Citrate clearance is a major function of aconitase 2 in the canonical TCA cycle

SUMMARY

The tricarboxylic acid (TCA) cycle couples nutrient oxidation with the generation of reducing equivalents that power oxidative phosphorylation. Nevertheless, the requirement for components of the TCA cycle is context-specific, raising the question of which TCA cycle outputs support cell fitness. Here, we demonstrate that citrate clearance is an essential function of the TCA cycle. As citrate production increases, so do TCA cycle activity and dependence upon aconitase 2 (ACO2), the enzyme that initiates citrate catabolism in the TCA cycle. Disrupting citrate catabolism activates the integrated stress response and impairs cell fitness, and these effects are reversed by preventing citrate production or promoting mitochondrial citrate efflux. In vivo, ACO2 deficiency induces citrate accumulation and triggers tubular degeneration in the kidney, a tissue that physiologically takes up circulating citrate. Thus, intracellular citrate accumulation can be a metabolic liability, and citrate clearance is a major function of ACO2 in the TCA cycle.

Graphical abstract

In brief

Beyond its traditional role in supporting bioenergetic and biosynthetic demands, the canonical TCA cycle is also essential for nutrient clearance. When pyruvate oxidation and citrate production increase, aconitase 2 becomes essential to prevent mitochondrial citrate accumulation, which can trigger the integrated stress response and impair cell fitness.

INTRODUCTION

The tricarboxylic acid (TCA) cycle is an evolutionarily ancient metabolic pathway coupling nutrient oxidation to the supply of essential biosynthetic precursors and reducing equivalents that drive energy production. Since Krebs’s discovery of the TCA cycle in pigeon breast muscle in 1937,¹ many studies have demonstrated notable flexibility in TCA cycle metabolism.^2,3 For example, when cells are starved of glutamine or other sources of anaplerosis, direct carboxylation of pyruvate to oxaloacetate allows cells to refill the TCA cycle intermediates required for continued TCA cycle turning.^4,5 Alternatively, when oxidative production of citrate is compromised by hypoxia or electron transport chain (ETC) disruption, cells reverse a portion of the TCA cycle to preserve citrate production for lipid synthesis through reductive carboxylation of alpha-ketoglutarate (αKG).^6,7 Cells may even bypass many steps of the canonical TCA cycle by exporting citrate to the cytosol in exchange for malate, and this citrate-malate shuttle represents an alternative TCA cycle that allows cells to shuttle acetyl-coenzyme A (CoA) to the cytosol for lipid synthesis and protein acetylation instead of combusting citrate within mitochondria.^8,9

Taken together, these studies demonstrate that in response to context-specific demands, cells engage different portions of the TCA cycle to maintain levels of metabolites beyond ATP. Perhaps as a consequence of this metabolic heterogeneity, patients with mutations in genes encoding components of the TCA cycle exhibit a wide range of phenotypes, from severe neurometabolic syndromes at birth to specific clusters of neuroendocrine and kidney tumors in adulthood.^10–12 The range in disease severity and affected organ system underscores the notion that components of the TCA cycle are variably essential across tissues. Identifying the distinct cellular functions supported by each node of the TCA cycle is a prerequisite for understanding diseases arising from TCA cycle disruption. In particular, the factors that dictate citrate consumption in the canonical TCA cycle, as opposed to alternative TCA cycle configurations such as mitochondrial citrate export and cleavage for lipid synthesis or glutamine-mediated anaplerosis to maintain bioenergetics and TCA cycle intermediates, are unknown. Here, we aimed to determine when cells engage in citrate oxidation in the full, canonical TCA cycle and which output(s) of the canonical TCA cycle support cell fitness.

RESULTS

Mitochondrial respiration activates the canonical TCA cycle

We first asked what determines whether cells oxidize citrate through the canonical TCA cycle. TCA cycle engagement can be monitored by incubating cells with [U-¹³C]glucose, which generates m+2-labeled citrate following oxidative decarboxylation of glucose-derived pyruvate (Figure 1A). Further metabolism through the first turn of the canonical TCA cycle produces m+2-labeled TCA cycle intermediates. Accordingly, the ratio of malate m+2 relative to citrate m+2 (mal+2/cit+2) reflects the degree to which citrate progresses through the canonical TCA cycle. Alternative configurations of the TCA cycle, such as the citrate-malate shuttle⁸ or glutamine-driven anaplerosis, will reduce the mal+2/cit+2 ratio. [U-¹³C]glucose tracing in 82 non-small cell lung cancer (NSCLC) lines¹³ demonstrated that cells exhibit a wide range of mal+2/cit+2 ratios even when grown in identical culture conditions (Figure 1B). Correlating mal+2/cit+2 ratios with gene expression data collated by the DepMap Project¹⁴ in 69 NSCLC lines revealed that genes encoding components of the ETC are highly enriched among genes correlating positively with the mal+2/cit+2 ratio (Figures 1C and S1A). The ETC accepts reducing equivalents generated by the TCA cycle and ultimately reduces oxygen. Accordingly, oxygen consumption serves as a readout of flux through the ETC. Consistently, oxygen consumption correlated with mal+2/cit+2 ratios in 11 NSCLC lines (Figure 1D). These results indicate that mitochondrial respiration correlates with carbon retention in the canonical TCA cycle, consistent with Krebs’s discovery that pyruvate oxidation closely mirrors cellular oxygen consumption.¹⁵

Figure 1. Mitochondrial respiration determines canonical TCA engagement.

(A) Schematic of [U-¹³C]glucose labeling. m+2-labeled citrate generated from pyruvate oxidation is further oxidized to m+2-labeled malate in the canonical TCA cycle.

(B) Fractional m+2 enrichment of malate relative to m+2 citrate (mal+2/cit+2) from 82 NSCLC lines cultured with [U-¹³C]glucose for 6 h, obtained from a published dataset.¹³

(C) Bubble plot of gene set enrichment analysis depicting gene sets enriched among genes positively correlating with the mal+2/cit+2 ratio in 69 NSCLC lines.

(D) Simple linear regression comparing mitochondrial oxygen consumption rates (average of 6–8 replicates/cell line) and mal+2/cit+2 from [U-¹³C]glucose (average of 3 replicates/cell line). Correlation was determined using Pearson (r).

(E) Mal+2/cit+2 from [U-¹³C]glucose in A549 cells treated with indicated concentrations of phenformin for 24 h. b.l.q, fractional enrichment below the limit of quantification.

(F and G) Mal+2/cit+2 from [U-¹³C]glucose in NSCLC lines treated with vehicle or 1 μM FCCP for 24 h (F) or expressing either empty vector or mitoLbNOX (G).

Data are mean ± SD with n = 3 independent replicates (E–G). Significance was assessed using two-way ANOVA with Sidak’s multiple-comparisons post-test relative to vehicle (F) or empty vector-expressing cells (G).

See also Figure S1 and Table S1.

To determine whether oxygen consumption is a cause or consequence of flux through the canonical TCA cycle, we titrated flux through the ETC using phenformin, a complex I inhibitor.^16,17 In line with work demonstrating that complex I activity enables entry of glucose-derived carbons into the TCA cycle in tumors in vivo,¹⁸ citrate m+2 labeling from glucose progressively declined alongside mitochondrial oxygen consumption following phenformin treatment in cultured cells (Figures S1B and S1C). Notably, malate m+2 labeling decreased disproportionally to citrate m+2 labeling, resulting in a dose-dependent decrease in the mal+2/cit+2 ratio following phenformin treatment (Figure 1E). The decline in the mal+2/cit+2 ratio was not due to label dilution from glutamine anaplerosis or increasing malate pool sizes, as both glutamine anaplerosis and overall TCA cycle metabolite pool sizes decreased with phenformin treatment as expected^7,19 (Figures S1D and S1E). Reciprocally, increasing electron flow through the ETC using carbonyl cyanide 4-(trifluoromethoxy)-phenylhydrazone (FCCP), a proton ionophore, enhanced citrate m+2 labeling from glucose and elevated the mal+2/cit+2 ratio in 6 NSCLC lines (Figures 1F and S1F). Importantly, phenformin reversed the increased mal+2/cit+2 triggered by FCCP, indicating that the effect of FCCP on TCA cycle metabolism is driven by increased ETC flux rather than reduced mitochondrial membrane potential (Figure S1G). Thus, both entry of glucose-derived carbon into the TCA cycle and retention of glucose-derived carbon within the TCA cycle depend upon electron flow through the ETC.

Complex I of the ETC accepts reducing equivalents from NADH, regenerating NAD⁺. To determine whether mitochondrial NAD⁺ regeneration drives TCA cycle activity, we expressed mitochondrially targeted Lactobacillus brevis NADH oxidase (mitoLbNOX), which catalyzes NADH-dependent oxygen reduction to regenerate NAD⁺ ,²⁰ in 6 NSCLC lines (Figure S1H). Cells exhibited a range of ETC-independent oxygen reduction, reflecting mitoLbNOX oxidase activity, that correlated with transgene expression (Figures S1H and S1I). Consistently, mitoLbNOX increased citrate m+2 from glucose and elevated the mal+2/cit+2 ratio in the four lines with the highest mitoLbNOX expression (Figures 1G and S1J). Of note, changes in the mal+2/cit+2 ratio were not driven by decreasing glutamine anaplerosis or metabolite pool sizes, as both FCCP and mitoLbNOX tended to increase, rather than decrease, glutamine entry into the TCA cycle and overall levels of TCA cycle metabolites (Figures S1K, S1L, S1N, and S1O). Accordingly, the overall amount of m+2-labeled malate increased alongside the mal+2/cit+2 ratio in all scenarios (Figures S1M and S1P). Thus, mitochondrial NAD⁺ regeneration determines not only entry of glucose-derived carbon into the TCA cycle but also retention within the canonical TCA cycle.

Pyruvate oxidation dictates carbon retention in the TCA cycle

Multiple TCA cycle enzymes (pyruvate dehydrogenase [PDH], isocitrate dehydrogenase 3, oxoglutarate dehydrogenase, and malate dehydrogenase 2) require NAD⁺ as an essential cofactor, raising the question of which NAD⁺ -responsive nodes control metabolite retention within the TCA cycle (Figure 2A). Unique among the NAD⁺ -dependent TCA cycle dehydrogenases, PDH is controlled by NAD⁺ both as a cofactor and as a determinant of phosphorylation by a family of PDH kinases (PDKs). As mitochondrial NAD⁺ regeneration declines, the association between PDH and PDKs increases, favoring inhibitory phosphorylation on any 1 of 3 serine residues on PDH.^21–24 Consistently, phenformin increased PDH phosphorylation in two NSCLC lines (Figure 2B), although the impact on specific residues was context-specific, in line with the known variability in residue preferences among the four differentially expressed PDK isoforms.^25–29 For example, A549 cells do not express PDK1, the exclusive kinase for serine 232,²⁵ which accordingly remains unphosphorylated in this cell line (Figures 2B and S2B). As expected, mitoLbNOX expression reduced basal PDH phosphorylation and prevented phenformin-induced phosphorylation²⁰ (Figure 2B). Accordingly, whereas phenformin blocked PDH flux—assessed as the generation of citrate m+2 from glucose-derived carbons—this effect was eliminated in mitoLbNOX-expressing cells, demonstrating that enforced mitochondrial NAD⁺ regeneration is sufficient to maintain PDH activity despite complex I inhibition, consistent with known regulatory mechanisms^18,30 (Figure 2C).

Figure 2. Pyruvate oxidation dictates downstream carbon retention in the canonical TCA cycle.

(A) Schematic depicting pyruvate dehydrogenase (PDH) regulation by NAD⁺/NADH and PDH kinase 1 (PDK1). DCA, dichloroacetate; MPCi, mitochondrial pyruvate carrier inhibition.

(B) Immunoblot of empty vector- and mitoLbNOX-expressing cells treated with vehicle or 2.5 μM phenformin for 24 h.

(C) Fractional m+2 enrichment of citrate from [U-¹³C]glucose in empty vector- and mitoLbNOX-expressing Calu1 cells treated with vehicle or 2.5 μM phenformin for 24 h.

(D and E) Fractional m+2 enrichment of citrate (D) and mal+2/cit+2 (E) from [U-¹³C]glucose in Calu1 cells expressing either control sgRNA (sgAAVS1) or sgRNA targeting PDK1 treated with vehicle or 2.5 μM phenformin for 24 h. b.l.q, fractional enrichment below the limit of quantification.

(F) Citrate pools plotted against mal+2/cit+2 from [U-¹³C]glucose in 16 NSCLC lines, each averaged across three replicates. Correlation was determined using Pearson (r).

(G) Metabolites ranked by Pearson correlation (r) comparing levels of 225 metabolites (DepMap) and mal+2/cit+2 from [U-¹³C]glucose in 60 NSCLC lines from a published dataset.¹³

(H) Simple linear regression comparing the fold change in citrate levels and the change in mal+2/cit+2 from [U-¹³C]glucose with mitoLbNOX expression compared with empty vector (average of 3 replicates per line). Correlation was determined using Pearson (r).

(I) Log₂ (fold change) of indicated metabolites in Calu1 PDK1-edited cells relative to the average of control cells (sgAAVS1), shown in triplicate.

(J) Mal+2/cit+2 from [U-¹³C]glucose in A549 and Calu1 cells.

Data are mean ± SD with n = 3 (C, I, and J), n = 6 (D), or n = 5–6 (E) independent replicates. Significance was assessed using two-way ANOVA with Sidak’s multiple-comparisons post-test relative to vehicle (C and J) or to sgAAVS1 control cells (D and E).

See also Figure S2.

Given the notable sensitivity of PDH to NAD⁺, we asked whether PDH activity was sufficient to determine not just entry of carbons into the TCA cycle but also downstream carbon retention in the canonical TCA cycle. Pharmacologic inhibition of all PDKs with dichloroacetate increased the mal+2/cit+2 ratio. Reciprocally, reducing pyruvate entry into mitochondria by inhibiting the mitochondrial pyruvate carrier (MPC) decreased the mal+2/cit+2 ratio in NSCLC cells, consistent with our previous results in mouse embryonic stem cells (mESCs) and myoblasts⁸ (Figure S2A). Across NSCLC lines, PDK1 expression significantly negatively correlated with the mal+2/cit+2 ratio, raising the possibility that PDK1 activity limits canonical TCA cycle activity in NSCLC cells (Figure S2B). As expected, PDK1 editing increased fractional m+2 labeling of citrate from [U-¹³C]glucose, indicative of enhanced PDH activity following PDK1 loss, and elevated the mal+2/cit+2 ratio without affecting glutamine anaplerosis, reflecting increased canonical TCA cycle engagement (Figures 2D, 2E, S2C, and S2D). Consistent with increasing overall substrate oxidation through the TCA cycle, disrupting PDK1 also increased mitochondrial oxygen consumption (Figure S2E).

To determine whether PDK1 mediates the impact of mitochondrial NAD⁺ regeneration on the canonical TCA cycle, we treated control or PDK1-deficient cells with phenformin to block NAD⁺ regeneration. Phenformin treatment eliminated production of m+2-labeled citrate in control cells but not PDK1-deficient cells (Figure 2D). Whereas phenformin dramatically reduced the mal+2/cit+2 ratio in control cells, PDK1-deficient cells retained mal+2/cit+2 ratios comparable to control cells despite phenformin treatment (Figure 2E), indicating that persistent PDH flux is sufficient to maintain canonical TCA cycle activity despite impaired mitochondrial NAD⁺ regeneration.

How could PDH, which lies upstream of the canonical TCA cycle, affect whether citrate is captured for oxidation in the canonical TCA cycle? Notably, the enzyme that initiates citrate catabolism in the canonical TCA cycle, aconitase 2 (ACO2), is highly reversible, with equilibrium constants favoring citrate production over citrate consumption at a 10:1 ratio.^31,32 We therefore hypothesized that interventions that increase citrate production would favor forward flux through the canonical TCA cycle. Suggestively, total citrate levels—but not levels of other TCA cycle intermediates—correlated with canonical TCA cycle activity across 16 NSCLC cell lines (Figures 2F and S2F). More broadly, in a dataset of 225 metabolites generated through the DepMap project, citrate emerged as one of the metabolites most closely correlated with the mal+2/cit+2 ratio across 60 NSCLC lines¹³ (Figure 2G). We therefore asked whether interventions that increase canonical TCA cycle usage also increase citrate production. Indeed, mitoLbNOX expression increased citrate pools, with the increase in citrate closely correlating with the increase in the mal+2/cit+2 ratio (Figure 2H). Specifically increasing PDH activity through PDK1 deletion also increased citrate pools, while having a more modest effect on other TCA cycle intermediates (Figure 2I). Furthermore, simply adding pyruvate to induce flux through PDH^33,34 was also sufficient to increase citrate pools and stimulate canonical TCA cycle activity (Figures 2J and S2G). Thus, mitochondrial NAD⁺ regeneration sets pyruvate oxidation to citrate, ultimately driving forward carbon capture and retention in the complete TCA cycle.

ACO2 supports fitness by clearing citrate

We next asked whether enhanced forward flux through the canonical TCA cycle contributes to cell fitness. To disrupt citrate catabolism in the canonical TCA cycle, we engineered ACO2 loss in two representative NSCLC lines (Figure S3A). At baseline, ACO2 loss only modestly increased cellular citrate pools in A549 cells, which exhibit one of the highest mal+2/cit+2 ratios (indicating high canonical TCA cycle use) among NSCLC cell lines but not Calu1 cells, which harbor a low mal+2/cit+2 ratio (Figure 3A). Furthermore, ACO2 deletion had no significant effect on proliferation in either cell line (Figure S3B). The muted effect of ACO2 loss on citrate pools and cell proliferation is consistent with the notion that under conventional culture conditions, cells exhibit relatively limited canonical TCA cycle engagement.

Figure 3. ACO2 activity supports fitness by clearing citrate.

(A and B) Citrate levels (A) and population doublings (B) in cells expressing indicated sgRNA.

(C and D) Population doublings (C) and citrate levels (D) of control (sgAAVS1, –) or ACO2-edited cells (+) that additionally express either control sgRNA (sgAAVS1) or sgRNA targeting CS, treated as indicated for 72 h (C) or 24 h (D).

(E and F) Schematic depicting competition experiment (left). A549 cells with sgAAVS1 expressing mCherry were mixed 1:1 with cells expressing BFP and edited with control (sgAAVS1), sgACO2, sgCS, or both sgACO2 and sgCS and assessed over time (E) or assessed after 21 days in culture with vehicle or indicated concentrations of pyruvate (F).

(G) Population doublings of ACO2-edited cells expressing empty vector (–) or SLC25A1 cDNA (+).

Data n = 1 (E) or are mean ± SD with n = 3 (A–D, F, and G) independent replicates. Significance was assessed using two-way ANOVA with Sidak’s multiple-comparisons post-test relative to sgAAVS1 control cells (A), relative to vehicle (C, D, and F) or relative to sgACO2/empty vector cells (G).

See also Figures S3 and S4.

To increase mitochondrial citrate production and canonical TCA cycle use, we supplemented cells with pyruvate, which led to a modest 1.3–1.6-fold increase in citrate in control cells. By contrast, pyruvate supplementation in ACO2-deficient cells induced a 3-fold increase in citrate, reflecting citrate buildup following ACO2 loss (Figure 3A), consistent with the model that forward flux through ACO2 is activated by pyruvate. Pyruvate supplementation was also sufficient to induce ACO2 dependence: ACO2-deficient A549 and Calu1 cells exhibited significant growth defects when grown with pyruvate (Figure 3B), and these effects were reversed following re-expression of sgRNA-resistant cDNA encoding Aco2 (Figures S3C–S3E). Citrate accumulation and ACO2 dependence were not simply a consequence of providing additional nutrients to the culture medium: equimolar supplementation of alanine only modestly increased citrate pools and induced no fitness defect in ACO2-deficient cells, consistent with the fact that alanine is more reduced than pyruvate³⁵ and does not induce canonical TCA cycle engagement (Figure S3F). These effects were not limited to cancer cells, as ACO2-deficient mouse embryonic fibroblasts likewise demonstrated enhanced citrate accumulation and reduced proliferation when provided exogenous pyruvate (Figures S3G and S3H). Similarly, ESCs lacking ACO2 exhibited notable growth deficits when cultured in medium containing 1 mM pyruvate, a standard additive to most stem cell media formulations³⁶ (Figures S3I and S3J).

The vulnerability of ACO2-deficient cells to pyruvate supplementation was surprising because pyruvate is a common cell culture additive that compensates for mitochondrial dysfunction by serving as an electron acceptor.^19,37 To determine whether pyruvate induces dependency upon ACO2 because of its contribution to cytosolic redox maintenance or because of its role as a substrate for the TCA cycle, we inhibited mitochondrial pyruvate import with UK5099.³⁸ MPC inhibition blocked the effects of pyruvate supplementation, preventing both citrate accumulation and growth inhibition in ACO2-deficient cells (Figures S3K and S3L). These results demonstrate that increasing pyruvate entry into the TCA cycle increases dependence upon ACO2 to support cell fitness.

We next asked why mitochondrial pyruvate metabolism imposes reliance on the canonical TCA cycle. Because pyruvate induces production of mitochondrial citrate, whose catabolism requires ACO2 (Figure 3A), we asked whether citrate production is sufficient to induce ACO2 dependence. To this end, we deleted citrate synthase (CS) to prevent mitochondrial citrate production from pyruvate in control and ACO2-deficient NSCLC cells (Figure S3M). Neither control nor ACO2-deficient cells exhibited growth defects following CS deletion in control conditions (Figure 3C). As expected, CS deletion prevented the accumulation of citrate following pyruvate addition in ACO2-deficient cells (Figure 3D). Strikingly, CS deletion completely reversed the growth defects induced by pyruvate in ACO2-deficient cells (Figure 3C). These results demonstrate that functions downstream of ACO2, including provision of reducing equivalents or downstream TCA cycle intermediates, do not drive fitness defects in ACO2-deficient cells. Rather, mitochondrial citrate production itself is sufficient to impose dependence on ACO2 in the context of high nutrient oxidation.

To test if citrate production in excess of citrate clearance impairs fitness, we determined the effect of ACO2 disruption on cellular fitness over a longer time period. When mixed 1:1 with cells transduced with a control sgRNA, cells transduced with sgACO2 exhibited no competitive disadvantage (Figure 3E). By contrast, pyruvate addition selectively led to the ACO2-disrupted cells being consistently outcompeted by controls over time, and co-deletion of CS completely rescued the fitness of ACO2-deficient cells in the presence of pyruvate (Figure 3E). Across a range of concentrations, including those reported in normal circulation,³⁹ pyruvate led to a dose-dependent competitive disadvantage in ACO2-deficient cells that was reversed by CS co-deletion (Figure 3F). These findings demonstrate that ACO2 activity supports cell fitness and growth by preventing citrate accumulation.

Citrate in the cytosol can allosterically inhibit phosphofructokinase, a rate-limiting enzyme in glycolysis,^40,41 and therefore could lead to a bioenergetic deficit and impaired proliferation. Indeed, citrate accumulation coincided with a reduced extracellular acidification rate, consistent with impaired glycolysis in pyruvate-supplemented ACO2-deficient cells (Figure S3O). To determine if citrate accumulation predominantly exerts antiproliferative effects in the cytosol, we overexpressed SLC25A1, the mitochondrial antiporter that mediates citrate efflux from mitochondria (Figure S3N).^42,43 SLC25A1 overexpression partially restored the proliferation of pyruvate-treated ACO2-deficient cells despite persistent inhibition of the extracellular acidification rate (Figures 3G and S3O). These results uncouple the effects of citrate on glycolysis and proliferation and indicate that mitochondrial citrate efflux rescues ACO2-deficient cells.

SLC25A1 transports citrate along a concentration gradient.⁴² Accordingly, we tested whether interventions that increase cytosolic citrate consumption alleviate the impact of ACO2 loss. Like SLC25A1 overexpression, overexpression of cytosolic aconitase (ACO1) modestly improved proliferation of ACO2-deficient cells in response to pyruvate (Figures S4A and S4B). Reciprocally, ACO1 depletion exacerbated the impact of ACO2 loss specifically in the context of pyruvate supplementation (Figures S4C and S4D). Cytosolic citrate can also be metabolized by ATP citrate lyase (ACL), which produces oxaloacetate and acetyl-CoA to fuel de novo lipogenesis and histone acetylation. Depriving cells of exogenous lipids present in serum to increase demand for de novo lipogenesis^44,45 was sufficient to mitigate the impact of pyruvate on ACO2-deficient cells (Figure S4E). Notably, so long as ACO2 remained intact, neither ACL nor ACO1 depletion induced citrate accumulation or growth defects, even in the presence of exogenous pyruvate (Figures S4D and S4F–S4I). These observations demonstrate that ACO2 activity prevents detrimental accumulation of citrate within mitochondria and that increasing cytosolic citrate catabolism improves cell fitness when mitochondrial citrate metabolism is impaired.

Citrate accumulation activates the ISR

To understand the impact of citrate accumulation, we assessed parameters of mitochondrial function in ACO2-deficient cells. ACO2 loss had no reproducible impact on mitochondrial mass, membrane potential, or oxygen consumption, even in the presence of exogenous pyruvate (Figures S5A–S5C). We therefore took an unbiased approach to identify consequences of ACO2 loss. RNA sequencing revealed that ACO2-deficient cells exhibited a unique transcriptional response to pyruvate supplementation that was reversed by CS co-deletion (Figure 4A). Gene set enrichment analysis of transcripts upregulated in pyruvate-treated ACO2-deficient cells relative to controls demonstrated consistent enrichment of gene sets related to the integrated stress response (ISR) (Figures 4B and S5D). Indeed, ISR-related genes were among the most induced genes in pyruvate-treated ACO2-deficient cells, and the induction of ISR genes was reversed by CS deletion (Figures 4C and S5E).

Figure 4. Citrate accumulation activates the ISR.

(A) Principal component (PC) analysis of the top 500 genes with the highest row variance from RNA sequencing data of A549 cells with control (sgAAVS1) or ACO2 editing with additional control (sgAAVS1) or CS editing.

(B) Dot plot depicting Reactome gene sets enriched among genes increased in ACO2-edited cells treated with pyruvate compared with vehicle.

(C) Violin plot depicting Z scores of ISR-related genes⁴⁶ in A549 cells with control (sgAAVS1; –) or ACO2 editing (+) with additional control (sgAAVS1; –) or CS editing (+) cultured in vehicle (–) or pyruvate (+) for 24 h.

(D) Immunoblot of A549 cells with control (sgAAVS1; –) or ACO2 editing.

(E) Median fluorescent intensity (MFI) depicting L-HPG incorporation of A549 cells with control (sgAAVS1) or ACO2 editing cultured as indicated for 24 h. Dotted line depicts L-HPG incorporation in cells treated with cycloheximide (CHX).

(F) MFI depicting OPP incorporation in A549 cells with control (sgAAVS1) or ACO2 editing cultured as indicated for 48 h. The dotted line depicts OPP incorporation in cells treated with CHX. Values are multiplied by 10E− 4.

(G) Immunoblot of A549 cells with control (sgAAVS1; –) or ACO2 editing (+) with additional control (sgAAVS1; –) or CS editing cultured as indicated for 24 h. (H and I) Percent EdU-positive cells (H) and cell cycle distribution (I) in A549 cells with control (sgAAVS1; –) or ACO2 editing cultured as indicated for 48 h. ****p < 3E− 6 (I).

RNA sequencing data reflect n = 2 independent replicates, shown as the average frequency distribution in (C). Data are mean ± SD with n = 2 (H for just sgAAVS1 + vehicle condition) or n = 3 (E, F, H, and I) independent replicates. Significance was assessed using two-way ANOVA with Sidak’s multiple-comparisons post-test (C), relative to vehicle treatment (E and H) or relative to sgAAVS1+pyruvate (I).

See also Figure S5.

The ISR is initiated by phosphorylation of eukaryotic initiation factor 2 subunit alpha (eIF2α), leading to global reduction of mRNA translation via inhibition of ternary complex assembly and selective translation of transcripts such as activating transcription factor 4 (ATF4).^47,48 Consistent with ISR activation, eIF2α was phosphorylated only in ACO2-deficient cells treated with pyruvate (Figure 4D). Accordingly, pyruvate-treated ACO2-deficient cells exhibited notable decreases in global translation, measured by the incorporation of the methionine analog L-homopropargyl glycine (L-HPG), and this phenotype was completely rescued by CS co-deletion (Figures 4E and S5F). Reduced translation started within hours of pyruvate addition, co-occurred with the peak of intracellular citrate, and was sustained through up to 48 h of pyruvate treatment (Figures S5G and S5H). Incorporation of O-propargyl puromycin (OPP), a puromycin analog incorporated into nascent polypeptides, was likewise reduced in ACO2-deficient cells treated with pyruvate (Figure 4F). Furthermore, pyruvate treatment increased ATF4 protein expression in ACO2-deficient cells, and ATF4 induction was reversed by CS co-deletion or SLC25A1 overexpression (Figures 4G, S5I, and S5J). Together, these data demonstrate that mitochondrial citrate accumulation is associated with ISR activation.

ISR activation can inhibit cell cycle progression by impeding synthesis of cell cycle proteins and by ATF4-dependent activation of cell cycle inhibitors.^49–54 Overall, pyruvate supplementation had no overt impact on cell viability but instead reduced proliferation, as assessed by 5-ethynyl-2’-deoxyuridine (EdU) incorporation (Figures 4H and S5K). Reduced proliferation was associated with fewer cells in S and G2/M and more cells accumulating in G1 and was completely reversed by CS co-deletion (Figures 4I and S5L). The impact of pyruvate on the cell cycle progression of ACO2-deficient cells was phenocopied by treatment with cycloheximide, a translation elongation inhibitor (Figure S5L). Together, our data support a model where citrate production in excess of clearance in the mitochondria triggers ISR activation, global inhibition of translation initiation, and subsequent impairment of cell cycle progression.

Citrate uptake leads to ACO2 dependence

To determine whether ACO2 limits citrate accumulation in vivo, we generated a mouse model of conditional Aco2 deficiency (Aco2^fl/fl ) (Figure S6A) and crossed Aco2^fl/fl animals to animals expressing tamoxifen-inducible Cre-recombinase expression driven by the human ubiquitin C promoter (UBC-Cre-ERT2). Ten days following tamoxifen administration, kidney, heart, lung, liver, skeletal muscle, and brain exhibited a 74%–98% decline in Aco2 expression (Figure S6B) accompanied by increased citrate levels in the serum and in select tissues, reaching the highest levels in the kidney (Figures 5A and 5B). Other TCA cycle metabolites were largely unchanged, underscoring the specific role of ACO2 in preventing citrate accumulation (Figure S6C). To examine the physiological consequences of Aco2 disruption, we subjected tissues harvested approximately 3 weeks post-tamoxifen-induced recombination to blinded pathological examination. While the brain, liver, lung, muscle, and heart displayed no overt histopathological abnormalities, kidneys from Aco2^fl/fl mice displayed a striking pattern of vacuolar degeneration in epithelial cells located in the cortex (Figures 5C, 5D, and S6D). Furthermore, kidneys from Aco2^fl/fl mice were strongly positive for neutrophil gelatinase-associated lipocalin (NGAL), a marker of renal tubular injury,^55,56 indicating that ACO2 loss induces kidney tissue damage (Figures 5C and 5D).

Figure 5. ACO2 deficiency leads to citrate accumulation and kidney damage in vivo.

(A and B) Citrate concentration in serum (A) and citrate levels in tissues (B) from indicated mice 10 days post tamoxifen administration (serum: n = 11 mice/genotype; other tissues: n = 5–11 mice/genotype).

(C and D) Representative hematoxylin & eosin (top) and immunohistochemistry for NGAL (bottom) in kidneys from indicated mice 20–25 days post tamoxifen administration (C). Black arrows indicate areas of renal tubular degeneration. Quantification of tubular pathology and NGAL immunolabeling (D) (n = 6 kidneys).

(E) Relative metabolite levels in kidneys of indicated mice harvested 21 days post tamoxifen administration (n = 9 Aco2^+/+ and n = 8 Aco2^fl/fl).

(F) Quantification of p-eIF2α relative to total eIF2α from immunoblot of kidneys harvested 21 days post tamoxifen administration, shown in Figure S6F (n = 6 Aco2^+/+ and n = 7 Aco2^fl/fl).

(G) Gene expression of selected ATF4 target and proximal tubule cell marker genes in kidneys harvested 21 days post tamoxifen administration (n = 9 Aco2^+/+ and n = 8 Aco2^fl/fl).

Data are mean ± SD with independent replicates as indicated for each graph. Significance was assessed using an unpaired two-tailed t test (A and F), a two-way ANOVA with Sidak’s multiple-comparisons post-test relative to Aco2^+/+ control (B, E, and G), or Fisher’s exact test (D).

See also Figure S6.

Further examination of kidneys at the point when pathology was observed revealed that citrate levels—increased by approximately 2-fold at day 10—reached 8.5-fold higher at day 21 (Figure 5E). This increase in kidney citrate was mirrored by a progressive increase in circulating citrate over time (Figure S6E). Notably, other TCA cycle metabolites remained unchanged in the kidney even at this late time point (Figure 5E). Citrate accumulation was associated with a significant increase in p-eIF2α relative to total eIF2α in kidneys from Aco2^fl/fl mice (Figures 5F and S6F). Consistent with ISR activation, many ATF4 target genes, including the kidney injury marker Lcn2 (encoding NGAL),⁵⁷ were significantly elevated in ACO2-deficient kidneys (Figure 5G). The observed induction of ATF4 target genes coincided with a downward trend in genes associated with proximal tubule identity, a common molecular signature of kidney injury.^{58 ,59}

The kidney is the only organ that exhibits net uptake of citrate from the circulation, and citrate is a major substrate for the TCA cycle uniquely in the kidney.^60,61 Consistently, both ACO2 and SLC13A2, the high-affinity concentrative citrate transporter, are highly expressed in proximal tubule cells^62,63 primarily situated in the renal cortex (Figure S7A). Although the liver expresses SLC13A5, a lower-affinity citrate transporter,^64,65 this is not sufficient to induce net citrate uptake under physiologic conditions.^60,61 The unique uptake and catabolism of citrate in proximal tubule cells raises the possibility that citrate uptake drives dependence upon ACO2 for citrate clearance. To determine whether circulating citrate increases dependence on ACO2, we supplemented drinking water with 3% sodium citrate or pH- and molarity-matched sodium chloride. Whereas control mice exhibited a non-significant trend toward increased circulating citrate when provided citrate-supplemented drinking water, ACO2-deficient mice dramatically increased circulating citrate under these conditions, suggesting that ACO2 is required for systemic citrate clearance (Figure 6A). Moreover, citrate-supplemented drinking water halved the lifespan of ACO2-deficient animals from 22 to 10.5 days, while having no impact on the lifespan of control mice on the timescales analyzed (Figure 6B). Together, these results support the model that increasing citrate levels impose dependence upon ACO2 for citrate clearance in vivo.

Figure 6. Citrate uptake increases reliance on ACO2.

(A) Serum citrate concentration in mice provided drinking water supplemented with NaCl (control) or 3% citrate 10 days post tamoxifen administration (control water: n = 12 Aco2^+/+, 11 Aco2^fl/fl; citrate water: n = 14 Aco2^+/+, 10 Aco2^fl/fl).

(B) Survival curves of indicated mice post tamoxifen (TAM) administration (control water: n = 4 Aco2^+/+, 9 Aco2^fl/fl; citrate water: n = 9 Aco2^+/+, 12 Aco2^fl/fl).

(C) Violin plot depicting Z scores of 122 ISR-related genes⁴⁶ in kidneys harvested from mice provided regular water or water supplemented with 3% citrate 10 days post tamoxifen administration (n = 6 for all conditions except n = 7 for Aco2^fl/fl mice on 3% citrate).

(D) Normalized enrichment scores of gene sets related to kidney injury or proximal tubule (PT) identity from RNA sequencing of kidneys described in (C), comparing Aco2^fl/fl vs. Aco2^+/+ animals on regular water (left column) or 3% citrate (right column). Selected gene sets represent genes increased or decreased following acute kidney injury (AKI),⁶⁶ associated with AKI in PT cells specifically,⁶⁷ or clusters associated with normal PT identity (segments 1–3) or injury.⁵⁹

(E) SLC13A2 and SLC13A5 expression in cancer cell lines of kidney (n = 51) or liver (n = 27) lineage (TPM, transcripts per million).

(F) Citrate levels in HepG2 cells cultured as indicated for 24 h.

(G and H) Percent DAPI-positivity (G) or population doublings (H) in HepG2 cells cultured as indicated for 3 days.

(I) Percent DAPI-positive cells of indicated genotype cultured in 5 mM (A549) or 1 mM (Calu1) citrate for 3 days.

(J) Cellular competition experiment mixing ACO2-edited cells expressing GFP (“sgACO2-GFP”) 1:1 with cells expressing SLC13A2 cDNA and either sgAAVS1 or sgACO2 and passaged with vehicle or indicated concentrations of citrate for 21 days. Shown are the percent changes in green fluorescent protein (GFP)-negative cells cultured with citrate relative to vehicle at each time point.

(K) Immunoblot of cells with control (sgAAVS1; –) or ACO2 editing (+) expressing EGFP (–) or SLC13A2 cDNA (+), cultured as indicated for 24 h.

Data are n = 1 (J) or mean ± SD with n = 3 (F–I) independent replicates. Significance was assessed using two-way ANOVA with Sidak’s multiple-comparisons post-test as indicated (C), relative to control water (A), relative to vehicle (F–H), or relative to sgAAVS1 + SLC13A2-expressing cells (I), using log-rank (Mantel-Cox) test (B), or using the fgsea package in R (D).

See also Figure S7.

Within the kidney, citrate levels increased in both control and ACO2-deficient mice exposed to citrate-supplemented drinking water (Figure S7B). RNA sequencing revealed that 10 days post-recombination, ACO2 deficiency induced expression of ISR-related genes (Figure 6C). Citrate supplementation enhanced this effect, and the ISR gene signature was most strongly activated in kidneys from ACO2-deficient mice provided citrate-supplemented drinking water (Figure 6C). Unbiased analysis of gene sets enriched in ACO2-deficient mice relative to control mice supplemented with citrate revealed ISR-related gene sets among the most significantly enriched pathways (Figure S7C). Citrate supplementation also accelerated signatures of kidney injury: whereas ACO2-deficient mice exhibited a modest upregulation of gene sets associated with kidney injury relative to control mice 10 days following recombination, these signatures were exacerbated when mice were provided citrate-supplemented drinking water (Figure 6D). Specifically, genes associated with healthy proximal tubule cells were downregulated, while genes associated with acute kidney injury and injured proximal tubules were upregulated in ACO2-deficient kidneys relative to control kidneys following citrate supplementation (Figure 6D). Together, these results link citrate availability with ISR activation and signatures of kidney damage in ACO2-deficient mice.

To determine whether cell-autonomous citrate uptake is sufficient to increase ACO2 dependence, we examined DepMap for liver and kidney cancer cell lines expressing SLC13A5 or SLC13A2. While no kidney cancer lines expressed SLC13A2, HepG2 hepatocellular carcinoma cells exhibit robust SLC13A5 expression and demonstrated citrate uptake in vitro (Figures 6E and S7D). As with NSCLC cells, HepG2 cells exhibited a modest increase in citrate following ACO2 depletion, and pyruvate exacerbated the increase in citrate following ACO2 loss (Figures 6F and S7E). We therefore asked whether increasing citrate availability, either by pyruvate or citrate supplementation, was sufficient to induce ACO2 dependence. Indeed, while neither pyruvate nor citrate notably affected fitness in control cells, both pyruvate and citrate were each sufficient to induce cell death and dramatically impair proliferation in ACO2-deficient HepG2 cells (Figures 6G and 6H). To assess the contribution of mitochondrial citrate accumulation to the proliferation defects observed in the context of citrate uptake, we depleted SLC25A1 to block mitochondrial citrate import^68,69 (Figure S7F). Although SLC25A1 loss alone led to a proliferation defect, SLC25A1 co-deletion reversed the effect of ACO2 loss and increased the proliferation of citrate-supplemented cells, consistent with blocking the import of citrate into mitochondria (Figure S7G). These results demonstrate that cells harboring concentrative citrate transporters become reliant on ACO2 for mitochondrial citrate catabolism in the presence of extracellular citrate.

To determine whether citrate uptake is sufficient to induce ACO2 dependence, we expressed SLC13A2 in two ACO2-edited NSCLC cell lines that do not endogenously express citrate transporters and confirmed that SLC13A2 expression enabled citrate uptake in all cells (Figures S7H and S7I). In SLC13A2-expressing cells grown under control conditions (without citrate or pyruvate), ACO2 loss did not affect proliferation or viability, as expected (Figures 6I and S7J). Furthermore, SLC13A2 expression did not affect growth in control conditions but did impair viability and proliferation in the presence of extracellular citrate, demonstrating that enforcing citrate uptake alone decreases cell fitness (Figures 6I, S7J, and S7K). Calu1 cells appeared more sensitive to extracellular citrate than A549 cells, consistent with both higher levels of SLC13A2 expression and with the notion that lower canonical TCA cycle engagement results in lower tolerance for citrate accumulation (Figures 1D and S7H). Notably, when SLC13A2-expressing cells were provided citrate in the absence of ACO2, both A549 and Calu1 cells exhibited an additional 2-fold increase in death and concomitant loss of proliferation compared with conditions of citrate uptake alone (Figures 6I and S7K). Accordingly, whereas the capacity to take up citrate led to a modest competitive disadvantage only at supraphysiological levels of citrate, citrate uptake in the absence of ACO2 led to a pronounced dose-dependent competitive disadvantage across a range of citrate concentrations, including concentrations as low as 0.5 mM, which approaches physiologic levels in circulation⁷⁰ (Figure 6J). As with kidneys in vivo, engineered citrate uptake in the absence of ACO2 in vitro led to further accumulation of intracellular citrate and signatures of ISR activation, including stabilization of ATF4 and suppression of translation initiation (Figures 6K, S7L, and S7M). Thus, cell-autonomous citrate uptake is sufficient to increase dependence on ACO2.

DISCUSSION

The TCA cycle is most famous for its role in harnessing the reducing power of nutrients to fuel oxidative phosphorylation and providing essential metabolic intermediates. Here, we demonstrate that metabolite clearance is an additional essential function of the TCA cycle. Citrate produced in excess of its clearance through the canonical TCA cycle induces the ISR, suppresses translation, and impairs proliferation. The observation that CS deletion reverses ACO2 dependence demonstrates that functions downstream of ACO2 in the canonical TCA cycle—provision of reducing equivalents or substrates for alternative metabolic pathways—are not responsible for growth defects in the absence of ACO2. Rather, it is specifically the production of citrate within mitochondria that imposes reliance on the initial steps of the canonical TCA cycle in this context, although the specific outputs of the canonical TCA cycle that drive reliance on ACO2 likely vary by cell type and context. In yeast, aconitase or isocitrate dehydrogenase deficiency impairs growth on specific carbon sources, and these phenotypes are likewise rescued by CS co-deletion,⁷¹ suggesting that allowing cells to cope with citrate production may be a broadly conserved function of the canonical TCA cycle. Consistently, the kidney—a major site of citrate uptake and metabolism in mammals—demonstrated the most sensitivity to ACO2 loss in adult mice. While we cannot rule out that other tissues would likewise exhibit pathological changes over longer time scales, our results underscore the role of citrate accumulation in driving reliance on canonical TCA cycle metabolism, both in vitro and in vivo.

Furthermore, our results demonstrate that mitochondrial citrate production is carefully calibrated to keep pace with citrate clearance. The tight link between citrate production and mitochondrial NAD⁺ regeneration, enabled by exquisite regulation of PDH, ensures that citrate is only produced at high levels when cells harbor sufficient oxidative capacity to maintain forward TCA cycle flux and continued deposition of reducing equivalents onto the ETC. Because the equilibrium constant of ACO2 favors citrate production over consumption, forward flux through the conventional TCA cycle is only favored either under conditions where citrate production is high or where isocitrate and downstream metabolites remain low, whether because of impaired glutamine anaplerosis or increased flux through isocitrate or αKG dehydrogenases. In addition to forward flux through the canonical TCA cycle, citrate can also be exported to the cytosol via SLC25A1. In the cytosol, citrate can be cleaved by ACL in a thermodynamically favored reaction, producing acetyl-CoA for lipid synthesis and histone acetylation, and oxaloacetate, which contributes to cytosolic redox regulation.^8,72,73 Suggestively, SLC25A1’s affinity for citrate exceeds that of ACO2 by at least an order of magnitude (7.5–130 μM for SLC25A1^42,43,74,75 vs. 0.95–3.6 mM for ACO2^32,76,77 ), raising the possibility that SLC25A1 outcompetes ACO2 for citrate under basal conditions, ensuring citrate efflux from mitochondria and safeguarding cytosolic citrate metabolism. In this scenario, increased citrate production triggered by elevated NAD⁺ regeneration could saturate SLC25A1, trapping citrate within mitochondria for further metabolism through the canonical TCA cycle.

We find that mitochondrial citrate accumulation triggers the ISR and impairs cellular fitness. Why cells sense mitochondrial citrate remains to be determined, but citrate is both negatively charged and acidic, two properties that will impair the ability of mitochondria to maintain the electrochemical gradient that sustains myriad mitochondrial functions. Citrate also readily chelates divalent ions, including iron, calcium, magnesium, and zinc,^78–80 all of which play important roles in mitochondrial biology. High levels of cytosolic citrate can also be toxic, likely through glycolytic inhibition and/or ion chelation impairing signaling pathways.^81,82 In the cytoplasm, elevated citrate allosterically inhibits glycolysis, thus tuning down carbon entry into the TCA cycle in the setting of sufficient cytosolic citrate. Our findings raise the possibility that cells also sense imbalances in mitochondrial citrate production and oxidation. Whether ISR activation helps relieve citrate accumulation, analogous to allosteric inhibition of glycolysis, or otherwise adapt to the sequelae of citrate accumulation are important questions for future investigation.

Consistent with the notion that intracellular citrate is a metabolic liability, only select tissues, including liver and kidney, express plasma membrane citrate carriers that enable citrate uptake, raising the possibility that only these tissues are primed to cope with high citrate levels. Uptake of citrate by proximal tubule cells paired with sufficient TCA cycle oxidative activity provides a mechanism for organisms to use circulating citrate as a fuel for the energetically demanding process of nutrient reabsorption in the kidney, and citrate unused by the TCA cycle in proximal tubule cells is ultimately discarded into urine. In healthy individuals, citrate is excreted in urine at levels ranging from 1.7 to 4.1 mM, significantly exceeded only by urea and creatinine.⁸³ Although urinary citrate is known to antagonize kidney stone formation, the notable abundance of citrate in urine, coupled with the lack of citrate uptake by most tissues, raises the possibility that citrate is excreted at such high levels because it is also an organismal waste product. While the mechanisms enabling citrate efflux from cells and tissues remain to be determined, the presence of citrate at high levels in mouse and human serum suggests that some tissues either discard or excrete citrate into the circulation.

More broadly, these findings illustrate that cellular usage and dependency on the TCA cycle are highly context-specific. In patients, ACO2 mutations are associated with phenotypes ranging from lethal infantile cerebellar-retinal degeneration syndrome accompanying severe loss of enzyme function to isolated optic atrophy from hypomorphic mutations.^84,85 Whether intracellular citrate accumulation or other canonical TCA cycle outputs contribute to these phenotypes remains to be determined. Indeed, while there are likely tissues that rely on the oxidative TCA cycle for energy production, others may specifically require TCA cycle components to regulate metabolite levels—not only in supplying anabolic intermediates but also in clearing metabolites whose accumulation may otherwise be detrimental to cellular fitness. Thus, the complete suite of metabolic reactions that define the “citric acid cycle” may only be required in select contexts. Delineating the specific metabolic outputs that support cell fitness is a prerequisite to designing more effective treatments for diseases driven by mutations in TCA cycle components or leveraging metabolic liabilities to impair cancer growth.

Limitations of the study

Here, we focus on the role of citrate clearance as an essential function of ACO2. ACO2 likely plays many important roles, and further examination of Aco2^fl/fl animals will reveal the context-specific outputs of the TCA cycle that support the function of various cell types. Although only the kidney exhibited gross pathology in our analyses, we cannot exclude the possibility that other organs or cell types exhibit functional defects in the absence of ACO2. Indeed, UBC-Cre-ERT2 Aco2^fl/fl animals reached a humane endpoint 3 weeks following recombination, but which organ system(s) contribute to this deterioration remains to be determined. Future studies using tissue-specific Cre drivers will be required to assess the impact of ACO2 on the function of specific organs, such as the kidney, without the complication of systemic weight loss. Our study demonstrates that mitochondrial citrate induces ISR activation and impaired proliferation, and exogenous citrate likely also induces additional detrimental effects in cells through both mitochondrial and non-mitochondrial roles. How mitochondrial citrate activates the ISR and what concentration of mitochondrial citrate leads to fitness defects remain to be determined.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Lydia W.S. Finley (finleyl@mskcc.org).

Materials availability

All unique reagents generated in this study are available upon request from the lead contact with a completed materials transfer agreement.

Data and code availability

• RNA sequencing data have been deposited at GEO: GSE295123 and GSE295125.

• NSCLC gene expression data are available from the DepMap portal (https://depmap.org/portal/).

• Single-cell RNA sequencing data across cell types are available at the Human Protein Atlas (https://www.proteinatlas.org/about/download).

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice

All animal work was approved by Memorial Sloan Kettering Cancer Center (MSKCC) Institutional Animal Care and Use Committee (protocol no. 19-09-014, Finley PI) in compliance with all relevant ethical regulations. Animals were housed in a pathogen-free facility maintained at 21.5 ± 1 °C and relative humidity 30–70% under a 12 h/12 h light/dark cycle. Animals received a standard pellet diet and tap water ad libitum unless otherwise indicated.

Aco2^fl/fl animals were generated by Taconic/Cyagen. In brief, CRISPR/Cas9-based editing was used to insert loxP sites around exon 3 of the Aco2 gene. The gRNA targeting the Aco2 locus, the donor vector containing loxP sites, and Cas9 mRNA were co-injected into fertilized mouse eggs to generate targeted conditional knockout offspring (gRNA1: GCTGAAGCAAGGTAAGCGAAGGG; gRNA2: AGAGTTACGGGCGCCTGATGTGG). F0 founder animals were identified by PCR followed by sequence analysis and were bred to wild-type mice to test germline transmission and F1 animal generation. Animals were generated in the C57BL/6N background and backcrossed for at least 5 generations to C57BL/6J. To induce full body recombination, Aco2^fl/fl animals were crossed to B6.Cg-^{Ndor1Tg(UBC-cre/ERT2)1Ejb/1J} (“UBC-Cre-ERT2”) animals (Jackson Laboratory strain 007001). Experiments included roughly equal numbers of male and female animals and were pooled from multiple litters. Animals of both genotypes were cohoused. The following genotyping primer sequences were used:

Primer ID	Sequences 5’ → 3’
Aco2 floxed F	AGCCTTTCAAACAGTAGAGAAGCA
Aco2 floxed R	ACCGATGTTAATAGCTCAGCTCAC
UbiquitinC-CreERT2 F	GACGTCACCCGTTCTGTTG
UbiquitinC-CreERT2 R	AGGCAAATTTTGGTGTACGG
Cre wild-type F	CTAGGCCACAGAATTGAAAGATCT
Cre wild-type R	GTAGGTGGAAATTCTAGCATCATCC

Tamoxifen (Sigma, T5648) was dissolved in sterile sunflower oil (Sigma, S5007) containing 10% ethanol (v/v) and stored at − 20 °C for a maximum of 2 weeks. To induce full body Aco2 ablation, 6 to 7-week-old Aco2^fl/fl UBC-Cre-ERT2 and control Aco2^+/+ UBC-creERT2 mice were administered 50 mg/kg tamoxifen intraperitoneally for 5 consecutive days. For citrate drinking water experiments, animals were provided pH 7 and molarity-matched drinking water supplemented with 3% citrate (Sigma, W302600) or NaCl (Sigma, S9888) for 7 days prior to tamoxifen administration. Animals continuously received NaCl or 3% citrate water until experimental endpoint. Endpoint for survival experiments was weight loss >80% from weight at first dose of tamoxifen. Indicated organs were harvested at the indicated times post first tamoxifen dose.

Cell lines

Cell lines were kept at 37 °C and 5% CO₂. A549, Calu1, H1975, H1993, H2122, H2170, HepG2, and C3H/10T1/2 (MEFs) cell lines were obtained from ATCC. Calu3, PC9, H441, H1299, H1792, H1944, H2009, H2087, H2172 and H2347 were a gift from Gina DeNicola.⁹⁷ Mouse embryonic stem cells (ESCs) were previously generated from C57BL/6 × 129S4/SvJae F₁ male embryos.⁸⁷ ESCs were maintained on gelatin-coated plates in medium containing a 1:1 mix of DMEM (11965092; Gibco) and Neurobasal-A medium (A2477501; Gibco) supplemented with 10% dialyzed fetal bovine serum (dFBS; Gemini), 0.1 mM 2-mercaptoethanol, 20 mM D-glucose, 2 mM l-glutamine, 1,000 U ml⁻¹ LIF (Gemini). All other cells were cultured in DMEM supplemented with 10% fetal bovine serum (FBS; Gemini) or 10% delipidated FBS (Gemini, 900–123) and 4 mM L-glutamine. All cells routinely tested negative for mycoplasma. Where indicated, cells were treated with DMSO (Sigma) or water as vehicle control, phenformin (Sigma, P7045), FCCP (Sigma, C2920), UK-5099 (Tocris, 4186), sodium dichloroacetate (Sigma, 347795), sodium pyruvate (Sigma, P2256), L-alanine (Sigma, A7469), or sodium citrate dihydrate (Sigma, W302600). Experimental medium was refreshed every 48 h and sodium pyruvate and sodium citrate each used at a concentration of 5 mM unless otherwise stated.

METHOD DETAILS

DepMap data analysis

The malate m+2/citrate m+2 ratio was calculated from isotope tracing data generated by Chen et al.¹³ in which 82 NSCLC cell lines were incubated with [U-¹³C]glucose for 6 h. RNA-sequencing data (16,382 genes, release 23Q4) was downloaded from the DepMap project (Broad Institute). Pearson correlation coefficients were calculated to describe the correlation between expression of each gene and the mal+2/cit+2 ratio across 69 overlapping cell lines. Genes were ranked by their Pearson correlation coefficient (r) and gene set enrichment analysis was performed using the ranked list with the fgsea package⁹⁸ (version 1.28.0) and Hallmark and Reactome gene sets using the msigdbr package (version 7.5.1) in R (version 4.3.2). For correlation of metabolites with the malate m+2/citrate m+2 ratio, metabolomics data was downloaded from DepMap (225 metabolites; release 23Q4) and Pearson correlation coefficients were calculated using the cor function in R to describe the correlation between levels of each metabolite and the mal+2/cit+2 ratio across 60 NSCLC lines. Data was exported and graphed in Prism version 10.3.1 to obtain a P value for correlation between citrate and mal+2/cit+2.

Generation of genetically edited lines

sgRNA sequences targeting PDK1, ACO2, ACLY, Aco2 or a safe harbor locus (human AAVS1, mouse Chr8) were cloned into lentiCRISPRv2 (Addgene, 52961). sgRNAs targeting ACO2 were also cloned into lentiCRISPRv2-neo (Addgene, 98292). sgRNA sequences targeting CS, ACO1, or a safe harbor locus (AAVS1) were cloned into the pUSEPB plasmid (gift from S. Lowe).⁸⁸ For dual sgRNA expression, sgRNA targeting SLC25A1 was first cloned into pmU6-gRNA (Addgene, 53187), and then the U6 promoter together with sgSLC25A1 were inserted into the lentiCRISPRv2 backbone carrying sgRNA targeting ACO2. The following sgRNA sequences were used:

Chr8: 5’ GACATTTCTTTCCCCACTGG 3’

AAVS1: 5’ GGGGCCACTAGGGACAGGAT 3’

ACO2 (homo sapiens) guide 1: 5’ AGCGAGGCAAGTCGTACCTG 3’

ACO2 (homo sapiens) guide 2: 5’ CCAGCCAGGAAATTGAGCG 3’

PDK1 (homo sapiens) guide 1: 5’ GAACTGCTTCATGGAGAGCG 3’

PDK1 (homo sapiens) guide 2: 5’ TTGCCGCAGAAACATAAATG 3’

CS (homo sapiens) guide 1: 5’ CAACATGGCAAGACGGTGGT 3’

CS (homo sapiens) guide 2: 5’ AACTGGACATATCCCAACAG 3’

ACL (homo sapiens) guide 1: 5’ ACCAGCTGATCAAACGTCG 3’

ACL (homo sapiens) guide 2: 5’ AGAATCGGTTCAAGTATGCT 3’

ACO1 (homo sapiens) guide 1: 5’ CCATTGGATCCTGTACAACC 3’

ACO1 (homo sapiens) guide 2: 5’ TCCTGCAGGATGACACGAGC 3’

SLC25A1 (homo sapiens): 5’ CGTCTTCACGTACTCGGTG 3’

ACO2 (mus musculus) guide 1: 5’ GCCAACCAGGAGATCGAGCG 3’

ACO2 (mus musculus) guide 2: 5’ ACTGATTCGCACACCCCCAA 3’

To stably introduce expression of mitochondrially-targeted Lactobacillus brevis NADH oxidase,²⁰ mitoLbNOX cDNA (Addgene, 74448) was cloned into pCDH-CMV-MCS-EF1α-puro (System Biosciences, CD510B-1). Human wild-type SLC25A1 cDNA (Horizon Discovery, MHS6278–202826294) was cloned into the N174-MCS backbone (Addgene, 81061) or the piggyBac backbone (pCAGGS-IRES-Neo, a gift from H. Niwa). Human wild-type SLC13A2 cDNA (Horizon Discovery, MHS6278–211689542), human ACO1 cDNA (Horizon Discovery, MHS6278–202757152) or EGFP (insert from LT3GEPIR; Addgene, 111177) were cloned into the pCDH-CMV-MCS-EF1α-neo backbone (System Biosciences, CD514B-1). Lentivirus was generated by co-transfection of viral vectors expressing the sgRNA or cDNA of interest with packaging plasmids psPAX2 (Addgene, 12260) and pMD2.G (Addgene, 12259) into 293T cells using calcium phosphate transfection. Viral-containing supernatants were cleared of cellular debris by 0.45 μm filtration and mixed with 6 μg/mL polybrene (Sigma). Target cells were exposed to viral supernatant for two 24 h periods. Following a 12 h incubation in fresh medium, cells were subjected to antibiotic selection using 1 μg/mL puromycin or 1 mg/mL neomycin until untransduced cells were eliminated. Mouse wild-type Aco2 cDNA (Horizon Discovery, MMM1013–202765893) was cloned into piggyBac (pCAGGS-IRES-Neo, a gift from H. Niwa). Cells were transfected with control or Aco2-containing piggyBac plasmid and piggyBac transposase at a 3:1 ratio using Fugene HD (Promega). 48 h later, cells were selected with 1 mg/mL neomycin until untransfected cells were eliminated.

Western blotting

Protein lysates were extracted in 1X RIPA buffer (Cell Signaling Technology) and protein concentration was determined by BCA assay (Thermo Fisher Scientific). Mouse kidneys were freeze clamped on dry ice. Tissues were ground on dry ice to fine powder and proteins extracted overnight at −80 °C in 1X RIPA buffer. Lysates were mixed with 4X LDS sample buffer (Thermo Fisher NP0007) and 10X sample reducing agent (Thermo Fisher NP0009) and boiled for 5 min. 20–30 μg of protein were separated by SDS-polyacrylamide gel electrophoresis and transferred to nitrocellulose membranes (Bio-Rad). Membranes were blocked in 3% milk in Tris-buffered saline with 0.1% Tween 20 (TBST) and incubated at 4 °C with primary antibodies overnight. Membranes were then washed in TBST and incubated in horseradish peroxidase (HRP)-conjugated secondary antibodies (mouse, NA931; rabbit, NA934; Cytiva) for at least 1 h. Proteins were detected using Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific, 32106) and imaged using HyBlot CL Autoradiography Film (Denville Scientific, E3018) and SRX-101A X-ray Film Processor (Konica Minolta). The antibodies used (at 1:1000 dilution unless otherwise indicated) were: anti-PDK1 (32702; Invitrogen), anti-PDH (3205, Cell Signaling Technology), anti-PDH-E1α (pSer293; AP1062; Sigma), anti-PDH-E1α (pSer232; AP1063; Sigma), anti-PDH-E1α (pSer300; AP1064; Sigma), anti-vinculin (1:10,000; V9131; Sigma), anti-ACO2 (6571S, Cell Signaling Technology for human lysates or ma1-029, Invitrogen for mouse lysates), anti-CS (16131–1-AP; ProteinTech), anti-ACL (15421–1-ap, ProteinTech), anti-ACO1 (12406–1-AP, ProteinTech), anti-ATF4 (11815s, Cell Signaling Technology), anti-phospho-eIF2α (Ser51, 3597, Cell Signaling Technology), anti-eIF2α (9721S, Cell Signaling Technology), anti-SLC13A2 (21722–1-AP, ProteinTech) and anti-SLC25A1 (15235–1-AP; ProteinTech).

Viability assay

Cells were seeded in triplicate in a 12-well plate. Two or three days prior to harvest, cells were changed to DMEM containing vehicle, pyruvate, or citrate. At harvest, cells, media, and all washes were collected, and cells were pelleted and resuspended in flow buffer (PBS with 2% FBS, 0.5 mM EDTA and 0.05% sodium azide) containing DAPI (1 μg/mL, Thermo Fisher, D1306) or propidium iodide (1 μg/mL, BD Biosciences 556463). Viability was measured on an LSRFortessa flow cytometer using FACSDiva software v.8.0 (BD Biosciences). Analysis was performed using FlowJo 10 (BD). The gating strategy applied to these data can be found in Document S1.

MitoTracker and TMRE assays

Cells were seeded in triplicate in a 12-well plate. The next day, cells were changed to DMEM containing vehicle or 5 mM pyruvate. 24 h later, a control well was treated with 10 μM FCCP for 30 min, then all cells were stained with MitoTracker Deep Red (100 nM, Thermo Fisher M22426) or TMRE (100 nM, Thermo Fisher T669) for 30 min. Cells were washed twice with PBS, collected, pelleted, and resuspended in flow buffer (PBS with 2% FBS, 0.5 mM EDTA and 0.05% sodium azide) containing DAPI (1 μg/mL, Thermo Fisher, D1306). Viability was measured on an LSRFortessa flow cytometer using FACSDiva software v.8.0 (BD Biosciences). Analysis was performed using FlowJo 10 (BD). The gating strategy applied to these data can be found in Document S1.

L-HPG nascent protein synthesis assay

Cells were seeded in triplicate in a 12-well and changed to media supplemented with vehicle or 5 mM pyruvate at indicated time points. 90 min before harvest, cells were washed with PBS and incubated in methionine-deficient medium with or without pyruvate for 30 min. Methionine-deficient media containing 50 μM L-homopropargylglycine (HPG, Invitrogen, C10186) with or without pyruvate was then added and cells were incubated for 1 h. One control well received cycloheximide (Sigma 01810, 10 μg ml− 1) during this final hour. Adherent cells were collected and stained with Zombie NIR^™ Fixable Viability Kit (Biolegend, 423105), followed by fixation with 4% PFA in PBS and permeabilization with 0.25% Triton-X-100. Fixed and permeabilized cells were stained using the Click-iT Cell Reaction Buffer Kit (Thermo Fisher Scientific, C10269) and AZDye 405 Picolyl Azide (Vector Labs, CCT-1308) or AZDye 647 Picolyl Azide (Vector Labs, CCT-1300) according to the manufacturer’s instructions and analyzed on the LSRFortessa flow cytometer using FACSDiva v.8.0 (BD Biosciences). Analysis of L-HPG incorporation was performed with FlowJo 10 (BD).

OP-puro assay

Cells transduced with sgRNA expressed from the neomycin-resistant lentiCRISPRv2-neo backbone were seeded 3 days prior to harvest and switched to medium containing vehicle or pyruvate 48 h prior to harvest. Cells were changed to media containing 20 μM O-propargyl-puromycin (OP-puro, HY-15680; MedChemExpress) with vehicle or with pyruvate and incubated for 60 min, 30 min, or 15 min. Cycloheximide (Sigma 01810, 10 μg ml− 1) treatment was added to a control well 30 min prior to the start of OPP. Cells were collected and stained with Zombie NIR^™ Fixable Viability Kit (BioLegend, 423105), followed by fixation with 4% PFA in PBS and permeabilization with 0.25% Triton-X-100. Fixed and permeabilized cells were stained using the Click-iT Cell Reaction buffer Kit (Thermo Fisher Scientific, C10269) and AZDye 405 Picolyl Azide (Vector Labs, CCT-1308) according to the manufacturer’s instructions and analyzed on the LSRFortessa flow cytometer using FACSDiva v.8.0 (BD Biosciences). Analysis of OP-puro incorporation was performed with FlowJo 10 (BD).

EdU incorporation flow cytometry assay

Cells were seeded and 24 h later changed to treatment media (vehicle or 5 mM pyruvate). After 48 h, cells were washed with PBS and changed into treatment media containing 10 μM EdU (Vector Labs, CCT-1149) for 30 min. Cells were collected and stained with Zombie NIR^™ Fixable Viability Kit (Biolegend, 423105) or Zombie Green^™ Fixable Viability Kit (Biolegend, 423111), followed by fixation with 4% PFA in PBS and permeabilization with saponin-based permeabilization and wash reagent (Thermo Fisher Scientific, C10419). Fixed and permeabilized cells were stained using the Click-iT Cell Reaction Buffer Kit (Thermo Fisher Scientific, C10269) and AZDye 405 Picolyl Azide (Vector Labs, CCT-1308) according to the manufacturer’s instructions. Samples were resuspended in FxCyclePI/RNase Staining Solution (Thermo Fisher Scientific, F10797) and analyzed on an LSRFortessa flow cytometer using FACSDiva v8.0 (BD Biosciences). Analysis of EdU and PI incorporation was performed with FlowJo 10 (BD).

Flow-based competition assay

BFP⁺ A549 cell lines, expressing either a control sgRNA (AAVS1) or an sgRNA targeting ACO2 and additionally expressing either sgAAVS1 or sgCS, were generated as described above. Separately, mCherry⁺ A549 cells were generated expressing sgAAVS1 using LentiCRISPRv2-mCherry (Addgene, 99154). On Day 0 of the experiment, mCherry-expressing cells were mixed at a ratio of 1:1 with BFP-positive lines (harboring sgACO2, sgCS, or both) and seeded in a 6-well plate, with a portion of the cells analyzed by flow cytometry as a day 0 timepoint. The next day, cells were provided medium with vehicle or pyruvate at the indicated concentrations. Every three days, cells were counted and 100,000 cells per condition were reseeded into a 6-well with or without pyruvate for continued culture. Every six days, a portion of cells were analyzed by flow cytometry. Zombie Green fixable viability dye (BioLegend, 423111) was used to exclude dead cells. To assess competition in cells that take up citrate, A549 cells expressing an sgRNA targeting ACO2 and additionally expressing GFP were mixed at a ratio of 1:1 with cells expressing SLC13A2 cDNA and either a control sgRNA (sgAAVS1) or sgACO2 on day 0. Cells were then provided medium with vehicle or citrate at the indicated concentrations, and the assay was conducted as described above, with DAPI (Thermo Fisher, D1306) used to exclude dead cells.

The gating strategy applied to these data can be found in Document S1.

Proliferation analyses

Cells were seeded in a 12-well plate at the following densities per well: A549, 25,000 cells/well, Calu1 and HepG2, 30,000 cells/well, C3H/10T1/2, 13,000 cells/well, and ES cells at 40,000 cells/well. The next day, three wells of each line were counted to determine the starting cell number. The remaining cells were changed to medium containing 5 mM pyruvate, 1 or 5 mM citrate, or vehicle control. Cells were counted 72 h later using a Beckman Coulter Multisizer 4e with a cell volume gate of 400–10,000 fL (A549, Calu1, 10T1/2, and ESCs) or 400–20,000 fL (HepG2). The number of population doublings per day was calculated using the following equation:

$$\text{Doublings}\text{per}\text{day}=\left[\text{log}2\left(\text{final}\text{cell}\text{count}/\text{initial}\text{cell}\text{count}\right)\right]/\left(\text{number}\text{of}\text{days}\right)$$

For growth exceeding 72 h, cells from each well were counted and re-seeded in a 12-well plate, and samples were counted again at the indicated timepoints. For relative growth, fold change in cell number relative to starting cell number was calculated, and each condition was normalized to the average fold change in the vehicle condition. All curves were performed at least two independent times.

Oxygen consumption and extracellular acidification rates

Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) were measured using a Seahorse XFe96 Extracellular Flux Analyzer (Agilent Technologies). NSCLC cells were plated on tissue-culture-treated XFe96 96-well plates (Agilent Technologies, 102416) in DMEM at the following densities: 1.25 × 10⁴ cells per well for A549 and H1299; 1.75 × 10⁴ cells per well for Calu1, Calu3, H1993, and H2170; 1.5 × 10⁴ cells per well for H1975; 1.6 × 10⁴ cells per well for H2122 and PC9; 2 × 10⁴ cells per well for H1944 and H1792; and 2.2 × 10⁴ cells per well for H441. Densities were determined by identifying seeding numbers that gave rise to OCR measurements within the linear range. For experiments testing the impact of pyruvate, cells were seeded in the XFe96 plate and 6 h later medium was refreshed with medium containing vehicle or 5 mM pyruvate. The next day, cells were washed twice with Seahorse XF DMEM medium supplemented with 10 mM glucose and 2 mM L-glutamine (with vehicle or 5 mM pyruvate as noted) and incubated for 2 h before performing the Seahorse XF Cell Mito Stress test kit (Agilent, 103015–100) according to the manufacturer’s instructions. During the measurements, oligomycin (1 μM), FCCP (1 μM) and rotenone/antimycin A (500 nM) were injected in sequence. For ECAR measurements, cells were incubated for 2 h in Seahorse XF DMEM medium supplemented with glutamine and with vehicle or 5 mM pyruvate and then ECAR was measured with sequential injections of glucose (10 mM), oligomycin (1 μM), and 2-deoxy-glucose (50 mM; Sigma D3179). Raw data were analyzed using Wave Desktop Software (Agilent, version 2.6.1). After the assay, OCR and ECAR measurements were normalized to either the initial seeding number for each cell line (Figure 1D) or the average protein content for each condition (Figures S1I, S2E, S3O, and S5A), determined via BCA assay. Basal mitochondrial OCR was calculated as the difference between the first baseline OCR measurement and the first OCR measurement following rotenone/antimycin A injection. ETC-independent OCR was calculated as the average of the 3 measurements post-rotenone/antimycin A injection. Glucose-dependent ECAR was calculated as difference between the first baseline measurement and the first measurement following glucose injection.

Isotope tracing and steady-state metabolite measurement

For isotope tracing experiments, NSCLC cells were seeded in 6-well plates and medium was refreshed after 24–48 h. The next day, cells were washed with PBS and changed into glucose- and glutamine-free DMEM (Gibco) supplemented with 10% dialyzed FBS and [¹²C]glucose (Corning) and [¹²C]glutamine or the [U-¹³C]-labelled versions of each metabolite (Cambridge Isotope Labs) to a final concentration of 20 mM (glucose) and 4 mM (glutamine) for 4 h before collection. Cell lines treated with inhibitors were processed as described above but medium was supplemented with vehicle (DMSO), doses ranging from 0.625 μM - 10 μM of phenformin, 1 μM FCCP, 10 mM dichloroacetate, or 10 μM UK-5099 for a total of 24 h. For steady-state measurement of intracellular metabolite pools, cells were seeded in 6-well plates, and 24 h prior to collection, cells were changed to fresh medium or medium containing vehicle or 5 mM pyruvate. At collection, media was completely removed, and metabolites were extracted with 1 mL ice-cold 80% methanol containing 2 μM deuterated 2-hydroxyglutarate (d-2-hydroxyglutaric-2,3,3,4,4-d5 acid (d5-2HG)). After overnight incubation at − 80 °C, lysates were collected and centrifuged at 21,000g for 20 min to remove protein.

For mouse tissues, lung, brain, heart, liver, kidney and skeletal muscle (gastrocnemius) were isolated and freeze clamped on dry ice. Tissues were ground on dry ice to fine powder and weighed; each sample contained roughly 10 mg tissue. For serum, blood was collected in BD Microtainer^™ Capillary Blood Collector tubes (BD, 365967) and spun down at 10,000g for 10 min and serum collected. To tissue powder or 10 μL serum, 1 mL ice-cold 80% methanol containing 2 μM d5-2HG was added, samples vortexed and stored at −80 °C. The next day, lysates were collected and centrifuged at 21,000g for 20 min to remove protein. All extracts were further processed using gas chromatography coupled with MS (GC-MS) as described below. Absolute concentrations of metabolites were determined by including a citrate standard curve of known concentration prepared in H₂O.

Citrate uptake assays

For experiments assessing cellular uptake of metabolites from media, cells were seeded in 6-well plates. 24 h prior to harvest, cells were washed with PBS and changed to 1 mL of DMEM containing 10% dialyzed FBS and defined concentrations of citrate (50 μM or 100 μM). Parallel plates without cells were used as media-only controls for normalization. 24 h after the media change, media samples were collected, centrifuged at 1,700g for 5 min, and 10 μL of supernatant per sample was mixed with 1 mL ice-cold 80% methanol containing 2 μM d5-2HG. After overnight incubation at − 80 °C, extracted media samples were cleared at 21,000g for 20 min and were further processed for GC-MS analysis as described below.

GC-MS analysis

Metabolite extracts were dried in an evaporator (Genevac EZ-2 Elite) and resuspended by incubating with shaking at 30 °C for 2 h in 50 μL of 40 mg/mL methoxyamine hydrochloride in pyridine. Metabolites were further derivatized by adding 80 μL of N-methyl-N-(trimethylsilyl) trifluoroacetamide (Thermo Fisher Scientific) and 70 μL ethyl acetate (Sigma-Aldrich) and then incubated at 37 °C for 30 min. Samples were analyzed using the Agilent 7890A gas chromatograph coupled to an Agilent 5975C mass selective detector. The gas chromatograph was operated in splitless injection mode with constant helium gas flow at 1 mL/min; 1 μL of derivatized metabolites was injected onto an HP-5ms column and the gas chromatograph oven temperature increased from 60 °C to 290 °C over 25 min. Peaks representing compounds of interest were extracted and integrated using MassHunter v.B.08 (Agilent Technologies) and then normalized to both the internal standard (d5-2HG) peak area and the protein content of triplicate samples as determined using the BCA assay (Thermo Fisher Scientific). Steady-state metabolite pool levels were derived by quantifying the following ions: d5-2HG, 354 m/z; aspartate, 334 m/z; αKG, 288 or 304 m/z; citrate, 465 m/z; fumarate, 245 m/z; malate, 335 m/z; and succinate, 247 m/z. All peaks were manually inspected and verified relative to known spectra for each metabolite. Enrichment of [¹³C] was assessed by quantifying the abundance of the following ions: aspartate, 334–346 m/z; citrate, 465–482 m/z; fumarate, 245–254 m/z; and malate, 335–347 m/z. Correction for natural isotope abundance was performed using IsoCor (v.2.0).⁹⁶ For [U-¹³C]glucose-labelled samples in which overall labeling in citrate was less than the parallel unlabeled samples, the mal+2/cit+2 ratios are considered below the limit of quantification and are indicated on graphs by “b.l.q.” For total isotopologue pool calculations, the sum of all isotopologues of either citrate or malate was calculated, normalized to the internal standard and protein concentration, and multiplied by fractional enrichment for each isotopologue. For GC-MS experiments using mouse tissues, peaks were normalized to internal standard d5-2HG and to the exact tissue weight collected for each sample. For quantification of amount of citrate uptake relative to media or citrate concentrations in tissues, sodium citrate standard solutions were made in water and extracted in 80% methanol containing d5-2HG, dried, and derivatized in parallel with the relevant experiment.

Pathology

Mice submitted for tissue postmortem examination 20–25 days following tamoxifen administration were euthanized by CO₂ asphyxiation and cardiac exsanguination by the Laboratory of Comparative Pathology (MSKCC, the Rockefeller University, and Weill Cornell Medicine). Representative segments from kidneys, hearts, lungs, liver, skeletal muscle and brain were fixed in 10% neutral-buffered formalin, processed in alcohol and xylene, embedded in paraffin, sectioned (5 μm thick) and stained with hematoxylin and eosin. The skull was decalcified in a formic acid and formaldehyde solution (Surgipath Decalcifier I, Leica Biosystems) before processing. Kidney pathology (observed at 10 and 20X magnification) was blindly graded as follows: 0 = Normal, no tubular degeneration (vacuolization and cytoplasmic swelling); 1 = Minimal, multifocal rare individual tubular epithelial degeneration in renal cortex and/or corticomedullary junction; 2 = Mild, multifocal occasional individual or small clusters of tubular degeneration in renal cortex and/or corticomedullary junction; 3 = Moderate, multifocal frequent small clusters of tubular degeneration in renal cortex and/or renal corticomedullary junction; 4 = Severe, diffuse (70% or above) exhibit of tubular degeneration in renal cortex and renal corticomedullary junction.

Immunohistochemistry

Immunohistochemistry (IHC) staining for the neutrophil gelatinase-associated lipocalin (NGAL) was performed on representative kidney sections by the Laboratory of Comparative Pathology. Following deparaffinization and heat-induced epitope retrieval in a citrate buffer at pH 6.0, the primary antibody against NGAL (Invitrogen, PAS-79590) was applied at a dilution of 1:1000. A goat anti-rabbit secondary antibody (Vector Labs, BA-1000) and a polymer detection system (Novocastra Bond Polymer Refine Detection, Leica Biosystems, DS9800) was then applied to the kidneys. The chromogen was 3,3’-diaminobenzidine tetrachloride (DAB), and the sections were counterstained with hematoxylin and examined by light microscopy. Kidneys from naïve mice or mice with chronic nephropathy were used as a negative and positive control, respectively. NGAL positivity was blindly graded as follows: 1 = Multifocal immunolabeling of renal tubules, weak granular immunostaining of the apical surface of tubular epithelium; 2 = Multifocal immunolabeling of renal tubules, intermediate and strong granular immunostaining of the apical surface of tubular epithelium; 3 = Multifocal immunolabeling of renal tubules, strong granular immunostaining of the apical surface and cytoplasm (or lumen) of tubular epithelium.

RNA isolation and qPCR

Cells were seeded in 6-well plates. Two days later, regular medium was refreshed or cells were changed to medium supplemented with vehicle or 5 mM pyruvate. 24 h after media change, cells were washed with PBS, and 1 mL of TRIzol (Thermo Fisher Scientific, 15596018) was added to each well. For mouse organs, dissected tissues were freeze clamped on dry ice. Tissues were ground on dry ice to fine powder and weighed into roughly 10 mg samples to which 1 mL TRIzol reagent was added. RNA was extracted according to the manufacturer’s instructions and was used for RNA-sequencing or for qPCR assays. For qPCR, 200 ng RNA was reverse transcribed using the iScript cDNA Synthesis Kit (Bio-Rad). Quantitative PCR with reverse transcription (RT–qPCR) analysis was performed using the Power SYBR Green Master Mix (Thermo Fisher Scientific) in technical triplicate using QuantStudio 5 or 6 Flex (Applied Biosystems). Data points shown represent cDNA generated from three independent wells for each condition or individual animals. Eukaryotic translation elongation factor 1 alpha 1 (Eef1a1) and 36B4 (RPLP0) were used as endogenous controls for mouse and human RNA, respectively, for all experiments. A detailed list of the RT–qPCR primer sequences is provided below:

Primer ID	Sequences 5’ → 3’
Eef1a1 Forward Primer	GCAAAAACGACCCACCAATG
Eef1a1 Reverse Primer	GGCCTGGATGGTTCAGGATA
Aco2 Forward Primer	GCGCAGGGCCAAGGACATAAA
Aco2 Reverse Primer	CCCACACCATACTTGGCACC
36B4 Forward Primer	GCTCCAAGCAGATGCAGCA
36B4 Reverse Primer	CCGGATGTGAGGCAGCAG
mitoLbNOX Forward Primer	AGGCTCTGATTGAGGACGCC
mitoLbNOX Reverse Primer	CTTGGGCATGACCCTAGCCA
Trib3 Forward Primer	GGCACTAGCGTGCAGGAGAC
Trib3 Reverse Primer	GGTGTAGCTCGCATCTTGTCCT
Chac1 Forward Primer	TTCGGGTACGGCTCCCTAGT
Chac1 Reverse Primer	CACTCGGCCAGGCATCTTGT
Asns Forward Primer	CCAAGTTCAGTATCCTCTCCAG
Asns Reverse Primer	CTTCATGATGCTCGCTTCCA
Phgdh Forward Primer	CTTACCAGTGCCTTCTCTCCAC
Phgdh Reverse Primer	GCTTAGGCAGTTCCCAGCATTC
Lcn2 Forward Primer	TCTGTCCCCACCGACCAATG
Lcn2 Reverse Primer	TCTCTGCGCATCCCAGTCAG
Atf3 Forward Primer	GGTCGCACTGACTTCTGAGG
Atf3 Reverse Primer	CTCTGGCCGTTCTCTGGA
Ppp1r15a Forward Primer	TCCTCTAAAAGCTCGGAAGGTACAC
Ppp1r15a Reverse Primer	CGGCTTCGATCTCGTGCAAA
Ddit4 Forward Primer	TTCGGGCCGGAGGAAGACTC
Ddit4 Reverse Primer	CAGCAGCTGCATCAGGTTGG
Aqp1 Forward Primer	ACACCTGCTGGCGATTGACT
Aqp1 Reverse Primer	TGAGAAGTTGCGGGTGAGCA
Cdh6 Forward Primer	AACCTTCCCATCCCGAGAGC
Cdh6 Reverse Primer	GCTTTCAGAGGGTACCTCGGTT
Cubn Forward Primer	GTCCACAGCAGTGCCAACTC
Cubn Reverse Primer	ATGCCGCCACACTCTGGTAG
Lrp2 Forward Primer	CCAGTCTTCCCTGACGCCTT
Lrp2 Reverse Primer	CACGGATTGGTGGCATTGGG

RNA-sequencing and analysis

After RiboGreen quantification and quality control by Agilent BioAnalyzer, 500 ng of total RNA with RIN values of 7.3–10 underwent polyA selection and library preparation using the TruSeq Stranded mRNA LT Kit (Illumina, RS-122–2102) with 8 cycles of PCR according to instructions provided by the manufacturer. Samples were barcoded and run on a NovaSeq 6000 in a PE100 run, using the NovaSeq 6000 S2 or S4 Reagent Kit (200 Cycles) (Illumina). An average of 15 million paired reads was generated per sample. Ribosomal reads represented 0.4–2.5% of the total reads generated and the percent of mRNA bases averaged 92%. Raw FASTQ files were aligned to hg38 v2 (Human) or GRCm39 (Mouse) using Dragen v3.10 (Illumina). A matrix of raw counts was generated using featureCounts/subread (version 2.16.1).⁹⁹ Principal component analysis was performed on vst-transformed counts using the plotPCA function in R (version 4.4.3), factoring in the top 500 genes with highest row variance. Differentially expressed genes were determined using DESeq2 (version 1.46.0).¹⁰⁰ Genes with raw count row sums <10 for all 12 A549 samples were excluded; or genes with raw count row sums <25 across all 25 mouse kidney samples were excluded. Z scores were calculated using the scale function on vst-transformed counts in R. Z scores of genes in an ISR-related gene set⁴⁶ were graphed and human orthologs of mouse genes were identified using the MGI list (http://www.informatics.jax.org/downloads/reports/HOM_MouseHumanSequence.rpt). Genes were ranked by log₂ fold change of the comparisons indicated in each figure legend and gene set enrichment analysis (GSEA) was performed using fgsea (version 1.32.4) in R with Reactome gene sets using the msigdbr function or with custom kidney injury-related gene sets. GSEA dot plots were graphed using ggplot2 (version 3.5.2) in R. Differentially upregulated and downregulated genes in mouse kidneys after drug-induced kidney injury (Chen et al.⁶⁶), in human proximal tubule cell populations after severe injury from critical illness, severe respiratory infections, or systemic inflammation (Hinze et al.⁶⁷), and in mouse kidneys after ischemia-reperfusion injury (Kirita et al.⁵⁹) were used for custom gene set enrichment analysis.

Single cell transcriptomics (Human Protein Atlas)

Normalized expression (nTPM) data across 81 cell types was downloaded from Human Protein Atlas version 23.0 (proteinatlas.org; section titled “RNA single cell type data”). A complete list of primary references for single cell RNA sequencing datasets can be found at https://www.proteinatlas.org/about/assays+annotation#singlecell_rna.

Statistics and reproducibility

All statistical analysis was performed using Prism 10 (GraphPad), with the exception of DepMap gene expression and RNA sequencing GSEA analysis, which was performed in R (version 4.3.2) as described above. Error bars, P values and statistical tests are reported in the figures and figure legends. No statistical methods were used to predetermine sample sizes. Experiments were performed in biological triplicate or as indicated in figure legends and were performed independently at least two times. Animal experiments were pooled from multiple independent cohorts to ensure reproducibility. Gating strategies used are reported in Document S1.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.cell.2026.01.028.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse IgG HRP Linked Whole Ab	Sigma-Aldrich/Cytiva	Cat#GENA931-1ML
Rabbit IgG HRP Linked Whole Ab	Sigma-Aldrich/Cytiva	Cat#GENA934-1ML; RRID: AB_2722659
PDK1 Recombinant Rabbit Monoclonal Antibody (JA67-30)	Invitrogen	Cat#MA5-32702; RRID: AB_2809979
Pyruvate Dehydrogenase (C54G1) Rabbit Monoclonal Antibody	Cell Signaling Technology	Cat#3205; RRID: AB_2162926
PhosphoDetect Anti-PDH-E1α(pSer293) Rabbit pAb	Sigma-Aldrich	Cat#AP1062
PhosphoDetect Anti-PDH-E1α(pSer232) Rabbit pAb	Sigma-Aldrich	Cat#AP1063; RRID: AB_10616070
PhosphoDetect Anti-PDH-E1α(pSer300) Rabbit pAb	Sigma-Aldrich	Cat#AP1064
Anti-Vinculin Antibody	Sigma-Aldrich	Cat#V9131; RRID: AB_477629
ACO2 (D6D9) Rabbit Monoclonal Antibody	Cell Signaling Technology	Cat#6571; RRID: AB_2797630
Aconitase 2 Monoclonal Antibody (7G4)	Invitrogen	Cat#MA1-029; RRID: AB_11157026
Citrate synthase Polyclonal antibody	ProteinTech	Cat#16131-1-ap; RRID: AB_1640013
ACLY Polyclonal antibody	ProteinTech	Cat#15421-1-AP; RRID: AB_2223741
Aconitase 1 Polyclonal antibody	ProteinTech	Cat#12406-1-AP; RRID: AB_10642942
ATF-4 (D4B8) Rabbit Monoclonal Antibody	Cell Signaling Technology	Cat#11815; RRID: AB_2616025
Phospho-eIF2 alpha (Ser51) (119A11) Rabbit Monoclonal Antibody	Cell Signaling Technology	Cat#3597; RRID: AB_390740
Phospho-eIF2 alpha (Ser51) Antibody	Cell Signaling Technology	Cat#9721; RRID: AB_330951
SLC13A2 Polyclonal antibody	ProteinTech	Cat#21722-1-AP; RRID: AB_2878912
SLC25A1 Polyclonal antibody	ProteinTech	Cat#15235-1-AP; RRID: AB_2254794
NGAL Polyclonal Antibody	Invitrogen	Cat#PA5-79590; RRID: AB_2746705
Goat Anti-Rabbit IgG Antibody (H+L), Biotinylated	Vector Labs	Cat#BA-1000
Chemicals, peptides, and recombinant proteins
DMSO	Sigma-Aldrich	Cat#D2650-100ML
Puromycin Dihydrochloride	ThermoFisher Scientific/Gibco	Cat#A1113803
G418 Sulfate Solution	GeminiBio	Cat#400-113
Doxycycline hyclate	Sigma-Aldrich	Cat#D9891
Hexadimethrine bromide (polybrene)	Sigma-Aldrich	Cat#107689
RPMI	MSKCC media core	N/A
DMEM, high glucose	MSKCC media core	N/A
DMEM, high glucose, minus methionine	MSKCC media core	N/A
L-Glutamine	MSKCC media core	N/A
BenchMark™ Fetal Bovine Serum	Gemini Bio	Cat#100-106
Dialyzed Fetal Bovine Serum	Gemini Bio	Cat#100-108-500
Delipitated Fetal Bovine Serum	Gemini Bio	Cat#900-123
DMEM, no glucose, no glutamine, no phenol red	Gibco	Cat#A1443001
D-(+)-Glucose solution 45%	Corning	Cat#45001-116
RIPA Buffer (10x)	Cell Signaling Technologies	Cat#9806
NuPAGE™ LDS Sample Buffer (4X)	Invitrogen	Cat#NP0007
NuPAGE™ Sample Reducing Agent (10X)	Invitrogen	Cat#NP0009
Pyridine	Thermo Fisher Scientific	Cat#TS-27530
Methoxyamine hydrochloride	Sigma-Aldrich	Cat#226904
Thermo Scientific™ MSTFA	Thermo Fisher Scientific	Cat#PI48910
Methanol, Optima™ for HPLC, Fisher Chemical™	Fisher Scientific	Cat#A454SK-4
Ethyl acetate, ACS, 99.5+%	Fisher Scientific	Cat#AA31344M1
Carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone (FCCP)	Sigma-Aldrich	Cat#C2920
Sodium pyruvate	Sigma-Aldrich	Cat#P2256
L-Alanine	Sigma-Aldrich	Cat#A7469
Phenformin hydrochloride	Sigma, Supelco	Cat#P7045
UK-5099	Tocris	Cat#4186
Sodium dichloroacetate	Sigma-Aldrich	Cat#347795
TRIzol™ Reagent	Invitrogen	Cat#15596018
Power SYBR™ Green PCR Master Mix	Applied Biosystems	Cat#4367659
D-Glucose (U-13C6, 99%)	Cambridge Isotopes Laboratories, Inc.	Cat#CLM-1396-10
L-Glutamine (13C5, 99%)	Cambridge Isotopes Laboratories, Inc.	Cat#CLM-1822-H-1
Pierce™ ECL Western Blotting Substrate	Thermo Fisher Scientific	Cat#32106
Tamoxifen	Sigma-Aldrich	Cat#T5648
Sunflower seed oil	Sigma-Aldrich	Cat#S5007
Sodium citrate dihydrate	Sigma-Aldrich	Cat#W302600
Sodium chloride	Sigma-Aldrich	Cat#S9888
FuGENE® HD Transfection Reagent	Promega	Cat#E2311
DAPI and Hoechst Nucleic Acid Stains	Invitrogen	Cat#D1306
BD Pharmingen™ Propidium Iodide Staining Solution	BD Biosciences	Cat#556463
MitoTracker™ Dyes for Mitochondria Labeling	Invitrogen	Cat#M22426
Tetramethylrhodamine, Ethyl Ester, Perchlorate (TMRE)	Invitrogen	Cat#T669
Click-IT™ L-Homopropargylglycine (HPG)	Invitrogen	Cat#C10186
Cycloheximide	Sigma-Aldrich	Cat#01810
AZDye 405 Picolyl Azide	Vector Labs	Cat#CCT-1308
AZDye 647 Picolyl Azide	Vector Labs	Cat#CCT-1300
O-Propargyl-Puromycin	MedChemExpress	Cat#HY-15680
EdU (5-Ethynyl-2′-deoxyuridine)	Vector Labs	Cat#CCT-1149
2-Deoxy-D-glucose	Sigma-Aldrich	Cat#D3179
Critical commercial assays
Pierce™ BCA Protein Assay Kits	Thermo Fisher Scientific	Cat#23225
Seahorse XF Cell Mito Stress Test Kit	Agilent	Cat#103015-100
iScript™ cDNA Synthesis Kit	Biorad	Cat#1708891
QIAGEN Plasmid Maxi Kit	Qiagen	Cat#12163
QIAGEN Plasmid Mini Kit	Qiagen	Cat#12125
QIAquick PCR & Gel Cleanup Kit	Qiagen	Cat#28506
QIAEX II Gel Extraction Kit	Qiagen	Cat#20051
Zombie NIR™ Fixable Viability Kit	Biolegend	Cat#423105
Zombie Green™ Fixable Viability Kit	Biolegend	Cat#423111
Click-iT™ Cell Reaction Buffer Kit	Invitrogen	Cat#C10269
FxCycle™ PI/RNase Staining Solution	Invitrogen	Cat#F10797
BOND Polymer Refine Detection	Leica Biosystems	Cat#DS9800
TruSeq Stranded mRNA LT	Illumnia	Cat#RS-122–2102
Deposited data
DepMap Portal R23Q4	Broad; Arafeh et al.86	https://depmap.org/portal
RNA sequencing	This paper	GEO: GSE295123, GSE295125
Metabolomics data	Chen et al.13	N/A
Single cell RNA sequencing	Proteinatlas	https://www.proteinatlas.org/about/download
Raw isotopologue values	This paper	Table S1
Experimental models: Cell lines
C3H/10T1/2, Clone 8	ATCC	Cat#CCL-226
C2C12	ATCC	Cat#CRL-1772
293T	ATCC	Cat#CRL-3216
A549	ATCC	Cat#CRM-CCL-185
Calu1	ATCC	Cat#HTB-54
NCI-H1975	ATCC	Cat#CRL-5908
NCI-H2122	ATCC	Cat#CRL-5985
NCI-H2170	ATCC	Cat#CRL-5928
NCI-H1993	ATCC	Cat#CRL-5909
HepG2	ATCC	Cat#HB-8065
Calu3	Gift from Gina DeNicola	N/A
PC9	Gift from Gina DeNicola	N/A
H441	Gift from Gina DeNicola	N/A
H1299	Gift from Gina DeNicola	N/A
H1792	Gift from Gina DeNicola	N/A
H1944	Gift from Gina DeNicola	N/A
H2009	Gift from Gina DeNicola	N/A
H2087	Gift from Gina DeNicola	N/A
H2172	Gift from Gina DeNicola	N/A
H2347	Gift from Gina DeNicola	N/A
Mouse embryonic stem cells	Carey et al.87	N/A
Experimental models: Organisms/strains
Aco2 fl/fl	This paper	N/A
UBC-Cre-ERT2: B6 Cg-Ndor1Tg(UBC-cre/ERT2)1Ejb/1J	Jackson Laboratory	Strain 007001
Oligonucleotides
Guide RNAs (STAR Methods)	This paper; Morris et al.88 and Wang et al.89	N/A
qPCR primers (STAR Methods)	This paper; Lercher et al.90 and Han et al.91	N/A
Recombinant DNA
lentiCRISPR v2	Sanjana et al.92	Addgene Plasmid #52961
lentiCRISPRv2-neo	Stringer et al.93	Addgene Plasmid #98292
LentiCRISPRv2-mCherry	Agata Smogorzewska	Addgene Plasmid #99154
pUC57-mitoLbNOX	Titov et al.20	Addgene Plasmid #74448
N174-MCS	Adam Karpf	Addgene Plasmid #81061
pUSEPB	Morris et al.88	N/A
pCAGGS-IRES-Neo	Gift from H. Niwa	N/A
pmU6-gRNA	Kabadi et al.94	Addgene Plasmid #53187
pCDH-CMV-MCS-EF1α-puro	System Biosciences	Cat#CD510B-1
pCDH-CMV-MCS-EF1α-neo	System Biosciences	Cat#CD514B-1
psPAX2	Didier Trono	Addgene Plasmid #12260
pMD2.G	Didier Trono	Addgene Plasmid #12259
LT3GEPIR	Fellmann et al.95	Addgene Plasmid #111177
SLC25A1 cDNA	Horizon Discovery	Cat#MHS6278-202826294
SLC13A2 cDNA	Horizon Discovery	Cat#MHS6278-211689542
ACO1 cDNA	Horizon Discovery	Cat#MHS6278-202757152
Aco2 cDNA	Horizon Discovery	Cat#MMM1013-202765893
Software and algorithms
R	Cran	https://www.r-project.org
Prism 9–10	Graphpad	https://www.graphpad.com
IsoCor v2.0	Millard et al.96	N/A
Quant Studio 5–6	Thermo Fisher Scientific	https://www.thermofisher.com/us/en/home/life-science/pcr/real-time-pcr/real-time-pcr-instruments/quantstudio-systems/features.html#design-and-analysis-software
MassHunter	Agilent	https://www.agilent.com/en/product/software-informatics/mass-spectrometry-software/data-analysis
Wave Desktop	Agilent	https://www.agilent.com/en/product/cell-analysis/real-time-cell-metabolic-analysis/xf-software/seahorse-wave-desktop-software-740897?srsltid=AfmBOoojwquEuqNM6xcIjeIHptR1Aj2AEeiKGlha-32Itn1iQsuyp1LH
FlowJo 10	BD Biosciences	https://www.bdbiosciences.com/en-us/products/software/flowjo-software?tab=flowJo-v11-software
FACSDiva v.8.0	BD Biosciences	https://www.bdbiosciences.com/en-us/products/software/instrument-software/bd-facsdiva-software

Highlights.

• Nutrient conditions that increase citrate production activate forward TCA cycle flux

• Increasing citrate drives dependence upon the TCA cycle enzyme aconitase 2 to clear citrate

• Citrate accumulation activates the integrated stress response and impairs cell fitness

• Cells and tissues, such as the kidney, that net uptake citrate rely on aconitase 2