent subsets.

Results
High DLST expression predicts poor overall and recurrencefree survival among TNBC patients. We previously reported that DLST contributes to MYC-driven leukemogenesis 11 . To study the role of DLST in breast cancer pathogenesis, we performed in silico Kaplan-Meier survival analysis using an online tool (http://kmplot.com/analysis/) 12 . Patients were categorized into three groups based on the receptor's expression in tumor cells: ER+, ER−, and TNBC 12 . We then correlated the expression of DLST and the other two KGDHC subunits, dihydrolipoamide dehydrogenase (DLD) and oxoglutarate dehydrogenase (OGDH), with survival probability among each patient group. We found that DLST expression is associated with poor overall survival with a similar trend for recurrence-free survival in the ER+ patients ( Fig. 1a, b). Different from ER-patients, which showed inconsistent overall and recurrence-free survival, high DLST expression reliably predicted poor overall and recurrence-free survival among TNBC patients (Fig. 1a, b). A similar observation for recurrence-free survival was made for TNBC patients when analyzing another publicly available database (GSE2034) 13 ( Supplementary Fig. 1a), despite somewhat different criteria applied in defining receptor positivity compared to the KMplot dataset. Different from DLST, neither DLD nor OGDH expression could similarly predict TNBC patient prognosis in both datasets, although OGDH showed a trend as DLST in predicting TNBC patient prognosis ( Supplementary Figs. 1-3). These findings demonstrate that high expression of DLST predicts disease relapse and poor survival among TNBC patients, indicating a different role of DLST in TNBC pathogenesis when compared to leukemia.
Human TNBC cells exhibit differential dependence on DLST. Next, we utilized cell line models to study how DLST contributes to TNBC pathogenesis. Our western blotting analysis revealed that DLST protein expression fluctuates among breast cancer cell lines (Fig. 2a). Interestingly, DLST protein expression in all of the ER+ cell lines tested is lower than that of the non-transformed MCF10A cells (Fig. 2a). However, DLST expression in TNBC cell lines differs, with some having similar protein levels compared to the non-transformed MCF10A cells and others with lower expression of DLST, which mimics its protein and transcript levels in both TNBC patient samples and cell lines ( Fig. 2a; Supplementary Fig. 4a-c). Among TNBC cell lines tested, Hs578T and SUM159PT were the only two that showed relatively lower DLST expression on both transcript and protein levels ( Fig. 2a; Supplementary Fig. 4c). Next, we knocked down DLST by shRNA in breast cancer cell lines and found that all of the ER+ cell lines were sensitive to DLST depletion, while TNBC cell lines exhibited a differential dependence on DLST ( Fig. 2b; Supplementary Fig. 5a, b). DLST knockdown significantly reduced the growth of several TNBC cell lines: HCC1806, BT-549, MDA-MB-231, and MDA-MB-436 ( Fig. 2b; Supplementary  Fig. 5b). However, DLST depletion minimally impacted SUM159PT and Hs578T TNBC cells, which had the lowest DLST expression, as well as the non-transformed MCF-10A cells (Fig. 2a, b; Supplementary Fig. 5a, b).
To understand the cellular mechanisms underlying TNBC's differential response to DLST depletion, we analyzed DLSTdependent BT-549 and MDA-MB-231 cell lines, as well as independent Hs578T and SUM159PT cell lines. At 5 days posttransduction of shLuciferase, shDLST1, or shDLST2 virus, we assessed the apoptosis and necrosis of these TNBC cells by measuring Annexin V and Ethidium homodimer III stainings. DLST depletion by shRNA significantly increased apoptosis and necrosis of BT-549 and MDA-MB-231 cells while minimally impacting Hs578T and SUM159PT cells ( Fig. 2c; Supplementary  Fig. 6a). Cell cycle analysis was then performed for these TNBC cells by propidium iodide staining and flow cytometry. DLST depletion caused a significant change in G1-phase and an increase in G2-phase arrest of BT-549 and MDA-MB-231 cells but only a moderate S or G1-phase change in Hs578T and SUM159PT cells ( Fig. 2d; Supplementary Fig. 6b). Taken together, the above data support that subsets of the TNBC cell lines depend on DLST for growth and survival.
Human DLST-dependent TNBC cells possess an intact TCA cycle function and utilize glutamine anaplerosis. Since DLST is a TCA cycle transferase 11 , we next evaluated the ability of these TNBC cell lines, BT-549, MDA-MB-231, SUM159PT, and Hs578T in utilizing the TCA cycle by galactose replacement assay. When galactose replaces glucose in the medium, cells will produce ATP preferentially through the TCA cycle instead of glycolysis 14,15 . Hence, if a cell can maintain its cellular ATP levels in the galactose medium, its TCA cycle function should be intact, and vice versa. Consistent with their dependence on DLST, BT-549 and MDA-MB-231 cells continuously produced ATP in the galactose medium with cellular ATP levels minimally affected ( Supplementary Fig. 7). In contrast, SUM159PT and Hs578T, the two cell lines with lower dependency on DLST, experienced a significant decrease in ATP levels when cultured in the galactose medium ( Supplementary Fig. 7). Next, we measured the oxygen consumption rates (OCR) by seahorse technology to further document the differential utilization of the TCA cycle in BT-549 and Hs578T cells. DLST depletion significantly decreased all aspects of OCR in DLST-dependent BT-549 cells, including basal, maximal, ATP-linked, and proton-leak OCR ( Fig. 3a and Supplementary Fig. 8a). However, OCR is minimally impacted in DLST-independent Hs578T cells upon DLST knockdown ( Fig. 3a and Supplementary Fig. 8a).
Given that TNBC cells are addicted to glutamine and that DLST/KGDHC governs the entry of glutamine into the TCA cycle [16][17][18][19][20] , we asked whether TNBC cells utilize glutamine anaplerosis. All TNBC cell lines tested are sensitive to glutamine withdrawal; however, only DLST-dependent TNBC cells can be partially rescued by the addition of exogenous TCA-cycle intermediates, dimethyl-2-oxoglutarate or mono-methyl hydrogen succinate ( Supplementary Fig. 8b). Next, we examined the effect of DLST depletion on glutamine anaplerosis in both DLSTdependent and independent TNBC cells. Using 13 C-labeled glutamine as a tracer, we measured the percentage changes of glutamine-derived metabolites upon DLST knockdown in DLSTdependent BT-549 and independent Hs578T cells (Fig. 3b). Both cell lines incorporated over 90% of the labeled glutamine in the presence and absence of DLST knockdown ( Supplementary  Fig. 9a). In addition, over 80% of α-ketoglutarate were labeled with 13 C-glutamine carbons in both cell lines without significant differences between control and DLST knockdown cells (Supplementary Fig. 9b). These results indicate that 13 C-labeled glutamine was incorporated into these cells efficiently and the α-ketoglutarate carbons were mainly from glutamine. Despite similar glutamine contribution to the TCA cycle, DLST knockdown in BT-549 cells significantly decreased the percentage of M +4 and M + 5 citrate, M+2 succinate, M+4 fumarate, and M+4 malate, while increasing the percentage of M+0/M+4 succinate, M+0/M+1 malate, and M+2 fumarate (Fig. 3c). In contrast, DLST knockdown had minimal effect on the contribution of glutamine to cycle intermediates in Hs578T cells (Fig. 3d). Collectively, these results demonstrate that the TNBC cell lines that depend on DLST for growth and survival have a higher capacity to utilize the TCA cycle, glutamine anaplerosis in particular, compared to the independent ones.
DLST knockdown induces discrete metabolic alterations in human DLST-dependent TNBC cells. Next, we performed unbiased metabolomics profiling to assess the broad metabolic changes induced by DLST inactivation in human TNBC cells. Specifically, BT-549 and Hs578T cells were transduced with the shLuciferase or shDLST hairpin. At early day 4 post-transduction when we could not detect apparent cellular changes, total metabolites were extracted from these TNBC cells for mass spectrometry analysis as described 21 . Pathway enrichment analysis revealed that DLST inactivation significantly impacted multiple metabolic pathways in DLST-dependent BT-549 cells while minimally impacting DLSTindependent Hs578T cells (Fig. 4a). Among those altered pathways, prominent changes were found in aminoacyl-tRNA biosynthesis, pyrimidine metabolism, and amino-acid metabolism pathways (Fig. 4a). As expected, the TCA cycle was significantly altered in BT-549 cells, with decreased succinyl-CoA and significantly increased succinate, phosphoenolpyruvate, isocitrate, and citrate levels ( Fig. 4b; Supplementary Fig. 10a).
Besides the changes observed in the TCA cycle, several pathways regulating redox balance, such as glutathione metabolism, cysteine, and methionine metabolism, and the pentose phosphate pathway, were also significantly altered in BT-549 cells ( Fig. 4b-e). DLST downregulation in BT-549 cells led to a significant increase of NADPH, glutathione disulfide, and glutamate levels, as well as a mild increase of glycine levels ( Fig. 4c, f). However, cysteine showed a decrease under DLST knockdown (Fig. 4c, d), accompanied by apparent changes of cysteine metabolism, with significantly increased levels of Sadenosyl-L-methionine, methionine, and cystine (Fig. 4d). The disturbance of the pentose phosphate pathway upon DLST knockdown led to a significant increase of glucose-6-phosphate, D-erythrose-4-phosphate, and 6-phospho-D-gluconate levels, as well as a decrease in 5-phosphoribosyl-1-pyrophosphate levels (Fig. 4e). In sharp contrast to changes in BT-549 cells, DLST knockdown only caused minimal metabolic changes in Hs578T cells ( Fig. 4a; Supplementary Fig. 10a, b). Hence, DLST depletion induces differential metabolic alterations in human TNBC cells based on their dependency on DLST.
DLST depletion induces ROS production in human DLSTdependent TNBC cells. Since the metabolomics data revealed pathway alterations related to redox balance in BT-549 cells, we asked how DLST knockdown influenced the production of ROS in human TNBC cell lines, in the presence or absence of a ROS inducer, tert-butyl hydroperoxide (TBHP). DLST knockdown significantly increased the ROS production in DLST-dependent BT-549 and MDA-MB-231 cells but not in the independent SUM159PT and Hs578T cells (Fig. 5a). To assess the source of ROS production upon DLST knockdown, we stained BT-549 and Hs578T cells with a mitochondrial superoxidase (mitoSOX) dye. DLST inactivation induced a significant increase of mitoSOX signals in BT-549 cells but not in Hs578T cells, indicating mitochondria as the location of ROS production (Fig. 5b, c). To understand whether the increased ROS levels in DLST-dependent TNBC cells contributed to decreased cell growth upon DLST inactivation, we supplemented the media of TNBC cell lines with 2 mM of N-acetyl-L-cysteine (NAC) starting at 2 days post-transduction. Supplement with NAC partially rescued the growth of BT-549 and MDA-MB-231 cells upon DLST inactivation (Fig. 5d). These findings indicate that mitochondrial ROS production serves as one mechanism to impair cell growth upon DLST depletion in TNBC cells.  Supplementary Fig. 5a, c. c Apoptosis and necrosis of TNBC cells were assessed at day 5 post-transduction by Annexin-V (green) and Ethidium homodimer III (red) staining, respectively, with DAPI (blue) staining for all cells. Representative images of BT-549 and Hs578T cells (left) together with their quantification (right) were shown (n = 6 per group). Scale bars = 20 µm. d Cell cycle distribution of BT-549 and Hs578T cells after DLST knockdown was shown as the percentage of cells in each cell-cycle phase (n = 3 biological samples). Data in (a-d) are presented as mean ± s.e.m. One-way analysis of variance (ANOVA) in (b, d) and an unpaired two-tailed t-test in (a, c) were used for statistical analyses. * or # P ≤ 0.05, ** or ## P ≤ 0.01, *** or ### P ≤ 0.001.
DLST depletion inhibits the migration, growth, and invasion of human DLST-dependent TNBC cells. TNBC is an aggressive disease and often metastasizes to distant organs in patients 22 . To understand whether DLST plays a role in TNBC aggression, we performed in vitro transwell migration assays using human BT-549 and MDA-MB-231 cells. DLST depletion significantly slowed the migration of these TNBC cells (Fig. 6a). To assess the contribution of DLST to TNBC aggression in vivo, we transduced BT-549 cells with Doxycycline-inducible shDLST2 and subsequently transplanted BT-549 cells into 2-day-old zebrafish, which lack adaptive immunity, to generate zebrafish xenografts. At 4 days of post-transplantation, DLST depletion significantly reduced tumor burden and invasion, which can be rescued by NAC treatment (Fig. 6b). Immunofluorescent staining of zebrafish xenografts revealed that DLST depletion induced mitochondrial ROS, necrosis, and apoptosis while reducing proliferation in BT-549 cells, all of which were again rescued by NAC treatment (Fig. 7a, b). The above data show that DLST supports the growth, survival, and invasion of TNBC cells that possess the intact TCA cycle function.
DLST-dependent TNBC cells are sensitive to CPI-613 treatment. Since subsets of TNBC depend on DLST for growth and disease invasion, we next asked whether they are sensitive to compounds targeting DLST-mediated TCA functions. CPI-613 is a lipoate derivative that can inhibit KGDHC activity and is currently in clinical trials for treating relapsed/ refractory lymphoma, metastatic pancreatic cancer, and clear cell sarcoma (NCT03793140; NCT03699319; NCT04217317; and NCT04593758). We thus applied CP-613 to treat zebrafish xenografts transplanted with human TNBC cells. We found that similar to the effect of DLST depletion, CPI-613 treatment effectively decreased tumor cell invasion in zebrafish xenografts transplanted with BT-549 cells (Fig. 8a, b). Consistent with their sensitivity to DLST depletion, CPI-613 treatment also reduced the tumor burden of these zebrafish xenografts (Fig. 8a, c). However, CPI-613 did not reduce the invasion or tumor burden of zebrafish xenografts transplanted with Hs578T cells, which are insensitive to DLST depletion (Fig. 8a-c). Interestingly, BT549 cells are more invasive in zebrafish xenografts compared to Hs578T cells (Fig. 8a, b). Hence, our data demonstrate that DLST-dependent TNBC cells are sensitive to the TCA-cycle inhibitor CPI-613.

Discussion
TNBC is highly heterogeneous in its usage of metabolic pathways 6 . However, no viable markers exist to categorize  (Supplementary Fig. 7). b Scheme of 12 C and 13 C labeled glutamine contribution to TCA-cycle metabolites. c, d Percentage of 13 C labeled glutamine-derived cycle metabolites was measured upon DLST knockdown in BT-549 (c) and Hs578T cells (d) (n = 3). Data are presented as mean ± s.e.m, and an unpaired two-tailed t-test was used for statistical analysis in (a, c, d). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001.  25,26 . Despite this knowledge, whether TNBC utilizes the TCA cycle remains elusive. Here we performed DLST inactivation studies in a panel of TNBC cell lines to understand the metabolic heterogeneity and dependence on the TCA cycle. Our studies show that subsets of TNBC are dependent on DLST and the TCA cycle for growth and invasion. Galactose replacement and bioenergetic assays demonstrate that DLST-dependent TNBC cells utilize the TCA cycle and OXPHOS, while the less dependent TNBC cells minimally do. In DLST-dependent TNBC cells, DLST depletion inhibits OCR, leads to prominent alterations of TCA cycle metabolites, slows TNBC cell migration, and decreases disease burden and invasion in zebrafish xenografts. Hence, our studies demonstrate the differential utility of the TCA cycle in TNBC cells and unveil DLST-dependency as a valid molecular marker to predict TCA cycle usage in TNBC.
Glutamine is an important carbon source for TNBC 17 , but little is known about how glutamine is utilized in human TNBC cells. Our 13 C labeled glutamine tracing data show that glutamine contributes to cycle intermediates in both DLST-dependent and independent TNBC cells. However, DLST depletion significantly reduces glutamine-contributed cycle intermediates in DLSTdependent TNBC cells, while minimally affecting independent ones. Consistent with our metabolomics data, DLST-dependent TNBC cells can be partially rescued upon glutamine withdrawal through supplementation of cycle intermediates, while the independent ones cannot. Hence, glutamine metabolism in TNBC is highly heterogeneous, and DLST/KGDHC mediates glutamine anaplerosis only in DLST-dependent TNBC cells. Interestingly, DLST depletion does not enhance the reductive carboxylation pathway, as no increased M+3 malate or M+5 citrate was detected in BT-549 or Hs578T cells. Instead, α-ketoglutarate in DLST-independent Hs578T cells can be converted into glutamate then glutamine to support cell survival under glutamine deprivation conditions 27 . On the other hand, DLST depletion in BT-549 cells leads to increased M+0/M+4 succinate, M+0/M+1 malate, and M+2 fumarate, indicating metabolic flexibility and compensatory effects. Our unbiased metabolomics profiling analysis reveals that DLST inactivation in BT-549 cells induces significant alterations in the TCA cycle, as well as nucleotide and amino-acid metabolic pathways, such as purine, pyrimidine, and serine and threonine metabolism. These data further support that glutamine anaplerosis replenishes cycle intermediates to support macromolecule and nucleotide synthesis in DLST-dependent TNBC cells. The aggressive clinical course, dismal prognosis, and lack of effective targeted agents for TNBC render the identification of predictive biomarkers and the development of novel therapeutic approaches critical to improve patients' survival. In particular, it is difficult to predict the sensitivity of chemotherapy or targeted therapy on TNBC patients, due to its intrinsic molecular and metabolic heterogeneity 28,29 . Our analysis of clinical data shows that high expression of DLST, a TCA cycle enzyme, is associated with an increased incidence of distant recurrence and poor survival among TNBC patients. Additionally, the majority of human DLST-dependent TNBC cell lines utilize the TCA cycle for growth and survival. Interestingly, TNBC cells with relatively high DLST expression use the TCA and are more invasive yet sensitive to DLST depletion, while most of those with relatively low DLST expression are not. Hence, our findings predict that TNBC patients with relatively high DLST expression in their tumor cells should be sensitive to inhibitors targeting the TCA cycle. CPI-613, an inhibitor targeting DLST/KGDHC in the TCA cycle, is in clinical trials for treating other types of cancers and shows promising safety and efficacy profiles (NCT03793140; NCT03699319; NCT04217317; and NCT04593758) 30,31 . Our data demonstrate that DLST-dependent TNBC cells are sensitive to CPI-613. Therefore, CPI-613 could serve as a targeted therapy for treating the aggressive TNBC with high DLST expression alone or in combination with chemotherapy.

Methods
Patient sample analysis. To assess the role of DLST in breast cancer pathogenesis, we performed in silico Kaplan-Meier analysis to determine the association of DLST expression with the overall and recurrence-free survival of breast cancer patients using the online tool and database (http://kmplot.com/analysis/) 12 . We applied the best cut-off, array, and immunohistochemistry (IHC) for ER, IHC for PR, and array for HER2 to stratify patients, which provides the optimal curve segregation. To further verify the findings, we re-analyzed the publicly available dataset (GSE2034) from NCBI/Genbank GEO database 13 . We also applied the best cut-off and receptor expression (ER < 1000, PR < 20, and HER2 < 3700 as being negative) to categorize patient samples into three categories: ER+ (n = 200), ER− (n = 86), and TNBC (n = 41). Cox regression analysis was utilized to determine the association of DLST expression with recurrence-free survival among breast cancer patients.
Cell cycle, apoptosis, and necrosis assay. Two million cells were fixed with 95% precooled ethanol, incubated overnight at −20°C, washed with Dulbecco's Phosphate Buffered Saline (DPBS, SH30028LS, Corning), and stained with propidium iodide/RNase staining buffer (BDB550825, BD Biosciences) containing 0.1% sodium citrate for 45 min (min) on ice in the dark. These cells were then analyzed by flow cytometry using BD LSRII SORP (BD Biosciences). ModFit LT software (Verify Software House) was used to generate DNA histograms. Each experiment was performed in triplicates.
The Apoptosis and Necrosis Quantification Kit (30017, Biotium) was used to quantify apoptotic and necrotic TNBC cells at 5 days post-transduction, according to the manufacture's protocol. Images were taken using a Revolve microscope (ECHO) and data were analyzed by Image J software. Apoptotic and necrotic cells were counted and normalized relative to the total number of cells in each well.
ROS detection. ROS detection was performed using 2,7′-dichlorofluorescin diacetate (DCFDA) according to the manufacturer's instruction (ab113851, Abcam) with modifications. Briefly, on day 4 post-transduction, TNBC cells were trypsinized from 6-well plates and incubated with DCFDA (20 µM) with a cell density of 500,000 ml −1 for 45 min at 37°C in the dark. These cells were then washed with DBPS, seeded into a 96-well plate at a density of 25,000 per well, and treated with or without 50 µM tert-butyl hydroperoxide (TBHP) for 16 h. The ROS signals were detected using a microplate reader (BMG LABTECH) under the excitation and emission wavelength at 485 nm and 535 nm, respectively.
Metabolomics profiling and analysis. Glutamine-free DMEM or RPMI medium was supplemented with 12 C or 13 C isotope-labeled L-Glutamine (G8540 and 605166, Sigma) to prepare media for metabolomics experiments. Lentivirus transduction was performed as described above. On day 3 post-transduction, the medium was replaced with the 2 mM 13 C or 12 C glutamine medium for BT-549 cells and 4 mM 13 C or 12 C glutamine medium for Hs578T cells. 24 h later, cells were washed with DPBS twice and extracted for polar metabolites. Specifically, prechilled 80% methanol (MeOH, A412P-4, Fisher Scientific) at −80°C was added to the cells. After 20 min of incubation at −80°C, the resulting mixture was scraped, collected into a 15-ml centrifuge tube, and centrifuged at 4000 rpm for 5 min at 4°C. Insoluble pellets were re-extracted with 1 ml of MeOH. The supernatants from two rounds of extraction were combined then dried with a Savant DNA 120 speedVac (Thermo Scientific). The unlabeled and 13 C labeled metabolites were analyzed by LC-MS at the Beth Israel Deaconess Medical Center.
The cell number of each sample was used to normalize the raw value of polar metabolites, which were then visualized and analyzed for statistical differences in R v3.5.3. Metabolites with missing values in more than half of the samples were removed and the missing data for the retained metabolites were imputed with the half of minimum values of its corresponding metabolite. Metabolites with either P values less than 0.05 from a two-sided t-test or absolute fold change greater than 1.5 were selected for pathway enrichment analysis by MetaboAnalyst 4.0 (https:// www.metaboanalyst.ca/). Isotope-labeled data were also processed similarly. FDR (false discovery rate) adjusted P values of pathway enrichment results for BT-549 and Hs578T cells with and without DLST knockdown were used to generate the pathway heatmap using the heatmap.2 function in the R package gplots. Heatmap plots for metabolites in the chosen pathway were generated using the levelplot function in the R package lattice.
Zebrafish xenograft assays. Zebrafish (Danio rerio) husbandry was performed as described 34 , in the aquatic facility at Boston University School of Medicine (BUSM), following the protocols approved by the Institutional Animal Use and Care Committee at BUSM.
Zebrafish embryos were gradually reared to 37°C and injected with~200 TNBC cells into the perivitelline space of each embryo at 2-day post-fertilization (dpf). BT-549 cells were transduced with doxycycline-regulated shDLST2 and stained with a cell tracker dye (C7001, Thermo Fisher). At 3 h posttransplantation, zebrafish xenografts were sorted into four groups and incubated with fish water containing either vehicle (distilled water) or 10 μg ml −1 of doxycycline (D9891, Sigma) to induce shDLST2 expression in the presence and absence of 300 μM NAC (A7250, Sigma). Fish water was refreshed daily. At day 2 post-transplantation, fish were observed under the microscope and their images were captured. For the CPI-613 treatment experiment, BT-549 and Hs578T cells transduced with pWPI-RFP were injected to fish embryos that were treated with 5 μM CPI-613 (HY-15453, MCE) or vehicle. On day 2 post-transplantation, zebrafish were euthanized and fixed with 4% Paraformaldehyde (158127, Thermo Scientific) overnight. The fixed fish were then imaged under the MVX10 microscope (OLYMPUS) or stored in 100% MeOH at −20°C.
Statistics and reproducibility. The association of DLST expression with overall and recurrence-free survival among patients with different subtypes of breast cancer was assessed by Kaplan-Meier analysis. The comparison of the statistical difference between the survival curves was done with the log-rank test. The student's t-test was used to analyze differences in DLST protein expression, OCR, ATP levels, metabolites, and isotope-labeled glutamine contribution to metabolic derivates for TNBC cells with control vs. DLST knockdown. One-way analysis of variance (ANOVA) was utilized to assess differences in cell growth rates, apoptosis, necrosis, cell cycle distribution, and ROS levels among TNBC cells in the presence or absence of DLST knockdown. Two-way ANOVA was used to analyze differences in mitoSOX staining, cell growth rates for NAC rescue experiments, tumor cell migration, tumor burden, and tumor invasion in zebrafish xenografts in the presence or absence of DLST knockdown. All experiments except re-analysis of the publicly available datasets were conducted at least three times independently. P values equal to or less than 0.05 were considered statistically significant without being adjusted for multiple comparisons.  52a (a, c). Fish number for BT-549: n = 27 for vehicle and 20 for CPI-613, respectively; and Hs578T: n = 24 for vehicle and 22 for CPI-613, respectively. Scale bar = 0.5 mm. Statistical difference was calculated using an unpaired two-tailed t-test. *P ≤ 0.05, ***P ≤ 0.001. Fig. 7 DLST depletion induces ROS, apoptosis, and necrosis while slowing the proliferation of human DLST-dependent TNBC cells, which can be rescued by NAC treatment. a, b Representative images (a) and quantitation (b) of mitoSOX (red, a marker for mitochondrial ROS), ethidium homodimer III (red, a marker for necrosis), active-caspase 3 (blue, a marker for apoptosis), and Ki-67 (blue, a marker for cell proliferation) stainings, which were overlaid either with GFP or DAPI signals of tumor cells in the presence or absence of NAC treatment (n = 3-10 fish per group). Scale bars = 20 μm. Data in (b) presented as mean ± s.e.m. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001. Statistical differences were calculated using one-way ANOVA.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
We confirm that all relevant data and methods are included in the main Article and the Supplementary Information section. The source data for the graphs and charts in the figures is available as Supplementary Data file and any remaining information can be obtained from the corresponding author upon reasonable request.